Split (3)

train · 4.35k rows

article stringlengths 1.98k 169k	summary stringlengths 1.01k 4.15k	section_headings listlengths 2 38	keywords listlengths 0 12	year stringclasses 11 values	title stringlengths 30 189
The early-life intestinal microbiota plays a key role in shaping host immune system development . We found that a single early-life antibiotic course ( 1PAT ) accelerated type 1 diabetes ( T1D ) development in male NOD mice . The single course had deep and persistent effects on the intestinal microbiome , leading to altered cecal , hepatic , and serum metabolites . The exposure elicited sex-specific effects on chromatin states in the ileum and liver and perturbed ileal gene expression , altering normal maturational patterns . The global signature changes included specific genes controlling both innate and adaptive immunity . Microbiome analysis revealed four taxa each that potentially protect against or accelerate T1D onset , that were linked in a network model to specific differences in ileal gene expression . This simplified animal model reveals multiple potential pathways to understand pathogenesis by which early-life gut microbiome perturbations alter a global suite of intestinal responses , contributing to the accelerated and enhanced T1D development . The human gastrointestinal ( GI ) tract contains a microbiome of enormous cell number ( Sender et al . , 2016 ) and complexity ( Lozupone et al . , 2012 ) , which plays important roles in shaping development of host immunity ( Honda and Littman , 2016; Hooper et al . , 2012; Kinnebrew and Pamer , 2012 ) . Altered composition of the GI microbiota modifies risk of inflammatory conditions—including type 1 diabetes ( T1D ) , asthma , and inflammatory bowel disease—by perturbing immune system development ( Eberl et al . , 2015; Flak et al . , 2013; Fujimura and Lynch , 2015; Gensollen et al . , 2016; Kostic et al . , 2015; Kriegel et al . , 2011; Livanos et al . , 2016; Rooks and Garrett , 2016; Schulfer et al . , 2018 ) . T1D is characterized by T-cell-mediated destruction of pancreatic β-cell-containing islets ( Wilson et al . , 1998 ) , but the triggers and intermediary molecular mechanisms remain unclear . Since numbers of intestinal Treg cells are significantly reduced in T1D , altered gut microbiota might play an initiating role ( Badami et al . , 2011 ) . The worldwide increasing incidence of T1D , with decreasing age of onset ( Patterson et al . , 2012; Paun et al . , 2017; Wändell and Carlsson , 2013 ) , coincides with the widespread use of antibiotics in children ( Hersh et al . , 2011; Lee et al . , 2014 ) . Since antibiotic exposure affects the intestinal microbiota , potentially changing interplay with immune systems , it could contribute to the rise in T1D; recent studies in the NOD ( non-obese diabetic ) mouse model support this hypothesis ( Brown et al . , 2016; Candon et al . , 2015; Hu et al . , 2017; Livanos et al . , 2016 ) . In male NOD mice , three courses of a pulsed ( macrolide ) antibiotic treatment ( 3PAT ) altered the intestinal microbiota and reduced intestinal lamina propria Th17- and Treg-populations , accelerating T1D development ( Livanos et al . , 2016 ) . That isolated cecal contents from the antibiotic-exposed NOD mice transferred to germ-free recipient mice produced parallel immunological effects , further supports a causal role of the antibiotic-perturbed microbiota in T1D pathogenesis ( Livanos et al . , 2016 ) . Enhanced T1D induction depended on the antibiotics used ( Brown et al . , 2016; Candon et al . , 2015; Hu et al . , 2017 ) , suggesting that differences in their activities influenced overall effects . Although the roles of antibiotics perturbing the microbiome and promoting T1D are becoming defined , the underlying molecular mechanisms require better resolution . A single early-life PAT course altered the intestinal microbiota and specific intestinal T-cell populations and effectors in C57BL/6 mice; experiments involving germ-free mice showed that the perturbed microbiota was both necessary and sufficient for the effects ( Ruiz et al . , 2017 ) . Here we asked whether the same single early-life ( pup day of life P5-P10 ) antibiotic pulse was sufficient to enhance T1D in NOD mice . This work now shows that the extensive early-life effects of the brief antibiotic course on the microbiota initiate a global cascade of effects flowing from the gut lumen via metabolites and specific interactions with host cells that change the developmental program of innate and adaptive immunity , leading to accelerated and enhanced T1D . To evaluate the effects of antibiotic exposure on T1D incidence , NOD mice were given a single ( 1PAT ) or three courses ( 3PAT ) of a macrolide antibiotic , or not ( controls ) ( Figure 1A ) . In both control groups , females had higher T1D incidence than males , as expected ( Bao et al . , 2002; Livanos et al . , 2016; Markle et al . , 2013 ) . In males , T1D development in the two control groups was similar , but both antibiotic-exposed groups had significantly accelerated and enhanced T1D rates ( Figure 1B ) ; their similarity indicates the sufficiency of the first exposure for the full effect . In the female controls , the spontaneous T1D rates approached 80% , and neither antibiotic exposure significantly increased the rates . As such , we focused on male mice especially the 1PAT group in whom the exposure ended by P10 . That the median T1D development was at P147 provided a prolonged window to understand the intermediary mechanisms . Examining the pancreatic islets , at P42 , there were no significant differences between the PAT and control groups , but by P70 , significantly more islets showed inflammation in both PAT groups than controls ( Figure 1C ) , confirming that the enhanced pathological process was well-advanced by P70 . 1PAT exposures significantly reduced body weight for both male and female mice continuing to P70 ( Figure 1—figure supplement 1 ) , and similar to 3PAT exposure ( data not shown ) , but not at later time points . There was no relationship between early-life body weight and risk or timing of T1D development . We then examined the antibiotic effects on the intestinal microbiome at points prior to the observed insulitis . On P12 , two days after 1PAT ended , there were significant changes in the cecal and fecal microbiota persisting until at least P49 ( Figure 2 and Figure 2—figure supplement 1 ) . Community structure ( β-diversity ) markedly differed between controls and both PAT groups ( Figure 2A and Figure 2—figure supplement 1B ) ; further studies focused on the simpler 1PAT experiment . In all cecal and fecal samples tested , 1PAT suppressed α-diversity weeks after the exposure ended ( Figure 2B and Figure 2—figure supplement 1A , C ) . Thus , the single early life pulse led to a persistent change in the microbial community . 1PAT also increased the inter-subject microbial heterogeneity . Although the composition of the gut microbiomes of control males and females were nearly identical , there were significant differences in microbiome heterogeneity amongst the 1PAT-exposed mice ( Figure 2—figure supplements 2 and 3 ) . In the ileum , PAT and control differed less from each other than in the cecum ( Figure 2—figure supplement 4 ) , and there were no significant differences between males and females for each treatment ( data not shown ) . The 3PAT exposure accelerated T1D development , which is consistent with our previous observations ( Livanos et al . , 2016 ) . Also consistent with the prior observations , we saw significant changes in fecal microbiota after course 1 ( at P21 ) , course 2 ( at P35 ) , and course 3 ( at P49 ) . β-diversity was markedly different between control and 3PAT , and α-diversity was significantly decreased at all three early timepoints ( Figure 2—figure supplement 5 ) . Whereas the control males and females were nearly identical in inter-subject heterogeneity , there were significant differences in the 3PAT-exposed mice at all three timepoints . There were no significant differences in α-diversity between males and females for either the control or 3PAT mice ( Figure 2—figure supplement 6 ) . Comparing across the 1PAT and 3PAT experiments based on Shannon index analysis , α-diversities were similar in the two control groups , but were significantly lower in the 1PAT than the 3PAT group in both males ( Figure 2—figure supplement 7 ) , and in females ( data not shown ) . With the observed inter-subject heterogeneity ( Figure 2A and Figure 2—figure supplement 5A ) , we next asked which specific taxa were associated with accelerated T1D . In total , across the 518 fecal samples at the three early time points studied , we identified 76 individual taxa . Using the stringent ANCOM algorithm ( Mandal et al . , 2015 ) , findings in males and females were similar , as were comparisons of the 1PAT and 3PAT mice with their controls ( Figure 2C ) ; thus the major effects on taxa were conserved across the treatments and both sexes were affected similarly ( Figure 2—figure supplements 3 and 6 ) . These results pointed to a differential transduction of the effects of the altered microbiome in the male and female hosts ( see below ) . In essentially all of the 12 individual timepoint/group comparisons , the relative abundances of four taxonomic groups ( Enterococcus , Blautia , Enterobacteriaceae , and Akkermansia ) were significantly over-represented in PAT . In contrast , the relative abundances of four taxa ( S24-7 , Clostridiales , Oscillospira , and Ruminococcus ) were significantly under-represented in ≥ 7 time-point/treatment comparisons between PAT and control . No other representational differences were reproducible across treatment and time , nor were significant differences identified between males and females ( Figure 2—figure supplements 3 and 6 ) . Analysis of the mixed effect of time in different groups further confirmed the over-represented taxa ( Enterococcus , Blautia , Enterobacteriaceae , and Akkermansia ) and the under-represented taxa ( S24-7 and Ruminococcus ) in PAT , and revealed another under-represented taxon Anaeroplasma . This analysis revealed representational differences prior to the phenotypic events in a small group of taxa strongly linked with accelerated T1D or with protection in the male mice . To determine whether the PAT-altered microbiota differed in metabolic functions or whether the taxonomic differences merely led to functional substitutions , we examined the fecal metagenome in 24 1PAT and control mice at both P21 and P49 . Based on Bray-Curtis analysis of 300 metabolic pathways identified in the shotgun sequencing results , 1PAT significantly affected metagenomic composition ( Figure 2D ) . Notably , of the 131 pathways differentiating 1PAT and control mice , 97% were overrepresented in the 1PAT samples , significantly deviated from chance at both times in both males and females ( p<0 . 001 for each subanalysis ) ( Figure 2—figure supplements 8 and 9 ) . In unsupervised hierarchical clustering of the pathways with the highest inter-individual variance , we identified a cluster strongly enriched for 1PAT ( 20/25 samples ) , that contains 34 specific pathways ( Figure 2—figure supplement 8 ) , encoding genes involved in long chain carbohydrate degradation , specific amino acid biosynthesis , and bacterial cell structural components ( e . g . peptidoglycan synthesis and maturation ) ( Figure 2—figure supplement 8 ) . In males , of 50 pathways that differed significantly between 1PAT and control , 22 ( 44% ) were significant at both P21 and P49 . For females , of 51 significant pathways , only 9 ( 18% ) were shared ( Figure 2—figure supplement 9 and Supplementary file 1 ) . Of 49 pathways significant at P21 , 10 ( 20% ) were shared between males and females , but at P49 , 19 ( 35% ) of 54 were shared . At P21 , of 39 pathways that were significantly differential in males , 100% were greater in PAT than controls . In females , 20 were differential , but 95% were greater in PAT than controls . Thus , there was a marked asymmetry in the pathways in both males and females , with overrepresentation in PAT; 10 ( 53% ) of the 19 over-represented pathways in females also were over-represented in males . Those 10 pathways were related to bacterial biosynthesis of the amino acids lysine , methionine , and homoserine and related to the bacterial degradation of rhamnose , aspartate , and inositol . Lysine provides one basis for arginine metabolism , which plays important roles in immune regulation ( Peranzoni et al . , 2007 ) ; plasma lysine is substantially produced by intestinal microbes ( Metges et al . , 1999 ) . Homoserine is an essential part of the Gram-negative enteric bacterial quorum-sensing auto-inducer , homoserine lactone , which mediates communication between bacteria and could have immunodulatory roles ( Gaida et al . , 2016 ) . As such , up-regulation of these intestinal bacterial amino acid pathways by 1PAT could affect intestinal and systemic homeostasis in the pups , and may affect subsequent T1D onset . An alternative way to consider these pathways is to assess how closely the representation for one matches the others . Pairwise correlations between 14 fecal metagenomics pathways were significantly different at P21 ( Figure 2—figure supplement 10 ) , identifying a close and significant correlation between superpathways of sulfate assimilation and cysteine biosynthesis , phospholipid biosynthesis , purine nucleotide salvage , oleate biosynthesis , ( saturated ) fatty acid elongation , and hexitol fermentation in all male samples at P21 ( Figure 2—figure supplement 10A ) . The remaining eight metagenomic pathways showed no significant associations . The majority of these relationships remained when considering 1PAT males at P21 independent of controls , but did not reach significance , likely due to the reduced sample size ( Figure 2—figure supplement 10B , C ) . These relationships were not seen in the control-only analysis . Since secondary metabolites are bioactive small molecules affecting microbial community structure and/or host physiology ( Dorrestein et al . , 2014; Sharon et al . , 2014 ) , we then asked whether the metagenomic analysis could also identify biosynthetic gene clusters ( BGCs ) encoding significantly differential secondary metabolites . Using an accelerated optimal gapped alignment algorithm , we mapped the metagenomic reads against a BGC database and identified 228 BGCs with high rates of within-sample metagenomic coverage . The number of BGCs per sample varied widely ( Figure 2—figure supplement 11 ) , and alone did not significantly distinguish 1PAT in male or female mice ( p>0 . 05 , Mann Whitney U test ) . Then asking whether the presence of particular BGCs were distinguishing at the earliest time point , P21 , we found 23 BGCs significantly enriched in 1PAT male mice ( Figure 2E ) . One enriched BGC mapped to a polysaccharide product in the MIBiG database , and is produced by Enterococcus species ( Medema et al . , 2015; Xu et al . , 1998 ) . Clustering the other 22 uncharacterized BGCs by sequence homology ( Rashidi et al . , 2018; Shields-Cutler et al . , 2018b ) , collapsed them to three functional units predicted to produce a triscatecholate siderophore biosynthesis pathway member , a siderophore secondary metabolite , and an arylpolyene , all annotated to the family Enterobacteriaceae . In total , these studies indicate a directional ( not substitutional ) effect of PAT on the metabolite profiles as detected by metagenomic analyses , and are consistent with changes in taxa that were independently identified in the analysis of 16S relative abundances . To determine whether the altered metagenome affects important microbial products , we examined production of seven short chain fatty acids ( SCFA ) in cecal samples at P23 and P42 . At P23 , the 1PAT mice had significantly reduced levels of butyric and propionic acid ( p<0 . 05 for both ) compared to controls ( Figure 3A ) ; none of the other SCFAs were significantly different . By P42 , there no longer were significant differences in any of the tested SCFA ( data not shown ) . Thus , antibiotic exposure , by altering the taxonomic and metagenomic composition , reduced two important host-signaling microbial metabolites in early life . We next asked whether the 1PAT-altered microbiome affected host metabolic phenotypes . Using samples obtained from P15 to P42 , we identified 30 metabolites that were significantly different between 1PAT and control in serum , and 12 in liver ( Figure 3B and Supplementary file 2 ) . These included three metabolites ( uracil , citric acid , and isoleucine ) at significantly higher levels in both serum and liver , and a fourth ( valine ) that was less abundant in hepatic samples from both the P23 and P42 PAT-exposed mice . The altered amino acid levels , consistent with the metagenomic representation of amino acid biosynthetic pathways ( Supplementary file 1 ) , provide direct evidence that the PAT-altered microbiome produces a metabolic signal that is transduced into the host . Given that 1PAT altered microbial populations and taxa , microbial genes , and metabolites , we next asked how 1PAT affects the microbial interaction with host tissues that led to accelerated T1D . We focused on the immunologically active ileum since our prior 3PAT studies showed altered gene expression in the NOD mouse ileum , and that immunological effects could be transferred to recipient mice only using the perturbed microbiome ( Livanos et al . , 2016 ) . Further , experiments in C57BL/6 mice confirmed that a 1PAT-altered microbiome is both necessary and sufficient for such immunological changes ( Ruiz et al . , 2017 ) . We began by characterizing ileal gene expression in P2 pups , asking how males and females differ in earliest life , prior to antibiotic exposure . Bray-Curtis analysis of RNA-Seq data indicated that ileal gene expression profiles are distinct at a global level between male and female P2 mice , and significantly differed from P12 ( Figure 4A and Figure 4—figure supplement 1 ) . At P2 , we identified the ~1500 genes that comprised the significant sex-specific differences ( Figure 4B ) , including KEGG pathways involved in the mTOR , MAPK , B cell receptor , and T cell receptor signaling . These sex-specific differences at post-natal day 2 , preceding any antibiotic exposure , may provide an important opportunity in future studies to better understand the basis of the T1D sex dimorphism in NOD mice before puberty . Next , we examined mice at P12 to assess the maturation of gene expression ( defined as the significant differences compared with P2 ) , and the effects of PAT on that maturation . At a global level , the expression profiles were significantly different between PAT and control in both males and females ( Figure 4A ) ; thus by P12 , the antibiotic effects on the microbiome were already being transduced into the tissues . As expected with normal development , expression of several thousand genes significantly changed in the control mice from P2 to P12 , however , in the mice receiving PAT between P5–10 , ~17% of the differences were lost in males and ~21% in females ( Figure 4—figure supplement 1 ) . KEGG analysis of the male mice highlighted significant changes between PAT and control in cell adhesion molecules , cytokine–cytokine receptor interaction , T cell receptor signaling , B cell receptor signaling , intestinal immune network for IgA production , and leukocyte transendothelial migration . ( Figure 4C ) . By P23 , we found striking differences between males and females; in males , 1511 ileal genes showed significant differential expression between PAT and control vs . only 124 in females , a 12 . 2-fold difference ( p<0 . 001 ) ( Figure 4D , Figure 4—figure supplement 2 and Supplementary file 3 ) . Thus , in a functional sense , the female ileum had greater resilience against the same microbiological perturbation ( Figure 2—figure supplement 9 ) . In the males , the genes that had significantly reduced expression included Nos2 , Saa1 , Runx1 , and Muc4 , all involving host defenses ( see below ) . By P42 , nearly half of the significant differences in males between 1PAT and controls were lost; nevertheless , 32 days after the antibiotic exposure had ended , there remained abnormal expression of hundreds of ileal genes ( Figure 4—figure supplements 2 and 3 ) . Thus , RNA-Seq studies indicated a broad effect of the microbiome changes on the maturation of early-life intestinal gene expression , with asymmetric effects in males and females , consistent with their differential development of accelerated T1D later in life . Since RNA-Seq provided a global assessment of ileal gene expression , we next sought to focus on a specific subset of 547 genes related to inflammation and immunity as captured on the NanoString nCounter mouse immunological assay ( Figure 5 ) . First , Bray-Curtis distance matrix analysis indicated strong and directional age-related effects from P12 to P42 ( Figure 5A ) , independently confirming and extending the RNA-Seq findings indicating maturation of gene expression . Using the time-series of samples , we compared the age-associated normal ( control ) immune/inflammatory gene expression with that after the brief PAT exposure . The period from P12 to P23 was more developmentally dynamic than that from P23 to P42 , consistent with the broader RNA-Seq findings ( Figure 5—figure supplement 1 top panel ) . However , although the number of genes with changed expression status in the 1PAT mice was similar to control ( Figure 5—figure supplement 1 bottom panel ) , many of the maturing genes differed . As such , we defined five classes of ileal immune expression maturation ( Figure 5B ) , including genes that significantly matured in: ( I ) . Controls; ( II ) . after 1PAT; ( III ) . maturation was resistant to 1PAT; ( IV ) . maturation was altered by 1PAT; or ( V ) . represented a new pattern of 1PAT-induced maturation ( dysmaturation ) . Such a classification strategy may have general utility in interpreting data from global genetic approaches in other models involving developmental perturbations . Using these individual gene-level expression differences to understand which KEGG pathways were differentially affected by 1PAT permitted identification of important innate pathways , for example , Toll-Like Receptor ( TLR ) signaling ( Figure 5C ) . Although the changes induced by PAT were widespread , analyses also indicated effects , for example , on pathways involved in NOD-like receptor signaling and Th17 cell differentiation ( Figure 5—figure supplements 2 and 3 ) . Next , to confirm and extend these broad RNA-Seq and NanoString observations , we explored individual genes of particular interest at the host-microbial interface . Using qPCR studies to validate the global findings , we found that in the PAT mice , expression was reduced for genes encoding four of the five transcription factors regulating the inducible nitric oxide synthase ( Nos2 ) , for Nos2 itself , and the related Calm3 ( Figure 6A ) . Runx1 , an early life transcription factor showed reduced expression with PAT , as did two of its downstream genes ( Foxp3 and Cd3g ) that are involved in the development of adaptive immunity ( Figure 6B ) , but effects on the innate Saa genes were bimodal over time ( Figure 6—figure supplement 1A ) . There was marked reduction of expression of the two major genes involved in intestinal mucin synthesis ( Muc2 and Muc4 ) ( Figure 6—figure supplement 1B ) . Ido1 , encoding indoleamine 2 , 3-dioxygenase , the key enzyme catalyzing tryptophan catabolism along the kynurenine pathway and having a major role in immune modulation by mediating T-cell inhibition depending on bone marrow stromal cell activation ( Meisel et al . , 2004; Munn and Mellor , 2013 ) , showed reduced expression with PAT ( Figure 6—figure supplement 1C ) . The RT-qPCR studies confirmed the gradual normalization of most but not all of the candidate genes ( Figure 6A , B and Figure 6—figure supplement 1 ) . In total , these studies provide evidence that the 1PAT-altered microbiota interferes with immune pathways during a critical developmental window . Based on the strong 1PAT-induced differences in early life intestinal gene expression , including transcription factors , we next considered an epigenetic basis for the changes . We examined global histone post-translational modification ( PTM ) states to assess whether differential chromatin regulation contributes to the enhanced T1D phenotypes in the 1PAT-exposed male mice . We focused on ileum and liver , based on the 3PAT-induced sex-specific differences in ileal gene expression ( Livanos et al . , 2016 ) , on the effect of the early-life antibiotic exposure on gene expression and metabolism we report here , and on prior identification of robust gut microbiota-driven global changes in intestinal and hepatic histone PTM states ( Krautkramer et al . , 2016 ) . Measuring 65 acetylated and methylated histone PTM states at P23 , we found differential effects between PAT and control ( Supplementary file 5 ) , but notably greater effects in males than females ( Figure 7 ) , consistent with both the differences in gene expression and the T1D phenotypic enhancement . The PAT effects on hepatic chromatin were more robust than for the ileum ( Figure 7A ) , with greater responses again in males than in females ( Figure 7B ) . Further , the RNA-Seq studies from the P23 male mice showed dysregulated expression of histone modifying genes including anti-silencing histone chaperone Asf1b , and HADC-binding transcription factors Mier1 and Mier3 . These results , in conjunction with the altered SCFA production ( affecting transcription of multiple histone-modifying genes ( Krautkramer et al . , 2016; Terova et al . , 2016 ) , provide a further connection between the PAT-altered microbiota and the altered intestinal gene expression . Given that T1D involves immune-mediated destruction of pancreatic islets , we next asked whether the 1PAT-altered microbiome and its downstream effects on metabolism and innate immunity had differential effects on adaptive immune mediators . First , we examined whether there might be differential B-cell effects between PAT and control Assessing fecal IgA levels , we found consistent decreases at least to P70 in both the PAT-exposed males and females ( Figure 8A ) , with findings paralleling those observed in C57BL/6 mice ( Ruiz et al . , 2017 ) . To assess changes in other immunological loci , lymphocytes of the pancreas , and spleen were immunophenotyped in P42 PAT and control mice , after many innate differences had normalized , but still prior to the insulitis observed at P70 . In the spleen , there were increased frequencies of CD62L+ CD4+ and CD62L+ CD8+ T cells ( Figure 8B ) , indicating that the PAT-induced changes systematically increase the pool of naïve T cells in early life and decrease T cell differentiation . The 1PAT mice had greater numbers of B cells in the spleen and pancreas compared with controls ( Figure 8B ) . These results indicate that following PAT , there are altered frequencies of adaptive immune cells downstream of the intestine in the spleen and in the target organ ( pancreas ) , preceding T1D development . Differences in absolute count of lymphocytes were not observed in T cells or B cells , indicating there were no changes in lymphocyte quantity ( data not shown ) . However , the change in frequency of naive and activated T cells suggests that the observed differences in the systemic differentiation of these cells does not impact total cellularity . To more directly examine the relationship of microbiota changes with differential gene expression , we studied 43 ileal immune genes significantly differing in expression at P23 between 1PAT and control ( Supplementary file 6 ) in relation to the 15 differentially represented taxa ( Figure 2C ) . CompPLS modeling ( Ramanan et al . , 2016 ) indicated that 31 ( 72% ) genes differentially expressed after 1PAT exposure were significantly correlated with the differential taxa ( Figure 6C ) , creating a model of the interactions between the dominant differentiating microbial taxa and the affected intestinal genes . The patterns of connections between taxa and ileal gene expression in males and females markedly differed ( Figure 6—figure supplement 2 ) , consistent with the phenotypic differences . Particular taxa had relationships across a conserved group of specific families of host genes ( Figure 6—figure supplement 2 ) . This analysis linked the effects of 1PAT on the microbiota with the downstream ileal gene expression , and separating the males and females . Our prior studies in C57/Bl6 mice that show that PAT has no effect on gene expression in the ileum in the absence of a microbiota ( Ruiz et al . , 2017 ) indicate that the ileal gene expression effects we observed in the PAT-exposed mice were due to the microbiota/metagenemic shifts and were not direct antibiotic effects . We now show in NOD mice that the gut microbiome was substantially remodeled by a single early-life PAT exposure , losing diversity without recovery over the entire window relevant to the development of auto-immunity , and with selection for taxa that may be highly metabolically active . One hypothesis to explain the heightened auto-immunity is that reduced diversity of the gut microbiome compromises the ability to control metabolically active opportunistic bacteria . Such organisms , as represented by the short list we identified ( Figures 2C and 7 ) , might dysregulate early-life immune responses . Alternatively , the loss of particular beneficial organisms that participate in normal host metabolic and immune maturation , for example , through stimulating mucus production , might trigger the pathogenic pathway . These hypotheses are not exclusive , and rather may be ecologically linked , as suggested by our network model ( Figure 9 ) . This study revealed a small consortium of pathobionts ( Enterococcus , Blautia , and Enterobacteriaceae species ) with consistently increased relative abundances in the PAT-exposed male mice that developed accelerated T1D . Akkermansia , a taxon that has been inversely correlated with obesity , inflammation , and metabolic syndrome ( Cani and de Vos , 2017 ) , also is present in increased relative abundance in PAT-exposed male mice . The significant overrepresentation of these four taxa in both 1PAT and 3PAT mice confirms their prior disease-association using the same model in a different mouse facility ( Livanos et al . , 2016 ) , and they also have been linked to pathological processes in other intestinal microbiome studies ( Hänninen et al . , 2018; Kim et al . , 2017; Minter et al . , 2017 ) : ( i ) , Enterococcus species , common in both the human and murine gut , secrete products that alter host immune responses through NFkB/TLR pathways ( Tien et al . , 2017 ) , perturb other commensals ( Fisher and Phillips , 2009 ) , translocate to the liver , inducing autoimmunity , including autoantibodies ( Manfredo Vieira et al . , 2018 ) , and also induce infiltration of CD42+ MPO+ cells into the rat pancreas ( Korsgren et al . , 2012 ) ; ( ii ) , Children who later developed T1D have had altered gut microbiota representation of Blautia ( Murri et al . , 2013; Qi et al . , 2016 ) ; ( iii ) , Akkermansia mucinophila , intestinal mucin-degrading bacteria , may modulate host innate immunity ( Derrien et al . , 2011 , 2004 ) , including endocannabinoids , affecting inflammation , gut barrier permeability and peptide secretion ( Everard et al . , 2013 ) . Long-term exposure of older female NOD mice to vancomycin identified Akkermansia as a T1D-protective taxon ( Hansen et al . , 2012 ) , but that experiment differed substantially from ours; ( iv ) , in our model , an Enterobacteriaceae taxon other than E . coli was induced by 1PAT across P21 to P49 , consistent with identification of an unclassified Enterobacteriaceae member as the most significantly increased taxon in T1D children compared to healthy controls ( Soyucen et al . , 2014 ) . Enterobacteriaceae induce innate immune responses via Toll-like receptor 4 ( TLR4 ) ( Tapping et al . , 2000; Vasselon and Detmers , 2002 ) , which was dysregulated by PAT perturbation ( Figure 5C ) . Enhanced TLR responses to the Enterobacteriaceae could propagate immunopathology , consistent with microbiota regulating T1D development through MyD88-dependent TLRs ( Burrows et al . , 2015 ) . Biosynthetic gene cluster ( BGC ) analysis from the metagenomic sequencing revealed several enriched secondary metabolite pathways in 1PAT mice which map to an Enterococcus polysaccharide pathway , which could affect host immune responses ( Teng et al . , 2002; Xu et al . , 1998 ) , and also the Enterobacteriaceae siderophores , aryl polyenes , and NRPS family products ( Figure 2E ) . Bacterial siderophore and related transition-metal scavenging responses are often associated with pathobiont organisms and activity , and have been shown to be important mediators of bacterial community structure and of bacterial persistence within human hosts during infection . Furthermore , certain bacterial siderophores interact with the immune defense and immunomodulatory protein Lipocalin 2 in mice and humans , and could be relevant for normal immune development and tolerance ( Berard et al . , 2012; Raffatellu et al . , 2009 ) . Enterobacteriaceae ( e . g . E . coli ) aryl polyenes protect bacteria from oxidative stress from immune cells during colonization ( Cimermancic et al . , 2014 ) , which could interfere with bacteria-host interaction and immune development . NRPS family products are broad in structure but include metabolites with demonstrated anti-inflammatory and immunosuppressant activity such as cyclosporin A ( Felnagle et al . , 2008 ) . The small pathobiont consortium found to bloom in this study has many requisite features for T1D immunopathogenesis . We also identified S24-7 , Clostridiales , Oscillospira , Ruminococcus , and Anaeroplasma as taxa that were potentially protective against enhanced T1D . Unclassified Bacteroidales of the S24-7 family were associated with T1D protection in this study , in our prior 3PAT model in male NOD/ShiLtJ mice ( Livanos et al . , 2016 ) , and in studies in female NOD/BomTac mice ( Krych et al . , 2015 ) . The under-represented Anaeroplasma in PAT is consistent with the recent report that in NOD mice , this genus also was under-represented in the gut of those with low-diabetes-frequency compared with those at high-diabetes-frequency ( De Riva et al . , 2017 ) . Prior findings related to Clostridia , Oscillospira , and Ruminococcus were less consistent ( Krych et al . , 2015 ) . In total , our findings support the hypothesis that by selectively diminishing particular ( beneficial ) taxa , PAT exposure permits emergence of a less diverse microbiota ( Figure 2B ) , dominated most likely by highly metabolically active host-interactive taxa . Male and female NOD mice , even from the same litter , have substantially different T1D rates ( Markle et al . , 2013 ) , and differing immune response profiles ( Bao et al . , 2002 ) . Sex-specific differences in intestinal gene expression profiles present at P2 , prior to any antibiotic exposure , provide one explanation for the differential disease rates . We found that the signals from the 1PAT-altered microbiome are differentially transduced in the intestinal tissues of sibling male and female mice , despite their same mothers , diet , mouse facility , and microbiome composition . The very high rate of T1D development in the control females may have fully saturated the major pathogenetic pathway . The histone , RNA-Seq , and NanoString-based analyses clearly indicate that young male and female NOD mice differ in their responses to the same PAT-induced gut microbiome perturbations . Our studies of early life sex-specific ileal gene expression differences provide opportunities to better understand the sex-based differences in autoimmune disease pathogenesis , extending other recent work ( Thion et al . , 2018 ) . Exposure to 1PAT decreased the total numbers of genes maturing in both early life males and females , disproportionately affecting those involved in innate and adaptive immune pathways , including those in Toll-like receptor ( TLR ) signaling , NOD-like receptor signaling , and Th17 cell differentiation; each of these pathways have been implicated in prior T1D studies ( Burrows et al . , 2015; Lien and Zipris , 2009; Livanos et al . , 2016; Wen et al . , 2008 ) . Th17 roles in NOD mice are not clear-cut; blocking Th17 generation successfully inhibited T1D development ( Lee et al . , 2013 ) , while inducing Th17 cells with particular gut microbiota protected from T1D development in the same animal model ( Bedoya et al . , 2013 ) . Our study revealed that multiple components of the Th17 pathway are dysregulated in opposing directions , indicating the complexity of this immunological pathway in the disease phenotype . The pro-inflammatory serum amyloid protein A ( SAA ) is significantly induced in T1D patients ( Zhi et al . , 2011 ) . The biphasic response of Saa1 to the microbiota changes that we observed may reflect the dominance of highly host-interactive organisms early , and the altered innate signaling pathways later . Perturbation of the mucin layer , a major barrier to the interaction of the microbiota with intestinal cells has been associated with intestinal diseases ( Crost et al . , 2016; Rokhsefat et al . , 2016; Tailford et al . , 2015 ) . Nos2 is critical for cellular signaling and immune defense responses ( Bogdan , 2015 ) , since NO activates Th1 phenotypic responses , regulates anti-inflammatory pathways , and mediates tissue restoration ( Wink et al . , 2011 ) . Nos2 expression , down-regulated early in 1PAT recipients , may be a good sensor of intra-luminal events , transducing the microbial signals to downstream host functions ( Atarashi et al . , 2015 ) . Given that binding of Epidermal Growth Factor ( EGF ) to the EGF receptor ( EGFR ) subsequently activates multiple signaling pathways—including PAR2 , TLR , Ras/MARK , PI3K/AKT and STAT—which regulate the intestinal barrier and permeability ( Good et al . , 2012; Tang et al . , 2016; Wang et al . , 2015 ) , we examined their maturation with age , and the effects of sex and PAT exposure on expression . Egfr expression was tightly conserved across the 47 specimens examined from P2 to P23 ( relative Mean expression 4 . 37 ± 0 . 62 × 10–5 ) . Very young ( P2 ) males had significantly higher levels of Egfr expression than young females . Subsequently , all rates normalized by P12 and thereafter ( Figure 6—figure supplement 3A ) . For Egf , levels did not significantly differ according to treatment or sex ( Mean 1 . 95 ± 0 . 58 × 10–6 ) , but diminished between P12 and P23 ( Figure 6—figure supplement 3B ) . Findings were similar if Egf was examined as a ratio in relation to EgfR ( Figure 6—figure supplement 3C ) . Thus , we observed that there is a very early in life ( P2 ) increased level of Egf in males , with rapid normalization , but that Egf levels significantly fell with weaning . Egfr is an example of a gene with sex-specific early life expression differences . In contrast , Egf is a gene whose expression in the ileum undergoes age-related changes , but which are independent of both sex and treatment effects [Class III ( PAT-resistant ) in Figure 5B] . Given that Egfr affects tight junctions and adherence in the colon ( Basuroy et al . , 2005 ) , the fall at weaning could explain later bacterial translocation and downstream immunological effects . In total , the altered expression of the identified/validated 1PAT-dysregulated genes may participate in the subsequent intestinal immune system dysmaturation . The histone modification data are consistent with the dysregulation of ileal gene expression in this and our prior study ( Livanos et al . , 2016 ) . The histone data also are in line with sexual dimorphism in how the gut microbiota communicate early-life environmental exposures to host chromatin ( Figure 7B ) , consistent with both the disease risk dimorphism ( Yurkovetskiy et al . , 2013 ) , and the early-life transcriptional phenotypes we observed . To link 1PAT-induced changes in global histone PTMs to genomic loci and ultimately to downstream effects on gene expression , both the GEO ( Gene Expression Omnibus ) ( Barrett et al . , 2013; Edgar et al . , 2002 ) and ENCODE ( ENCyclopedia Of DNA Elements ) ( ENCODE Project Consortium , 2012 ) repositories were mined for publicly available ChIP-seq data sets targeting either histone H3 K36me2 or acetylated histone H4 , and most specifically histone H4 K16ac . These modifications were selected for further analysis due to the relatively high magnitude of differences and sex-specificity of responses at these sites ( Figure 7 , Supplementary file 5 ) . For histone H3 , the variant histone H3 . 3 , and histone H4 provide an unsupervised and global assessment of common PTMs . In the male 1PAT ileum , there was a modest but statistically significant decrease in acetylation of histone H4 and a concomitant increase in abundance of unmodified histone H4 compared to controls , suggesting that 1PAT leads to a net loss of acetylated H4 . These results are consistent with our previous observations in the C57BL/6 proximal colon that the gut microbiota induces histone acetylation ( Krautkramer et al . , 2016 ) , suggesting that 1PAT exposure decreases key microbial populations important for this chromatin response . There also were significant decreases in H3 K27me1 K36me1 , H3 K27un K36me1 , and H3 K27un K36me2 peptides in 1PAT ileum relative to controls ( Figure 7A ) . Importantly , and in contrast , the effects of 1PAT in the female ileum were minimal ( Figure 7B ) . Coupled with the differential gene expression asymmetry between male and female mice , both a priori ( P2 ) and PAT-induced , these studies link epigenetic changes with the differential disease phenotypes . Similar to the chromatin signatures in male ileum , the 1PAT exposure resulted in hepatic hypoacetylation of histone H4 and hypomethylation of histone H3 K36-containing peptides in males and females and a concomitant increase in the abundance of peptides containing highly methylated H3K27 ( Figure 7A ) . Acetylation of histone H4 and methylation of H3K36 are both associated with active transcription , whereas methylation of H3K27 is associated with transcriptional repression ( Pasini et al . , 2010; Robinson et al . , 2008; Zhao and Garcia , 2015 ) . Other search parameters explored include limiting data to those samples from either mouse or human origin and either intestine or liver tissues or cell lines ( including primary cultures or transformed cell lines ) . However , this search returned no publicly available data sets that met these criteria , precluding further analyses . Since SAA can induce expression of Jmjd3 ( Kdm6b ) , a histone H3 lysine 27 ( H3K27 ) demethylase , reducing H3K27 trimethylation ( Yan et al . , 2014 ) , Saa dysregulation by PAT-induced microbiome perturbation may alter epigenetic status in early life ( Figure 7 and Figure 6—figure supplement 1 ) . H3K27 methylation is associated with repression of both Runx1 through enhancer of zeste homolog 2 EZH2 ( Takayama et al . , 2015 ) , and iNOS ( Dreger et al . , 2016 ) . Thus , our observation that histone H3K27me3K36me2 ( K27 trimethylation ) was > 2 fold up-regulated by 1PAT while histone H3K27un ( unmodified ) was significantly down-regulated ( Figure 7A ) is consistent with the RNA-Seq and RT-qPCR analyses showing early up-regulation of Saa and down-regulation of Nos2 and Runx1 in the 1PAT male mice ( Figure 6A , B , Figure 6—figure supplement 1A , and Supplement file 3 ) . Finding 1PAT-reduction of cecal butyric acid confirmed our prior result with 3PAT ( Livanos et al . , 2016 ) . By inhibiting histone-modifying genes in intestinal cells , SCFAs including butyrate , affect epigenetic status and thus gene expression ( Fellows et al . , 2018; Schilderink et al . , 2013 ) ; the transcriptional analyses we performed provide direct evidence of such dysregulation . Diminished butyrate levels are consistent with the decreased ileal Runx1 and Foxp3 expression that we observed , which could differentially shift T-helper cell maturation away from Treg-cells ( Furusawa et al . , 2013; Geuking et al . , 2013; Smith et al . , 2013 ) . Our analyses show that antibiotic-induced gut microbiome remodeling reduced intestinal SCFA levels , dysregulated histone PTM status , and repressed mucin genes- all of which contribute to perturbing innate intestinal immunity development . Since butyrate and propionate both increase histone H3 and H4 acetylation and H3 methylation , affecting the Muc2 promoter and increasing Muc2 mRNA levels ( Burger-van Paassen et al . , 2009 ) , the low levels we observed are consistent with the decreased Muc2 expression . In total , our investigation suggests the following model to explain the 1PAT effects on T1D in the male NOD mice ( Figure 9 ) . Antibiotic administration in early life selected for particular intestinal microbial populations , continuing weeks after the antibiotic stopped , including small groups of significantly over- and under-represented taxa . Changes in the representation of these taxa , especially those predicted to have highly active metabolism and thus host-interaction , can initiate the primary events in the disease enhancement . The altered microbial populations and their products , including SCFAs , differentially interact with ileal epithelial cells , affecting histone modification , and changing gene expression and its normal maturation . The downstream effects on specific innate genes and pathways and the metabolic changes would then influence how adaptive immunity develops . Together , the broad cascade of events altering the expression of specific host genes and pathways appears sufficient to trigger and accelerate T1D in males . These studies contribute to a growing body of evidence on the effects of early life antibiotic exposures in mouse models of disease ( Cox et al . , 2014; Nobel et al . , 2015; Ruiz et al . , 2017; Schulfer et al . , 2018 ) , and particularly of T1D ( Candon et al . , 2015; Livanos et al . , 2016; Pearson et al . , 2016 ) , and are consistent with some ( Pflüger et al . , 2010; Yallapragada et al . , 2015 ) , but not all ( Hviid and Svanström , 2009; Kemppainen et al . , 2017 ) epidemiologic studies in humans identifying early life antibiotic exposure as a risk factor for T1D development . This simplified animal model , the taxonomic , metagenomic , and metabolic leads we have identified , and the approach and classification system for gene expression maturation advance understanding of the mechanisms by which gut microbiome perturbations contribute to the pathogenesis of T1D and to other immune-associated diseases . NOD/ShiLtJ mice ( 6 weeks old ) were purchased from Jackson Laboratory ( Bar Harbor ME ) , and bred in an SPF vivarium at the New York University Langone Medical Center ( NYUMC ) Skirball animal facility . All animal procedures were approved by the NYUMC Institutional Animal Care and Use Committee ( IACUC protocol no . 160623 ) . The dams and their litters were randomly assigned to control or PAT groups . At postnatal ( P ) day 23 , the pups were weaned and housed to separate males and females . All mice received acidified drinking water supplied by the facility routinely except for the periods when some litters were receiving antibiotic treatment . A therapeutic dose of the macrolide tylosin tartrate ( Sigma-Aldrich , Billerica MA ) was given to mice in their non-acidified drinking water with 333 mg/L ( about 50 mg/kg body weight/day ) ( Livanos et al . , 2016 ) on P5-10 for 1PAT mice or in three courses ( P10-15 , P28-31 , and P37-40 ) ( 3PAT ) , exactly as described ( Livanos et al . , 2016 ) . All mice were monitored for diabetes by weekly measurement of tail blood glucose using the FreeStyle Lite meter and blood glucose test strips ( Abbott Diabetes Care Inc . , Abbott Park IL ) . The measurement was started at week 11 of age and continued to week 30; diabetes onset was defined as two consecutive values > 250 mg/dl ( Livanos et al . , 2016; Wen et al . , 2008 ) . Kaplan-Meier analysis was applied for evaluating diabetes progression of treatments ( Kaplan and Meier , 1958 ) , and the Log-rank ( Mantel-Cox ) test was applied to detect the difference significance between treatment ( Harrington and Fleming , 1982 ) . Individual mice were placed in an empty clean beaker for 2–5 min to allow them to defecate normally to obtain 3–4 pellets , which were frozen at –80°C for further analysis . At mouse sacrifice , the distal ileum ( 1 cm long ) was collected , ileal contents removed , and tissue and contents separately introduced into RNAlater ( Qiagen , Valencia CA ) . The next most distal 1 cm ileal segment with contents was frozen at –80°C for 16S rRNA analysis , and the subsequent segment without contents was frozen at –80°C for histone modification analysis . Cecum samples with contents were frozen at –80°C for 16S rRNA and metabolic analyses . After removal of the colonic contents , colonic tissues were collected into RNAlater , and the more proximal tissues and contents frozen at –80°C for microbiome 16S rRNA analysis . The liver of each mouse was collected and frozen at –80°C for metabolic analysis and histone modification analysis . From mice sacrificed at P42 and P70 , the pancreas was removed and fixed in freshly prepared modified Bouin’s fixative ( Leiter , 2001 ) , paraffin-embedded , sectioned , stained , and scored , as described ( Livanos et al . , 2016 ) with modification by using methyl green as counterstain ( Forestier et al . , 2007 ) . Fecal and intestinal microbiota DNA was extracted using the PowerLyzer PowerSoil DNA Isolation Kit ( MoBio , Carlsbad CA ) and the PowerSoil-htp 96 Well Soil DNA Isolation Kit ( MoBio ) . The amplicon library of V4 regions of the bacterial 16S rRNA genes were obtained by triplicate PCR with barcoded fusion primers , quantification with the Qubbit 2 . 0 Fluorometer ( Life Technologies , Carlsbad , CA ) , and combination of each DNA sample at equal concentrations as previously described ( Livanos et al . , 2016 ) . The library was sequenced with the Ilumina MiSeq 2 × 150 bp paired end platform ( Ilumina , San Diego CA ) at the NYUMC Genome Technology Center . QIIME 2 . 0 was used as the amplicon read processing pipeline as described ( https://qiime2 . org/ ) ( Caporaso et al . , 2010 ) . Reads with more than three consecutive low-quality bases ( Phred score < 20 ) were filtered , and only reads with > 75% of the original length were retained . OTUs were picked using the open reference picking strategy based on the Greengenes database ( DeSantis et al . , 2006 ) . Reads were first clustered into 97% identity OTUs using UCLUST program ( Edgar , 2010 ) , and taxonomy assignment was performed using the RDP Classifier with a confidence interval of 50% , and chimeras were removed using ChimeraSlayer ( Haas et al . , 2011 ) . Microbial diversities within samples ( α-diversity ) and between samples ( β-diversity ) , and taxa relative abundance were analyzed using QIIME2 . 0 . α-diversity was evaluated with phylogenetic diversity ( Faith , 1992 ) , and mean values and statistical significance tests were calculated using Prism ( GraphPad Software , La Jolla CA ) . β-diversity was evaluated with unweighted UniFrac ( Lozupone and Knight , 2005 ) . Statistical significance of the inter- and intra-group β-diversity was determined by permutation testing using R . Assessment of significantly different taxa between different treatment groups was performed using the ANCOM program in QIIME2 . 0 ( Mandal et al . , 2015 ) . Mixed effects models ( Laird and Ware , 1982 ) were fitted to test the disparity in relative abundance for each taxon ( from phylum to genus ) between two groups of mice ( 1PAT vs . 1PAT control and 3PAT vs . 3PAT control , respectively ) for both males and females . The mixed effects models included relative abundances for each taxon as outcomes , and group indicator and time as fixed effects , and the intercept and slope of the linear time trend for each mouse as random effects . Unclassified taxa and taxa which were monotonic or singletons in any of the groups ( 1PAT , 1PAT control , 3PAT and 3PAT control , for males and females , separately ) , were excluded in the analysis . For multiple testing correction , the Benjamini-Hochberg ( BH ) procedure ( Benjamini and Hochberg , 1995 ) was applied for each taxonomic level . A total of 48 fecal samples from 24 mice ( 6 males and 6 females each from wither the 1PAT and 1PAT control groups at P12 and P49 were chosen for metagenomic study , along with each inoculum . Extracted genomic DNA ( 5 ng ) from each sample was used for library preparation and subsequent whole genome sequencing ( WGS ) using the Illumina HiSeq 2500 platform . Samples were sequenced over 6 flow cell lanes as 100 bp paired-end reads . The metagenomes were pre-processed for quality metrics using Trimmomatic ( Bolger et al . , 2014 ) and aligned to the mouse genome ( mm10 ) to reduce host-contaminated sequences using KneadData as previously described ( Schulfer et al . , 2018 ) . After filtering low-quality and contaminated sequences , we performed functional profiling to detect microbial genes and pathways using HUMAnN2 v0 . 9 . 5 ( Abubucker et al . , 2012 ) with default settings and screened against the EC-filtered UniRef90 database . Raw shotgun reads were quality controlled using SHI7 ( Al-Ghalith et al . , 2018 ) and aligned using exhaustive gapped alignment at 95% identity ( Al-Ghalith and Knights , 2017; Needleman and Wunsch , 1970 ) against a reference database of 21 , 186 putative BGCs predicted by antiSMASH or deposited in the MIBiG database ( Blin et al . , 2017; Medema et al . , 2015; Weber et al . , 2015 ) . The per-sample metagenomic coverage of each BGC was calculated using in-house Python and R code and filtered to pathways with a ratio of actual coverage to expected coverage ( expected coverage probability is defined as 1-exp⁡ ( NLread/LBGC ) , where N = number of reads , Lread = median read length , and LBGC = BGC sequence length ) of at least 0 . 75 . Differentiating BGCs were identified by comparing BGC presence/absence frequency between the treatment groups using Fisher’s exact test with FDR correction at q < 0 . 15 . To collapse homologous BGCs we used custom Python and C code to hierarchically cluster the pathways based on amino acid identity and open reading frame composition ( Rashidi et al . , 2018; Shields-Cutler et al . , 2018a , Shields-Cutler et al . , 2018b ) . Cluster annotations and taxonomic assignments were derived from their antiSMASH references . The metabolite pathways were discovered directly from the metagenomic data presented in the study , by analyzing coverage of the DNA pathways . The metagenomic data were annotated to particular pathways and taxa by DNA sequence homology ≥95% to BGC pathways present in the antiSMASH database ( Blin et al . , 2017 ) , including Enterobacteriaceae reference strains . Cecal contents ( ~10 mg ) were subjected to a targeted GC-MS analysis to quantify short chain fatty acid levels , as described ( Lucas et al . , 2018; Tangerman and Nagengast , 1996 ) , with modifications . Metabolite extraction was carried out by addition of 50:1 extraction solvent [80% methanol in water ( LCMS Grade ) with 0 . 5 mm zirconium/silica beads ( Research Products International ) ] to the measured sample mass ( ±0 . 01 mg ) in a tared bead blaster tube . Each vial underwent two homogenization cycles at 30 s on/30 s off at 6 . 0 m/s ( 4°C ) and insoluble matter was pelleted by centrifugation at 21 , 000 g for 3 min at 4°C . The supernatant was transferred to a gas-tight 1 . 5 mL glass GC vial ( Agilent Technologies ) for analysis . Using aThermoTM TRACE 1310 gas chromatograph , a split injection method used a split ratio of 25 , 1 μL injection volume , and split flow of 25 . 0 mL/min , with constant 250°C temperature for the injector with a Topaz low pressure drop precision liner with wool ( RestekTM ) . Helium carrier gas was used at a constant flow of 1 . 0 mL/min with a ThermoTM TG-WAXMS A column ( 30m × 0 . 25 mm ) . The 15 min thermal gradient profile included equilibration of 2 min followed by a 1 min hold at 100°C , a ramp from 100 to 145°C at 20°C /min , a 4 min hold at 145°C , a ramp from 145 to 165°C at 15°C /min , finishing with a 3 min hold at 165°C . The GC system was coupled to a Thermo Q ExactiveTM mass spectrometer operating in electron ionization positive mode at 70KeV . MS1 scan range from 32 to 350 m/z was used at resolution 120 , 000 with AGC target 1e6 and maximum IT 100 ms . SCFA intensities were quantified at 5 ppm tolerance within a 0 . 1 min retention time window , using a 5-point standard curve ( from 10 to 1000 pg/μL ) , with a randomized acquisition order of samples and standards , run in duplicate . Standard curve analyte intensities were fit to a linear regression and sample values reported as picogram analyte/mg cecal material ( ppm ) . Aliquots of approximately 100 mg of liver tissue were placed in a 2 mL Magna Lyser tube ( Roche ) and mixed with 500 µL of ice-cold 50% acetonitrile , pulse sample 2 × 30 s @ 2000 in Magna Lyser ( Roche ) . The homogenized mixture was centrifuged at 18 , 600 g for 5 min at 4°C to yield supernatant . The quality control samples for liver and serum were made by pooling individual liver supernatant into one tube and pooling individual serum sample into another tube . Aliquots of 100 µL liver supernatant or 30 µL serum sample were mixed with 1000 µL of a cold degassed acetonitrile-isopropanol-water solution ( 3:3:2 ) , and centrifuged for 4 min at 18 , 600 g , and then dried by vacuum in new tubes . Study samples and quality control samples were processed identically . Study samples were randomized , and quality control pool samples ( prepared under identical conditions ) were interspersed . Samples were derivatized using a two-step method with acquisition parameters similar to Fiehn et al . ( Fiehn et al . , 2008; Kind et al . , 2009 ) . To summarize , a 0 . 5 µl volume was injected into an Agilent 6890 gas chromatograph ( Agilent Technologies ) with a 30 m long , 0 . 25 mm i . d . Rxi5Sil-MS column with 0 . 25 µm film thickness ( Restek , Bellefonte PA ) , using a 250°C injector temperature in splitless mode with 25 s splitless time , at a constant flow of 1 ml/min . The oven temperature was ramped ( 20°C/min ramp ) from 50°C to 330°C ( Chou et al . , 2017 ) . Data were acquired using a Leco Pegasus 4D TOF-MS ( Leco , Saint Joseph MI ) with a 280°C transfer line temperature , electron ionization at –70 V , and an ion source temperature of 250°C . Spectra were acquired from m/z 50–750 at 20 spectra s-1 and 1850 V detector voltage . GC-TOF-MS data were deconvoluted by ChromatTOF ( Leco , St . Joseph MI ) and then further processed by BinBase ( Skogerson et al . , 2011 ) for peak retention index calculations , spectral identification , and generation of a table of peak identifications and intensities . Multivariate data analysis was conducted using SIMCA 13 . 0 ( Umetrics , Sweden ) for data normalized to the sum of intensities . Mean-centered and pareto-scaled data were analyzed by Principal Component Analysis ( PCA ) and Orthogonal Projections to Latent Structures Discriminant Analysis ( OPLS-DA ) . Peaks with Variable Influence on Projections ( VIP ) ≥ 1 . 0 were deemed important for differentiating the study groups . Significant changes in pairwise comparison were evaluated by Wilcoxon Rank-Sum test by SAS 9 . 4 ( SAS Institute Inc . , Cary NC ) with p value ≤ 0 . 05 denoting statistical significance . Total RNA was extracted from mouse tissues using the PureLink RNA Mini Kit ( Invitrogen , Carlsbad CA ) , and contaminating genomic DNA was removed by treatment with DNase I ( Qiagen ) . Total RNA quality and quantity were determined using the NanoDrop ND-1000 UV-Vis Spectrophotometer ( NanoDrop Technologies , Inc . , Wilmington DE ) , and Agilent 2100 Bioanalyzer ( Agilent Technologies , Santa Clara CA ) . For RNA-Seq , we used methods , as described ( Ruiz et al . , 2017 ) . Reads were aligned to the mouse GENCODE GRCm38 . p5 ( M14 release ) genome using STAR v2 . 5 . 2b ( Dobin et al . , 2013 ) , and the read summarization program FeatureCounts ( Liao et al . , 2014 ) was used to count mapped reads against annotated genes . Differential expression analysis between different treatments and KEGG pathways visualization was performed using DESeq2 in the R-package ( Love et al . , 2014 ) and Pathview , respectively ( Luo and Brouwer , 2013 ) . p values were corrected for multiple comparisons , based on the false discovery rate ( FDR ) ( Benjamini and Hochberg , 1995 ) with significance considered by the adjusted p value<0 . 05 . Differentially expressed pathways and functions were interpreted using with Ingenuity Pathway Analysis ( IPA , QIAGEN Redwood City , http://www . ingenuity . com ) . Immune gene expression profile of each of the above RNA samples was evaluated by using the nCounter GX mouse immunology kit ( NanoString Technologies , Seattle WA ) . Counts were normalized using DESeq2 ( Love et al . , 2014 ) . P values were corrected for multiple comparisons , based on the false discovery rate ( FDR ) ( Benjamini and Hochberg , 1995 ) , with significance considered by the adjusted p value<0 . 05 . Heat maps were generated using the pheatmap package in R ( Kolde and Vilo , 2015 ) . To evaluate specific gene expression , cDNA was synthesized through reverse transcription from 1 µg of each total RNA sample above with the Verso cDNA Synthesis kit ( Thermo Scientific , Waltham MA ) using random hexamer primers provided . qPCR was run in a LightCycler 480 system ( Roche , Branchburg NJ ) using 10 ng cDNA , target gene-specific primer pairs ( Supplementary file 7 ) and Power SYBR Green PCR Master mix ( Roche ) . Target mRNA was normalized to 18S rRNA or housekeeping gene HPRT as an internal control in each sample ( Saha and Blumwald , 2014 ) . For group mean comparisons , the Mann Whitney test was performed with p value < 0 . 05 as significant difference . Fecal samples were resuspended in PBS at a concentration of 50 mg/mL by extensive vortexing , allowed to stand for 20 min , and centrifuged at 16 , 000 g for 10 min to collect supernatant as described ( Haneberg et al . , 1994 ) . Each supernatant was assessed for IgA using the mouse IgA ELISA kit with suitable dilutions , according to the manufacturer’s instructions ( Bethyl , Montgomery TX ) , and the absorbance was measured at a wavelength of 450 nm using the Dynex MRX TC Revelation microplate reader ( Dynex Technologies , Chantilly VA ) ( Ruiz et al . , 2017 ) . The pancreas or splenic tissue of each mouse was placed into 5 mL of PBS supplemented with 2% fetal calf serum ( Corning , Tewksbury MA ) , 1 mg/mL trypsin inhibitor ( Sigma , St . Louis MO ) , 1 mg/mL of Collagenase IV ( Worthington Biochemical , Lakewood , NJ ) , and 0 . 5 mg/mL DNaseI ( Sigma ) and the spleen in DMEM ( Corning ) supplemented with 10% fetal calf serum ( FCS ) . The pancreas or spleen was physically disrupted for 5 min and dissociated using a GentleMACS Dissociator ( Miltenyi Biotec Inc . , Auburn CA ) . Following dissociation , the tissues were enzymatically digested for 20 min at 37°C . Following enzymatic digestion , the cell suspension was filtered , washed , and resuspended in cold PBS with 2% FCS . Spleens were physically disrupted over 70-micron nylon mesh ( Corning ) on ice . Cells were pelleted at 300 g for 5 min , supernatants were removed , followed by red blood cell lysis with ACK ( ammonium-chloride-potassium ) lysis buffer ( Lonza , Basel , Switzerland ) . Single cell suspensions were washed and resuspended 1 mL of cold PBS with 1% FCS . Cells were stained using anti- CD45 , CD4 , CD8 , CD3 , CD44 , CD62L , CD19 and B220 ( BioLegend , San Diego CA ) . Data were acquired on a BD LSRII ( BD Bioscience , San Jose CA ) and analyzed using FlowJo v10 . 2 ( Tree Star Inc . , Ashland OR ) . Histones were isolated from flash-frozen whole post-mortem ileum and liver from P23 mice ( n = 4 or 5 per sex and treatment group ) and prepared for analysis by liquid chromatography coupled to tandem mass spectrometry ( LC-MS/MS ) . Histone extraction , label-free chemical derivatization , and data acquisition using a Dionex Ultimate3000 nanoflow HPLC with a Waters nanoAcquity UPLC C18 column ( 100 µm × 150 mm , 3 µm ) online with aThermo Q-Exactive mass spectrometer using a data- independent acquisition method , as previously described ( Krautkramer et al . , 2016 ) . Following data acquisition , normalization and quantification of histone PTM abundance was performed as previously described ( Krautkramer et al . , 2015 ) . Isobaric and co-eluting peptides were not deconvoluted , and are denoted as such ( e . g . K18ac+K23un and K18un+K23ac are isobaric and co-eluting and are denoted as a single value for K18ac/K23ac since their MS1 ions are identical and thus representative of both peptide species ) . Normalized percent of total values were then used to calculate fold-changes and statistics . All p values were generated using Welch’s t test , with statistical significance set at p<0 . 05 . Using the compPLS framework ( Ramanan et al . , 2016 ) , we aimed to detect associations between taxa in the gut microbiome and host immune genes . To prevent detection of spurious associations , we: ( i ) performed a centered log-ratio ( clr ) transformation on the OTU relative abundance data; ( ii ) applied a variance decomposition to extract within-subject variation; and ( iii ) estimated a sparse linear model via sparse Partial Least Squares ( sPLS ) regression to detect associations between a sparse set of multi-collinear features ( OTUs ) and responses ( host covariates ) ( Bastien et al . , 2005; Chun , 2010 ) . As our host covariates or response variables , we used expression levels representing significantly differentially expressed genes . We first filtered OTUs at the genus level with relative abundance > 0 . 01% . Taxa were then selected if present in at least one of the intestinal samples . This filtering resulted in a species-level OTU table . We decomposed the clr-transformed OTUs and host response data using a two-factor variance decomposition to account for differences in sex ( M or F ) and treatment group ( 1PAT and 1PAT control ) . For each sPLS run , we set the number of latent components to the number of non-zero singular values in the cross-covariance matrix . To find a sparse set of significant associations between OTUs and genes , we ( i ) applied sPLS and used a stability approach to regularization selection ( StARS ) ( Liu et al . , 2010 ) to select the sparsity level; and ( ii ) used bootstrap-based empirical p value calculation to assess the significance of associations of the StARS-selected support ( Lê Cao et al . , 2009 ) . We calculated empirical p values over 5000 bootstraps and set a p value threshold of 0 . 05 after FDR multiple testing correction . We visualized significant associations as a network using the igraph package in R ( Csárdi and Nepusz , 2006 ) . RNA-Seq data that support the findings of this study have been deposited in ArrayExpress database ( www . ebi . ac . uk/arrayexpress ) with the accession code E-MTAB-6826 ( https://www . ebi . ac . uk/arrayexpress/experiments/E-MTAB-6826 ) . 16S rRNA data have been deposited in QIITA ( https://qiita . ucsd . edu/ ) with the identifier 11242 ( https://qiita . ucsd . edu/study/description/11242 ) . Ileal NanoString data have been deposited in NCBI's Gene Expression Omnibus ( https://www . ncbi . nlm . nih . gov/geo/ ) and are accessible through GEO Series accession number , GSE101721 ( https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE101721 ) . Shotgun metagenomics data have been deposited in the European Nucleotide Archive ( ENA ) ( https://www . ebi . ac . uk/metagenomics/ ) under the accession number , PRJEB26585 ( http://www . ebi . ac . uk/ena/data/view/PRJEB26585 ) . Metabolomics data have been deposited at the NIH Common Fund Metabolomics Workbench ( www . metabolomicsworkbench . org; doi: 10 . 21228/M8C39R ) . Type 1 diabetes ( T1D ) , pulsed therapeutic antibiotic treatment ( PAT ) , non-obese diabetic mouse ( NOD mouse ) , histone post translational modification ( histone PTM ) .	The human body contains many microbes that play important roles in our health . These microbes begin to live in the intestines , skin , and mouth shortly after birth . They form complex communities called the microbiome , which changes as babies develop . The microbiome works with organs to maintain human health . For example , the lower intestinal tract is home to the most numerous and active microbes in the body . The intestines provide microbes with food and a welcoming environment , and the microbes make products the body needs , influence immune system development , and help maintain a balance of beneficial microbes . Use of antibiotics to treat infections , particularly early in life , disrupts intestinal microbe communities . Recent studies show that such microbiome disturbances may affect how the immune system develops and the rate at which type 1 diabetes develops . Type 1 diabetes is an autoimmune disease in which the immune system destroys cells in the pancreas that produce insulin . Scientists would like to learn more about how use of antibiotics in early life may contribute to the development of this disease . Now , Zhang et al . show that a single course of antibiotics administered early in life accelerates the development of type 1 diabetes in mice prone to develop the disease . In the experiments , a strain of laboratory mice that spontaneously develops type 1 diabetes were either given a single course of antibiotics , three courses of antibiotics , or no antibiotics in their first weeks of life . After one single course , the gut microbiome was different in mice treated with antibiotics compared with mice who were never exposed . The antibiotics also changed the molecules produced by these microbes . These alterations in the microbiome turned on or off certain genes in the intestine , affecting the development of the immune system . Zhang et al . identified some microbes that appear to protect against type 1 diabetes and others that seem to speed it up and how they do so . Antibiotic use in children is very common , so finding ways to reduce its potentially harmful effects on development are critical . The experiments provide one way to study how antibiotics may contribute to autoimmune disease . It also may allow scientists to test ways to reverse harmful change .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "immunology", "and", "inflammation" ]	2018	Antibiotic-induced acceleration of type 1 diabetes alters maturation of innate intestinal immunity
The Rising Star cave system has produced abundant fossil hominin remains within the Dinaledi Chamber , representing a minimum of 15 individuals attributed to Homo naledi . Further exploration led to the discovery of hominin material , now comprising 131 hominin specimens , within a second chamber , the Lesedi Chamber . The Lesedi Chamber is far separated from the Dinaledi Chamber within the Rising Star cave system , and represents a second depositional context for hominin remains . In each of three collection areas within the Lesedi Chamber , diagnostic skeletal material allows a clear attribution to H . naledi . Both adult and immature material is present . The hominin remains represent at least three individuals based upon duplication of elements , but more individuals are likely present based upon the spatial context . The most significant specimen is the near-complete cranium of a large individual , designated LES1 , with an endocranial volume of approximately 610 ml and associated postcranial remains . The Lesedi Chamber skeletal sample extends our knowledge of the morphology and variation of H . naledi , and evidence of H . naledi from both recovery localities shows a consistent pattern of differentiation from other hominin species . The Rising Star cave system ( 26°1′13′′ S; 27°42′43′′ E , Figure 1 ) in the Cradle of Humankind World Heritage Site , Gauteng Province , South Africa , is known for the discovery in 2013 of more than 1 , 550 fossils representing a novel hominin species , Homo naledi ( Berger et al . , 2015; Dirks et al . , 2015 ) . These remains , representing at least 15 individuals of various ages at death , were recovered from a deep chamber ( 30 m below ground surface ) , named the Dinaledi Chamber . 10 . 7554/eLife . 24232 . 003Figure 1 . Geographical location of the Rising Star cave in the Cradle of Humankind UNESCO World Heritage Site . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 003 Additional fossil hominin material was subsequently discovered in the Lesedi Chamber of the cave system in November 2013 by Rick Hunter and Steven Tucker . The deposition of sediment and skeletal remains in the Lesedi Chamber has no direct geological connection to the Dinaledi Chamber . In the time following the first discovery of hominin material in the Lesedi Chamber , excavators have recovered 131 hominin specimens within three discrete collection areas . The sedimentary context of the three collection areas is broadly similar , but we have not yet established whether the fossil material resulted from a single depositional episode or from multiple distinct events . We approached the hominin skeletal remains from the Lesedi Chamber with the aim of identifying elements , assessing the number of individuals represented by the material , and determining the taxonomic identity of the sample . Preliminary examination of the hominin remains suggested that they are morphologically consistent with H . naledi . To test this hypothesis , we carried out systematic comparisons , employing the taxonomic diagnosis of this species ( Berger et al . , 2015 ) and focusing upon those characters that distinguish H . naledi from other hominin taxa . We also present essential contextual information to place the specimens within the Lesedi Chamber and provide descriptions of the hominin specimens , focusing upon those features that contribute to the taxonomic diagnosis of the sample . All identifiable hominin fragments , including those that do not present information useful to taxonomic diagnosis , are listed in Table 1 . 10 . 7554/eLife . 24232 . 004Table 1 . Hominin fossil material from the Lesedi Chamber . All diagnostic hominin specimens are listed , with attribution to element . Specimens that have been refitted are not listed separately . Most Locality 102a cranial fragments are presumed to be part of LES1 and are not listed separately . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 004Specimen numberElementNotesLOCALITY 102aLES1craniumconstituted of 57 specimens , not listed separatelyU . W . 102a-001proximal right femurU . W . 102a-002proximal right humerusU . W . 102a-003proximal left femurU . W . 102a-004distal left femurU . W . 102a-010right scapula fragmentacromionU . W . 102a-013humeral head fragmentsU . W . 102a-015right proximal ulnaU . W . 102a-018long bone fragmentimmatureU . W . 102a-019partial ribU . W . 102a-020right ulna fragmentU . W . 102a-021right clavicleU . W . 102a-025right radius shaft fragmentU . W . 102a-028right fourth metacarpalU . W . 102a-036T10 vertebraU . W . 102a-039rib fragmentsU . W . 102a-040long bone shaft fragmentU . W . 102a-117right scaphoidU . W . 102a-138right ilium fragmentsimmatureU . W . 102a-139L5 vertebra fragmentsU . W . 102a-148sternum fragmentU . W . 102a-151T11 vertebraU . W . 102a-152rib fragmentsU . W . 102a-154T12 and L1 vertebraefound in articulationU . W . 102a-155mid-thoracic vertebral bodyU . W . 102a-171atlas fragmentU . W . 102a-172atlas fragmentU . W . 102a-189rib fragmentU . W . 102a-195rib fragmentU . W . 102a-206left clavicle fragmentU . W . 102a-207rib fragmentU . W . 102a-210sacral elementimmature , possibly S1U . W . 102a-231rib fragmentU . W . 102a-232rib fragmentU . W . 102a-236humerus head fragmentU . W . 102a-239left clavicle fragmentU . W . 102a-247right scapula fragmentcoracoid processU . W . 102a-250right first ribU . W . 102a-252rib fragmentU . W . 102a-256left scapula fragmentportion of body , spine , and acromionU . W . 102a-257left proximal humerusU . W . 102a-279left scapula fragmentpartial glenoid fossaU . W . 102a-280rib fragmentU . W . 102a-300vertebral fragmentU . W . 102a-306L4 vertebra bodyU . W . 102a-322L2 vertebra bodyU . W . 102a-337vertebral fragmentneural archU . W . 102a-348right pubic ramus fragmentU . W . 102a-349vertebral fragmentneural archU . W . 102a-358rib fragmentsU . W . 102a-360vertebral fragmentU . W . 102a-455ulna shaft fragmentU . W . 102a-456ulna shaft fragmentU . W . 102a-470rib fragmentsU . W . 102a-471right distal radius fragmentU . W . 102a-474long bone fragmentimmatureU . W . 102a-476right capitateU . W . 102a-477partial right lunateU . W . 102a-479rib fragmentLOCALITY 102bU . W . 102b-178LI2U . W . 102b-437rdm2U . W . 102b-438right mandibular corpus fragmentimmature , RP4 in cryptU . W . 102b-502cranial fragmentsU . W . 102b-503RP4 crownU . W . 102b-506cranial fragmentU . W . 102b-507cranial fragmentU . W . 102b-509cranial fragmentU . W . 102b-511LC1 crownU . W . 102b-514cranial fragmentU . W . 102b-515LI2U . W . 102b-516cranial fragmentLOCALITY 102cU . W . 102 c-589left mandibular fragmentLM1 and LM2 in place Following the University of the Witwatersrand’s fossil-numbering system ( Zipfel and Berger , 2009 ) , this second H . naledi locality has been designated U . W . 102 . The chamber itself has been named the Lesedi Chamber , a word meaning ‘light’ in Setswana . By contrast , the Dinaledi Chamber was numbered site U . W . 101 . Excavations in the Lesedi Chamber have been carried out in three areas , designated U . W . 102a , U . W . 102b , and U . W . 102c . The Lesedi Chamber is in the central sector of the Rising Star system ( Figure 2 ) , at a depth of ~30 m from the surface directly above the chamber . All measurements reported here are approximate . The first fossil deposit to be recognized ( U . W . 102a ) is located just off the southwest corner of the North-South Fracture Passage , a northern arm of the Lesedi Chamber . This fossil deposit is approximately 60 m NNE in a straight line from the Dinaledi Chamber . There is no straight-line route between the Dinaledi and Lesedi Chambers , and the shortest traversable route between the two areas is almost 145 m . There are currently four access routes from the surface to the Lesedi Chamber . The most accessible of these currently follows an 86 m downward-sloping path with several narrow passages and short climbs , but only one squeeze and no significant crawls . This has been the main access route for excavators . The other three routes are each substantially more challenging . 10 . 7554/eLife . 24232 . 005Figure 2 . Location of the Lesedi Chamber ( U . W . 102 ) in the Rising Star system ( red circle ) . The Dinaledi Chamber ( U . W . 101 ) is marked by a yellow circle , while three surface entrances into the system are marked by blue circles . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 005 In addition to the first fossil deposit to be recognized in the chamber , two additional concentrations of skeletal material have been identified to date ( Figure 3 ) , and we have designated these as areas 102a , 102b , and 102c . We began investigating each of these areas because team members noticed hominin fossil material exposed on sediment surfaces . The discovery of 102a by Rick Hunter and Steven Tucker led to the initial scientific investigation of the chamber; discoveries of both 102b and 102c were made by Hannah Hilbert-Wolf during the course of geological sampling of the chamber . These three areas do not represent a systematic sampling of the chamber’s contents and we have excavated only a very small sediment volume , less than 200 L ( <0 . 2 m3 ) in total from all three areas . The chamber contains a much greater volume of sediment and we do not know what density of fossil bone it may contain beyond our samples . 10 . 7554/eLife . 24232 . 006Figure 3 . Schematic of the Lesedi Chamber , showing the three hominin-bearing collection areas: U . W . 102a , 102b , and 102c . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 006 U . W . 102a is located at the entrance of a 20–50-cm-wide blind tunnel , which is 1 . 8 m long in total . The blind tunnel leads off of the southwest corner of the North-South Fracture Passage ( Figure 3 ) . Fossil material was exposed on the surface within this blind tunnel at the time of discovery . We have excavated the proximal 1 . 5 m of this blind tunnel , which has a tapering width of less than 50 cm in our excavation unit . The depth of excavation in this area is a maximum of 40 cm . The deposit in this area is a weakly stratified , unlithified mud-clast breccia . Most hominin material has been recovered from an approximately 10-cm-thick horizon of fine-grained mud-clast breccia , beneath a surface layer of ~2 cm of lighter brown-colored mudstone . This deposit is the source of at least some of the sediments that slope from the blind tunnel into the Antechamber . Fossil material attributed to 102a has also been recovered from the surface within the North-South Fracture Passage . U . W . 102b is a sediment deposit on a horizontal chert shelf 80 cm above the cave floor along the western wall of the Antechamber . It is also dominated by unlithified mud-clast breccia . The 102b deposit is located ~3 . 8 m to the south and 1 . 8 m below the 102a deposit . After the discovery of hominin fossil material on the surface here , we undertook limited excavations , with a total volume of ~20 L . U . W . 102c is a small unlithified sediment deposit within an irregular dissolution cavity on the north wall of the east–west-running Cake-Icing Fracture . This deposit is 1 . 3 m above the current cave floor . It is 11 . 6 m from U . W . 102a , and 0 . 3 m below the level of the 102a fossils . We have excavated this small sediment pocket in its entirely , with a total volume of approximately 2 L . Geological work to characterize the Lesedi Chamber depositional history is underway . The stratigraphy is complex , with some hominin and faunal material concentrated in deposits of poorly consolidated mud-clast breccia , generally similar to the facies in the Dinaledi Chamber ( Dirks et al . , 2015 ) . Notably , the fossil material in the Lesedi Chamber is concentrated in minor side fractures , dissolution cavities , or on chert shelves well above the current chamber floor . Our working hypothesis is that the chamber once held a greater volume of sediment than is present today , and when sediment eroded from the chamber , erosional remnants remained in protected fractures , wall cavities , and on chert shelves along the chamber walls . This and other indications of reworking of the deposits make it uncertain how much of the hominin assemblage may remain in its primary depositional context . Hominin material from the 102a area includes 118 identifiable specimens ( Table 1; Figure 4 ) . Fifty-seven of these are cranial and dental specimens that either refit directly or are morphologically compatible with a nearly complete fossil cranium , designated LES1 ( Figure 5 ) . Hominin postcranial remains from locality 102a include 61 identified specimens that represent a minimum of 31 postcranial elements , not counting ribs . These include a minimum of two partial femora , two partial humeri , one complete clavicle and two clavicular fragments , two partial ulnae , several fragments of scapula and radius , many rib fragments , a near-complete first rib , a partial sternum , four hand and wrist elements , an immature ilium and sacrum fragment , and a partial thoracic and lumbar vertebral column . Every anatomical region of the skeleton is represented with the notable exceptions of tibia , fibula and pedal remains . 10 . 7554/eLife . 24232 . 007Figure 4 . Skeletal material from locality 102a provisionally assigned to the LES1 skeleton . The adult cranial material from 102a all belongs to a single cranium; most of the adult postcranial material probably belongs to the same individual . The adult cranial and postcranial material is shown here , except for the U . W . 102a-001 femur . The possibility that the femora represent two adult individuals makes it unclear which femur may be attributable to the skeleton; for the purposes of illustration , the U . W . 102a-003/U . W . 102a-004 femur is included in this photograph . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 00710 . 7554/eLife . 24232 . 008Figure 5 . LES1 cranium . Clockwise from upper left: three-quarter , frontal , superior and left lateral views . Fragments of the right temporal , the parietal and the occipital have also been recovered ( not pictured ) , but without conjoins to the reconstructed vault or face . Scale bar = 5 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 008 The LES1 cranium is fragmented but is represented by most of the vault and part of the face ( Figure 5 ) . To date , we have successfully refitted the near-complete mandible , the near-complete right maxilla , a partial palate and a partial left maxillary dental row , and a partial vault including the near-complete frontal , left and right nasal and left lacrimal bones , near-complete left parietal and temporal , partial right parietal , and a portion of left occipital . LES1 has a complete adult dentition except for the crowns of the lower left central and lateral incisors . The face is reconstructed from the partial right maxillary bone , including the frontal process , which refits to the right nasal bone and frontal . The left mandibular ramus is well-enough preserved to allow a rough estimation of the condyle position , enabling an approximation of the midsagittal contour of the face ( Figure 5 ) . All additional cranial fragments in the present 102a collection are non-duplicative with this refitted vault and face , and where they represent the opposite side of the vault , they match in morphological detail . However , many of the fragments lack clear refits with the existing vault or maxillary portions . Further physical reconstruction of the cranium will await fragments that may emerge from excavation in the future . The refitted vault , with the application of virtual mirror reconstruction , is sufficient to allow an estimate of endocranial volume of approximately 610 ml ( Figure 6 ) . 10 . 7554/eLife . 24232 . 009Figure 6 . Digital reconstruction of endocranial volume in LES1 . The refitted calvaria was mirrored and filled , resulting in a volume estimate of 610 ml . Scale sphere = 10 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 009 Most of the features of the LES1 vault are characteristic of H . naledi from the Dinaledi Chamber ( Supplementary file 1; Figure 7 ) . The LES1 vault is relatively short anteroposteriorly , without the elongation and sharp occipital angulation found in H . erectus . LES1 exhibits mild frontal and parietal bossing , similar to H . naledi DH3 . Other features on the vault that are consistent with H . naledi include prelambdoidal flattening , limited postorbital constriction , widely spaced temporal lines , a continuous supraorbital torus with a supratoral sulcus , an occipital torus , and a marked angular torus . In the temporal region , LES1 has an anteroinferiorly oriented root of the zygomatic process of the temporal , a medially positioned mandibular fossa , a small and obliquely oriented external auditory meatus , a projecting Eustachian process , a small vaginal process , a weak crista petrosa , a triangular-shaped mastoid process , and a small suprameatal spine . Each of these traits characterizes the Dinaledi H . naledi sample ( Berger et al . , 2015; Laird et al . , 2017 ) . Some of these traits occur individually in other species , including H . erectus , H . habilis , H . rudolfensis , and Australopithecus sediba , but they have never been found in combination except in H . naledi ( Figure 7 ) . 10 . 7554/eLife . 24232 . 010Figure 7 . Frontal and vault morphology in H . naledi compared to that in other hominin species . Several of the crania pictured here are similar to H . naledi in endocranial volume , including Sts 5 , MH1 , KNM-ER 1813 , and D2282 , representing four different species . However , these skulls contrast strongly in other features . H . erectus is highly variable in size , as illustrated here by D2282 from Dmanisi , Georgia , one of the smallest and earliest H . erectus crania , and the L2 cranium from Zhoukoudian , China , one of the largest and latest H . erectus specimens . The relatively early KNM-ER 3733 has a size and endocranial volume close to the mean for H . erectus . Cranial remains that are attributed to H . erectus share a combination of anatomical features despite their diversity in size . Many such features of H . erectus are also shared with H . naledi , H . habilis , or Au . sediba , and notably , the differences in the frontal and vault between KNM-ER 1813 ( H . habilis ) and KNM-ER 1470 ( H . rudolfensis ) are mostly features that the smaller KNM-ER 1813 shares with H . naledi , H . erectus , and Au . sediba . The H . naledi skulls share some aspects of frontal morphology with Au . sediba , H . habilis and H . erectus that are not found in Au . africanus or H . rudolfensis , including frontal bossing and a supratoral sulcus . Two additional traits of the H . naledi anterior vault are shared with Au . sediba and H . erectus:slight postorbital construction and a posterior position of the temporal crest on the supraorbital torus . More posteriorly on the vault , H . naledi further shares an angular torus with H . erectus , and some individuals also have sagittal keeling . Both of these traits are also present in some archaic humans . Some H . naledi crania , such as DH3 , are substantially smaller than any H . erectus cranium , and the small size and thin vault bone of even the largest H . naledi skull , LES1 , are outliers compared to H . erectus , matched only by some Dmanisi crania . The facial morphology of H . naledi is more distinct from those of H . erectus and H . habilis . The nasal bones of LES1 do not project markedly anteriorly , although like many specimens of H . erectus , LES1 has a projecting nasal spine . LES1 has a relatively flat lower face , with a transversely concave clivus and incisors that project only slightly past the canines . This morphology is similar but less extreme than that found in KNM-ER 1470 of H . rudolfensis , and is not shared with the other species pictured here . H . naledi has several distinctive features of the temporal bone that are absent from or found in only a few specimens of the other species pictured , including a laterally inflated mastoid process ( comparable to some specimens of Au . afarensis ) , a weak or absent crista petrosa ( comparable to Au . afarensis ) , and a small external auditory meatus ( comparable to KNM-WT 40000 of Kenyanthropus platyops [Leakey et al . , 2001] ) . In this illustration , KNM-ER 1813 , KNM-ER 1470 , KNM-ER 3733 , and ZKD L2 are represented by casts . Because these images are in a nonstandard orientation , scale is approximate . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 010 The maxilla and mandible of LES1 are also consistent with the Dinaledi H . naledi sample ( Supplementary file 1; Figures 8 , 9 , 10 and 11 ) . The maxilla has a mediolaterally flattened subnasal region , a parabolic dental arcade , and an anteriorly shallow palate . The mandible of LES1 has a gracile mandibular corpus , a vertical mandibular symphysis with weak mentum osseum , a steeply inclined lingual alveolar plane , weak inferior and absent superior transverse tori , continuous and deeply excavated anterior and posterior subalveolar fossae , mental foramina positioned above mid-corpus height , well defined ectoangular and endoangular tuberosities , and a root of the ascending ramus that originates at the mesial border of the M2 . Again , many of these traits can be found individually in other hominin species , but in combination , they are uniquely found in H . naledi . 10 . 7554/eLife . 24232 . 011Figure 8 . LES1 mandible compared to the DH1 holotype mandible of H . naledi . In each pair , LES1 is on the left and DH1 on the right . Top left: anterior view . Top right: occlusal view . Bottom left: left lateral view . Bottom right: posterior view . Scale bar = 2 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 01110 . 7554/eLife . 24232 . 012Figure 9 . Comparison of LES1 maxilla to the DH1 holotype maxilla of H . naledi . In each pair , LES1 is on the left and DH1 on the right . Top left: anterior view . Top right: right ( LES1 ) and left ( DH1 ) lateral view . Bottom: occlusal view . Scale bar = 2 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 01210 . 7554/eLife . 24232 . 013Figure 10 . Mandibular and dental anatomy in H . naledi compared to other species of Homo . Right demi-mandibles attributed to H . rudolfensis , H . habilis , H . naledi , H . erectus , and H . sapiens are pictured . All mandibles are aligned using the line marking the distal edge of the first molar . Each of the six horizontal lines corresponds to the edges of teeth in the DH1 mandible , the holotype specimen of H . naledi , with corresponding teeth labeled to the left . Using these lines , it is apparent which specimens have longer premolars and first molars , and which have longer second and third molars compared to DH1 . The dentition of the LES1 mandible has been affected by interproximal wear , resulting in shorter mesiodistal measurements . Mandibular morphology and dental proportions vary slightly among most species of Homo , particularly in comparison with the large differences in dental proportions among species of Australopithecus and Paranthropus . Still , H . naledi is clearly distinguishable from other species of Homo ( Berger et al . , 2015; Laird et al . , 2017 ) . Fossils of H . rudolfensis , H . habilis , and H . erectus differ from H . naledi in the proportions of different parts of the postcanine tooth row and in features of the mandibular corpus . H . erectus . While fossils attributed to H . erectus vary in dental proportions , the early African and Georgian fossil specimens ( here represented by KNM-ER 992 , D211 and D2600 ) have larger first molars than H . naledi , comparable premolar sizes , and highly variable second and third molar sizes . The mandibles attributed to H . erectus mostly have greater corpus height than H . naledi mandibles and are highly diverse in corpus breadth , symphyseal thickness , and robusticity . Many have a strong post-incisive planum , most obvious in D2600 ( shown ) . All three also differ from H . naledi in the crown complexity of their molars and premolar morphology , as illustrated in more detail in Figure 12 . Some specialists would attribute these three mandibles of H . erectus to three different species . H . habilis . The two Olduvai mandibles of H . habilis are themselves quite different from each other in size . Both have similar dental proportions to H . naledi with bigger teeth across the postcanine dentition . Tobias ( 1967 ) viewed O . H . 13 as being similar to H . erectus and described it as an ‘evolved H . habilis’ . Its occlusal morphology and dental proportions do resemble KNM-ER 992 ( Wood , 1991 ) , although the mandibular corpus is thinner and shallower , with a curved base in lateral profile . A strong post-incisive planum is evident in both mandibles . H . rudolfensis . The KNM-ER 1802 and Uraha ( UR 501 ) mandibles have often been attributed to H . rudolfensis , although both attributions may be doubtful ( Leakey et al . , 2012 ) . However , both lack any special similarities with contemporary australopiths and represent a megadont early Homo morphology with corpus size and robusticity much greater than those of H . naledi . Au . sediba . Molar sizes in the MH2 mandible are around 1 mm larger than the average for H . naledi , but the proportions are very similar to those of H . naledi , and like H . naledi , MH2 has a weak post-incisive planum and a small symphysis area . H . sapiens . The modern human mandible shown here , from a recent South African individual , has similar first molar size to the H . naledi mandibles , but much smaller premolars and second and third molars . The crown complexity in this individual , which is not unusual for African population samples , is substantially greater than evidenced in H . naledi . The mandibular corpus is smaller and much less robust than H . naledi . KNM-ER 1802 , UR 501 , O . H . 13 , O . H . 7 , and KNM-ER 992 are illustrated here with casts; the remainder are original specimens . The left side of O . H . 7 is shown here mirrored . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 01310 . 7554/eLife . 24232 . 014Figure 11 . Comparison of H . naledi mandibles to other hominin species , from lateral view . The DH1 holotype mandible and the LES1 mandible of H . naledi have a moderately deep mandibular corpus compared to other species of Homo; the LES1 mandible has a slightly greater corpus height anteriorly ( at P3 ) than posteriorly ( at M2 ) . LES1 has rather a high coronoid process; the height of the condyle was probably lower than this . The mental foramen is at the midpoint or slightly higher in both H . naledi mandibles , and in both , the symphysis is nearly vertical . These features vary substantially within Homo . Modern humans ( bottom ) typically have a chin , but otherwise vary substantially in corpus height , whether the base of the corpus is parallel with the alveolar portion or with the occlusal surfaces of the teeth . Here that variability is illustrated with two modern human mandibles of male individuals , one from island Melanesia ( left ) , and one from southern Africa ( right ) . H . erectus exhibits very extensive variation in corpus height and thickness . D2600 ( shown ) is extremely thick and robust , but is not an outlier; other H . erectus mandibles approach or equal its corpus dimensions . The position of the mental foramen also varies , as does the relative anterior versus posterior corpus height and the symphyseal profile , from more sloping to near vertical ( as illustrated by KNM-ER 992 , although this specimen is damaged at the symphysis ) . MH2 ( Au . sediba ) has comparable corpus height and robusticity to the H . naledi mandibles , with a more sloping symphysis . O . H . 13 is a more gracile mandible than the H . naledi specimens in many respects . It has a curved base and a sloping symphysis . The more complete left side of LES1 is shown here and mirrored for comparison to other specimens . KNM-ER 992 and O . H . 13 are represented here by casts . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 014 The teeth exhibit moderate occlusal wear on the second and third molars , trending toward near-complete dentine exposure on the occlusal surfaces of first molars and substantial removal of occlusal detail of the anterior dentition . The dental morphology of LES1 is entirely consistent with the Dinaledi sample of H . naledi ( Figures 10 and 12 ) . The mesiodistal and buccolingual ( or labiolingual ) crown dimensions of all the LES1 teeth fall within the range of the Dinaledi dental sample , except for those teeth where interproximal wear has clearly reduced the mesiodistal dimension ( Table 2; Figure 13 ) . The P3 crowns are worn , but they are roughly symmetrical about their mesiodistal axis in occlusal view; they are fully bicuspid and multirooted , with a smaller circular mesiobuccal root and larger , more platelike , distal root . This configuration is repeated throughout the Dinaledi dental assemblage . The shared overall P3 morphology of LES1 and the Dinaledi sample is distinctive in H . naledi and not observed in other species of hominins ( Figure 12; Berger et al . , 2015 ) . The P3 and P4 are both three-rooted , with two ovoid roots present buccally and a larger ovoid root present lingually . The roots are not widely splayed as in some other multi-rooted hominins , and especially for the P4 , the buccal roots are closely packed in buccal view . This root configuration is seen in the H . naledi type specimen , U . W . 101–1277 . The mandibular canine crowns have asymmetrically placed crown shoulders , with the mesial more apically placed than the distal . Further , the distal shoulder is formed by an accessory cuspule . These features are strongly distinctive in H . naledi ( Berger et al . , 2015 ) , with only a few specimens of H . erectus approaching this canine configuration . None of the molars exhibit any evidence of supernumerary cusps , and cingular features , such as the protostylid and Carabelli’s feature , are either absent or weakly developed and are expressed independently of the grooves of the crown . The molar size gradient in the LES1 mandible is M1 < M2 < M3 as in the Dinaledi Chamber sample of H . naledi ( Figure 10 ) . The Dinaledi Chamber includes no maxillary dentition with all three molars in place , but U . W . 101–1269 is a LM3 that exhibits a mesial interproximal facet that matches the distal facet of the LM2 present in the U . W . 101–1277 ( DH1 ) maxilla . If these specimens do represent a single individual , then the maxillary molar gradient for this specimen would be M1 < M3 < M2 , which is also seen in the LES1 maxilla . In total , these dental features are within the known range for H . naledi in every instance and distinguish LES1 clearly from all other hominin species . 10 . 7554/eLife . 24232 . 015Figure 12 . Occlusal view of H . naledi mandibular teeth compared to those of other hominins . Teeth from the canine to the third molar are shown , if present , in the orientation in which they are found within the mandible . All individuals are aligned vertically by the distal margin of the first molar . Mandibles from the Lesedi Chamber , U . W . 102 c-589 and LES1 are shown next to DH1 and U . W . 101–377 ( Berger et al . , 2015 ) . The mandibles illustrated from H . erectus have relatively little occlusal wear , so their morphology can be seen more clearly than that of worn mandibles . The immature U . W . 101–377 ( H . naledi ) is comparable in developmental age and wear to O . H . 7 ( H . habilis ) , as well as to D2735 and KNM-WT 15000 ( H . erectus ) . When compared to H . habilis , H . erectus , and australopiths , H . naledi is notable for its relatively small first molars , its relatively small canines , and its lack of supernumerary cusps and crenulation on the molars . The complexity of molar cusp and groove patterns is especially evident in the chronologically early H . erectus specimens from Africa and Georgia shown here . For example , the unworn M2 of the immature U . W . 101–377 mandible of H . naledi has a relatively simple crown anatomy with very little wrinkling or crenulation . By comparison , the M2 of D2735 , D211 , and KNM-WT 15000 , all with minimal occlusal wear , show extensive crenulation and supernumerary cusps . Canine size and molar crown complexity vary substantially among modern human populations , but the southern African individual illustrated on the right is not atypical for its population , and has greater molar crown complexity and larger canine dimensions than any of the H . naledi mandibular dentitions . The morphology of the third premolar varies extensively among these hominin species and within H . erectus . The H . naledi P3 anatomy can be seen clearly in the immature U . W . 101–377 individual . It is characterized by roughly equally prominent lingual and buccal cusps and an expanded talonid . In H . naledi , this tooth is broadly similar in morphology and size to the P4 . This configuration of the P3 is not present in the other species , with only KNM-WT 15000 exhibiting some expansion of the lingual cusp in what remains an asymmetrical and rounded P3 . A . L . 400–1 , O . H . 7 and KNM-WT 15000 are represented by casts; The left dentition of U . W . 102 c-589 and O . H . 7 have been mirrored to compare to right mandibles . Images have been scaled by measured first molar dimensions . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 01510 . 7554/eLife . 24232 . 016Figure 13 . Metric comparisons of the Lesedi Chamber dental material . H . naledi is clearly differentiated in first molar and canine dimensions from other species with broadly similar cranial and dental morphology , including Au . sediba , H . habilis , H . rudolfensis , and early H . erectus samples from Africa and Georgia . The material from the Lesedi Chamber is within the range of or similar to H . naledi in these dimensions and well differentiated from the other samples . Top left: mandibular first molar dimensions . Top right: maxillary first molar dimensions . Bottom left: mandibular canine dimensions . Bottom right: maxillary canine dimensions . The LES1 first molars and maxillary canines have a substantial degree of interproximal wear , and the values plotted here are not corrected for this wear , which shortened the mesiodistal dimension by as much as a millimeter . The values plotted here should thus be regarded as minimum values . The H . erectus sample here includes specimens from the Lake Turkana area , Konso , Tighenif ( Ternifine ) , Thomas Quarry , and Dmanisi; Asian H . erectus specimens are omitted . Attributions of H . habilis and H . rudolfensis specimens are indicated in the Materials and methods . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 01610 . 7554/eLife . 24232 . 017Table 2 . Dental measurements for Lesedi Chamber specimens . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 017SpecimenMesiodistal diameterBuccolingual ( or labiolingual ) diameterU . W . 102b-437 ldm210 . 78 . 7U . W . 102b-503 RP48 . 410 . 9U . W . 102b-515 LI26 . 86 . 5†U . W . 102b-178 LI25 . 65 . 9U . W . 102b-511 LC16 . 86 . 8†U . W . 102 c-589 LM111 . 410 . 6U . W . 102 c-589 LM213 . 111 . 3LES1 maxillaryRI17 . 66 . 9RI26 . 87 . 0RC17 . 58 . 7RP38 . 110 . 8RP48 . 111 . 3RM110 . 611 . 8RM211 . 712 . 7RM311 . 412 . 7LI17 . 46 . 9LI26 . 16 . 8LC17 . 48 . 7LP38 . 010 . 9LP48 . 111 . 3LM110 . 711 . 9LM212 . 112 . 8LM311 . 413 . 6LES1 mandibularRI15 . 86 . 3RI25 . 46 . 1RC17 . 17 . 7RP38 . 49 . 3RP48 . 29 . 1RM110 . 810 . 6RM212 . 311 . 5RM313 . 311 . 7LC17 . 87 . 5 †LP38 . 49 . 3LP48 . 29 . 1LM111 . 210 . 6LM212 . 311 . 5LM313 . 311 . 7*Denotes measurements where the tooth is extremely worn , and mesiodistal diameter reported here has not been corrected for the degree of wear . †Denotes instances where we report a minimum value for labiolingual measurements because the crown is not complete or is broken . The LES1 cranium does exhibit some traits that differ from comparable examples in the Dinaledi Chamber . The cranium is slightly larger overall , with an estimated endocranial volume of approximately 610 ml , and this larger size is reflected in the external vault measurements . Previously , the largest known H . naledi endocranium was DH1 at approximately 560 ml ( Berger et al . , 2015 ) . LES1 contrasts in morphological features with the small DH3 cranium in ways that have been observed when comparing male individuals with female individuals in other hominin species . The supraorbital torus of LES1 is more pronounced than that in the small DH3 individual . LES1 has a stronger supramastoid/suprameatal crest , a larger mastoid process , and a more marked pterygoid insertion when compared to the U . W . 101–361 mandible . Although LES1 is outside of the endocranial volume range of specimens presently attributed to H . naledi , the larger size and more robust features of LES1 are consistent with the combination of cranial and mandibular characters in H . naledi . U . W . 102a-021 is a nearly complete right clavicle , missing only the articular surface of the sternal end , where trabecular bone is exposed over the entire articular area , including a small bit of the anterior surface ( Figure 14 ) . The shaft is broken into two pieces near the midshaft but the two pieces conjoin cleanly . There is also a small bit of damage to the acromial end . On the posterior surface , the medial part of the crest for the conoid tubercle is broken off . The specimen exhibits a dark surface coating on the anterior aspect of the sternal half and patchy areas of black staining on its acromial half . There are fine hairline longitudinal cracks on much of the acromial half of the bone . 10 . 7554/eLife . 24232 . 018Figure 14 . U . W . 102a-021 right clavicle from the Lesedi Chamber . Left , from top: superior , anterior , inferior , posterior views . Right , from top: medial and lateral views . Scale bar = 2 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 018 U . W . 102a-206 is a ca . 41-mm-long fragment of left clavicular shaft , preserving the midshaft region ( based on anatomical comparisons with U . W . 102a-021 ) . The shaft anteroposterior ( AP ) and superoinferior ( SI ) dimensions are slightly smaller than those of the right side clavicle at this position . The fragment compares favorably to U . W . 102a-021 in overall size , curvature , and shaft morphology . U . W . 102a-239 is the acromial end of a left clavicle , including the lateral 51 . 5 mm , preserving the conoid tubercle ( but not its medial crest ) and the articular surface for the acromion of the scapula . This is slightly larger than the acromial end of U . W . 102a-021 , but otherwise fairly similar in morphology . U . W . 102a-002 is a proximal shaft fragment of a right humerus ( Figure 15 ) . The head and greater tubercle are missing , as is all but the very distal base of the lesser tubercle . From this metaphyseal region , the fragment preserves approximately 50–60% of the shaft , with a total fragment length of 85 mm . U . W . 102a-013 includes two fragments identified as humeral head , each with some articular subchondral bone , which may derive from the same element as U . W . 102a-002 . They appear to be consistent in curvatures of the articular surface with U . W . 102a-257 . The specimen is mostly coated with a brown to dark-brown mineral patina , the surface is unweathered with only slight surface removal on the distal end of the anterior surface . The breaks , both proximal and distal , are sharp . 10 . 7554/eLife . 24232 . 019Figure 15 . U . W . 102a-002 right humerus fragment . From left: posterior , medial , anterior and lateral views . Right , from top: Scale bar = 5 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 019 U . W . 102a-257 is a fragment of left humerus , including the head and proximal shaft , and is largely complete from head to just around midshaft ( Figure 16 ) . There is corrosion to the superior aspect of the proximal articular surface ( which precludes an accurate measurement of the superoinferior diameter of the head ) , and to the articular margin of the superolateral head . The surface of the specimen is otherwise very well preserved with no signs of weathering . A dark-brown to black patina covers much of the posterior surface of the shaft , wrapping around to the anterior surface on the most distal part . The proximal 40 mm or so of the lateral crest of the deltoid tuberosity is present . U . W . 102a-257 is consistent with the morphology and size of U . W . 102a-002 , and they may represent left and right humeri of the same individual . 10 . 7554/eLife . 24232 . 020Figure 16 . U . W . 102a-257 left proximal humerus fragment . From left: posterior , medial , anterior and lateral views . Top: proximal view . Bottom: distal view . Scale bar = 5 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 020 The proximal humerus fragments from the Lesedi Chamber have morphology consistent with the Dinaledi Chamber collection of H . naledi , both in the size of the head and in the shaft diameter . In both assemblages , the bicipital groove appears deep and narrow , and the lesser tubercle is projecting . The most distinctive aspect of the humerus material of H . naledi in comparison with other hominin species is the very low humeral torsion angle in the adult U . W . 101–283 humerus , in which the head faces nearly directly posteriorly ( Feuerriegel et al . , 2017 ) . This aspect cannot be assessed directly in the fragments from the Lesedi Chamber . U . W . 102a-015 is a right proximal ulna , on which much of the trochlear notch and olecranon process are preserved in addition to the proximal half of the diaphysis ( Figure 17 ) . There is erosion to the anterior tips and margins of the coronoid and olecranon process . The surface of the shaft is well-preserved and exhibits very slight hairline longitudinal cracks . The break to the distal end is sharp and cleanly transverse . The olecranon process is mediolaterally narrow and the trochlear notch appears to have opened anterosuperiorly , as in modern humans but not Neandertals . While there is only one fragmentary ( and probably immature ) proximal ulna from the Dinaledi Chamber , U . W . 102a-015 generally compares well with U . W . 101–560 in terms of morphology and gracility . 10 . 7554/eLife . 24232 . 021Figure 17 . U . W . 102a-015 ulna fragment . From left: anterior , medial , posterior and lateral views . Right from top: proximal and distal views . Scale bar = 2 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 021 Four adult hand and wrist elements have been recovered from the 102a locality ( Figure 18 ) . U . W . 102a-028 is a right fourth metacarpal ( RMc4 ) , with a small base and a relatively radioulnarly broad head ( Figure 15 ) . The metacarpal shaft shows substantial curvature and is relatively robust for its length , although it still falls within the upper range of variation found in modern humans ( Figure 16 ) . U . W . 102a-117 is a complete right scaphoid; U . W . 102a-476 is a complete right capitate; and U . W . 102a-477 is a partial right lunate . The scaphoid , lunate , and capitate are consistent in size and appear to match each other when placed in anatomical articulation; the RMc4 is likewise a good match in size , with a lateral base matching in dorsopalmar contour the base of the capitate . Thus , these four bones may represent the right hand of one individual . 10 . 7554/eLife . 24232 . 022Figure 18 . U . W . 102a-028 right fourth metacarpal . From left: dorsal , ulnar , palmar and radial views . Right from top: distal and proximal views . Scale bar = 1 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 022 These four bones are qualitatively similar in overall shape to that described for H . naledi , but they are absolutely larger in most of their overall dimensions ( Kivell et al . , 2015 ) . The lunate is missing a large portion of its articular surface for the radius and adjacent areas , precluding quantitative comparisons of its morphology . A canonical variates analysis of scaphoid and capitate comparative morphology in African apes and hominins demonstrates that the Dinaledi and Lesedi scaphoids and capitates fall together within a distinct space relative to other fossil and extant hominids . Along the first canonical axis , Dinaledi and Lesedi wrist remains cluster with modern humans and Neandertals because they all share derived features relative to those of African apes ( Figure 19; Supplementary file 3 ) . For instance , the scaphoid’s trapezium facet extends further onto the tubercle , and together , the trapezium and trapezoid facets are relatively large , as in modern humans and Neandertals ( Supplementary file 3 ) . The Dinaledi and Lesedi scaphoid and capitate morphology are distinguished from those of modern humans and Neandertals on the second canonical axis because the Mc2 facet orientation in H . naledi is roughly intermediate between that of modern humans and Neandertals on the one hand and that of African apes on the other . In this respect , the H . naledi capitates are more similar to those of H . floresiensis and several australopiths . 10 . 7554/eLife . 24232 . 023Figure 19 . Quantitative comparisons of hand and wrist material from the Lesedi Chamber . Left: ratios of fourth metacarpal dimensions in H . naledi compared to those in other hominin and great ape samples . Right: canonical variates analysis of capitate and scaphoid morphology in humans , chimpanzees , gorillas , and fossil hominins . H . naledi from the Dinaledi Chamber occupies a unique position in scaphoid and capitate joint configurations , which is closely matched by the capitate and scaphoid from the Lesedi Chamber . In this analysis , no a priori groups are assumed; we also examined the scenario in which Homo naledi and other fossil specimens are included as a priori groups and the results are essentially identical . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 023 Seven vertebrae have been recovered in the 102a assemblage , all from the lower thoracic and lumbar region of the spine . These vertebrae are roughly equivalent in preservation . The thin cortical bone of the vertebral bodies is eroded in large patches on these elements with exposure of underlying trabeculae . They have minimal surface staining or patination , and where the vertebral arches are present , the cortical surface is well-preserved . U . W . 102a-036 is a largely complete antepenultimate thoracic vertebra , inferred as T10 , with limited erosion to the anterior surface of the vertebral body and some damage distally on the transverse processes , particularly on the right side , and to the posterior end of the spinous process ( Figure 20 ) . The vertebral body is ovoid to kidney-shaped and the ring apophyses are relatively thick , covering approximately three-quarters of the vertebral body surface . The spinal canal is ovoid in shape and about one-third the size of the vertebral body . Facets for the tenth rib are posteriorly positioned , almost entirely on the pedicles . The pedicles themselves are transversely thick , as are the transverse processes , which are strongly posteriorly oriented . Together , the transverse processes and pedicles form nearly continuous , robust lateral structures for anchoring epaxial muscles and for transmitting forces to and from the vertebral body , respectively . 10 . 7554/eLife . 24232 . 024Figure 20 . U . W . 102a-036 vertebra , T10 . Clockwise from top left: posterior , superior , inferior , left , right , and anterior views . Scale bar = 2 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 024 U . W . 102a-151 is a nearly complete penultimate thoracic vertebra , inferred as T11 , with some erosion and loss to the left side of the vertebral body and missing the left-side transverse process ( Figure 21 ) . Portions of the superior vertebral body surface are eroded away , revealing trabeculae . The superior articular facets are planiform and posteriorly oriented . The inferior articular facets are asymmetrical – the right side is curved and anterolaterally oriented , whereas the left side is planiform and oriented anteriorly on the coronal plane , as in the transitional vertebra . Costal facets are large , extending from the posterior aspect of the body inferiorly and posteriorly onto the pedicle . The vertebral body is kidney- to heart-shaped and the spinal canal is ovoid , with a slightly triangular shape . The spinous process is relatively long and relatively horizontal in its orientation , with its major axis deflecting inferiorly at an angle of approximately 20° from the surface of the superior vertebral body . 10 . 7554/eLife . 24232 . 025Figure 21 . U . W . 102a-151 vertebra , T11 . Clockwise from top left: posterior , superior , inferior , left , right , and anterior views . Scale bar = 2 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 025 U . W . 102a-154a is a nearly complete last thoracic vertebra , inferred as T12 . The right inferior articular facet , distal spinous process , and anterior aspect of the inferior vertebral body are broken away . The anterior portion of the body is eroded on the right side , as are the lateral aspects of the superior vertebral body . The superior articular facets are asymmetrical , matching the inferior articular facets of the superjacent vertebra ( U . W . 102a-151 ) : the left superior articular facet is planiform and posteriorly oriented on the coronal plane , whereas the right superior articular facet is curved and posterolaterally oriented . The right superior articular facet is comparatively diminutive in size , particularly in transverse dimension . The vertebral body is kidney- to heart-shaped and transversely wide . The costal facets are positioned at the body-pedicle border but are eroded on both sides; thus , their morphology cannot be fully appreciated . The pedicles themselves are anteroposteriorly short and contribute to a wide , ovoid spinal canal . U . W . 102a-154b , U . W . 102a-322 , and U . W . 102a-306 are vertebral bodies associated with little or no vertebral arch structures . U . W . 102a-139 is a lumbar vertebra preserving most aspects of the vertebral body and neural arch , but it is broken into five pieces that refit reasonably well , although the spinous process is missing . None of the bodies or preserved aspects of pedicles bear costal facets . U . W . 102a-154b nicely articulates with U . W . 102a-154a superiorly and U . W . 102a-322 inferiorly , and U . W . 102a-306 and U . W . 102a-139 articulate with each other; however , U . W . 102a-322 and U . W . 102a-306 do not articulate . The lumbar transverse processes of U . W . 102a-139 are anteroposteriorly wide , emerging anteriorly from the posterior aspect of the vertebral body along the pedicles and posteriorly to the bases of the superior articular processes . Its body is clearly posteriorly wedged in lateral view . Together , these features indicate that U . W . 102a-139 is the last lumbar vertebra . Therefore , the likely seriation is as follows: U . W . 102a-154b is L1 , U . W . 102a-322 is L2 , U . W . 102a-306 is L4 , U . W . 102a-139 is L5 , and L3 is missing . U . W . 102a-250 is a nearly complete right first rib , with erosion and breakage to the head , tubercle , lateral border and distal end . The neck is flattened in its superior-inferior dimension and descends in the vertebro-inferior direction . The tubercle and the posterior angle coincide . The facet of the articular tubercle was damaged post-mortem . Two partial first ribs ( U . W . 101–083 and U . W . 101–621 ) of H . naledi are preserved in the Dinaledi Chamber hominin sample , but neither rib has its head nor enough of the shaft preserved to allow accurate estimation of curvature ( Williams et al . , 2017 ) . The angulation and shape of these fragments appears comparable to those of MH2 Au . sediba and A . L . 288–1 Au . afarensis . U . W . 102a-250 is more complete than the Dinaledi first rib fragments , and is similar in morphology in the overlapping regions . This rib is slightly more curved than the Sterkfontein first rib , StW 670 ( Tawan et al . , 2016 ) . The anatomy of U . W . 102a-250 is entirely compatible with attribution to H . naledi , although the bone is also similar in morphology and size to known australopith first ribs . Thirteen additional specimens from 102a are partial ribs or rib fragments , none are identifiable to element and none present anatomical information that is useful for testing the taxonomic affiliation of the sample . U . W . 102a-138 ( Figure 23 ) is a fragmentary right ilium of an immature individual ( as evident by the presence of triradiate cartilage , by an unfused apophysis at the anterior inferior iliac spine , and by very small overall size ) . The fragment is very light , with thin cortical bone , and is eroded around margins of the acetabular portion . The iliac blade is mostly missing , but the auricular surface , greater sciatic notch , acetabulosacral buttress , and anterior margin of the iliac blade are present . Despite the thin and fragile nature of this element , the surface is well-preserved . 10 . 7554/eLife . 24232 . 027Figure 23 . U . W . 102a-138 immature right os coxa fragment . The medial view is at the center . Clockwise from top: superior , lateral , inferior and anterior views . The unfused triradiate suture is notable . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 027 The adult pelvic material of H . naledi from the Dinaledi Chamber is notable in combining an Au . afarensis-like degree of iliac flare , a weak and anteriorly placed iliac pillar , and a narrow tuberoacetabular sulcus on the ischium ( Berger et al . , 2015; VanSickle et al . , personal communication ) . U . W . 102a-138 represents the most complete immature ilium fragment of H . naledi found to date , and its morphology is comparable to that of the juvenile U . W . 101–486 ilium fragment , and thus consistent with the morphology seen in H . naledi . It lacks the diagnostic characters that could differentiate it clearly from ilium fragments from other hominin species , as the iliac blade and iliac pillar are both poorly preserved . The lack of an accompanying ischial fragment precludes an evaluation of tuberoacetabular sulcus morphology in the 102a material . U . W . 102a-001 is a proximal right femur , in which much of the head and neck , and the proximal subtrochanteric shaft are preserved ( Figure 24 ) . The head is badly eroded , especially anteriorly , and only a few small patches of subchondral articular bone are preserved on the posterior aspect . The posterior side of the neck is fairly well preserved from the head all the way to the lesser trochanter , which is planed off , with only the base remaining . The anterior side of the neck is missing . Trabecular bone is exposed from the anterior head all the way to the lateral surface at the base of the greater trochanter . The greater trochanter is missing entirely , save for a small bit of its distal lateral surface . The surface overall is marred by areas of post-depositional damage , including a number of transverse scratches on the shaft . 10 . 7554/eLife . 24232 . 028Figure 24 . U . W . 102a-001 proximal femoral fragment . From left: posterior , medial , anterior and lateral views . Right from top: proximal and distal views . Scale bar = 5 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 028 U . W . 102a-003 is a left femoral shaft fragment , from the lesser trochanter proximally to about midshaft ( Figure 25 ) . Only the base of the lesser trochanter remains . The head and neck are not present . 10 . 7554/eLife . 24232 . 029Figure 25 . U . W . 102a-003 left proximal femur fragment . From left: posterior , medial , anterior and lateral views . Right from top: proximal and distal views . Scale bar = 5 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 029 U . W . 102a-004 is a fragment of left distal femur , preserved from roughly midshaft to the distal subchondral bone surface of the intercondylar notch ( Figure 26 ) . Both condyles are missing . The shaft has surficial markings similar to those present on U . W . 102a-001 . This fragment is morphologically compatible with U . W . 102a-003 in shaft diameter and cross-section , and the two specimens exhibit no morphological overlap , suggesting that they may represent the same femur . 10 . 7554/eLife . 24232 . 030Figure 26 . U . W . 102a-004 left distal femur fragment . From left: posterior , medial , anterior and lateral views . Right from top: proximal and distal views . Scale bar = 5 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 030 U . W . 102a-003 and U . W . 102a-004 may conjoin with each other . Both fragments are morphologically compatible in shaft diameter and cross-section , and at the broken distal end of U . W . 102a-003 and at the proximal end of U . W . 102a-004 , a small part of the circumference of the shaft ( approximately 10 mm in total ) on the posterolateral side appears to provide a refit . However , the edges of this apparent break are abraded , reducing the certainty of the association . Joining the bones at this point , U . W . 102a-003 and U . W . 102a-004 preserve 321 mm of a femoral shaft . Using the similarly sized U . W . 101–002 to represent the missing proximal end and KNM-ER 1481 to represent the distal end ( Figure 27 ) , we preliminarily estimate the femoral length of this individual to be ~375 mm . 10 . 7554/eLife . 24232 . 031Figure 27 . Length estimation of femur based on U . W . 102a-003 and U . W . 102a-004 . Two specimens were used to estimate the missing proximal and distal ends of the femur . Top: U . W . 101–215 is a distal femur fragment that presents a similar morphology to the U . W . 102a-004 distal femur , while preserving the distal articular surface . Middle: U . W . 102a-004 and U . W . 102a-003 conjoined , in posterior view . Bottom: U . W . 102a-001 is comparable in size with U . W . 102a-003 , and while the morphology of the muscle markings is different , the alignment of the lesser trochanters gives a good basis for estimating the proximal extent of the bone . The length estimate is 375 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 031 U . W . 102a-001 ( right proximal ) and U . W . 102a-003 ( left proximal ) are similar in size . They preserve an overlapping area of anatomy from just above the lesser trochanter down to around the midshaft area . However , despite their similarity in size , the two contrast in several anatomical details . U . W . 102a-001 is more platymeric in the subtrochanteric area , with a greater mediolateral ( ML ) breadth than U . W . 102a-003 . The lesser trochanter is abraded in both specimens , but the morphology of the inferior and medial aspects of it appear different in the two bones . U . W . 102a-001 has a shallow sloping border to the lesser trochanter medially , and the inferior aspect tails off into a broad , less marked line leading to a very slight linea aspera by midshaft . By contrast , U . W . 102a-003 has a steep medial aspect to the lesser trochanter , and it tails into a sharply defined crest that broadens around 15 mm down the shaft into a rugose , double crest , which narrows by midshaft into a strong linea aspera . The insertion for m . gluteus maximus is prominent and rugose in both femora but in U . W . 102a-001 , the rugosity extends further down the shaft . Overall , the asymmetry of these two bones would be very unusual in the left and right femora of a single individual . We accept this provisionally as evidence for a second adult individual in the 102a assemblage . The hominin material from 102a appears to represent a minimum of two adult individuals and one immature individual . The inference of two adults is based upon the morphological incongruence of the left ( U . W . 102a-003 ) and right ( U . W . 102a-001 ) femoral elements ( discussed above ) . Still , no adult element is clearly duplicated in the collection . The U . W . 102a-138 ilium , along with an immature sacrum fragment and two immature long bone fragments not described here , demonstrates the presence of at least one immature individual . The lack of duplication of elements suggests that much ( but not all ) of the adult material may represent a single individual skeleton , which parsimoniously would also include the LES1 cranium . We accept this hypothesis provisionally . All elements in the current 102a collection were recovered from within an excavation area less than 50 cm x 70 cm , and 40 cm deep . Two vertebrae ( U . W . 102a-154a and U . W . 102a-154b ) were in articulation in situ . The articular morphology and sizes of seven vertebrae suggest strongly that they represent a single individual; the remainder of these elements were recovered in close physical proximity but not in articulation . All fragments attributed to the LES1 cranium were likewise recovered from a small area . The hand and wrist material is consistent with a single right hand on the basis of articular morphology . None of the other elements lend themselves to an evaluation of articular compatibility , but they are consistent in size . We consider it unlikely that the number of elements in the current collection would be recovered from a commingled assemblage consisting of substantial parts of multiple skeletons without also introducing duplicate elements . This raises the problem of what seems to be a mismatch of the U . W . 102a-001 and U . W . 102a-003 femora . These two are similar in size , and the difference in their shaft dimensions is not greater than that found in 95% of a large sample of paired left and right human femora . However , they are different in muscle attachments and diaphyseal morphology , to the extent that would represent unusual asymmetry in a single individual . The data do not allow us to discard the hypothesis that one of the femora represents a second adult individual in 102a , albeit an individual of similar body size . The descriptive and measurement data do not indicate which ( if either ) of these femora may belong to the individual represented by LES1 , nor which ( if any ) of the other postcranial elements are associated with either femur . On the basis of the non-duplication of elements , it seems likely that if there is a second individual , this second individual is represented by only a small number of elements , possibly just the femur . The two humerus specimens , U . W . 102a-002 and U . W . 102a-257 , differ slightly in shaft diameter where it can be compared , but no other morphological differences are apparent on the preserved fragments , and a slight degree of upper limb asymmetry is not unusual in humans or in fossil Homo . We conclude that most of the 102a adult material probably represents a single skeleton , but we cannot assume that the ratios of either femur with other elements in the sample would reflect the proportions of a single individual . To be conservative , any consideration of proportions should allow for the possibility that multiple individuals are present . The material from area 102b includes 12 specimens identified as hominin . Most of this collection consists of small fragments of cranium , many of which are identifiable to element , but which preserve no diagnostic morphology that would assist in taxonomic assessment . One partial mandible and five possibly associated teeth preserve morphological characters that are useful for taxonomic attribution . U . W . 102b-438 is a fragment of right mandibular corpus from an immature individual ( Figure 29 ) . The alveoli for the RM1 are present; the alveoli for the rdm2 are also present and reveal the crypt with RP4 crown intact within . The fragment is broken anteriorly at the crypt for RP3 and posteriorly at the crypt for RM2 , neither permanent tooth is present . The preserved corpus height at the midpoint of rdm2 is 14 . 9 mm; this is perhaps 1–2 mm less than the true value because of the erosion of the alveolar bone surface . Corpus breadth at the anterior edge of M1 is 15 . 2 mm; total length of the fragment as preserved is 43 . 5 mm . 10 . 7554/eLife . 24232 . 033Figure 29 . U . W . 102b-438 immature mandibular fragment . From left: basal , lingual , occlusal , buccal and anterior views . The RP4 is within its crypt . Scale bar = 2 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 033 Several teeth excavated from 102b are compatible with the same approximate developmental stage as the U . W . 102b-438 mandibular fragment , and were recovered in close proximity to each other and to the mandible; no elements are duplicated . We hypothesize that these fragments represent the same individual , at least until the recovery of additional material makes us reassess this possible association . U . W . 102b-437 is the complete crown and two nearly complete roots of a ldm2 with moderate occlusal wear , including dentine exposure at the centra of cusps , and roots that appear to have begun to resorb at the tips . U . W . 102b-503 is a RP4 crown , nearly complete but not erupted . U . W . 102b-511 is a LC1 crown , nearly complete with no occlusal wear . U . W . 102b-515 is a LI2 that is nearly crown complete and unerupted . U . W . 102b-178 is a broken but apparently unworn probable RI2 crown that was recovered separately from these other teeth , which may also represent this individual . The teeth all fall within the size range of equivalent elements from the Dinaledi Chamber sample . The U . W . 102b-438 RP4 crown is partially obscured within its crypt and not complete , to the minimal extent it is visible at this time , it is consistent with H . naledi mandibular P4 morphology . The U . W . 102b-503 maxillary P4 is likewise consistent with comparable Dinaledi examples , although this tooth differs little among several species of Homo , including modern humans . Like the incisors of LES1 , U . W . 102b-515 lacks prominent cervico-incisal crown curvature , has no prominent crests on its lingual surface , and is not shoveled . The U . W . 102b-437 ldm2 is buccolingually narrow and mesiodistally long , with an ovo-rectangular shape . Five well-developed cusps are present in a Y-fissure pattern , and the talonid is wider than the trigonid . Compared to the two lower dm2 specimens from the Dinaledi chamber , the metaconid and protoconid are relatively small , although it is not clear whether or not this is an artifact of mesial wear . Although the mesial aspect is worn , it is clear that the mesial marginal ridge was thick; it forms the distal border of a fissure-like anterior fovea . A thick distal marginal ridge is also present and it borders a fissure-like posterior fovea . Both the mesio- and disto-buccal grooves are deep and are associated with a wide V-shaped furrow . No protostylid is present . The mesial and distal roots each have a buccal and lingual canal connected by a dentin plate . The crown morphology is nearly identical to that of the two analogous teeth from the Dinaledi chamber ( U . W . 101–655 and U . W . 101–1686 ) , and supports their taxonomic designation to H . naledi . Although few have been found , the lower dm2s of H . erectus s . l . are relatively longer and more rectangular than those of H . naledi . In addition , the substantially wider talonid compared to the trigonid of both the 102 tooth and the H . naledi lower dm2s differentiates these dm2s from those of most other Homo . The most important of the 102b teeth for morphological assessment is the LC1 . The U . W . 102b-511 crown preserves its occlusal morphology and replicates with better detail the same morphology observed in the LES1 mandibular canines ( Figure 30 ) . It has asymmetrically placed crown shoulders , with the mesial shoulder more apically placed than the distal . Further , the distal shoulder is formed by an accessory cuspule and the mesial crest is shorter and more convex than the vertically disposed distal crest . These features are identical to those found in the Dinaledi permanent mandibular canines and are among the defining features of H . naledi . The asymmetrical shoulders of the canine crown and distal accessory cuspule are found to some degree in a number of specimens attributed to H . erectus , H . rudolfensis , H . habilis and Australopithecus , but the small canine and incisor sizes are inconsistent with attribution to any of these other taxa . On the basis of these observations , the immature dentition from 102b is identifiable as H . naledi . 10 . 7554/eLife . 24232 . 034Figure 30 . U . W . 102b-511 left mandibular canine crown from locality 102b . Top row , from left: occlusal , lingual , distal , labial and mesial views . Bottom row: U . W . 102b-511 ( left ) , compared to U . W . 101–1126 ( middle ) and U . W . 101–985 ( right ) relatively unworn left mandibular canines from the Dinaledi Chamber . All three teeth share a distinctive morphology , which is also present in the other Dinaledi mandibular canines , that includes an asymmetrical crown , higher mesial shoulder , and distal accessory cuspule . Scale bar = 1 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 034 All of the cranial fragments recovered from 102b are compatible with an immature developmental stage , and may come from the same individual as U . W . 102b-438 . The minimum number of individuals ( MNI ) in 102b is therefore one individual at present . Without more information about the depositional history of the 102b locality in comparison with that of 102a , we cannot say for sure whether any of the specimens collected at this locality may represent the same immature individual that is represented by the pelvic fragments in 102a . The evidence does not currently exclude the hypothesis that the U . W . 102a-138 ilium and other immature pelvic fragments may represent the same individual as the U . W . 102b-438 mandible . The 102a and 102b areas are separated by approximately 3 m , and there is a possibility that slumping of material might have brought the remains of a single individual into both areas . Further work to establish the life-history stage of the immature remains may help to resolve this question . The 102c deposit is a very small ( ~2 L ) volume of sediment enclosed within a dissolution cavity in the cave wall . Only one morphologically identifiable hominin specimen has been recovered from this area . U . W . 102c-589 is a fragment of left mandibular corpus with worn LM1 and LM2 in situ ( Figure 31 ) . These teeth have less occlusal wear than those of the LES1 mandible , but are morphologically very similar . The mandibular corpus is broken irregularly at the alveoli and the base of the mandible is not present . Anteriorly , it is broken at the mesial M1 root; posteriorly , it is broken at the mesial alveolus for the M3 . The root of the ascending ramus is preserved and becomes independent at approximately the midpoint of M2 . Corpus breadth at M2 is approximately the anatomical value at 19 . 8 mm; the broken corpus does not permit this measurement elsewhere . 10 . 7554/eLife . 24232 . 035Figure 31 . U . W . 102 c-589 mandibular fragment . Top row , from left: lingual view , buccal view and occlusal view . Top right: two H . naledi teeth from the Dinaledi Chamber that present comparable occlusal morphology and wear to the U . W . 102 c-589 teeth: U . W . 101–297 RM1 ( top , reversed to represent left side ) and U . W . 101–284 LM2 ( bottom ) . Bottom row , from left: U . W . 102 c-589 posterior view , anterior view and basal view . Scale bar = 2 cm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 035 The morphology of the mandible and tooth crowns is similar to that of the mandibular remains of H . naledi from the Dinaledi Chamber . The size and preserved morphology of the mandibular corpus and root of the ascending ramus are very similar to that of the DH1 holotype of H . naledi . The morphology and wear stage of the U . W . 102 c-589 LM1are nearly identical to those of the U . W . 101–297 RM1 crown from the Dinaledi sample , and the LM2 is very similar in morphology to the U . W . 101–284 LM2 from Dinaledi ( Figure 31 ) . Neither of the U . W . 102 c-589 teeth exhibit supernumerary cusps , and both have simple crowns with a Y-5 cusp pattern without crenulation or complexity , similar to the Dinaledi teeth . The areas represented by the cusps appear similar to those in the Dinaledi molars , as do the crown heights . The relatively small size of the mandible and molars rule out attribution to Au . africanus or P . robustus , and the crown morphology rules out attribution to H . erectus or H . habilis . This mandible is entirely consistent with H . naledi . The U . W . 102 c-589 mandible clearly duplicates the LES1 mandible , thereby demonstrating the presence of a second adult individual in the Lesedi Chamber as a whole . Nevertheless , it is more difficult to determine whether the U . W . 102 c-589 individual also contributed a femur to 102a , accounting for the evidence of two adult individuals in that deposit . The 102c deposit is separated by a distance of approximately 12 m from the 102a assemblage . While this distance does not rule out the occurrence of the remains of a single individual in both locations , the context does not provide any reason for an assumption that they are the same . The general preservation of the Lesedi Chamber hominin material resembles that of the skeletal assemblage from the Dinaledi Chamber ( Dirks et al . , 2015 ) . The Lesedi skeletal material has a surface coloration that ranges from light grey to red-brown , and internal structures or cortex at fresh break points are colored pale buff to off-white , contrasting with unbroken adjacent surfaces . The remains present no evidence of calcite crystal formation . Specific comments on preservation have been included with the specimen descriptions above , and taphonomic observations on each specimen are summarized in Supplementary file 5 . The Lesedi skeletal assemblage exhibits varying degrees of post-mortem damage . Most specimens are fragmented or broken to some degree , with areas of cortical bone removal or abrasion ( Supplementary file 5 ) . All fractures observed in the assemblage are consistent with post-mortem ( dry bone ) failure ( Supplementary file 5 ) , and there are no spiral or incomplete fractures indicative of green or wet bone ( L’Abbé et al . , 2015; Symes et al . , 2014 ) . Fractures in the assemblage include transverse or right-angled breaks on the shafts of long bones , showing block-comminution between major breaks and step-fractures following the longitudinal grain of the long bone , which are distinct from cracking or crazing resulting from weathering ( Symes et al . , 2014 ) . The Lesedi remains display no evidence of sub-aerial weathering processes ( sensu Behrensmeyer , 1978; see also Lyman and Fox , 1989 and Junod and Pokines , 2014 , indicating that the dry bones were not subject to surface processes before deposition in the cave environment . Sub-aerial weathering is primarily characterized by both cracking and delamination . Within the Lesedi assemblage , there is some evidence of surface cracking , but it generally does not penetrate deep into the cortex . No indications of the secondary features of weathering , such as delamination , deep patination , bleaching or cortical exfoliation , were observed in any of the bone fragments from Lesedi , suggesting that the bones were not affected by surface exposure ( Supplementary file 5 ) . A comparison with human bone derived from forensic or sealed archaeological contexts ( Figure 32 ) suggests that the cracking and fracture patterns observed in the Lesedi assemblage can be explained most parsimoniously by the effects of the burial environment ( Pokines , 2014 ) , and were exacerbated in specimens where fluctuations in moisture content caused swelling and shrinkage of the cortical structure ( Conard et al . , 2008; Dirks et al . , 2016; Dirks et al . , 2015 ) . 10 . 7554/eLife . 24232 . 036Figure 32 . Surface cracking and pitting in Lesedi Chamber material . Top: right corpus and ramus of LES1 mandible prior to reconstruction , showing patterns of cracking consistent with the effects of sediment loading and wetting during the burial process and during skeletal decomposition . This pattern of taphonomic alteration is superficially similar to sub-aerial weathering processes , but is independent of surface exposure and occurs in both deep- and shallow-buried deposits . Middle: forensic known-history case for comparison . This specimen was recovered from a deep known-history inhumation . The fleshed body was deposited at a depth of 1 . 5 metres and recovered after a 30-year burial period . Note the similarities in surface texture , and superficial cracking , which follows the biomechanical stress lines ( grain ) of the bone . Bottom: U . W . 102a-010 acromial fragment showing surface modification with cortical pitting and punctate marks . Note that many of these marks ( circled in yellow ) penetrate pre-existing layers of manganese oxy-hydroxide deposited on the bone surface; the damage was therefore produced inside the Lesedi Chamber on bones that were already covered in coatings of manganese mineral . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 036 The assemblage presents no evidence of high-energy fluvial transport , such as smoothing and rounding , polish , frosting , cortex thinning , or aperture formation ( Bassett and Manhein , 2002; Behrensmeyer , 1988; Evans , 2014; Nawrocki et al . , 1997 ) . We have found no traces of carnivore or scavenger modification , with an absence of bone cylinders , tooth scores or traces of gastric corrosion ( Blumenschine et al . , 1996; Haynes , 1983; Hill , 1976; Pickering et al . , 2004; Thompson and Lee-Gorishti , 2007 ) . In addition , the profiles of damaged diaphyses show no evidence of end gnawing , scalloping or flaking ( Blumenschine et al . , 1996; Lyman , 1994; Wood , 1991 ) . In some cases , missing areas of epiphyses display patterns of localized cortical destruction , which may be consistent with a process referred to as ‘coffin wear’ ( Rogers , 2005; Schultz , 2012; Schultz et al . , 2003 ) whereby bones come into direct contact with an underlying substrate following soft tissue decomposition; this may lead to the loss of cortex or of portions of elements that are in contact with the substrate . This sub-surface process may completely remove processes or condyles from major elements such as the femur or humerus , leaving behind flattened or sheared areas of bone ( Rogers , 2005; Pokines and Baker , 2014 ) . As with the Dinaledi material , some of the Lesedi Chamber remains are stained by iron oxides and manganese oxy-hydroxide ( Dirks et al . , 2015; Randolph-Quinney et al . , 2016 ) . This staining primarily occurs as diffuse spots or mats of mineral , some of which cover and encapsulate the fractured ends of long bones , indicating a complex post-mortem history of mineral deposition . However , we have observed no mineral tide marks upon the Lesedi Chamber skeletal material which , unlike material from the Dinaledi Chamber , provides us with no information about the former position of skeletal remains relative to the sediment–air interface . Some of the Lesedi material exhibits minor pitting , striations , grooves , scratches , or gouges . Striations and grooves are consistent with abrasion marks , as defined by D’Errico and Villa ( 1997 ) and Fisher ( 1995 ) . Some of the elements have a palimpsest of taphonomic traces , with mineral deposition preceding invasive surface modification or fracturing ( Figure 32 ) . The pitting that is present on the Lesedi material shows that the modifying agents ( abiotic or biotic ) penetrated already-altered bone , suggesting that damage occurred on bone surfaces that were already covered in coatings of manganese and iron oxide ( Randolph-Quinney et al . , 2016 ) . Generally , the pattern of pitting and stripping is similar in composition and gross morphology to patterns observed in the Dinaledi Chamber assemblage , which were attributed to gastropod and other invertebrate activity ( Dirks et al . , 2015 , 2016; Randolph-Quinney et al . , 2016 ) , although these must be studied in more detail to be certain . Overall , the gross taphonomic signature of the hominin remains in the Lesedi Chamber appears consistent with sub-surface deposition with limited post-depositional dispersal . Particularly in the U . W . 102a locality , this is evidenced by the presence of articulated remains , the proximity of cranial and postcranial remains , and the recovery of small elements such as teeth and carpals . The bone surface condition , as well as the absence of other markers of transport and secondary modification , are generally consistent with sub-surface skeletal decomposition ( Carter and Tibbett , 2008; Carter et al . , 2007 ) . At present , we find no supporting evidence for sub-aerial weathering or post-mortem exposure outside of the cave environment ( Hill , 1976; Junod and Pokines , 2014; Lyman and Fox , 1989; Tappen and Peske , 1970 ) . Nor are there signs of carnivore or scavenger modification ( Pokines , 2013 ) , water transport ( Behrensmeyer , 1988; Evans , 2013 ) , or peri-mortem trauma such as may be expected in a natural death-trap scenario ( L’Abbé et al . , 2015; Symes et al . , 2014 ) . In addition to the hominin skeletal material , some faunal remains have been recovered from the Lesedi Chamber . Over 80 faunal elements or fragments were collected from U . W . 102a . U . W . 102b yielded 23 specimens of fauna , while a single rodent tooth was recovered from U . W . 102c . In-depth analyses of the faunal material are ongoing , and comprehensive descriptions are in preparation . We provide here a preliminary list of the taxa identified for reference ( Table 3 ) . 10 . 7554/eLife . 24232 . 037Table 3 . Mammal species recorded in the Lesedi Chamber . Several specimens from the classes Aves , Amphibia and Reptilia were also recovered , but individual counts and taxonomic identifications are pending further examination . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 037ClassOrderFamilySubfamilyGenus/speciesMNINISPMammaliaLagomorphaLeporidae11SoricomorphaSoricidae33CrocidurinaeCrocidura11Rodentia11BathyergidaeBathyerginae11Muridae11OtomyinaeOtomys11Murinae57Mus12NesomyidaeMystromyinaeMystromys11Dendromurinae34Steatomys36CarnivoraFelidaeFelis aff . F . sylvestris11Felis cf . sylvestris16Felis sp . 12Herpestidaecf . Mungos11cf . Herpestidae13CanidaeCanis aff . C . familiaris114Canis cf . mesomelas13Canis sp . 121Vulpes cf . chama13cf . Vulpes16Vulpes sp . 15 The Lesedi faunal assemblage includes micromammal , small to mid-size mammal , and non-mammalian remains . With respect to the micromammals ( prey body mass <500 g ) , 5 genera of rodents and 1 genus of shrew were identified out of 28 craniodental specimens ( 21 MNI ) . There are also potentially two additional murine genera and one soricid genus present in the assemblage , although lower-level identification is not possible at this time . Interestingly , all of the non-hominin fauna are of relatively small species . The largest mammalian specimens come from dental material attributed to Canis aff . C . familiaris . The size range of this material is outside the range of modern C . mesomelas but the morphology is definitively not Lycaon ( Hartstone-Rose et al . , 2010; Wayne , 1986 ) . The felid material is also small , falling in the size range of the African wildcat . None of these individuals is likely to have exceeded 10–15 kg ( Kingdon , 2015 ) and the rest of the assemblage consists almost entirely of animals smaller than 3 kg , including four specimens from the family Herpestidae . Aside from a single lagomorph specimen , the macro-mammalian material comes exclusively from the order Carnivora , a situation that is unusual in the fossil record ( Werdelin and Sanders , 2010 ) . We do not presently know whether some or all of these faunal remains may be contemporaneous with any of the hominin fossil material . Faunal remains have been recovered both on the surface and also from within sediments near hominin remains . However , the Lesedi Chamber is not a completely isolated environment , and sediment deposits are currently eroding from their original depositional contexts , with evidence for slumping and reworking in the chamber . Therefore , we cannot yet comment on the relative timing of deposition of the hominin and faunal material . Additional tests , including attempts to date both hominin and faunal elements directly , will help us answer these questions and to relate the faunal material to the chronological and environmental context of H . naledi in the Lesedi Chamber . The Dinaledi Chamber assemblage of H . naledi represents individuals who lived at roughly the same time as skeletal remains attributed to early modern or near-modern humans in some parts of Africa ( Figure 36; Berger et al . , 2017 ) . No derived features of the H . naledi skeleton require a close or exclusive relationship with modern humans . Overall , H . naledi resembles more primitive species of Homo such as H . erectus , H . habilis , or H . rudolfensis much more than it resembles archaic or modern humans ( Berger et al . , 2015; Laird et al . , 2017; Marchi et al . , 2017; Feuerriegel et al . , 2017; Williams et al . , 2016; Schroeder et al . , 2017 ) . H . naledi does , however , possess a number of derived features that are otherwise known only from modern humans and Neandertals ( Supplementary file 1; Figures 34 and 35; Berger et al . , 2015; Dembo et al . , 2016; Kivell et al . , 2015; Harcourt-Smith et al . , 2015; Williams et al . , 2016 ) . Some of these derived features , including features of the wrist , cannot be assessed in H . erectus because no fossils of the relevant bones exist for this species ( Kivell et al . , 2015 ) . But others , including features of the cranium and dentition , raise at least the possibility that H . naledi may be a sister to H . antecessor or to a clade including H . antecessor with other archaic and modern humans ( Dembo et al . , 2016 ) . It is also conceivable that , rather than indicating a recent branching of H . naledi from an archaic human lineage , such derived similarities may have resulted from introgressive hybridization between H . naledi and other hominin lineages ( Berger et al . , 2017 ) , although testing this hypothesis will likely require genetic data . 10 . 7554/eLife . 24232 . 042Figure 36 . Lateral cranial comparison of H . naledi crania to crania of other hominin species . H . naledi crania , DH1 , LES1 , and DH3 are in the center row . All crania are oriented as near as possible to the Frankfort plane , delineated by the light gray lines in the background of the figure . Compared to other hominin genera , including Australopithecus and Paranthropus , fossil Homo is often recognized by cranial and dental features such as a more vertical face profile , a reduced postcanine dentition , larger endocranial volume , a higher frontal , and a true supraorbital torus . Au . africanus ( Sts 5 , top left ) represents the ancestral hominin condition lacking these traits . The other crania in the top two rows vary substantially in these features . LB1 has a vertical face , reduced dentition , and high , rounded frontal , but has comparatively small endocranial volume . KNM-ER 1470 has a large volume , a high frontal , and a more vertical face profile , but also is inferred to have a large postcanine dentition and has no true supraorbital torus . MH1 ( Au . sediba ) has a small volume , but shares features with Homo that include the less sloping face profile , a supraorbital torus , and reduced postcanine dentition . O . H . 24 has a low , sloping frontal , and a concave facial profile , but a true supraorbital torus and reduced postcanine teeth . This variability among species that are interpreted as ‘primitive’ Homo , such as H . habilis and H . floresiensis , and Homo-like australopiths makes it difficult to delineate the genus Homo ( Wood and Collard , 1999; Dembo et al . , 2016 ) . H . erectus is also highly variable . It includes several crania with endocranial volumes below 700 ml , including KNM-ER 42700 and D2282 , but also many larger crania , here represented by Sangiran 17 and Zhoukoudian L2 ( ZKD L2 ) . Specimens attributed to H . erectus tend to share a series of traits first noted in Asian H . erectus samples , including a long , low cranial profile , thick cranial bone , sagittal keeling , prominent supraorbital , angular , and occipital tori , a sharply angled occiput , and a postbregmatic depression . The smallest H . erectus crania share most of these features , with a low cranial profile , angled occiput and postbregmatic depression visible here in D2282 . But these features do vary substantially and are less evident in the immature KNM-ER 42700 . The H . naledi crania are similar to KNM-ER 1470 in having a transversely flat clivus contour , but all are smaller , with a much smaller palate and with very different frontal morphology . Like O . H . 24 and KNM-ER 1813 , the H . naledi crania have relatively thin cranial bone and a thin and projecting supraorbital torus . But H . naledi manifests a different clivus shape , a projecting nasal spine , a greater cranial height , sagittal keeling and an angular torus . The H . naledi crania bear little resemblance to LB1 , differing in face profile , size , and their larger postcanine dentitions . Known African Homo specimens from the later Middle Pleistocene other than H . naledi , such as the Kabwe skull ( pictured ) , contrast strongly with H . naledi in cranial size and morphology . The Omo 2 skull , one of the earliest known modern human crania at approximately 196 , 000 years ( McDougall et al . , 2005 ) , is vastly larger and very different from any H . naledi specimen , despite being near the same geological age . In this figure , O . H . 24 , KNM-ER 1470 , LB1 , KNM-ER 42700 , ZKD L2 , and Omo 2 are represented by casts . Images have been adjusted to a common scale by maximum cranial length , or by glabella-bregma length where maximum length is not available . Photos of Sangiran 17 and D2282 are courtesy of Milford Wolpoff . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 042 The late Middle Pleistocene geological age of H . naledi , when considered together with its small endocranial volume , prompt comparisons with H . floresiensis . Skeletal remains of H . floresiensis are known from ~100–60-ka-old sediments at Liang Bua , Flores ( Brown et al . , 2004; Sutikna et al . , 2016 ) , and possibly also in Mata Menge , Flores sediments from the early Middle Pleistocene ( van den Bergh et al . , 2016; Brumm et al . , 2016 ) . The features shared by H . naledi and H . floresiensis are nearly all inferred to be primitive features shared with australopiths and/or in some cases with early Homo ( Figures 35 and 36 ) . For most features of the skeleton where H . naledi exhibits derived morphology , including aspects of the foot and hand , H . floresiensis exhibits morphology thought to be primitive within hominins ( Harcourt-Smith et al . , 2015; Kivell et al . , 2015; Tocheri et al . , 2007; Jungers et al . , 2009a; Orr et al . , 2013 ) . All known H . floresiensis long bones are small , reflecting small stature and mass ( Jungers et al . , 2009b; Larson et al . , 2009; Grabowski et al . , 2015 ) , and none approach the much larger size , particularly the taller stature , manifested in adult specimens of H . naledi . The mandibles of H . floresiensis are relatively robust , even for their small size , ( Brown and Maeda , 2009; Daegling et al . , 2014 ) and in this sense , are comparable to the robust mandibles of H . naledi , though the two species differ in symphyseal morphology ( Laird et al . , 2017 ) . The dentition of H . floresiensis displays a unique combination of primitive and derived characters ( Kaifu et al . , 2015 ) , which are different from the combination found in H . naledi . Homo floresiensis canines and premolars display a combination of features that are primitive for hominins ( Kaifu et al . , 2015 ) , while H . naledi mandibular canines and third premolars have a uniquely derived form . The H . floresiensis molars exhibit derived proportions and some other features similar to those of modern humans , which may reflect the extremely shortened molar crowns , particularly of the first molars ( Kaifu et al . , 2015 ) . By contrast , H . naledi has primitive proportions of the molars and occlusal morphology that is broadly primitive within Homo , although simplified in complexity . Shape analyses also place the cranium of H . floresiensis ( LB1 ) far from any H . naledi cranial specimen ( Schroeder et al . , 2017 ) . These differences in anatomy may reflect a distant phylogenetic relationship between the two species , although these data do not clearly resolve where on the Homo phylogeny either species should be placed . Some phylogenetic evidence suggests that the H . floresiensis branch preceded the node linking H . erectus and modern humans ( Argue et al . , 2009; Dembo et al . , 2015 ) . Cranial and dental evidence does not resolve whether H . naledi may also have branched early in the evolution of Homo , or whether it may instead be a sister taxon to modern and archaic humans ( Dembo et al . , 2016 ) . No data that are available at present answer whether either or both of these lineages may have retained small EVC from an ancestry among the earliest members of Homo , or whether their smaller brain sizes may have evolved secondarily from larger-brained ancestors . Together , they establish that diverse hominin lineages with varying brain and body sizes existed during the Middle Pleistocene , suggesting the influence of ecological factors that promoted diversity during the Pleistocene evolution of humans and great apes ( Berger et al . , 2017; Tocheri et al . , 2011 , 2016; Dunn et al . , 2014 ) . H . naledi shares many derived cranial and dental characters with H . erectus ( Figures 34 and 36; Dembo et al . , 2016; Laird et al . , 2017 ) , and Dmanisi specimens of H . erectus are among the closest in multivariate shape to H . naledi crania ( Schroeder et al . , 2017 ) . H . naledi overlaps with H . erectus in long bone lengths , inferred stature , and estimated body mass ( Figure 28; Berger et al . , 2015 ) . To the extent that the H . naledi postcranial skeleton is different from material attributed to H . erectus , it is mostly because H . naledi manifests primitive traits that are not otherwise seen in Homo , or because H . naledi manifests traits not otherwise seen in hominins ( Figure 35 ) . H . naledi also shares several derived postcranial features with archaic and modern humans , but a lack of postcranial evidence for these parts of the H . erectus skeleton has made it impossible to determine whether H . erectus may also have shared these derived features ( Figure 35 ) . With respect to the cranium and dentition , nearly all of the nonmetric traits shared by H . naledi and H . erectus are also shared with Au . sediba , H . habilis , or both ( Supplementary file 1; Figure 34 ) . H . naledi and H . erectus share at least three derived nonmetric features of the skull that are not also found in Au . sediba or H . habilis: an angular torus , sagittal keeling , and an anteriorly projecting nasal spine . But each of these three traits can also be found in some archaic or modern humans . H . naledi also possesses some cranial and dental traits not seen in any specimens of H . erectus; these include both primitive traits shared with Au . afarensis , Au . africanus or Au . sediba and derived traits shared either with H . rudolfensis or with archaic and modern humans . Even though H . naledi crania are most similar in shape to the smallest H . erectus crania from Dmanisi ( Schroeder et al . , 2017 ) , they are distinct from the Dmanisi sample in numerous aspects of cranial , dental , and postcranial morphology ( Berger et al . , 2015; Laird et al . , 2017; Marchi et al . , 2017; Feuerriegel et al . , 2017; Rightmire et al . , 2017 ) , and Bayesian analysis of cranial and dental morphology provides strong evidence against the hypothesis of a sister taxon relationship for these samples ( Dembo et al . , 2016 ) . In summary , no traits link H . naledi exclusively or specifically with H . erectus , and many traits distinguish the two . The evidence is not sufficient to say whether H . naledi may have evolved from an earlier population that resembled H . erectus , or whether earlier fossils representing the H . naledi lineage may already have been found and until now attributed to H . erectus ( Berger et al . , 2017 ) . Fossil samples from Malapa , Olduvai Gorge , and the Lake Turkana area have been attributed to lineages often thought to represent some of the earliest species of Homo or their near relatives among the australopiths . H . naledi shares many derived cranial and mandibular features with Au . sediba and H . habilis , although most of these are also shared with H . erectus ( Supplementary file 1; Figures 34 and 35 ) . Compared to these other species , H . naledi shares many fewer cranial and dental features with H . rudolfensis , except for its relatively flat and squared nasoalveolar clivus . Phylogenetic analysis does not reject the hypothesis of a sister taxon relationship between H . naledi and Au . sediba or H . habilis ( Dembo et al . , 2016 ) . However , in comparison to both Au . sediba and H . habilis , H . naledi is derived and similar to modern humans and Neandertals in several aspects of hand and wrist morphology ( Kivell et al . , 2015 ) . It is also derived in its foot morphology in comparison to both MH2 of Au . sediba and the O . H . 8 foot usually attributed to H . habilis ( Harcourt-Smith et al . , 2015 ) . H . naledi also lacks the derived configuration of the Au . sediba ilium ( Figure 35 ) . H . naledi further shares a number of derived cranial characters with H . erectus and with archaic or modern humans as discussed above ( Supplementary file 1 , Figure 34 ) . Whatever the relationship of these species , their evolution must have involved homoplasy of many features and the evidence does not yet make it clear how they are connected . If any one of these species were a possible ancestor of H . naledi , the branch leading to the Dinaledi sample of H . naledi would approach 1 . 5 Ma or longer in evolutionary time . With abundant opportunity for adaptive and nonadaptive evolution of the H . naledi lineage over this time , it would be very hard to distinguish a long branch connecting it to Au . sediba or H . habilis from a somewhat longer branch connecting it to the very base of Homo . The excavation of the Lesedi Chamber has added to our knowledge of the biology of H . naledi and has confirmed the presence of this species in a second depositional context . The skeletal material described here derives from a very small and limited excavation , and the total sediment volume of the chamber has not yet been sampled sufficiently to estimate the abundance of hominin-bearing deposits or the relationship of faunal and hominin species . Further resolution of how the material was originally deposited must await more detailed sedimentological analysis and more excavation work . The relative completeness of the morphological evidence from H . naledi has not resolved its phylogenetic placement within the genus Homo ( Dembo et al . , 2016 ) . The morphological evidence from the Lesedi Chamber hominin material reinforces the observations made on the basis of the Dinaledi hominin sample . In particular , the more complete LES1 cranium and the more fragmented cranial remains from multiple individuals already known for H . naledi share an almost identical pattern of derived features with other hominin species ( Laird et al . , 2017 ) . The discovery of H . naledi within the Dinaledi Chamber documented a pattern of anatomy that had not been anticipated by anthropologists on the basis of earlier fossil discoveries . Learning how H . naledi connects to other fossil evidence of human origins may require more unexpected discoveries . Formal excavations began in the Lesedi Chamber in May 2014 , after initial surface and ex situ material had been recovered . Like the Dinaledi Chamber , the access route to the Lesedi Chamber is very restricted . Additionally , the excavation area available in the Lesedi Chamber is considerably more confined than that in the Dinaledi Chamber . As a result , protocols followed those used in the Dinaledi Chamber ( Dirks et al . , 2015 ) , with a few exceptions . Specifically , attempts to use some of the surface-scanning methods applied in the Dinaledi Chamber ( Kruger et al . , 2016 ) failed due to the limited working distances of the scanner , the convoluted surfaces , and the poor contrast of the surrounding dolomite . However , high-resolution laser scan data have been acquired for the Lesedi Chamber and will form the basis of future spatial work . Consequently , documentation of the excavations and provenience of the material was conducted using traditional archaeological methods , including written descriptions , photography and drawings . Excavations were conducted in tandem with geological mapping and with sedimentologic and taphonomic analyses of the site . For excavations in U . W . 102a , the area was divided into 20–40 cm sections that were based on the contours of the sloped surface and the width of the tunnel at that point . Sections were excavated in 2–5 cm levels . In some areas where few specimens were being recovered , levels were expanded to 10 cm . In the U . W . 102b area , excavations began on the surface of the sediment deposit in which the first material was recovered , and proceeded in 2–5 cm levels until the chert layer on which the sediment had accumulated was reached . When this area was exhausted , excavations moved along the chert shelf , removing sediments from the surface , to the chert , in 20 cm horizontal sections . A sediment pocket above the fossil deposit ( on Chert 3 ) was also excavated to determine whether material had trickled down from above . Preliminary excavations were also begun on the antechamber floor below the primary fossil deposit . Owing to the confines of the dolomite recess in which the hominin material was found , U . W . 102c was not divided into sections . However , it was also excavated in 2–5 cm levels , from the top of the sediment accumulation to the chert shelf at its base . For all three areas , all sediments were collected for every section and level separately . Each bag of back dirt was labeled , removed from the cave and dry screened to recover small elements , fragments and microfaunal remains . All sediments were retained and are currently stored at the University of the Witwatersrand’s Evolutionary Studies Institute . Taxonomic identification of the Lesedi Chamber hominin material depended on a series of systematic comparisons to other hominin species . Many of these observations were carried out on the original fossil specimens by the authors; some have relied upon observations taken from research-grade casts of fossil specimens available at the University of the Witwatersrand and elsewhere , while in some cases , observations were only available from the literature . The fossil comparative samples employed in this study are essentially the same as those described in Berger et al . ( 2015 ) , with the addition of the Dinaledi Chamber sample of H . naledi ( Berger et al . , 2015; Kivell et al . , 2015; Marchi et al . , 2017; Feuerriegel et al . , 2017; Laird et al . , 2017; Williams et al . , 2017; Zipfel and Berger , 2009 ) . The gross morphological appearance and dental dimensions of the remains immediately made it clear that they were inconsistent with Paranthropus , and with Australopithecus afarensis , Australopithecus africanus , or Australopithecus garhi , and so we focused our comparisons upon Homo and Australopithecus sediba . The composition of samples of other hominin species is as follows , largely repeated from Berger et al . ( 2015 ) . Australopithecus sediba . The partial skeletons MH1 and MH2 from Malapa , South Africa were included in this study on the basis of examination of the original specimens by the authors . Homo habilis . Samples from Olduvai Gorge , East Lake Turkana , the Omo Shungura sequence , Hadar , and Sterkfontein were considered within the hypodigm of H . habilis for this study . Original Olduvai Gorge and East Lake Turkana fossils were examined first-hand , whereas for the Omo and Hadar materials , we relied on our original observations on casts and originals and published reports ( Tobias , 1991; Boaz et al . , 1977; Kimbel et al . , 1997 ) . As in the initial announcement of H . naledi ( Berger et al . , 2015 ) , in this paper we adopt a conservative approach that follows a more conventional hypodigm , thereby encompassing a maximum amount of variation in this taxon; for a more detailed discussion of the probable hypodigms of early Homo species , see de Ruiter et al . , 2017 . We therefore include the following fossils in the hypodigm of H . habilis: A . L . 666–1 , KNM-ER 1478 , KNM-ER 1501 , KNM-ER 1502 , KNM-ER 1805 , KNM-ER 1813 , KNM-ER 3735 , O . H . 4 , O . H . 6 , O . H . 7 , O . H . 8 , O . H . 13 , O . H . 15 , O . H . 16 , O . H . 21 , O . H . 24 , O . H . 27 , O . H . 31 , O . H . 35 , O . H . 37 , O . H . 39 , O . H . 42 , O . H . 44 , O . H . 45 , O . H . 62 , OMO-L894-1 , and Stw 53 . Homo rudolfensis . Samples from Olduvai Gorge , East Lake Turkana , and Lake Malawi were considered as part of the hypodigm of H . rudolfensis for this study . The East Lake Turkana fossils available prior to 2010 were examined first-hand , while for the Olduvai and Lake Malawi fossils and for KNM-ER 60000 , 62000 , and 62003 , we relied on original observations of fossils and casts as well as published reports ( Blumenschine et al . , 2003; Schrenk et al . , 1993; Leakey et al . , 2012 ) . As above , and in the initial announcement of H . naledi ( Berger et al . , 2015 ) , in this paper we adopt a conservative approach that follows a more conventional hypodigm , thereby encompassing a maximum amount of variation in this taxon; for a more detailed discussion of the probable hypodigms of early Homo species , see de Ruiter et al . ( 2017 ) . We include the following fossils in the hypodigm of H . rudolfensis: KNM-ER 819 , KNM-ER 1470 , KNM-ER 1482 , KNM-ER 1483 , KNM-ER 1590 , KNM-ER 1801 , KNM-ER 1802 , KNM-ER 3732 , KNM-ER 3891 , KNM-ER 60000 , KNM-ER 62000 , KNM-ER 62003 , O . H . 65 , and UR 501 . Homo erectus . Samples from Buia , Chemeron , Daka , Dmanisi , East and West Lake Turkana , Gona , Hexian , Konso , Mojokerto , Olduvai Gorge , Sangiran , Swartkrans , Trinil , and Zhoukoudian were included in the hypodigm of H . erectus for the purposes of this study . South African material is of special interest in this comparison because of the geographic proximity , and because of the difficulty of clearly identifying Homo specimens within the large fossil sample from Swartkrans . In particular , the following specimens from Swartkrans are considered to represent H . erectus: SK 15 , SK 18a , SK 27 , SK 43 , SK 45 , SK 68 , SK 847 , SK 878 , SK 2635 , SKW 3114 , SKX 257/258 , SKX 267/2671 , SKX 268 , SKX 269 , SKX 334 , SKX 339 , SKX 610 , SKX 1756 , SKX 2354 , SKX 2355 , SKX 2356 , and SKX 21204 . We considered ‘Homo ergaster’ ( and also ‘Homo aff . erectus’ from Wood , 1991 ) to be synonyms of Homo erectus for this study; Turkana Basin specimens that are attributed to H . erectus thus include KNM-ER 730 , KNM-ER 820 , KNM-ER 992 , KNM-ER 1808 , KNM-ER 3733 , KNM-ER 3883 , KNM-ER 42700 , KNM-WT 15000 . Olduvai specimens include O . H . 9 , O . H . 12 and O . H . 28 . Original fossil materials from Chemeron , Lake Turkana , Swartkrans , Trinil , and Dmanisi were examined first-hand by the authors , whereas the remainder were based on casts and published reports ( Abbate et al . , 1998; Gilbert and Asfaw , 2008; Wood , 1991; Weidenreich , 1943; Suwa et al . , 2007; Antón , 2003; Rightmire et al . , 2006 , 2017; Martinón-Torres et al . , 2008 ; Spoor et al . , 2007 ) . A large number of postcranial specimens have been collected from the Turkana Basin and appear consistent with the anatomical range otherwise found in Homo , and inconsistent with known samples of Australopithecus and Paranthropus from elsewhere . These include KNM-ER 1472 , KNM-ER 1481 , KNM-ER 3228 , KNM-ER 737 , KNM-ER 5881 ( Ward et al . , 2015 ) , and others . Specimens from the latest Lower Pleistocene and Middle Pleistocene of Europe and Africa that cannot be attributed to H . erectus were also included in the comparisons in this study . These include fossils that have been attributed to H . heidelbergensis , H . rhodesiensis , ‘archaic H . sapiens’ or ‘evolved H . erectus’ by a variety of other authors . Specimens include KNM-ES 11693 , Arago 2 , Arago 13 , Arago 21 , Atapuerca 1 , Atapuerca 2 , Atapuerca 4 , Atapuerca 5 , Atapuerca 6 , Cave of Hearths , Ceprano , SAM-PQ-EH1 , Kabwe , Mauer , Ndutu , Salé , Petralona , Reilingen-Schwetzingen , and Steinheim . We also included Neandertal samples from Krapina , Vindija , La Chapelle-aux-Saints , La Ferrassie , Monte Circeo , Saccopastore , and Feldhofer . Homo floresiensis . The hypodigm of H . floresiensis used in this study includes specimens from Liang Bua , Flores , as described by Brown et al . ( 2004 ) , Brown and Maeda ( 2009 ) , Tocheri et al . ( 2007 ) , Orr et al . ( 2013 ) , Jungers et al . ( 2009a , 2009b ) , Larson et al . ( 2009 ) , Morwood et al . ( 2005 ) , Falk et al . , 2005 , Kaifu et al . ( 2011 ) , and Kubo et al . ( 2013 ) . The LES1 ECV was estimated virtually using 3D surface scans of the reconstructed partial cranium . These methods were similar to those performed on the previously published Dinaledi H . naledi crania ( Berger et al . , 2015 ) . The LES1 partial cranium was scanned using a NextEngine 3D Scanner and associated ScanStudio HD Pro software . Two 360-degree scans were collected using 16 divisions and 1000 dpi with the cranium in different orientations in order to capture the maximum ectocranial and endocranial surfaces . The two 360-degree scans were then merged and exported as a . ply file . The . ply file was opened in GeoMagic Studio , converted to a point cloud , and re-wrapped to minimize the number of edges in the model ( created from the individual surface scans ) . The LES1 virtual model was then duplicated and mirrored , and the mirror-image ( right portion ) was aligned with the original model ( left portion ) using both the Manual and Global Registration functions . In doing this , the program uses an iterative process to find the greatest congruency between the region of overlap in the two models , minimizing deviations . Convergence was detected after 20 iterations , with an average deviation of 0 . 47 mm and a standard deviation of 0 . 43 mm . Some deviation is expected given the natural anatomical asymmetry and refitting of the fragments . A deviation map illustrates that areas with the highest deviation were fairly localized ( Figure 37 ) . We inspected the mirror image in comparison to the preserved but not refitted fragments that represent portions of the right parietal and right temporal bones of LES1 , finding that they are similar in size and curvature . The LES1 original images and mirror-image were then merged . 10 . 7554/eLife . 24232 . 043Figure 37 . LES1 model congruency . Congruency between the original LES1 3D scan and an aligned mirror-image . Scale bar = 2 cm . ( A ) Frontal view of a 3D scan of the original LES1 specimen ( brown ) aligned with the mirror-image ( grey ) , illustrating congruency between overlapping regions . ( B ) Deviation map of the internal view of the frontal region . Deviation scale bar in mm . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 043 The endocranial surface was then isolated by manually selecting the surfaces and creating a new model ( Figure 38 ) . Fragment edges were deleted and small holes in the model were filled using the ‘Fill by Curvature’ function . Gaps in the model remained in the posterior parietal , occipital , and cranial base regions . In order to compute a volume , the model must be completely closed; thus , these regions were carefully filled using either the ‘Fill by Curvature’ or ‘Flat Filling’ functions depending on the regions . This procedure was carefully monitored and the approximated surfaces were modified as necessary to ensure that they were congruent with the endocranial surfaces present . The volume of the closed endocranial model was then computed in GeoMagic Studio . 10 . 7554/eLife . 24232 . 044Figure 38 . LES1 endocast reconstruction . Virtual reconstruction of LES1 endocast for endocranial volume estimation . Scale bar = 3 cm . ( A ) Oblique posterior view illustrating missing portions in the endocranial reconstruction that were filled to close the model for volume estimation . ( B-E ) Left lateral ( B ) , superior ( C ) , anterior ( D ) , and inferior ( E ) views of the completed endocranial model . ( F ) Right lateral view of the completed endocranial model within the surrounding cranium . DOI: http://dx . doi . org/10 . 7554/eLife . 24232 . 044 Previous research suggests that using a model cranial base from another hominin taxon does not significantly alter virtual estimate results ( Berger et al . , 2015 ) ; however , given the uniformity in the cranial base reconstruction of the endocranial model presented here ( e . g . , the lack of consideration for sella turcica , etc . ) , the current estimate may be a slight overestimate . Holloway ( personal communication ) found that by manually constructing a cranial base , ECV estimates for the previously published DH1/DH2 and DH3/DH4 H . naledi composite endocasts decreased by 5 ml ( 0 . 9% and 1 . 1% , respectively ) . The hominin assemblage from the Lesedi Chamber was analysed using the taphonomic protocols and methods applied to the U . W . 101 Dinaledi fossils , and detailed in Dirks et al . ( 2015 ) : 22–24 and 32–33 . Specimens were viewed macroscopically , and fragments >50 mm diameter were imaged on all surfaces using a Canon 70D DSLR fitted with a Canon EF-S 60mm f2 . 8 macro lens and ring-flash . Criteria for scoring taphonomic observations are detailed in Supplementary file 5 . Additional taphonomic analyses are underway and will be described in future publications . The taxonomic classification of the micromammal remains follows that of Wilson and Reeder , 2005 , using published descriptions and images ( Avery , 2007; Coetzee , 1972; Davis , 1965; De Graaff , 1981; Meester and Setzer , 1971; Reppening , 1967; Skinner and Chimimba , 2005 ) , as well as comparisons with a photographic database of southern African rodents from various museum collections . As genus is the lowest taxonomic level at which most material can be identified accurately ( Reed , 2005 , ( 2007 ) ; Reed and Geraads , 2012 ) , specimens were identified to this level or higher . Macrovertebrate remains were also identified using published reference guides ( Walker , 1985 ) and visual comparisons with mammalian collections at the University of the Witwatersrand . Wherever possible , specimens were attributed to species . In cases where this was not possible , the next highest taxonomic level was used . Identification of the non-mammalian remains was limited to the level of class or order , pending more detailed analyses . All fossil material from the Lesedi Chamber is available for study by researchers upon application to the Evolutionary Studies Institute at the University of the Witwatersrand where the material is curated . Three-dimensional surface renderings and other digital data are available from the MorphoSource digital repository ( http://morphosource . org ) .	Species of ancient humans and the extinct relatives of our ancestors are typically described from a limited number of fossils . However , this was not the case with Homo naledi . More than 1500 fossils representing at least 15 individuals of this species were unearthed from the Rising Star cave system in South Africa between 2013 and 2014 . Found deep underground in the Dinaledi Chamber , the H . naledi fossils are the largest collection of a single species of an ancient human-relative discovered in Africa . After the discovery was reported , a number of questions still remained . These questions included: why were so many fossils from a single species found at the one site , and how did they come to rest so far into the cave system ? Possible explanations such as H . naledi living in the cave or being washed in by a flood were considered but ruled out . Instead , the evidence was largely consistent with intact bodies being deliberately disposed of in the cave and then decomposing . Now , Hawks et al . – who include many of the researchers who were involved in the discovery of H . naledi – report that yet more H . naledi fossils have been unearthed from a second chamber in the Rising Star cave system , the Lesedi Chamber . The chamber is 30 meters below the surface and there is no direct route between it and the Dinaledi Chamber . Again , the evidence is most consistent with the bodies arriving intact into the chamber , and there were no signs that the remains had been exposed to the surface environment . Also like the Dinaledi Chamber , no remains of other ancient humans or their relatives were found in the Lesedi Chamber . In total , 133 fossils of H . naledi have been found in this second chamber representing at least three individuals: two adults and a juvenile . However , and as Hawks et al . point out , only a small volume of the chamber has been excavated so far , and so there are likely more fossils still to be found . The fossils in the Lesedi Chamber are similar to those found before but include intact examples of bones , like the collarbone , that were previously known only from fragments . Perhaps the most impressive among the new fossils is a relatively complete skull that is part of a partial skeleton . The skull could have housed a brain that was 9% larger than the maximum estimate calculated from the previous H . naledi fossils . Though these new fossils provide us with yet more information about H . naledi , some questions still remain unanswered – the material from the Lesedi Chamber is undated , for example . However , a related study by Dirks et al . does give an estimate for the age of the fossils from the Dinaledi Chamber , while Berger et al . provide an explanation for why this date might be much younger than was previously predicted .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2017	New fossil remains of Homo naledi from the Lesedi Chamber, South Africa
One of the major unanswered questions in evolutionary biology is when and how the transition between diderm ( two membranes ) and monoderm ( one membrane ) cell envelopes occurred in Bacteria . The Negativicutes and the Halanaerobiales belong to the classically monoderm Firmicutes , but possess outer membranes with lipopolysaccharide ( LPS-OM ) . Here , we show that they form two phylogenetically distinct lineages , each close to different monoderm relatives . In contrast , their core LPS biosynthesis enzymes were inherited vertically , as in the majority of bacterial phyla . Finally , annotation of key OM systems in the Halanaerobiales and the Negativicutes shows a puzzling combination of monoderm and diderm features . Together , these results support the hypothesis that the LPS-OMs of Negativicutes and Halanaerobiales are remnants of an ancient diderm cell envelope that was present in the ancestor of the Firmicutes , and that the monoderm phenotype in this phylum is a derived character that arose multiple times independently through OM loss . The bacterial envelope is one of the oldest and most essential cellular components , involved in key housekeeping functions such as physical integrity , cell division , motility , substrate uptake and secretion , and cell-cell communication ( Silhavy et al . , 2010 ) . Yet , bacteria show substantial differences in their cell envelope architectures , among which the most dramatic one is the presence of one ( monoderm ) or two ( diderm ) membranes ( Sutcliffe , 2010 ) . The study of cell envelope architecture has been mostly narrowed to the Firmicutes and the Gammaproteobacteria as textbook examples of monoderm and diderm bacteria , respectively . In Bacillus subtilis , teichoic and lipoteichoic acids are embedded in a thick peptidoglycan wall , while in Escherichia coli a thin peptidoglycan layer is surrounded by an outer membrane ( OM ) whose biogenesis and functioning involve a complex system of synthesis and transport for LPS , lipoproteins , and OM proteins ( OMPs ) ( Silhavy et al . , 2010 ) . The transition between monoderm and diderm cell envelopes must have been a significant and complex process in the evolutionary history of Bacteria . Two major hypotheses have been largely discussed in the literature , which can be generally defined as diderm-first ( Cavalier-Smith , 2006 ) and monoderm-first ( Gupta , 2011; Lake , 2009 ) scenarios . The fact that the majority of phyla seem to possess two membranes might favor the diderm-first scenario , although the actual diversity of cell envelopes in Bacteria remains largely unexplored ( Sutcliffe , 2010 ) . However , the lack of a robustly resolved phylogeny for Bacteria , notably the uncertainty on its root and the nature of the earliest branches , has left the relationships between diderm and monoderm phyla unclear , and not allowed to define in which direction and how many times this transition occurred . In this respect , the Negativicutes ( Marchandin et al . , 2010 ) represent an interesting case: while belonging phylogenetically to the classical monoderm Firmicutes , they surprisingly display a diderm cell envelope with an OM and LPS ( Delwiche et al . , 1985; Vos et al . , 2009 ) . The Negativicutes have been identified in various anaerobic environments , such as soil and lake sediments , industrial waste , and animal digestive tract ( Vos et al . , 2009 ) . Their best-characterized member is Veillonella , first described in 1898 by Veillon and Zuber ( Veillon and Zuber , 1898 ) . Curiously , the very first observation and use of the term 'outer membrane' has been based on studies of Veillonella ( Bladen and Mergenhagen , 1964 ) . Veillonella is one of the most abundant components of the human oral flora ( Tanner et al . , 2011 ) , and a common inhabitant of the intestinal microbiome ( van den Bogert et al . , 2013 ) . Together with other gut microbes , it has been recently associated with maturation of the immune system and partial protection of asthma in infants ( Arrieta et al . , 2015 ) , but can also develop into an opportunistic pathogen ( Hirai et al . , 2016 ) . Several other Negativicutes members such as Dialister , Selenomonas , Mitsuokella , and Anaeroglobus show increased incidence in oral tract disease linked to biofilm formation ( Griffen et al . , 2012 ) and involvement in other infections ( Wang et al . , 2015 ) . Very little experimental data is available on the nature of the diderm cell envelope of Negativicutes . In Selenomonas ruminantium the abundant OmpM protein appears to replace the important function of Braun’s lipoprotein in anchoring the OM to the cell peptidoglycan through a link with cadaverine ( Kojima et al . , 2010 ) . How the OM originated in the Negativicutes represents an evolutionary conundrum . Recently , Tocheva and colleagues analyzed the sporulation process in the Negativicute Acetonema longum by cryoelectron microscopy ( Tocheva et al . , 2011 ) . They showed that , while an outer membrane forms only transiently during sporulation in classically monoderm Firmicutes such as Bacillus subtilis , it is retained in A . longum leading to its diderm phenotype ( Tocheva et al . , 2011 ) . This study provided the first experimental support for the hypothesis that the bacterial OM could have initially evolved in an ancient sporulating monoderm bacterium ( Dawes et al . , 1980; Errington , 2013; Vollmer , 2012 ) . Moreover , a phylogenetic tree of the essential Omp85 protein family for proteins insertion in the outer membrane , although largely unresolved , did not show the Negativicutes as emerging from any specific diderm phylum ( Tocheva et al . , 2011 ) . The authors speculated that the OM of Negativicutes was not acquired by horizontal gene transfer but was already present in the ancestor of Firmicutes and would have been lost in the other members of this phylum , although it remained unclear when and how many times this would have occurred ( Tocheva et al . , 2011 ) . In contrast , a recent analysis of the genome of the Negativicute Acidaminococcus intestini revealed that as much as 7% of the BLAST top hits were from Proteobacteria , the majority of which corresponded to functions related to OM biogenesis , concluding to a possible acquisition of the OM in Negativicutes by horizontal gene transfer ( Campbell et al . , 2014 ) . Interestingly , the Negativicutes are not the only diderm lineage in the Firmicutes . The Halanaerobiales are a poorly studied group of moderate halophilic , strictly anaerobic Firmicutes that were isolated from saline environments such as lake and lagoon sediments , and oil reservoirs ( Oren , 2006; Roush et al . , 2014 ) . Similarly to the Negativicutes , they display a diderm-type cell envelope , with a thin peptidoglycan and an outer membrane ( Cayol et al . , 1994; Zeikus et al . , 1983; Zhilina et al . , 1992 , Zhilina et al . , 2012 ) . When analyzing the first sequenced genome of a member of Halanaerobiales , Halothermothrix orenii , Mavromatis and colleagues identified a number of OM markers , suggesting the presence of an LPS-diderm cell envelope homologous to the one of Negativicutes and other diderm bacteria ( Mavromatis et al . , 2009 ) . In contrast to the few analyses on Negativicutes , no experimental data are available on the characteristics of the OM in the Halanaerobiales . The existence of two diderm lineages in the Firmicutes provides a fantastic opportunity to clarify the monoderm/diderm transition in this major bacterial phylum . However , the origins and evolutionary relationships between the OM of Halanaerobiales and Negativicutes have been unclear . In fact , no Halanaerobiales were present in the analysis of Tocheva ( Tocheva et al . , 2011 ) . Mavromatis et al . built a tree from the combined analysis of the genes coding for LPS , which showed a clustering of Halanaerobiales and Negativicutes , leading the authors to propose a horizontal gene transfer of the OM between these two lineages ( Mavromatis et al . , 2009 ) . However , the sequenced genome of only one member of Halanaerobiales and one of Negativicutes were available at the time , and the LPS tree was largely unresolved ( Mavromatis et al . , 2009 ) . Moreover , current phylogenies of the Firmicutes have been unclear with respect to the relationships between Negativicutes and Halanaerobiales . The Negativicutes have been alternatively indicated as branching within Clostridia ( Yutin and Galperin , 2013; Mavromatis et al . , 2009; Vesth et al . , 2013 ) or at the base of Bacilli ( Kunisawa , 2015 ) . The phylogenetic placement of Halanaerobiales remains also uncertain , as they have been assigned either to Class Clostridia ( Cayol et al . , 1994 ) , as a deep branch in the Firmicutes ( Mavromatis et al . , 2009; Vos et al . , 2009; Kunisawa , 2015 ) , or left unresolved ( Yutin and Galperin , 2013 ) . Finally , no detailed genomic analysis has been carried out to infer and compare the characteristics of the cell envelopes of several Halanaerobiales and Negativicutes . The large number of Negativicutes and Halanaerobiales genomes currently available prompted us to carry out a global phylogenomic study . This allowed to robustly clarifying the relative placement of Negativicutes and Halanaerobiales within the Firmicutes , to assess the evolutionary relationships of their cell envelopes , and to perform in depth comparative analysis to understand the characteristics of key OM-related processes in these two lineages . Our results provide robust support for an emergence of monoderm Firmicutes from diderm ancestors via multiple independent losses of the OM . Although the presence of an OM has been previously shown by electron microscopy for members of Negativicutes ( e . g . Tocheva et al . , 2011 ) and Halanaerobiales ( e . g . Zhilina et al . , 2012 ) , these images have been obtained separately and with different techniques , making difficult their comparison . We therefore obtained electron microscopy images of one representative of Negativicutes ( Megamonas rupellensis ) and one of Halanaerobiales ( Halanaerobium saccharolyticum ) . We used transmission electron microscopy ( TEM ) following high-pressure freezing , freeze substitution , plastic embedding and ultrathin sectioning of the samples ( see Materials and Methods ) . The application of high-pressure freezing in combination with appropriate freeze-substitution protocols facilitates the ultrastructural analysis of microorganisms and their membranes and also results in densely and homogeneously packed cytoplasm ( McDonald , 2007; McDonald et al . , 2007a; Rachel et al . , 2010 ) . Ultrathin sections of high-pressure frozen cells of M . rupellensis and H . saccharolyticum confirmed the presence of clearly diderm-type cell envelope architecture in both strains ( Figure 1 ) . In a cross section from inside to outside , the densely packed cytoplasm of M . rupellensis is surrounded by a cytoplasmic membrane followed by the periplasm with a thin peptidoglycan layer and an OM ( Figure 1A ) . Furthermore , pilus-like structures could be detected , as well as another electron dense layer outside the OM , which might correspond either to the lipopolysaccharide ( LPS ) or an S-layer . Thin sections of H . saccharolyticum also revealed a diderm cell envelope with a densely packed cytoplasm enclosed by a membrane surrounded by a relatively electron lucent periplasm so that the thin line representing the peptidoglycan is clearly visible ( Figure 1B ) . For both organisms , in some cases the periplasm appeared inflated ( Figure 1C and D ) , which was most likely caused by a preparation artifact due to swelling of the cells in the freeze substitution process . This effect nevertheless enabled us to observe the peptidoglycan much better as compared to cells without that artifact . 10 . 7554/eLife . 14589 . 003Figure 1 . Transmission electron microscopy of a member of Negativicutes and a member of Halanaerobiales . Ultrathin sections of high-pressure frozen cells of the Negativicutes member Megamonas rupellensis ( A , C ) , and the Halanerobiales member Halanaerobium saccharolyticum ( B , D ) . A Gram-negative like cell wall architecture is visible for both taxa ( A , B ) : a cytoplasmic membrane ( CM ) surrounding the cytoplasm ( C ) , a thin peptidoglycan layer ( PG ) , and an outer membrane ( OM ) . Pili-like structures ( P ) are also visible in M . rupelllensis . In some cases and due to a preparation artifact caused by swelling of the cells , the OM detaches from the IM creating an enlarged periplasmic space ( PP ) between two dividing cells ( C , D ) . In these cases , the peptidoglycan becomes more apparent as it is also the case for an electron dense surface coat ( SC ) , which might represent lipopolysaccharide ( LPS ) or a potential S-layer . Scale bars: 200 nm ( A , C ) and 100 nm ( B , D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14589 . 003 We gathered homologues of 47 ribosomal proteins from a local database of 205 Firmicutes taxa and 13 bacteria belonging to eight major phyla as outgroup ( Materials and methods ) . We did not include the Tenericutes in the analysis , because their reduced genomes and fast evolutionary rates are likely to cause artifacts in deep phylogenies , but it is known that they phylogenetically belong to the Bacilli ( Davis et al . , 2013 ) . We assembled the 47 ribosomal proteins into a large concatenated dataset ( 5551 amino acid characters ) and carried out Bayesian analysis with a sophisticated site-heterogeneous model of protein evolution ( CAT ) that allows each site to evolve under its own substitution matrix and is robust against tree reconstruction artifacts that frequently affect deep phylogenies ( Lartillot and Philippe , 2004 ) . The Bayesian tree was well resolved at most nodes ( Posterior Probabilities ( PP ) > 0 . 95 , Figure 2 ) . Despite the weak signal and the stochastic errors frequently associated to small proteins such as ribosomal ones , topology congruence tests on individual markers showed a largely congruent phylogenetic signal , especially at high rank taxonomy level , justifying their combined analysis ( Figure 2—figure supplement 1 and Materials and methods ) . Maximum Likelihood ( ML ) analysis of the same concatenated dataset and the site-homogeneous LG model ( Le and Gascuel , 2008 ) gave a largely consistent topology although it was much less resolved , especially at deep nodes ( Figure 2—figure supplement 2 ) . With respect to previous analyses , the relative placement of Negativicutes and Halanaerobiales in the Firmicutes phylogeny was robustly resolved ( Figure 2 ) . In fact , the Negativicutes branched within Class Clostridia , specifically related to Peptococcaceae and other incertae-sedis clostridial families ( PP = 1 , Figure 2 ) . This placement is consistent with previous analyses , although performed with less taxa ( Yutin and Galperin , 2013; Mavromatis et al . , 2009 ) . As opposed to the diderm nature of Negativicutes , the members of Peptococcaceae have monoderm phenotype ( Vos et al . , 2009 ) and no homologues of OM markers . 10 . 7554/eLife . 14589 . 004Figure 2 . Phylum-level phylogeny of the Firmicutes . Bayesian phylogeny of the Firmicutes based on a concatenation of 47 orthologous ribosomal proteins comprising 5551 amino acid positions and the CAT+GTR+Γ4 model . Values at nodes represent Bayesian posterior probabilities . The scale bar represents the average number of substitutions per site . For details on analyses , see Materials and methods . DOI: http://dx . doi . org/10 . 7554/eLife . 14589 . 00410 . 7554/eLife . 14589 . 005Figure 2—figure supplement 1 . Results of IC congruence test for the 47 ribosomal proteins . IC values were mapped onto the ribosomal protein concatenation phylogeny shown in Figure 2 . Branches in red indicate congruence among markers according to IC tests . Raw results of the test are provided as Additional Data in Dryad ( Antunes et al . , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14589 . 00510 . 7554/eLife . 14589 . 006Figure 2—figure supplement 2 . Maximum likelihood phylogeny of the Firmicutes . The tree was obtained by PhyML 3 . 0 based from the same concatenation of 47 orthologous ribosomal proteins as the Bayesian tree in Figure 2 and the LG+Γ4 model . Values at nodes represent non-parametric bootstrap values calculated on 100 replicates of the original dataset . The scale bar represents the average number of substitutions per site . DOI: http://dx . doi . org/10 . 7554/eLife . 14589 . 00610 . 7554/eLife . 14589 . 007Figure 2—figure supplement 3 . Results of AU test for 12 alternative topologies . Full results of the test are provided as Additional Data in Dryad ( Antunes et al . , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14589 . 007 In contrast , the Halanaerobiales emerged as a distinct , well-supported , and deep-branching lineage of the Firmicutes ( PP = 0 . 99 , Figure 2 ) , robustly grouped with the order Natranaerobiales ( PP = 1 , Figure 2 ) . This clustering was also observed in a previous analysis performed with only one member of Halanaerobiales and one of Natranaerobiales , and its position in the Firmicutes phylogeny was left unresolved ( Yutin and Galperin , 2013 ) . Natranaerobiales are a poorly known group of moderately halophilic Firmicutes that appear monoderm under the microscope ( Mesbah et al . , 2007 ) and have no homologues of OM markers . In order to verify the robustness of the distinct branching of the two diderm Firmicutes lineages , we ran AU tests on 12 topologies alternative to the Bayesian ribosomal protein concatenate tree , where Negativicutes or Halanaerobiales were moved 'up and down' the six nodes separating them ( N1-N6 for the topologies involving moving the Negativicutes; H1-H6 for the topologies involving moving the Halanaerobiales Figure 2—figure supplement 3 , Materials and methods and Additional Data ) . Unfortunately , topology testing is currently only available in a Maximum Likelihood framework with site-homogeneous models . Accordingly , these tests should reflect the poor resolution of the deep nodes of the maximum likelihood tree ( Figure 2—figure supplement 2 ) . Nevertheless , the Bayesian topology of Figure 2 ( H0N0 ) was the preferred one ( Figure 2—figure supplement 3 ) . Two alternative topologies only ( N1 and N2 ) were not rejected by the data , where the Negativicutes branched earlier in the Clostridia . Importantly , the two alternative topologies presenting a clustering of Halanaerobiales and Negativicutes ( H6 and N6 ) were strongly rejected by the data , as well as all topologies where the Halanaerobiales were moved away from the root of the Firmicutes tree ( H2-H5 ) ( Figure 2—figure supplement 3 and Additional Data ) , consistent with the separate origins of the two diderm lineages . To sum up , our phylogenetic analysis shows that Halanaerobiales and Negativicutes have distinct evolutionary origins , and are each related to different monoderm Firmicutes lineages . In contrast to the distinct emergence of Halanaerobiales and Negativicutes in the Firmicutes , the presence in their genomes of markers related to OM biogenesis and functioning ( Campbell et al . , 2014; Mavromatis et al . , 2009; Tocheva et al . , 2011 ) clearly indicates that their diderm cell envelopes are homologous structures . However , as discussed in the Introduction section , the specific evolutionary relationships between the OMs of Halanaerobiales and Negativicutes have been unclear . Interestingly , synteny analyses revealed a large genomic locus that is conserved between Halanaerobiales and Negativicutes , and is not present in monoderm Firmicutes ( Figure 3 and Supplementary file 1 ) . Other than LPS synthesis and transport ( green ) , the genes belonging to this genomic locus encode a number of cell envelope systems , such as OMP assembly ( blue ) , motility ( light pink ) , OM-PG attachment ( red ) , efflux ( purple ) , but also a number of hypothetical proteins ( brown ) , and proteins not known to be specifically related to the OM ( white ) . 10 . 7554/eLife . 14589 . 008Figure 3 . Conserved genomic locus for cell envelope components . Co-localization of the genes coding for LPS synthesis and transport , OMP assembly and structural OMPs in the Negativicutes and the Halanaerobiales . Representatives of the 2 families of Negativicutes and the 2 families of Halanaerobiales are shown ( for full distribution and accession numbers see Supplementary file 1 ) . Genes are colored according to their functional class: LPS synthesis and transport ( green ) , OMP assembly ( blue ) , flagellum ( light pink ) , OM-PG attachment ( red ) , hypothetical ( brown ) , efflux ( purple ) ( see text for discussion ) . White boxes indicate proteins not known to being related to the OM or non-conserved proteins whose connection with the OM is unclear . The figure was obtained by EasyFig ( Sullivan et al . , 2011 ) , where vertical lines represent BLAST hits with a cutoff of 0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 14589 . 008 Such clustering is unusual as in E . coli for example these genes are scattered in different regions of the genome . However , the genes coding for the first four steps of LPS synthesis ( lpxABCD ) display a conserved synteny in diderm Bacteria at very large evolutionary distances , suggesting that they have similar evolutionary histories ( Opiyo et al . , 2010 ) . Accordingly , synteny is also conserved in Halanaerobiales and Negativicutes ( Figure 3 and Supplementary file 1 ) . We therefore searched for these four core LPS genes ( lpxABCD ) in a local databank of 121 genomes representative of 30 major bacterial phyla ( Materials and methods and Supplementary file 2 ) . As compared to what could be previously inferred from available genomic data ( Sutcliffe , 2010; Opiyo et al . , 2010 ) , we show the presence of homologues of the four core LPS coding genes in 26 major bacterial phyla , eight of which evidenced for the first time: Thermodesulfobacteria , Fibrobacteres , Ignavibacteria , Nitrospina , Chrysiogenetes , Cloacimonetes , Atribacteria , and Armatimonadetes ( Supplementary file 2 ) . This suggests that LPS-diderm cell envelopes might be even more widespread in Bacteria than currently thought , leaving only four major phyla that appear to lack the coding capacity for LPS: Thermotogae , Caldiserica , Chloroflexi/Thermomicrobia , and Actinobacteria ( Supplementary file 2 ) . We assembled the four core LPS protein homologues into a concatenated dataset ( 898 amino acid characters ) also including Halanaerobiales and Negativicutes , and obtained a Bayesian tree with the CAT+GTR+Γ4 evolutionary model ( Figure 4 , Materials and methods ) . In agreement with their conserved synteny , congruence tests showed that these four core LPS genes have a consistent phylogenetic signal at large evolutionary distances , in particular concerning the monophyly of major bacterial phyla , justifying their combined analysis ( Materials and methods and Figure 4—figure supplement 1 ) . Consistently with the notorious difficulty in resolving the global phylogeny of Bacteria , the tree is not completely resolved . However , it is largely in agreement with bacterial systematics , showing the monophyly of major phyla ( Figure 4 ) . This pattern indicates that the core LPS genes were present in the ancestor of each of these diderm phyla , and that inter-phylum horizontal gene transfers were surprisingly rare during bacterial evolution . Consistently , the Halanaerobiales and Negativicutes also form a well-supported monophyletic cluster ( PP = 1 , Figure 4 ) , with internal branching pattern matching their respective reference species phylogeny shown in Figure 2 . These results indicate that the LPS-OM of Halanaerobiales and Negativicutes do not have distinct origins , but rather that , similarly to the other main diderm bacterial phyla , they were inherited from their common ancestor , which is also the ancestor of all Firmicutes , in agreement with Tocheva et al . , 2011 . The inclusion of a second diderm lineage in our analysis allows us to strengthen and extend this scenario , and to infer that present-day monoderm Firmicutes would have emerged from diderm ancestors via not less than five independent losses of the OM ( Figure 4B ) . The two alternative topologies that were not rejected by AU tests do not affect the inference of a diderm ancestor and imply four and three independent OM losses , respectively ( Figure 2—figure supplement 3 and Additional Data ) . 10 . 7554/eLife . 14589 . 009Figure 4 . Phylogenetic tree of core LPS components . ( A ) Bayesian phylogeny based on a concatenation of orthologs of the four core components of the LPS biosynthesis pathway ( lpxABCD ) , comprising 898 amino acid positions and the CAT+GTR+Γ4 model . Values at nodes represent Bayesian posterior probabilities . The scale bar represents the average number of substitutions per site . For details on analyses , see Materials and methods . ( B ) Schematic representation of the Firmicutes phylum-level phylogeny from Figure 2 , onto which putative losses of the OM are mapped ( red crosses ) . See text for discussion . DOI: http://dx . doi . org/10 . 7554/eLife . 14589 . 00910 . 7554/eLife . 14589 . 010Figure 4—figure supplement 1 . Results of IC congruence test for the 4 LPS core proteins . IC values were mapped onto the LPS core proteins concatenation phylogeny shown in Figure 3 . Branches in red indicate congruence among markers ( IC values>1 ) . Full results of the test are provided as Additional Data in Dryad ( Antunes et al . , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14589 . 010 Our phylogenetic analyses suggest that the diderm cell envelopes of Halanaerobiales and Negativicutes might be the remnants of ancient bacterial structures that were inherited from the Firmicutes ancestor . In the absence of experimental characterization , exploration of genomic data can guide inferences on the nature of these atypical diderm cell envelopes . To this aim , we investigated a few key processes that are related to OM biogenesis and functioning and are shared between Negativicutes and Halanaerobiales . Because OM markers frequently display low sequence conservation or are part of large membrane-related protein families often preventing the building of robust phylogenies , we helped tentative annotation by merging information obtained from homology to known OM markers , the presence of specific protein domains , and genomic synteny . In this respect , the presence of the conserved OM locus helped annotation greatly and provided important insights into the unique nature of the cell envelopes of Negativicutes and Halanaerobiales , which show both specific characteristics as well as an intriguing combination of diderm and monoderm features ( Figure 5 ) . 10 . 7554/eLife . 14589 . 011Figure 5 . Sketched diagram of inferred characteristics of the diderm Firmicutes cell envelope . The main processes discussed in the text are shown schematically . Components that were not detected in the genomes of Negativicutes and Halanaerobiales are indicated with a dashed outline and grey font . DOI: http://dx . doi . org/10 . 7554/eLife . 14589 . 01110 . 7554/eLife . 14589 . 012Figure 5—figure supplement 1 . Flagellar gene cluster of Negativicutes and Halanaerobiales . Structure of the region coding for flagellar components in representative members of Negativicutes and Halanaerobiales , and its conservation with respect to their closely related monoderm relatives Therminicola potens , and Natranaerobius thermophiles , respectively . By comparison is shown the structure of the operon in Escherichia coli as representative of a classical diderm . Colors are only meant to highlight synteny . DOI: http://dx . doi . org/10 . 7554/eLife . 14589 . 01210 . 7554/eLife . 14589 . 013Figure 5—figure supplement 2 . Genomic context of the genes coding for flagellar rings in Halanaerobiales and Negativicutes . Structure of the region coding for components of the flagellar rings ( flgA , flgH , flgI ) in representative members of Negativicutes and Halanaerobiales , in comparison with their closely related monoderm relatives Therminicola potens , and Natranaerobius thermophiles , respectively . The genomic structures in Bacillus subtilis and Escherichia coli are also shown as the most studied models for monoderm and diderm flagella . Colors are only meant to highlight synteny . DOI: http://dx . doi . org/10 . 7554/eLife . 14589 . 01310 . 7554/eLife . 14589 . 014Figure 5—figure supplement 3 . Structure of the main Type IV pilus cluster in Negativicutes and Halanaerobiales . Structure of the region coding for components of the type IV pilus in representative members of Negativicutes and Halanaerobiales , in comparison with their closely related monoderm relatives Therminicola potens , and Natranaerobius thermophiles , respectively . The genomic structure in Pseudomonas aeruginosa is also shown as the most studied model for diderm type IV pili . Colors are only meant to highlight synteny . DOI: http://dx . doi . org/10 . 7554/eLife . 14589 . 014 The origin of the cell envelope has represented one of the most fascinating questions in evolutionary biology since decades , and has been widely discussed in the literature ( Blobel , 1980; Cavalier-Smith , 1987 , 2006; Errington , 2013; Griffiths , 2007; Gupta , 2011; Koch , 2003; Vollmer , 2012 ) . The main issue mostly revolves around the question of how and when an OM originated in Bacteria , and whether monoderm cell envelopes predate diderm cell envelopes or instead derived from them . The complexity of the diderm cell envelope with respect to a perceived more 'rudimentary' monoderm type , together with its higher resistance toward antibiotics , are arguments usually put forward to propose that the OM is a relatively late invention in Bacteria ( Koch , 2003; Gupta , 2011 ) . However , it is now evident that diderm phyla outnumber monoderm ones , and span a large fraction of bacterial diversity , including early emerging lineages ( Sutcliffe , 2010; Errington , 2013 ) . Unfortunately , the evolutionary relationships among monoderm and diderm bacterial phyla are presently unclear , and do not allow to clarify OM origins . In this respect , the existence of a major bacterial phylum –the Firmicutes- including both diderm and monoderm lineages , and whose evolutionary relationships can be resolved , provides a unique opportunity to address the issue . Our results provide support for the hypothesis that the Firmicutes are ancestrally diderm , and that the monoderm envelope is a derived cell structure that originated through OM loss , at least in this phylum . Although previously suggested ( Tocheva et al . , 2011 ) , the inclusion of both Halanaerobiales and Negativicutes in our analysis strengthens and extends this scenario . Our robust phylogeny of the Firmicutes indicates that Halanaerobiales and Negativicutes form two distinct lineages , each related to different monoderm relatives . This allows inferring that the OM was lost from three to five times independently in the Firmicutes , and is therefore not a unique event in the history of Bacteria that would have led to all present-day monoderm lineages , as proposed earlier ( Cavalier-Smith , 2006 ) . The deep branching of Halanaerobiales and the still limited genomic coverage for this group makes them a priority target for further exploration as their cell envelopes may retain ancestral characters . Our results confirm that the LPS-OMs of Negativicutes and Halanaerobiales are homologous structures that share a common origin and are evolutionarily related with those of other classically diderm bacteria , therefore excluding convergence . Indeed , we show that the core enzymatic apparatus for producing LPS is even more widespread than previously thought , and that the LPS-OM is an ancient feature that emerged once and was largely inherited vertically during bacterial evolution , including in Halanaerobiales and Negativicutes . This is unusual for cytosolic enzymes , and underlines the key importance of maintaining cell-envelope function . Clearly , the availability of genomic data from an ever-wider sampling of bacterial diversity is sensibly changing perspective on the evolution of bacterial cell envelopes . For example , by revealing the presence of an LPS-OM in the ancestor of Deinococcus/Thermus we show that this phylum does not represent a monoderm-to-diderm intermediate ( Gupta , 2011 ) , but rather that LPS was lost in some of their members , a process similar to what likely occurred in Thermotogae ( Cavalier-Smith , 2006 ) . This indicates that , although having an OM is surely advantageous in certain conditions , the diderm cell envelope is a flexible structure that can be modified/simplified during evolution . Although the presence of a large cluster coding for key OM functions might be seen as supporting the hypothesis of an acquisition of the OM of diderm Firmicutes via genetic transfer from a diderm bacterium ( Mavromatis et al . , 2009; Campbell et al . , 2014 ) , our data weaken this hypothesis . Given the distinct branching of the two diderm lineages in the Firmicutes phylogeny , and the pattern of LPS inheritance similar to all other diderm bacterial phyla , the gene transfer hypothesis would imply a complex scenario consisting of two independent transfers of a very large genomic region , a first one to the ancestor of Halanaerobiales or to the ancestor of Negativicutes , and a second one between these two ancestors , which would have had to coexist at the same time and in the same environment . The sudden acquisition of an OM has been already discussed as mechanistically complicated ( Cavalier-Smith , 2006 ) . This would have in fact required the dramatic modification of an originally monoderm cell envelope through the concerted acquisition of several complex systems at once , not to mention the replacement of the native inner membrane components of these systems or their coordination with the newly acquired ones . Moreover , not all OM systems involved in OM biogenesis and function in Negativicutes and Halanaerobiales are part of the gene cluster , and their detailed annotation suggests that their cell envelopes share characteristics with deep-emerging bacterial phyla ( e . g . the OmpM porin instead of Braun’s lipoprotein for OM tethering ) , and present a peculiar combination of monoderm/diderm features ( e . g . flagella , pili ) . The OM gene cluster may therefore represent an ancestral genomic locus for OM–related functions , and it becomes essential to obtain further experimental characterization of the OM of Halanaerobiales and Negativicutes , as well as many other poorly explored deep emerging bacterial phyla . Nevertheless , we have analyzed here only a few OM systems shared between Halanaerobiales and Negativicutes in order to infer the nature of the ancestral diderm cell envelope in the Firmicutes . There are surely additional components that are lineage-specific and may have been acquired from diderm bacteria thriving in the same environment . This is an important future area of investigation , as it could inform on how the presence of an OM in a Firmicutes background may have helped adaptation to specific niches , including the human environment . By which process the ancestral OM would have been lost multiple times independently in the majority of present-day Firmicutes remains to be understood . It has been proposed that a spontaneous mutation leading to hypertrophy of the peptidoglycan layer would be sufficient to transform a diderm into a monoderm , through disruption of the attachment of the OM , leading to its loss ( Cavalier-Smith , 2006 ) . Alternatively , we speculate that mutations may have affected the ancestral OmpM , causing a de-regulation of OM-PG attachment . This transition may have been made easier during the process of sporulation , where an OM is transiently formed and lost when the vegetative cell matures . Tocheva et al . ( 2011 ) proposed indeed that the OM might have been lost in the Firmicutes to increase sporulation and germination efficiency ( Tocheva et al . , 2011 ) . A link between OM loss and sporulation may explain why there is no current evidence of monoderm lineages within other diderm phyla that do not sporulate . Further genomic and experimental data on the closest monoderm relatives of Negativicutes and Halanaerobiales might provide key information on the process of OM loss . Our results suggest that the cell envelopes of diderm Firmicutes might be the remnants of an ancient type of cellular structure , adding up to the ones found in the major diderm bacterial phyla . Moreover , they seem to have ancestral and simpler cell envelope systems with respect to the well-studied Proteobacteria . Halanaerobiales and Negativicutes are therefore promising new experimental models that will provide precious insights into the processes that have shaped the diversity not only of diderm cell envelopes , but also of monoderm ones . Negativicutes strain Megamonas rupellensis DSM 19944T was grown anaerobically at 37°C to late exponential phase on TGY broth ( w/v; 3% tryptone , 2% yeast extract , 0 . 5% glucose , 0 . 05% L-cysteine hydrochloride ) as described previously ( Chevrot et al . , 2008 ) . Halanaerobiales strain Halanaerobium saccharolyticum subsp . saccharolyticum DSM 6643T was grown anaerobically to late exponential phase at 37°C on a synthetic medium as described in ( Zhilina et al . , 2012 ) . The ultrastructure of the respective bacterial strains was determined via transmission electron microscopy ( TEM ) following high-pressure freezing , freeze substitution , plastic embedding and ultrathin sectioning of the samples . Due to a relatively long transportation time to the high-pressure freezer , cells were pre-fixed with 2% glutaraldehyde . Afterwards , they were centrifuged for 10 min at 14 . 000 x g , the supernatant was discarded and the resulting pellet was resuspended in 50µl growth medium . From this cell suspension , 2µl were high-pressure frozen and freeze substituted as described in ( Peschke et al . , 2013 ) . For substitution , acetone containing 0 . 2% OsO4 , 0 . 25% uranyl acetate and 5% ( vol/vol ) H2O was used . Embedding of the cells , sectioning and post-staining was carried out as described in ( Rachel et al . , 2010 ) . For negative staining of bacteria , 2% uranyl acetate was used for contrast enhancement following pre-fixation with 2% glutaraldehyde ( Rachel et al . , 2010 ) . Finally , transmission electron microscopy was performed either on a JEOL JEM 2100 , operated at 120 kV in combination with a fast-scan 2k x 2k CCD camera F214 ( TVIPS , Gauting , Germany ) for negatively stained samples or on a Zeiss EM 912 equipped with an integrated OMEGA energy filter and operated at 80 kV in the zero-loss mode for ultrathin sections . We assembled a local databank of 205 complete genomes from a wide representative sampling of Firmicutes , including 38 Negativicutes and 7 Halanaerobiales genomes available at the beginning of this analysis ( Supplementary file 1 ) . Exhaustive HMM-based homology searches were carried out on this genome databank by using the HMMER package ( Johnson et al . , 2010 ) and as queries the HMM profiles of the complete set of 54 bacterial ribosomal proteins from the Pfam 29 . 0 database ( http://pfam . xfam . org , Finn et al . , 2016 ) . Additional searches with tblastn ( Altschul et al . , 1997 ) were used to identify eventually misannotated homologues in some genomes . Because it is unclear which bacterial phylum is closest to the Firmicutes , we included as outgroup 13 taxa representatives of eight major bacterial phyla ( 2 Actinobacteria; 2 Cyanobacteria; 1 Deinococcus; 2 Proteobacteria; 1 Spirochaetes; 3 Flavobacteria/Bacteroidetes/Chlorobi; 2 Plactomyces/Chlamydia ) . Seven ribosomal proteins ( S2 , S4 , S14 , S21 , L25 , L30 , L33 ) that were absent from >50% of the considered genomes or had paralogous copies making difficult the identification of orthologues were discarded from the analysis . The remaining 47 single protein data sets were aligned with MUSCLE v3 . 8 . 31 ( Edgar , 2004 ) with default parameters , and unambiguously aligned positions were selected with BMGE 1 . 1 ( Criscuolo and Gribaldo , 2010 ) and the BLOSUM30 substitution matrix . Trimmed datasets were concatenated by allowing a maximum of 11 missing sequences per taxon into a large character supermatrix ( 218 taxa and 5551 amino acid characters ) . PhyloBayes v3 . 3b ( Lartillot et al . , 2009 ) was used to perform Bayesian analysis using the evolutionary model CAT+GTR+Γ4 . Two independent chains were run until convergence , assessed by evaluating the discrepancy of bipartition frequencies between independent runs . The first 25% of trees were discarded as burn-in and the posterior consensus was computed by selecting one tree out of every two . A Maximum likelihood ( ML ) tree was also calculated from the ribosomal protein concatenate with PhyML v3 . 0 ( Guindon et al . , 2010 ) and the evolutionary model LG+Γ4 ( Le and Gascuel , 2008 ) as suggested by the Akaike Information Criterion implemented in ProtTest 3 ( Darriba et al . , 2011 ) . Branch supports were estimated by standard nonparametric bootstrap based on 100 replicates . In order to assess whether the ribosomal proteins carried a congruent phylogenetic signal , we compared each of the 47 corresponding individual gene trees with the Bayesian ribosomal protein concatenate tree by using the recently proposed 'Internode Certainty' ( IC ) criterion , which measures the existence of statistically supported conflicting splits between trees ( Kobert et al . , 2016 ) . ML phylogenetic trees of individual genes were inferred by IQ-TREE v1 . 3 . 12 ( Nguyen et al . , 2015 ) with evolutionary model selected by optimizing the Akaike information criterion . In order to minimize the negative impact on IC estimation of the large irresolution within most of the single gene trees caused by the small number of aligned characters , all branches displaying <70% bootstrap support were collapsed . IC values were then estimated by RAxML 8 . 2 . 8 ( Stamatakis , 2006 ) and reported on the concatenate tree . Significance of 12 alternative tree topologies was assessed by the approximately unbiased ( AU ) test ( Shimodaira , 2002 ) . Each alternative topology was obtained by moving specific nodes on the Bayesian concatenate ribosomal protein tree by using Seaview v4 . 6 ( Gouy et al . , 2010 ) . For each tree topology , log-likelihood per character was estimated by PhyML v3 . 0 with the evolutionary model LG+Γ4 . In order to estimate the AU test p-values associated to each topology , the resulting data were processed with CONSEL v0 . 20 ( Shimodaira and Hasegawa , 2001 ) with default parameters . For the LPS core gene analysis , homologues were searched by using Pfam HMM profiles for LpxA , LpxB , LpxC , and LpxD . The same approach as the one described above for ribosomal proteins was used to assemble a 4-gene supermatrix of 898 unambiguously aligned amino acid characters , which was analysed by PhyloBayes with the evolutionary model CAT+GTR+Γ4 . Congruence among the four markers was assessed by the IC criterion as described above . Given the small number of markers analyzed and their frequently limited conservation at the sequence level , we followed a semi-manual procedure for annotation based on a combination of profile-based homology searches , protein domain identification , and genomic synteny . Profile-based homology searches of specific OM markers were performed by using the HMMER package ( Johnson et al . , 2010 ) . Initial searches were conducted by HMM on the local Firmicute genome databank by using standard Pfam domain models corresponding to a given protein of interest . The top-scoring hits were used to build new HMM models and perform a novel search . This process was iterated until no new hits were found . The resulting homologues were aligned and manually inspected in order to confirm homology and to eliminate divergent , partial or non-homologous sequences . Additional searches with tblastn ( Altschul et al . , 1997 ) were used to identify eventually misannotated homologues in some genomes . Protein domains were inspected by querying the Conserved Domain Database ( CDD ) at NCBI ( Marchler-Bauer et al . , 2015 ) . Genomic synteny was investigated using the interactive web-based visualization tool SyntTax ( Oberto , 2013 ) . Local genomic alignments were generated using EasyFig ( Sullivan et al . , 2011 ) with a BLAST cutoff E-value of 0 . 0001 . For localization prediction , we used the PSORT v3 . 0 server ( http://www . psort . org/psortb/ , Yu et al . , 2010 ) with default settings for Gram-negative Bacteria . Protein families were assigned to Clusters of Orthologous Groups by searching the COG database , which was downloaded from the NCBI FTP server ( ftp://ftp . ncbi . nih . gov/pub/wolf/COGs/ ) ( Tatusov et al . , 2000 ) . While this manuscript was in the last phase of revision , Tocheva et al . published a Perspective paper ( Tocheva , EI , Ortega , DR & G Jensen . 2016 . Sporulation , bacterial cell envelopes and the origin of life . Nature Reviews Microbiology 14 , 533-542 . doi:10 . 1038/nrmicro . 2016 . 85 ) . It extends the discussion of their previous hypothesis ( Tocheva et al . , 2011 ) by focusing on the origin of the outer membrane , and prompts for further genomic and evolutionary analysis , which is timely addressed in the present work .	The cell envelope is one of the evolutionarily oldest parts of a bacterium . This structure – made up of a cell wall and either one or two cell membranes – surrounds the bacterial cell , maintaining the cell’s structure and providing an interface through which bacteria can sense their environment and communicate . Bacteria can be broadly classed based on the number of cell membranes that their envelope consists of . Bacteria that have a single cell membrane are known as “monoderm” , whereas those with two membranes are termed “diderm” . The number of membranes that bacteria have can affect how well they resist antibacterial compounds . When , how and why bacteria switched between monoderm and diderm cell envelopes are some of the major unanswered questions in evolutionary biology . The textbook example of a monoderm cell envelope can be found in bacteria called Firmicutes . This group includes some notoriously harmful bacteria such as Staphylococcus , which can cause conditions ranging from abscesses to pneumonia . However , some Firmicutes possess two cell membranes . It was unclear how these unusual diderm Firmicutes developed a second membrane , and how they are related to their monoderm relatives . Antunes , Poppleton et al . set out to answer these questions by analyzing the information contained in the thousands of bacterial genomes that have already been described . The results indicate that Firmicutes originally had diderm envelopes , and that species with monoderm envelopes arose independently several times through the loss of their outermost membrane . Future work is needed to investigate the driving forces and the precise mechanism that led most Firmicutes to lose their outer membrane . Also , further characterization of diderm Firmicutes will provide key information about the biology of these poorly understood bacteria .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "cell", "biology" ]	2016	Phylogenomic analysis supports the ancestral presence of LPS-outer membranes in the Firmicutes
Modification of the outer membrane charge by a polymyxin B ( PMB ) -induced PmrAB two-component system appears to be a dominant phenomenon in PMB-resistant Acinetobacter baumannii . PMB-resistant variants and many clinical isolates also appeared to produce outer membrane vesicles ( OMVs ) . Genomic , transcriptomic , and proteomic analyses revealed that upregulation of the pmr operon and decreased membrane-linkage proteins ( OmpA , OmpW , and BamE ) are linked to overproduction of OMVs , which also promoted enhanced biofilm formation . The addition of OMVs from PMB-resistant variants into the cultures of PMB-susceptible A . baumannii and the clinical isolates protected these susceptible bacteria from PMB . Taxonomic profiling of in vitro human gut microbiomes under anaerobic conditions demonstrated that OMVs completely protected the microbial community against PMB treatment . A Galleria mellonella-infection model with PMB treatment showed that OMVs increased the mortality rate of larvae by protecting A . baumannii from PMB . Taken together , OMVs released from A . baumannii functioned as decoys against PMB . Pathogenic Acinetobacter baumannii strains cause urinary tract infections and ventilator-associated pneumonia ( Wright et al . , 2014 ) . The incidence and prevalence of multidrug-resistant A . baumannii ( MRAB ) , which are resistant to various antibiotics including sulbactam and tigecycline , threaten the effective prevention and treatment of MRAB-related infections ( Brauers et al . , 2005; Perez et al . , 2007; Alsan and Klompas , 2010 ) . The broad-spectrum β-lactam carbapenem antibiotics , such as imipenem , meropenem , and doripenem , are mainly used to treat severe MRAB infections ( Papp-Wallace et al . , 2011; Kang et al . , 2012 ) . Unfortunately , studies have reported the spread of MRAB clonal complex 92 , which contains the blaOXA-23 gene encoding a carbapenem-hydrolyzing class D β-lactamase ( Antunes et al . , 2014; Yoon et al . , 2017 ) . Mobile genetic elements , particularly ISAba I upstream of the blaOXA-type genes , allow A . baumannii strains to transmit the antibiotic resistance ( AR ) gene to other bacteria and the environment through horizontal gene transfer ( Bahador et al . , 2015 ) . Carbapenem-resistant A . baumannii ( CRAB ) is a top-priority pathogen , for which new antibiotics or combination therapies are desperately needed ( Shlaes and Bradford , 2018 ) . Lack of an efficacious treatment for MRAB and CRAB infections has caused a reversion to initial treatments involving drugs that are considered dated now , such as colistin and polymyxin B ( PMB ) , despite the adverse effects of these drugs ( Velkov et al . , 2013 ) . Polymyxins are cationic polypeptide antibiotics that specifically kill Gram-negative bacteria by binding to the bacterial outer membrane ( OM ) . These last-resort antimicrobial agents interact with the negatively charged lipopolysaccharide ( LPS ) components of the OM by displacing positively charged ions , such as Mg2+ and Ca2+ . Consequently , the integrity of the cell membrane is impaired , leading to cell death ( Jeannot et al . , 2017; Moffatt et al . , 2019 ) . Modification of lipid A in LPS by pmrCAB and arnBCADTEF gene products , which are modulators of the bacterial OM surface charge and permeability , is generally recognized as a primary mechanism of resistance to polymyxins ( Trent et al . , 2001; Olaitan et al . , 2014; Jeannot et al . , 2017 ) . The pmrCAB and arnBCADTEF operons regulated by a PmrAB two-component system modify lipid A by adding 4-amino-4-deoxy-l-arabinose ( LAra4N ) and phosphoethanolamine ( PetN ) , respectively ( Kline et al . , 2008; Arroyo et al . , 2011; Pelletier et al . , 2013 ) . Activation of the arnT operon ( arnBCADTEF , also called as the pmrHFIJKLM operon ) adds LAra4N groups to lipid A only in several Gram-negative bacteria such as Escherichia coli , Yersinia pestis , Pseudomonas aeruginosa , and Ralstonia solanacearum , but not in A . baumannii or Citrobacter rodentium ( Trent et al . , 2001; Sinha et al . , 2019 ) . Because A . baumannii lacks the genes for the biosynthesis of LAra4N , it uses PetN addition as the main polymyxin-resistance mechanism ( Gerson et al . , 2019 ) . In PMB-resistant A . baumannii , the mutant PmrB protein is autophosphorylated , consequently phosphorylating the cytoplasmic transcriptional regulator PmrA . Phosphorylated PmrA then upregulates the PetN transferase PmrC and prevents PMB binding to lipid A ( Cannatelli et al . , 2014 ) . These modifications in lipid A decrease the negative charge of the cell membrane , reducing the membrane-binding affinity of PMB . Consequently , the cell is protected from the PMB-mediated disruption of its membrane and survives ( Trimble et al . , 2016 ) . In some A . baumannii clinical isolates , expression of the eptA gene is activated by the adjacent ISAba I sequence , and the gene product PmrC homolog contributes to PMB resistance ( Trebosc et al . , 2019 ) . Interestingly , E . coli producing outer membrane vesicles ( OMVs ) can survive in the presence of PMB ( Manning and Kuehn , 2011; Kulkarni et al . , 2015; Kim et al . , 2018 ) . Gram-negative bacteria release OMVs with diameters of 20–400 nm to their environments . These vesicles are composed of phospholipids ( PLs ) , outer membrane proteins ( OMPs ) , and LPSs or lipooligosaccharides ( LOS ) , but also contain periplasmic proteins and cell wall components ( Roier et al . , 2016a ) . Modulation of peptidoglycan ( PG ) cross-links is associated with membrane stability and rigidity ( Schwechheimer and Kuehn , 2015 ) . Approximately one-third of the alpha-helical OM lipoprotein Lpp ( also known as Braun’s lipoprotein ) anchored to the outer cell membrane are covalently cross-linked with PGs through connection between C-terminal lysine residue of Lpp and the diaminopimelic acid ( meso-DAP ) site of PG ( Dramsi et al . , 2008 ) . Reduction in Lpp production disrupts the cross-linkage of PG on the membrane , thereby significantly contributing to the formation of OMVs ( Wessel et al . , 2013 ) . Differential expression of OM protein A ( OmpA ) , which is a β-barrel porin that non-covalently interacts with the meso-DAP site of PG , can modulate OMV excretion ( Schwechheimer and Kuehn , 2015 ) . Many mutant studies have shown that the absence of inner membrane ( IM ) proteins , such as the Tol component of the Tol-Pal system and NlpA , can enhance OMV production ( Bernadac et al . , 1998; Toyofuku et al . , 2015 ) . Production of OMVs is also facilitated by a high periplasmic turgor pressure , which occurs when misfolded proteins , nucleotide fragments , or PG debris accumulate in the periplasm ( Brown et al . , 2015 ) . Additionally , in Haemophilus influenzae , OMVs are also released by PLs that accumulate on the OM when the PL transporters of the VacJ/Yrb ABC transport system ( also known as the Mla system ) are not produced ( Roier et al . , 2016b ) . Changes in membrane fluidity and rigidity , determined by the compositions of acyl chains and head groups of PLs , affect OMV production ( Mashburn-Warren et al . , 2008 ) . The proportion of saturated fatty acids appears to be higher in OMVs than in cellular membranes ( Baumgarten et al . , 2012 ) . In pathogenic bacteria , these spherical OMVs formed by various routes can have many different physiological functions , including biofilm formation , cell–cell communication , secretion of virulence factors , and survival under antibiotic stress ( Toyofuku et al . , 2015; Bonnington and Kuehn , 2016 ) . OMVs have attracted interest as putative vaccine agents or vehicles for drug delivery ( Jan , 2017 ) . They can reduce the effectiveness of antibiotics and other bactericidal agents , thereby posing a serious threat to the treatment of bacterial infections ( Koeppen et al . , 2016; Jan , 2017 ) . For example , the OMVs of β-lactam-resistant E . coli are enriched with proteins involved in degrading β-lactam antibiotics in comparison with those of susceptible bacteria ( Kim et al . , 2018 ) . Accordingly , OMV supplementation of microbial cultures increases the microbial resistance to antimicrobial peptides . Similarly , hypervesiculating mutants show increased resistance ( Manning and Kuehn , 2011 ) . OMVs from A . baumannii are known to possess various virulence factors such as OmpA , proteases , phospholipases , superoxide dismutase , and catalase ( Kwon et al . , 2009 ) . However , it is unclear whether A . baumannii can gain polymyxin resistance or confer such resistance onto its bacterial community through hypervesiculation . Here , the genomes and OMV-producing phenotypes of experimental evolution PMB-resistant A . baumannii cells were analyzed alongside other clinical isolates as well as the A . baumannii ATCC 17978 strain . Our multi-omics data and mutational studies suggested that OMVs functioned as decoys for PMB in A . baumannii . Accordingly , these vesicles can protect not only the OMV producer , but also the entire bacterial community from the bactericidal effects of PMB . PMB-resistant strains were generated by exposing wild-type cells ( Lab-WT; minimum inhibitory concentration [MIC] , 2 μg/mL ) to increasing concentrations of PMB . Consequently , strains that had 2-fold ( MIC , 4 μg/mL ) and 64-fold ( MIC , 128 μg/mL ) higher PMB resistance than Lab-WT were generated ( low and high PMB resistance , PMRLow and PMRHigh , respectively ) ( Figure 1A , Figure 1—source data 1 ) . The PMRHigh strain had a lower growth rate ( 1 . 386 ± 0 . 17 h−1 ) than both Lab-WT ( 2 . 081 ± 0 . 92 h−1 ) and PMRLow ( 1 . 985 ± 0 . 86 h−1 ) strains ( Figure 1—figure supplement 1A ) . Expression analyses of the pmrC gene , which encodes a PetN transferase in the absence of PMB , showed 1 . 2- and 4-fold higher expression in PMRLow and PMRHigh strains , respectively ( Figure 1—figure supplement 1B ) . The zeta-potential analyses suggested that the average negative surface charge of PMRLow and PMRHigh decreased from −27 mV ( Lab-WT ) to −20 mV and −10 mV , respectively ( Figure 1B , Figure 1—source data 1 ) . The reduced surface charge , which is partly due to PmrC-driven incorporation of PetN into lipid A , presumably decreases the initial binding of PMB , a cationic amphiphilic cyclic decapeptide , to the cell surface . Consequently , cells with more PetNs in lipid A acquire resistance to PMB . PmrC-mediated reduction of surface charges was not feasible in PMRLow because the expression level of the pmrC gene was not significant . These observations could not account for all the reasons underlying the high PMB resistance of PMRHigh strain because short-term ( 10 min ) PMB exposure ( 4 μg/mL , MIC of PMRLow ) decreased the surface charge in all tested strains , albeit to different extents ( Figure 1—figure supplement 1C , n = 10 per strain ) . OMV-associated biofilm was measured in the Lab-WT and PMRHigh strains ( Figure 1—figure supplement 1D , E ) . Biofilm formation observed using confocal microscopy with FilmTracer indicated that the PMRHigh strain produced thick and condensed biofilm within 24 hr , and the addition of PMB induced biofilm formation in both strains ( Figure 1—figure supplement 1D ) . Quantification of biofilm formation using the crystal violet assay also indicated that the PMRHigh strain produced more biofilm ( twofold higher ) than the Lab-WT strain ( Figure 1—figure supplement 1E ) . Interestingly , addition of OMVs derived from PMRHigh increased the initial stage of Lab-WT biofilm formation ( Figure 1—figure supplement 1F , G ) . The PMRLow and PMRHigh strains were resistant only to polymyxins , but not to the other nine tested antibiotics of different classes , including meropenem , rifampicin , chloramphenicol , and ampicillin ( Figure 1—figure supplement 2 ) . This PMB-specific phenotype prompted us to confirm hypervesiculation of the PMR-resistant strains through transmission electron microscopy ( TEM ) . OMV production appeared to be twofold higher in PMRHigh than in Lab-WT , and the treatment with 1/2 MIC PMB ( 1 or 128 µg/mL , respectively ) caused increased OMV production as quantified using the lipophilic dye FM4-64 ( Figure 1C ) . We observed markedly larger and greater production of OMVs in the PMRHigh strain than in our Lab-WT strain using TEM ( Figure 1D , Figure 1—figure supplement 1H ) . Dissociated OMVs from PMRHigh were larger and became entangled with other OMVs through hairy structured materials ( Figure 1E ) . To determine whether modified lipid A plays an important role in PMB resistance , lipid A was subjected to matrix-assisted laser desorption/ionization time-of-flight ( MALDI-TOF ) analysis in the positive-ion mode . The lipid A extraction method is summarized in Materials and methods . The main peak of Lab-WT was measured at 1490 m/z , corresponding to bis-phosphorylated penta-acylated lipid A ( Figure 2A , B ) . In the PMRHigh strain , the predominant spectrum was also measured at 1490 m/z , followed by 1630 m/z , which is predicted to be the peak when a PetN and a hydroxyl group were attached to the dominant lipid A ( Figure 2A , B ) . The PMRHigh strain overexpressed pmrC ( encoding a PetN transferase ) , which added a PetN to some lipid A molecules , but not all , leading to interference PMB binding to the OM ( see Figure 1—figure supplement 1B ) . Raman spectroscopy is used for microbial analysis at a single-cell level since it can rapidly measure the cell components , including lipids , nucleic acids , and proteins , by detecting the inelastic scattering of a molecule irradiated by monochromatic lights . The phenylalanine peak ( 1001–1004 cm−1 ) appears to be a signature signal for the presence of peptides and proteins in a sample . This phenylalanine peak at 1002 cm−1 appeared to be similar in both strains ( Figure 2C ) . However , the Raman peak corresponding to PL ( at 1334 , 1445 , and 1666 cm−1 ) increased notably in PMRHigh ( Figure 2C ) . The intensity of the other prominent peak at 1243 cm−1 assigned to an amide III β-sheet was higher in resistant strains ( Figure 2C ) . Increased magnitude of Raman intensity at amide III , along with hydroxyl , carboxyl , and phosphoryl groups , was reported in bacterial biofilm , indicating profound changes in the bacterial cell membrane ( Jung et al . , 2014; Kim et al . , 2019 ) . The Raman peaks of amide III were increased in clinical quinolone-resistant E . coli ( Kim et al . , 2019 ) . Principal coordinate analysis ( PCoA ) using the collected Raman spectra of bacterial cells revealed that dispersion of individual dots was clustered into two groups , with 38 . 2% explanatory power , although some dots were in shared areas between Lab-WT and PMRHigh ( Figure 2D ) . Individual cell heterogeneity analysis showed that Lab-WT strains were more heterogeneous than resistant strains , suggesting that PMRHigh strains were specifically adapted and unified by PMB stress ( Figure 2E ) . We hypothesized that the higher production of OMVs secreted by the PMRHigh strain functioned as a decoy target for PMB . In this scenario , OMVs are expected to protect their mother cells as well as the entire bacterial community from the effect of PMB . To gain a better insight into the mechanism underlying PMB resistance and OMV biogenesis in the PMRHigh strain , whole-genome sequencing ( WGS ) analyses of PMRLow and PMRHigh were conducted ( Supplementary file 1b ) . It has been known that A . baumannii ATCC 17978 strain possesses a single chromosome and two plasmids , namely pAB1 ( 13 , 408 bp ) and pAB2 ( 11 , 302 bp ) . The same small plasmid , pAB2 ( 11 , 299 bp , three bases were deleted ) , was detected only in the PMRHigh . The large plasmid was missing in both evolved strains ( Figure 3—figure supplement 1A ) . The chromosome sizes of ATCC 17978 , PMRLow , and PMRHigh were 3 , 976 , 747 bp , 3 , 971 , 618 bp , and 3 , 955 , 017 bp , respectively ( Figure 3—figure supplement 1A ) . Gene alignments and comparative genomic analyses were performed on the reference ATCC 17978 strain ( Lab-WT ) . The WGS data were assembled to construct draft genomes with median N50 values of 3971 kb for PMRLow and 3955 kb for PMRHigh ( Supplementary file 1b ) . The numbers of coding DNA sequences ( CDSs ) in the PMRLow and PMRHigh genomes were 3678 and 3762 , with mean CDS lengths of 947 . 5 and 930 . 2 , respectively ( Supplementary file 1b ) . Both MUMmer program-based genomic and additional PCR analyses were performed to confirm mutational positions ( Supplementary file 1a , b ) . No mutation was detected in the promoter ( all intergenic ) regions of either mutant strain genome . Two chromosomal regions mostly annotated as hypothetical proteins were lost during adaptive evolution of PMRLow and PMRHigh strains ( Supplementary file 1c ) . All 20 antibiotic-resistant genes and 78 IS elements remain unchanged in both evolved strains ( Supplementary file 1d , e ) . We focused on PMRHigh rather than PMRLow because the first strain produced more OMVs than the latter . In the PMRHigh strain , mutations were found in the protein damage repair gene surE , fimbrial gene fimT , cell division-related gene ftsL ( nonsense mutation ) , DNA repair gene udg , DNA integration gene xerD , and fatty acid synthesis-related gene fabH ( nonsense mutation ) ( Supplementary file 1a , Figure 3—figure supplement 1A ) . Additionally , comparative genomic analysis between Lab-WT and PMRHigh strains revealed point mutations in the pmrB gene of the PMRHigh strain . PmrB was the protein that showed different mutations in the PMRLow and PMRHigh strains , with amino acid substitutions at N353I and T2351 , respectively ( Supplementary file 1a , Figure 3—figure supplement 1A ) . The mutations in N353I and T2351 in PmrB are known to increase the expression of the pmrC , resulting in PetN modification of lipid A ( Adams et al . , 2009 ) . Due to high MIC under PMB and hypervesiculation of the PMRHigh strain , we focused on comparative transcriptome between lab WT and PMRHigh strain ( Supplementary file 1f ) . Our RNA-seq analysis showed that 250 and 166 genes were up- and downregulated , respectively , in PMRHigh in comparison with the Lab-WT strain without PMB exposure ( Figure 3—figure supplement 1B ) . In addition , 89 . 16% of all genes showed no change at the transcriptional level . The highly upregulated gene , encoding an aldehyde dehydrogenase ( aldB ) , showed a 3 . 43-fold increase in the PMRHigh in comparison with that in the control ( Figure 3—figure supplement 1B ) . It is worth noting that dynamic membrane lipid biosynthesis in the PMRHigh might generate long-chain fatty aldehydes that may be recycled through AldB ( Asperger and Kleber , 1991; Ishige et al . , 2000 ) . 8 of the 16 most upregulated genes were hypothetical proteins , and the remaining genes were annotated as a glutamin- ( asparagin ) -ase ( ansB ) , poly-beta-1 , 6-N-acetyl-d-glucosamine N-deacetylase ( pgaB ) , ribosomal protein ( rpmI ) , and an ABC transporter glutamine-binding protein ( glnH ) ( Figure 3—figure supplement 1B ) . In particular , the PgaB belonging to the pgaABCD operon functions for exopolysaccharide export across the OM . The overexpressed pga operon appeared to facilitate the production of PNAG-attached OMVs in E . coli through deacetylation of PNAG by the PgaB activity , which resulted in enhanced biofilm formation ( Stevenson et al . , 2018 ) . The PMRHigh strain showed upregulation of both pmrA and pmrC genes encoding a transcriptional regulator and a PetN transferase , respectively , which evidently affects the membrane charge leading to PMB resistance ( Figure 3—figure supplement 1B ) . Notably , the most downregulated genes except the hypothetical gene were related to two genes encoding the OMPs ompA and bamE ( Figure 3—figure supplement 1B ) . OmpA , a PG-linked non-specific OM porin , harbors a flexible periplasmic domain that is strongly associated with PG ( Iyer et al . , 2018 ) . Lack of OmpA in the OM is known to promote bacterial OMV production ( Schwechheimer et al . , 2014 ) . The bamE , encoding a β‐barrel assembly machine E , was shown to bind specifically to phosphatidylglycerol ( PG ) in the OM ( Knowles et al . , 2011 ) . Both OmpA and BamE are determinants for cell shape and OM integrity ( Ryan et al . , 2010; Choi and Lee , 2019 ) . Downregulation of both genes may lead to loosening of OM-PG linkages and is associated with hypervesiculation in the PMRHigh . An additional qRT-PCR assay confirmed that the pmr and pgaB genes showed more than fourfold higher expression , but all membrane-linkage-related genes ( ompA , bamE , lpp , and mlaC ) were downregulated in PMRHigh ( Figure 3A , Figure 3—source data 1 ) . Biogenesis of OMV in the PMRHigh was twice as much as that in Lab-WT and further increased under 1/2 MIC PMB ( Figure 1C ) . To identify the commonly expressed genes in the high OMV production condition , we compared three conditions ( Lab-WT + PMB vs . Lab-WT ) , ( PMRHigh + PMB vs . PMRHigh ) , and ( PMRHigh vs . Lab-WT ) ( Figure 3—figure supplement 2 ) . Our RNA-seq analyses confirmed that 29 and 12 genes were up- and downregulated , respectively , in the hypervesiculation conditions ( Figure 3—figure supplement 2A ) . High levels of gene expression involved in stress response and survival were also predominant in the transcriptomic analysis , in which ribosomal proteins ( rplW and rplO ) and chaperons ( groL ) were upregulated in the hypervesiculation conditions ( Supplementary file 1g ) . The pgpA gene encoding a lipid phosphatase was also upregulated , which can dephosphorylate phosphatidylglycerophosphate to form PG , a major constituent in OMVs ( Supplementary file 1g , Lu et al . , 2011 ) . Increased levels of PG in the OM induced changes in membrane integrity , causing higher OMV production ( Sohlenkamp and Geiger , 2016 ) . Notably , several genes belonging to the pmr and pga operons are included among the commonly upregulated gene categories , and the commonly downregulated genes included important membrane-associated genes ( mlaC , ompA , lpp , and bamE ) ( Supplementary file 1g ) . Purified OMVs were separated by two-dimensional gel electrophoresis ( 2-DE ) followed by Coomassie blue staining . Gel-based proteomics with MALDI-TOF analysis were performed under the same four conditions ( Figure 3B , Supplementary file 1h , Figure 3—source data 2 and 3 ) . 43 out of the 61 up- or downregulated proteins were designated with the MS-Fit program by using the NCBInr and UniPortKB databases , and 18 spots were not identified ( Supplementary file 1h ) . The commonly identified OMV components harbored several proteins that were noted in our RNA-seq analyses for OMV biogenesis , showing increased production ( PmrC , alanine racemase [Alr] , and PgaB ) and decreased production ( OmpA and OmpW , elongation factor Tu [TufB] ) . Both our RNA-seq and proteomics analyses strongly indicated that simultaneously identified genes and proteins are critical components of OMV biogenesis in both wild-type and PMRHigh strains . Interestingly , A . baumannii OMVs also contained many cytoplasmic cargo proteins such as ribosomal proteins and Ef-Tu , similar to the OMVs from other bacteria ( Dallo et al . , 2012; Deo et al . , 2018 ) . Elongation factor ( Ef-Tu ) , observed in our proteomics analyses , is one of the most abundant proteins ( >20% ) in bacteria and has a canonical role in translation . However , it is also found on the surface of bacterial OMs , which suggests a possible noncanonical function on the membrane ( Widjaja et al . , 2017; Harvey et al . , 2019 ) . Accumulation of Ef-Tu in the periplasmic space might induce pressure , which promotes blebbing of the OM resulting in OMV secretion although lower level of TufB was detected in PMRHigh strain ( Figure 3B , McBroom and Kuehn , 2007 ) . Reduced expression of OmpA and OmpW leads to diminished linkage between the OM and PG and subsequently causes more OMV generation in the PMRHigh . Our RNA-seq and proteomics data implied that PMRHigh changes the OM surface charge and membrane integrity by altering the expression of many OMV-related genes , which leads to PMB resistance and higher OMV production . Decreased membrane-linkage proteins induce overproduction of OMVs , which is also linked to enhanced biofilm formation . OMV production and MIC were measured in all strains , including several Campbell-type and CRISPR-Cas9-based knockout strains ( Figure 4—figure supplement 1A ) . We used several approaches for quantifying OMV production ( lipid staining [FM4-64 dye] , quantification of total proteins [Bradford assay] , total DNA measurement , and LPS quantitation using the purpald assay ) ( Figure 4 , Figure 4—source data 1 , Figure 4—figure supplement 1B ) . OMV production appeared to be 1 . 5- to 2-fold higher in PMRHigh than PMRLow or Lab-WT , as quantified by using the lipophilic dye FM4-64 ( Figure 4A , Figure 4—source data 1 ) . Total protein , DNA , and LPS amounts in OMVs collected from the same amounts of cell cultures were higher in PMRHigh than Lab-WT ( Figure 4B–D ) . Furthermore , when the pmrB gene was deleted in Lab-WT , OMV biogenesis was reduced ( 0 . 3-fold ) ( Figure 4A–D ) , indicating that OMV production is impaired in the absence of PmrB . This finding was supported by our observation in which an extra copy of pmrBH gene increased OMV production in Lab-WT ( Figure 4A–D ) . However , the ΔftsL , Δudg , Δlpp , and ΔmlaC mutants exhibited higher OMV production ( Figure 4A–D ) . As expected from our omics data , OMV production was reduced when the PNAG transporter-related gene , pgaB , was deleted in Lab-WT ( Figure 4A–D ) . The relative OMV production measured by FM4-64 , protein , DNA , and LPS quantification in the ΔfabH and ΔsurE mutants could be ignored because no statistically significant difference was found ( Figure 4—figure supplement 1C–F ) . Our mutational study clearly demonstrated that mutant strains lacking genes involved in membrane cross-linkages and PL management ( lpp , ftsL , and mlaC ) showed increased OMV secretion . The MICs of the pmrB-deleted strain ( ΔpmrB ) appeared to be reduced ( Figure 4E ) . Introduction of the complementary pmrB gene to Lab-WT ( Lab-WT + pRK-pmrB ) increased the MIC , whereas the empty vector ( pRK ) did not affect the MIC ( Figure 4—figure supplement 1B ) . The MICs of ΔftsL , Δudg , Δlpp , and ΔpgaB remained unchanged in comparison to that of Lab-WT ( Figure 4E ) . The MICs of all constructed mutants were not affected , although they showed higher OMV production due to exposure to PMB at the beginning stage of culture conditions , in which no OMV production was expected . However , tendency for protecting cells grown in stationary phase from high PMB concentration ( 1/2 MIC of each clinical strain ) was observed in our tested higher OMV producers ( Figure 4F ) . We hypothesized that the OMVs secreted by PMRHigh and mutant strains act as decoys to mask the effect of PMB on cells . To assess this hypothesis , we extracted OMVs from PMRHigh culture ( Figure 5—figure supplement 1A ) and confirmed their presence using flow cytometry ( Figure 5—figure supplement 1B ) . Concentrated OMVs were negatively stained with uranyl acetate and visualized using TEM ( Figure 5—figure supplement 1C ) . The protective effect of these PMRHigh-derived OMVs was assessed using the Luria–Bertani ( LB ) broth ( Figure 5A , Figure 5—source data 1 ) and LB agar plate tests ( Figure 5B ) , after confirming that additional purified OMVs had no effect on the bacterial growth . Interestingly , supplementation of PMB-containing liquid cultures ( Figure 5A ) or agar ( Figure 5B ) with these OMVs could increase the survival of all tested strains except for PMRHigh on agar plates . To visualize direct binding PMB to OMVs , we developed a dansyl fluorophore-PMB , which was synthesized by forming a chemical bond between the primary γ-amines on the diaminobutyric-acid residues of PMB and dansyl-chloride ( Soon et al . , 2011 ) . Confocal laser scanning microscopy ( CLSM ) images revealed that dansyl-PMB was less bound to PMRHigh cells than Lab-WT cells at their respective PMB MICs ( 2 or 4 µg/mL ) due to the fact that the PmrC-driven lipid A modification of PMRHigh cells decreased PMB binding to OMs ( Figure 5C ) . The lipid-selective dye ( FM4-64 ) and dansyl-PMB were employed to visualize binding of PMB to purified OMVs of the PMRHigh stain ( Figure 5D ) . Our data showed that the two dyes exhibited overlapping images , which clearly demonstrated the direct binding of PMB to OMVs . To test whether A . baumannii clinical isolates could be protected from PMB through OMV secretion , the linkages between PMB protection rates and OMV production in 40 MDR clinical isolates were assessed ( Figure 6A , Figure 6—source data 1 ) . The degree of OMV production in each clinical isolate varied . Although direct correlations between the MIC and OMV production could not be identified under our tested conditions due to the same reasons we provided earlier , higher OMV producers could show higher protection rates against PMB in the early stationary phase ( Figure 6A ) . Particularly , tendency was greater in the six clinical isolates with red-colored area ( Figure 6A ) . Interestingly , addition of PMRHigh-derived OMVs to the growth cultures of five randomly chosen MDR strains ( Supplementary file 1i; F-1025 , F-1208 , F-1379 , F-1410 , and F-1629 ) enhanced the microbial survival in the presence of PMB ( Figure 6B ) . PMB MIC tests were performed using four additional Gram-negative bacteria ( P . aeruginosa PAO1 , Pseudomonas putida KT2440 , E . coli K12 , and Acinetobacter oleivorans DR1 ) and the protective roles of their OMVs were tested ( Figure 6—figure supplement 1 ) . Our data suggested that OMV-driven protection under PMB conditions could be applicable to other Gram-negative bacteria . We next tested whether OMVs could mask the effect of PMB on the entire bacterial microbiota . To this end , we used human fecal microbiota and employed the survival of Galleria mellonella as an infection model ( Figure 7 ) . The fecal microbiota was incubated for 5 days under anaerobic conditions to verify the extended protective effect of OMVs . Cultivation was continuously performed after dividing the sample into three groups ( a control group without PMB , a second bottle with PMB , and a third bottle with PMB after addition of OMVs 3 hr earlier ) . PMB and OMVs were added daily for 5 days . The environmental MIC of PMB was determined and 1/2 environmental MIC was used in the following experiments ( ~50% reduction of aerobic bacterial microbiota in 24 hr ) ( Figure 7—figure supplement 1A ) . After 5 days , the community cells were spotted on an LB plate and cultured under aerobic conditions to check PMB toxicity and the protective effect of OMVs ( Figure 7—figure supplement 1B ) . The PMB-treated sample showed a 104-fold reduction in aerobic bacterial cells than the control , and the OMV plus PMB sample showed similar levels of cell counts as the control ( Figure 7—figure supplement 1B ) . Bacterial community analysis was conducted by extracting DNA from each sample anaerobically cultured for 5 days to determine the composition of the protected in vitro microbiota . Our taxonomic profile , along with rarefaction curves and the PCoA 3D plot , revealed that addition of OMVs extracted from the PMRHigh strain protected the bacterial community from PMB under our tested anaerobic conditions , resulting in recovery of a similar microbial composition ( Figure 7A–C , Figure 7—source data 1 ) . The total 16S rDNA genes per dried gram in the PMB-added sample ( 145 , 311 copies/dried gram ) decreased by ~16% in comparison with that in the 5 days incubated control sample ( 173 , 403 copies/dried gram ) , while the corresponding value in the OMV-added sample was reduced by 14% ( 149 , 643 copies/dried gram ) ( Figure 7A ) . Surprisingly , the PMB-exposed sample showed significantly reduced numbers of 16S rDNA copies corresponding to the Enterococcus faecium group belonging to Gram-positive bacteria ( 75% lower than the control ) and recovery of the E . faecium group occurred in the OMV-added PMB sample ( Figure 7A ) . No growth inhibition with PMB was observed in both aerobically grown Enterococcus type strains ( E . faecium and Enterococcus faecalis ) . Interestingly , severe growth retardation with PMB was especially observed in the anaerobic culture of E . faecium ( Figure 7D , Figure 7—source data 1 ) . Unlike aerobically grown cells , anaerobically grown Enterococcus cells might show the effects of PMB toxicity on membrane charges and composition , which warrants further examination . The survival of G . mellonella larvae infected with Lab-WT or PMRHigh was assessed in the presence or absence of OMVs ( Figure 7E ) . Larvae were infected with either Lab-WT ( 106 CFUs/mL ) or PMRHigh with or without PMB ( 1 μg/mL ) after OMV addition . Comparison of the mortality rates demonstrated that PMRHigh , which possessed higher virulence than Lab-WT , induced faster death of all tested larvae within 72 hr ( Figure 7E ) . In the presence of PMB , only 10% of the G . mellonella larvae infected with PMRHigh died within 24 hr , but 80% of the larvae died when OMVs were included ( Figure 7E ) . This experiment suggested that A . baumannii survived longer due to the protective effect of OMVs against PMB , leading to the higher mortality of G . mellonella . Collectively , OMVs can protect the human pathogen A . baumannii from PMB under in vivo and in vitro conditions by functioning as decoys for PMB binding . Production of OMVs has been linked to many important cellular behaviors , including quorum-sensing , AR , and biofilm formation in pathogenic bacteria ( Florez et al . , 2017 ) . OMV biogenesis was thought to be also linked to the elevated hydrophobicity of the cell surface in P . putida , which caused greater biofilm formation ( Baumgarten et al . , 2012 ) . OMVs are known to function as cargoes for many extracellular or intracellular materials such as virulence factors , proteins , DNA , RNA , and signaling molecules ( Bonnington and Kuehn , 2016 ) . Interestingly , the OMVs produced by A . baumannii have been shown to have a virulence factor ( OmpA ) and an antibiotic-resistance factor ( AmpC ) ( Weber et al . , 2017 ) . OMVs are composed of PLs , OMPs , periplasmic proteins , and LPS; however , the packaging of cytosolic molecules into OMVs remains unclear ( Roier et al . , 2016a; Jan , 2017 ) . Interestingly , OMVs released from CRAB could be responsible for the horizontal transfer of the carbapenem-resistance-related gene blaNDM-1 to carbapenem-susceptible A . baumannii ( Chatterjee et al . , 2017 ) . Notably , Klebsiella pneumoniae strains had OMs and larger OMVs harboring different lipid compositions under PMB treatment , which might contribute to the low permeability of PMB ( Jasim et al . , 2018 ) . The following factors have been proposed to modulate OMV biogenesis: ( i ) diminished linkage proteins between OM and PG; ( ii ) increased turgor pressure due to the accumulation of cellular materials in the periplasm; ( iii ) modulation of the production of transmembrane proteins; and ( iv ) changes in membrane fluidity and rigidity ( Toyofuku et al . , 2015 ) . The bacterial membranes of Gram-negative bacteria are also constantly modified via two-component systems , including PmrAB , in response to changes in the environment such as metals and PMB ( Tsang et al . , 2017 ) . LptA is the periplasmic protein of the LPS transport ( Lpt ) complex , which bridges the IM and OM , transporting LPS across the periplasm ( Powers and Trent , 2019 ) . A recent study has shown that reduced LptA connections lead to overproduction of OMVs ( Falchi et al . , 2018 ) . Decreased Lpp-PG covalent cross-linking or non-covalent OmpA-PG interactions also enhance the production of OMVs ( Schwechheimer and Kuehn , 2015 ) . Disruption of the Mla complex system , which transports mislocalized OM glycerophospholipids back to the IM , results in hypervesiculation by accumulation of glycerophospholipids in the outer leaflet ( Kamischke et al . , 2019 ) . The Bam complex is an OM complex that mediates the folding and insertion of OM proteins into the membrane ( Malinverni and Silhavy , 2011 ) . DolP , a lipoprotein , is known to promote proper folding of BamA , one of the essential subunits of the BAM complex , which might also play a role in OM integrity ( Ranava et al . , 2021 ) . Deletion of bamE in E . coli or in Salmonella enteritidis results in a more severe defect in OM and lower OM protein levels than does loss of bamC ( Sklar et al . , 2007; Fardini et al . , 2009 ) . Taken together , the accumulated data suggest that OMVs can be generated when decreased cross-links between the OM , PG , and IM occur in the bacterial membrane . Our omics and mutational analyses substantiated that many genes involved in the structural components of OM and IM , such as ompA , lpp , mlaC , and ftsL , play important roles in OMV biogenesis ( Figures 3 and 4 ) . However , it is still not clear how the DNA repair-related udg gene is linked to OMV production ( Figure 3 ) . Linkage between those deleted regions in evolved strains and OMV biogenesis remains to be investigated ( Figure 3—figure supplement 1A ) . It is worth noting that antibiotic-induced genomic alterations including deletions , rearrangements , and amplification have been reported in many laboratory-evolved antibiotic-resistant bacterial strains ( Sandegren and Andersson , 2009 ) . Several laboratory-evolved antibiotic-resistant E . coli strains also appeared to carry more than 20 mutations ( Maeda et al . , 2020 ) . Tetracycline-induced genome rearrangements led to the loss of 6 . 1 kb having seven full-length and two truncated genes in E . coli strains ( Hoeksema et al . , 2018 ) . Norfloxacin-treated E . coli had ~11 . 5 kb deletions possibly through IS1-mediated recombination ( Long et al . , 2016 ) . Deducing a direct link between OMV production and a single OMV-related gene might be an oversimplification because of the multifaceted control of bacterial OMV production; however , our expression analyses clearly demonstrated that the hypervesiculating PMRHigh strain showed reduced expression of many OMV-related genes , which might cumulatively affect OMV production . Several other data have also shown that bacterial OMVs contained IM proteins as well as cytoplasmic proteins ( Nagakubo et al . , 2019 ) . The unusual OMV biogenesis recently identified in a P . aeruginosa model system has clearly shown that OMV production through explosive cell lysis ended up including several cytosolic components of cells ( Schwechheimer and Kuehn , 2015 ) . Explosive cell lysis induced by phage-derived endolysin burst the outer and IMs , leading to the presence of cytoplasmic proteins and DNA in OMVs ( Turnbull et al . , 2016 ) . Two known mechanistic factors ( PQS and MucD ) for OMV production were not required for cycloserine-induced OMV formation in P . aeruginosa , indicating the existence of multiple pathways for OMV biogenesis ( Macdonald and Kuehn , 2013 ) . OMV biogenesis in PMRHigh causes PMB resistance via the decoy activity of OMVs . OMVs have been shown to protect bacterial cells against several classes of antibiotics and physical and chemical stresses ( Kulkarni et al . , 2015; Lee et al . , 2016 ) . The decoy activity of OMVs was proposed in an E . coli model under membrane-active antibiotics such as colistin and melittin ( Kulkarni et al . , 2015; Kim et al . , 2018 ) . Our study also supported the role of OMVs as decoys under PMB conditions in A . baumannii ( Figure 5A , B ) , clinical isolates ( Figure 6B ) , and in the microbial community ( Figure 7A ) . The in vitro human microbiota experiment clearly showed that the presence of OMVs can modulate the bacterial community under PMB conditions . The reasons underlying our unexpected observation of sensitivity of the Gram-positive Enterococcus strains to PMB under anaerobic conditions remain to be elucidated ( Figure 7A ) . Recently , more studies suggested that polymyxins can bind to teichoic acid , which is a negatively charged compound in the OM of Gram-positive bacteria ( Brown et al . , 2013; Rudilla et al . , 2018; Yu et al . , 2020 ) . We speculated that different degrees of teichoic acid-like production might occur in anaerobic culture conditions . Our G . mellonella-infection experiments also suggested the protective role of OMVs on A . baumannii cells against PMB during infection ( Figure 7C ) . Likewise , Vibrio cholerae cells use hyperproduction of OMVs to increase their colonization and adaptation during mammalian infection ( Zingl et al . , 2020 ) . Interestingly , the OMVs induced by the pga operon can stimulate the production of broad antimicrobial antibodies against P . aeruginosa in a mouse infection model ( Stevenson et al . , 2018 ) . In conclusion , our data suggested that greater attention must be paid to develop new antibiotics and vaccines because the multifaceted OMV production in pathogenic A . baumannii strains warrants extensive investigations . The bacterial strains used in this study are listed in Supplementary file 1i . We named A . baumannii ATCC 17978 wild-type strain as Lab-WT and constructed PMB-resistant strains ( PMRLow and PMRHigh ) from Lab-WT . All the strains were grown at 37°C in LB broth with aeration by shaking at 220 rpm . The PMRLow and PMRHigh strains were constructed by transferring wild-type cells onto LB agar plates with serially higher PMB concentrations . In total , 3 A . baumannii experimentally evolved isolates and 40 clinical isolates were investigated . The clinical isolates were obtained from the Samsung Medical Center , Sungkyunkwan University School of Medicine , and National Culture Collection for Pathogens ( NCCP ) . The complete genome sequence of the A . baumannii ATCC 17978 strain can be accessed in GenBank ( accession number: CP000521 , CP015122 . 1 ) . Sequence reads for the experimentally evolved isolates are accessible from the NCBI sequence archives under accession numbers PRJNA530195 ( Lab-WT ) , PRJNA530197 ( PMRLow ) , and PRJNA530202 ( PMRHigh ) . Nine different classes of antibiotics , including meropenem , rifampicin , chloramphenicol , and ampicillin , were selected to test the AR . The MICs of each antibiotic were examined in all tested cells ( Lab-WT , PMRLow , PMRHigh , P . aeruginosa , P . putida , E . coli , and A . oleivorans ) with the broth-dilution method using 96-well plates , as previously described ( Pérez-Cruz et al . , 2016 ) . For determination of protection tendency with OMV in clinical isolates , 40 clinical A . baumannii isolates were grown at 37°C in LB broth at 220 rpm until each culture reached an OD600 of 0 . 8–1 . 0 . The cell pellet was washed with phosphate-buffered saline ( PBS ) twice and each collected cells ( 106 CFU/mL ) was inoculated separately into LB medium with or without 1/2 MIC of each isolate for 1 hr at 37°C . After grown cells were serially diluted to get 10−4 dilution with PBS , 10 µL of suspension spotted on an LB agar plates and incubated for 16 hr at 37°C . The protection rate ( % ) was calculated survived colonies ( CFU ) in comparison with untreated conditions of each strain . The expression levels of our target genes were normalized using the expression level of 16S rDNA , as previously described ( Shin et al . , 2020 ) . The total RNAs from different A . baumannii strains were isolated from 5 mL of cell cultures in the mid-exponential phase ( optical density at 600 nm [OD600] of approximately 0 . 4 ) using the RNeasy Mini Kit ( QIAGEN , Germany ) . For qRT-PCR , the cDNA was synthesized from 3 μg of RNA using the primers listed in Supplementary file 1j . All qRT-PCR procedures were conducted in triplicate from at least three independent cultures . The relative expression levels were compared with those of the Lab-WT strains . Biofilms were grown in confocal dishes ( SPL Life Sciences , South Korea ) at 37°C for 24 hr in LB broth . The biofilm cells were stained with FilmTracer SYPRO Ruby ( Invitrogen , USA ) for 30 min at room temperature ( RT ) , protected from light , and then washed with distilled water . The obtained CLSM ( Carl Zeiss , Germany ) images were analyzed and modified using the Zen 2 . 1 ( Blue edition; Carl Zeiss , Germany ) software . LB broth in sterile 96-well microtiter plates ( SPL Life Sciences ) was inoculated in triplicate with each overnight LB-grown culture and further diluted 1:100 with LB broth . The volume of the cells was determined by converting the OD600 value of the O/N cells . Uninoculated LB broth was used as the negative control . The microtiter plates were then incubated at 37°C for 24 hr . After removing planktonic cells , the biofilm biomass was stained with crystal violet and solubilized with 95% ethanol ( v/v ) , after which its absorbance was measured at 595 nm . To measure OMVs-induced initial stage of biofilm formation in Lab-WT , cells were grown at 37°C in LB until an OD600 reached around 0 . 6 ( for 6 hr ) . Biofilm formation was checked after additional 4 hr incubation with PMRHigh-driven OMVs ( 0 , 12 . 5 , or 25 μg/mL ) , then both crystal violet assay and CLSM observation were used . A . baumannii strains were grown at 37°C in LB broth until each culture reached an OD600 of 1 . 0 . Cells were harvested via centrifugation at 7800 rpm for 15 min . Then , the cell pellet was washed with 30 mL of PBS . The supernatant was poured off and resuspended in 35 mL of a single-phase Bligh–Dyer mixture ( chloroform:methanol:distilled water , 1:2:0 . 8; v/v ) . The solution was mixed by inversion and incubated at RT for >20 min to ensure complete cell lysis . Then , the mixture was centrifuged at 2000 rpm for 20 min . LPS will pellet along with proteins and nucleic acids; however , PLs , isoprenyl lipids , and small , hydrophobic peptides remain in the supernatant . The supernatant was discarded . The LPS pellet was washed with 35 mL of the single-phase Bligh–Dyer mixture and centrifuged at 2000 rpm for 20 min . The supernatant was discarded; 8 mL of mild acid hydrolysis buffer ( 50 mM sodium acetate , pH 4 . 5; 1% SDS ) was added to the pellet , and it was mixed by pipetting . The solution was sonicated at a constant-duty 2× cycle for a 20 s/burst , with an interval of 5 s between bursts , at 50% output . Then , the sample was boiled in a water bath for 30 min , and the bottles were removed from the water bath and allowed to cool to RT . To extract lipids after hydrolysis , the mixture was formed by adding 9 mL of chloroform and 9 mL of methanol to the SDS solution . Then , the sample was mixed by inversion and centrifuged for 10 min at 2000 rpm . The lower phase was extracted into a clean 50 mL conical tube . A second extraction was performed by adding 30 mL of the lower phase from a pre-equilibrated two-phase Bligh–Dyer mixture ( chloroform:methanol:DW , 2:2:1 . 8; v/v ) to the upper phase from the previous step . The mixture was centrifuged at 2000 rpm for 10 min . The lower phase was extracted and pooled with the lower phase from the previous step . Then , the pooled lower phase ( 18 mL total ) was washed by adding 34 . 2 mL of pre-equilibrated two-phase Bligh–Dyer upper phase to create a two-phase Bligh–Dyer mixture ( chloroform:methanol:DW , 2:2:1 . 8; v/v ) . The solution was mixed and centrifuged at 2000 rpm for 10 min . The lower phase was removed to a clean glass rotary evaporator flask and the sample was dried using rotary evaporation . The lipid sample was added to 1 . 5 mL of chloroform:methanol ( 4:1 , v/v ) in a rotary flask , and ultrasonicated ( 40 s ) to facilitate suspension of lipid from the sides of flask . The lipid A was dried in a nitrogen dryer and transferred to small glass sample vial . The dried sample was stored at 4°C . Lipid A was analyzed by MALDI-TOF mass spectrometry in the positive-ion mode . The matrix was a saturated solution of 2 , 5-dihydroxybenzoic acid ( DHB ) . A 10 μL volume of the matrix solution was deposited on the sample plate , followed by 10 μL of the sample dissolved in chloroform:methanol ( 4:1 , v/v ) . WGS analyses were performed on three selective A . baumannii isolates using the PacBio sequencing technique ( ChunLab , South Korea ) . The genome sequences were downloaded from EzGenome ( http://ezgenome . ezbiocloud . net/ezg_browse ) and were analyzed using the CLgenomics program ( ChunLab ) . Long-read NGS sequencing techniques , such as the PacBio used in this study , may produce sequencing errors ( Rhoads and Au , 2015 ) . To compensate for this shortcoming , additional PCR reactions using Pfu DNA polymerase were performed to confirm mutational positions listed in Supplementary file 1a ( 18 genes ) . The primers used for PCR reactions are listed in Supplementary file 1k . Genomes were analyzed by using ClustalW 2 . 0 software , while neighbor-joining trees and amino acid compositions were constructed and calculated using MEGA6 software . The origin of replication ( oriC ) was identified using Ori-Finder ( http://tubic . tju . edu . cn/Ori-Finder/ ) . In addition to the amino acid sequence of PmrB , the LPS/lipid A biosynthesis components were compared between the isolates and the reference strain ATCC 17978 . Antibiotic-resistant genes were identified using the CLC Genomics Workbench v . 10 . 0 . 1 ( QIAGEN ) . We used the Find Resistance tool ( Microbial Genomics Module ) with the QMI-AR , CARD , ResFinder , and PointFinder . All parameters were set as default . Insertion sequence ( IS ) elements were analyzed with the IS finder ( https://isfinder . biotoul . fr/blast . php ) . Both Lab-WT and PMRHigh strains were grown to exponential phase ( OD600 ~ 0 . 5 ) in LB media . For antibiotic treatment conditions , both strains were grown to the exponential phase ( OD600 ~ 0 . 25 ) in 1/2 MIC ( 1 µg/mL or 64 µg/mL , respectively ) of PMB-supplemented LB media . Total RNA was isolated from 10 mL of cells by using the RNeasy Mini Kit ( QIAGEN ) according to the manufacturer’s instructions . All procedures for RNA sequencing were conducted by ChunLab . The RNA was subjected to a subtractive Hyb-based rRNA removal process using the MICROBExpress Bacterial mRNA Enrichment Kit ( Ambion , USA ) . RNA sequencing was performed with two runs of the Illumina Genome Analyzer IIx to generate single-ended 100 bp reads . Quality-filtered reads were aligned to the reference genome sequence using the CLC Genomics Workbench v . 10 . 0 . 1 ( QIAGEN ) . Mapping was based on a minimal length of 100 bp with an allowance of up to two mismatches . Relative transcript abundance was measured in RPKM . The RNA-seq data have been deposited in NCBI under Gene Expression Omnibus ( GEO ) accession number GSE163581 . To construct ΔftsL , Δudg , and Δlpp mutants by homologous recombination , target genes were disrupted using a single-crossover recombination method with several vectors , such as pVIK112 and pRK415 , as previously described ( Shin and Park , 2015 ) . The primers used in this study are listed in Supplementary file 1j . In general , EcoRI and KpnI restriction enzymes were used for all the knock-out and overexpression mutants . ΔpmrB is a pmrB mutant of Lab-WT strains . To construct the mutants , fragments from PCR amplification and gel extraction were inserted into the pVIK112 vector via ligation as the first step . The pVIK112-fragment unified vectors were then transformed into E . coli S17-1λ pir . Finally , the amplified plasmids were inserted into the Lab-WT and PMRHigh strains of A . baumannii ATCC 17978 by electroporation . To ensure that homologous recombination had occurred in A . baumannii ATCC 17978 , PCR verification was conducted using the pmrB OE-F/MCS-R primer pairs . The MCS-R primer was designed based on the sequence of the pVIK112 plasmid . To construct ΔpmrB , ΔmlaC , and ΔpgaB by the CRISPR-Cas9 method , the pCasAb-apr ( cat . no 121998 ) and pSGAb-km ( cat . no 121999 ) plasmids were obtained from Addgene ( Wang et al . , 2019 ) . A pair of 20 bp spacer oligos and 80-nt oligos were designed to target the genomic locus and donor repair template for editing . A designed 20 bp spacer was phosphorylated and annealed to the pSGAb-km plasmid with a Golden Gate assembly reaction . Then , the plasmids were transformed into E . coli DH5a-competent cells , followed by plating onto an LB agar plate containing 100 µg/mL , and incubated at 37°C for 16 hr . Successful cloning of the spacer was verified by PCR with the primers of Spacer-F/M13R and by sequencing with the primer of M13R ( 5′-CAGGAAACAGCTATGACC-3′ ) . Then , 1 M of IPTG was added into A . baumannii Lab-WT harboring pCasAb-apr to induce the expression of the RecAb recombination system and Cas9 nuclease . After incubating at 37°C for 2 hr , the same method was used to prepare the competent cells as described previously ( Wang et al . , 2019 ) . Then , 200 ng of the spacer-introduced pSGAb-km plasmid and 300 µM ssDNA ( donor repair template ) were co-transformed to the IPTG-induced A . baumannii Lab-WT cells harboring the pCasAb-apr plasmid by electroporation . The cells were plated onto an LB agar plate containing 100 µg/mL apramycin and 50 µg/mL kanamycin , and the plate was incubated at 37°C overnight . Successful editing was verified by PCR and sequencing . For plasmid curing , a colony was plated onto an LB agar plate containing 5% sucrose at 37°C overnight . Then , the colony were streaked onto LB agar plates without the antibiotics to confirm curing . A detailed genome-editing protocol has been provided in the file ( Figure 4—figure supplement 1 ) . TEM was conducted at the Korea University Medical Research Center ( https://medicine . korea . ac . kr/web/msrc/home ) . The experimental procedure was in accordance with the instructions of the manufacturer . Briefly , for bacterial cell imaging , cells were grown for 18 hr in LB media . Aliquots were transferred onto carbon-coated copper grids ( Formvar; Ted PELLA , Canada ) and fixed with 2 . 5% glutaraldehyde diluted in 50 mM sodium cacodylate ( pH 7 . 2 ) . After washing three times with 3% saccharose , the cells were observed . For OMV imaging , aliquots of purified OMV samples were dispensed on carbon-coated copper grids . Excess liquid was discarded from the grids and the samples were negatively stained with phosphotungstic acid ( 3% [wt/vol] ) for 5 min and then dried with filter paper . The specimens were examined using an H-7100 microscope ( Hitachi , Tokyo , Japan ) operating at an accelerating voltage of 75 kV and at magnifications of 40 , 000× for cells and 30 , 000× for OMVs . The lipid content of the OMV samples was indirectly quantified using the lipophilic dye FM4-64 ( N- ( 3-triethylammoniumpropyl ) −4- ( 4- ( dibutylamino ) styryl ) pyridinium dibromide; Molecular Probes , Eugene , OR , USA , and Life Technologies , Carlsbad , CA , USA ) . The lipophilic dye FM4-64 , which intercalates into the OM , is commonly used to stained and observed membrane of bacteria ( Toyofuku et al . , 2017; Rojas et al . , 2018 ) . To measure OMVs using this dye , cells were incubated for 16 hr at 37°C and 220 rpm in 5 mL of LB . The cultured cells were harvested , and the supernatant was filtered through a 0 . 22 μm syringe filter . Aliquots of the filtrate ( 100 μL ) containing OMVs were transferred to the wells of a 96-well dark plate . The FM4-64 dye was added to each well at a final concentration of 1 μg/mL . After excitation at 510 nm , the fluorescence emission at 610 nm was measured using a fluorescence spectrophotometer ( TECAN , Männedorf , Switzerland ) using 5 nm slit widths for both excitation and emission . To quantify the protein contents of OMVs , all the samples were standardized using the Bradford assay ( Bio-Rad Laboratories , Protein Assay Dye Reagent ) , and triplicates were used . For the DNA contents of OMVs , the NanoPhotometer ( Implen , Germany ) was used to quantify the concentration of DNA . To quantify the LPS content of OMVs , purpald assays using the 2-keto-3-deoxyoctonate ammonium salt ( Kdo ) standard ( Sigma-Aldrich ) were performed as previously described ( Lee and Tsai , 1999 ) . Isolates were grown in LB broth for the exponential phase at 37°C , and the cell pellet was washed with PBS twice . Each bacterial suspension was dropped into an aluminum dish ( width , 1 . 75 in; depth , 0 . 38 in ) for spectral collection . After drying the plate for 2 hr , each spot was washed with 10 µL of PBS . Raman spectra of the isolates were collected using a confocal Raman imaging system ( XperRam35V; Nanobase ) assembled with 3-port excitation 532 nm DPSS laser ( LTL-532RL; Leading Tech ) , microscope body ( Olympus BX43; Olympus ) , spectrometer ( XPE-35 VPHG; Nanobase ) , and charge-coupled device ( Atik 428EX; Atik ) . Each spectrum was the sum of 25 s of acquisition time with 2 mW of laser power and collected for 15–20 single cells of the A . baumannii isolate . The procedure used to isolate the OMVs is presented in Figure 5—figure supplement 1 . OMVs were isolated from PMRHigh bacteria . The isolation was confirmed using FM4-64 staining and flow cytometric analysis . To extract OMVs from PMRHigh strains , 600 mL of cells were grown for 16 hr to obtain a robust and quantifiable amount of OMVs . Overnight bacterial culture was grown to an OD600 of 0 . 8–1 . 2 and the culture was centrifuged at 7800 rpm at 4°C for 30 min . The supernatant of each culture was sequentially filtered through 5 μm hydrophilic polyvinylidene difluoride and 0 . 45 μm Millex membrane filters ( Millipore , Billerica , MA , USA ) . The resulting filtrates were concentrated with centrifugation ( 7800 × g ) at 4°C for 30 min using a 10 kDa molecular weight cutoff Amicon Ultra-15 centrifugal filter unit ( Millipore ) . Each concentrated filtrate was centrifuged ( 100 , 000 × g , 4°C , 3 hr ) in a tabletop Optima ultracentrifuge ( Beckman Coulter ) . Pelleted OMVs were suspended in 100 μL of PBS . The OMVs were quantified by using Bradford assay . The confirmed OMV pellet was spread on LB agar plates to confirm the cell-free status . The OMVs were counted by using a flow cytometer ( BD Accuri C6 Plus ) . The forward scatter threshold value was 10 , 000 . Three independent experiments were conducted for each condition , with 50 , 000 cells typically analyzed per experiment . PMB sulfate ( 40 mg , Sigma-Aldrich ) was dissolved in 1 . 2 mL of 0 . 1M NaHCO3 , and dansyl-chloride ( 10 mg , Sigma-Aldrich ) was dissolved in 0 . 8 mL of acetone . Dansyl-chloride was added to PMB and placed in the dark for 90 min at RT . After incubation , the mixture was loaded onto a Sephadex G-50 column ( 50 × 2 . 5 cm ) equilibrated with 10 mM Na-phosphate buffer ( pH 7 . 1 ) containing 0 . 145M NaCl , and 5–6 mL fractions from the column were assessed . The dansyl-PMB appears as a broad peak ahead of the unreacted dansyl-chloride peak . The location of the dansyl-PMB in the collected fractions was determined by holding a UV lamp over the fractions and looking for fluorescence . The fluorescence of the dansyl-PMB was yellowish , while the unreacted dansyl-chloride showed more blue-green fluorescence . The fractions containing dansyl-PMB were extracted into approximately 1/2 vol of n-butanol . The butanol was then evaporated to dryness in a glass Petri dish placed inside a desiccator , which was then evacuated and placed at 37°C for 24 hr . The dried dansyl-PMB was dissolved in 3 mL of buffer ( 5 mM HEPES , pH 7 . 0 ) and stored at −20°C . CLSM was performed using LSM 770 ( Carl Zeiss Microscope , Jena , Germany ) . The exponential phase of the 5 mL bacterial culture was washed and resuspended with PBS . The cells were stained with dansyl-PMB ( 2 or 4 µg/mL ) for 30 min at 37°C . To confirm the protective effect of OMVs on LB agar plates , cultured cells were inoculated on media containing 0 , 0 . 5 , 1 , 2 , or 4 μg/mL PMB in the absence or presence of OMVs from PMRHigh . Each culture was serially diluted 1:10 and inoculated beginning in the first spot . In the liquid medium , the protective effect of OMVs against PMB was monitored by assessing the growth curves of OMV-treated Lab-WT , PMRLow , and PMRHigh cells and clinical isolates ( designated F-1025 , F-1208 , F-1379 , F-1410 , and F-1629 ) from patients treated at Sungkyunkwan University Hospital . PMB was used at concentrations of 0 , 1 , 2 , and 4 μg/mL . The cultured cells ( 106 CFU/ml ) were inoculated separately into LB medium containing different concentrations of PMB ( 0 , 1 , 2 , and 4 μg/mL ) and PMRHigh -derived OMVs ( 25 μL/mL ) . Bacterial growth was monitored at OD595 at 2 hr intervals for 72 hr . Experiments were performed using bacterial cultures from three independent batches . The bacterial cultures were shaken for 10 s inside the instrument ( TECAN ) every 10 , 000 s . The fecal sample was obtained from a woman aged 25 years , collected in sterilized plastic tubes , and suspended at a dilution of 1/10 in PBS ( pH 7 . 4 ) . The fecal solution ( 5 mL ) was added to an anaerobic bottle , injected with 80% N2 + 20% CO2 gas , and sterilized . Subsequently , fecal microbiota ( 50 μL , 1/100 of the fecal solution ) were inoculated and cultured at 37°C . Analysis of bacterial communities revealed whether the protected bacteria depended on the type of OMV in the bacterial community ( Figure 6A ) . PMB ( 4 μg/mL ) and OMVs ( 25 μL/mL ) of equal concentrations were administered daily into the fecal samples suspended in PBS; the sample was stabilized for 3 hr by applying OMVs first , and then treated with PMB and incubated at 37°C for 5 days . Then , total community DNA was extracted using a FastDNA spin kit for soil ( MP Biomedicals , USA ) , and the DNA yield was quantified using a NanoDrop spectrophotometer ( BioTek , USA ) . The V3-V4 hypervariable region of 16S rRNA gene from the genomic DNA was amplified using the primers 341F ( 5′-TCGTCGGCAGCGTC-AGATGTGTATAAGAGACAG-CCTACGGGNGGCWGCAG-3′ ) and 805R ( 5′-GTCTCGTGGGCTCGG-AGATGTGTATAAGAGACAG-GACTACHVGGGTATCTAATCC-3′ ) . The amplified products were confirmed by agarose gel electrophoresis . The amplicons were purified by Agencourt AMPure XP ( Beckman Coulter , Republic of Korea ) and quantified using a Quanti-iT Picogreen dsDNA Assay kit . Equimolar concentrations of each amplicon from the different samples were pooled and purified using Agencourt AMPure XP ( Beckman Coulter ) . All sequencing procedures were conducted by ChunLab . The sequences obtained were compared and classified using the EzTaxon Database ( http://www . ezbiocloud . net ) . The operational taxonomic units ( OTUs ) among the samples were obtained with a taxonomic composition using the CLcommunity program ( ChunLab ) . The bacterial community data have been deposited in NCBI under Sequence Read Archive ( SRA ) accession numbers SRX9819399 PRJNA689940 , SRX9819397 PRJNA689944 , and SRX9819398 PRJNA689944 . G . mellonella larvae were obtained from SWORM ( Cheonan , Republic of Korea ) . Healthy G . mellonella larvae weighing 200 mg were starved for 4 hr at 20°C before injection . Then , the larvae were placed on ice and injected via the last left proleg with 10 µL of OMV solution ( 25 μL/mL ) using 31-gauge , 6-mm-long needles ( BD Ultra-Fine insulin syringes ) . Infected larvae were divided into the following four experimental groups with 15 larvae per group: ( i ) inactive control groups , which received 10 µL of PBS or OMVs; ( ii ) bacterial treatment groups , which received either 1 × 106 CFU/larvae of A . baumannii ATCC 17978 Lab-WT or PMRHigh; ( iii ) bacteria and antibiotic treatment groups , which received 1 × 106 CFU/larvae of Lab-WT or PMRHigh and 1 µg/mL PMB; and the ( iv ) OMV treatment group , which received 1 × 106 CFU/larvae of Lab-WT or PMRHigh , 1 µg/mL PMB , and OMVs . PBS was added to each group up to 10 µL . The larvae were incubated in a Petri dish ( 90 × 15 mm , SPL Life Sciences ) with 100 mg of wheat bran powder ( MG Natural , Republic of Korea ) at 37°C in air for 96 hr and inspected and scored every 12 hr for death , failure to move in response to touch , and melanization .	Wrapped in a thick , protective outer membrane , Acinetobacter baumannii bacteria can sometimes cause serious infections when they find their way into human lungs and urinary tracts . Antibiotics are increasingly ineffective against this threat , which forces physicians to resort to polymyxin B , an old , positively-charged drug that ‘sticks’ to the negatively-charged proteins and fatty components at the surface of A . baumannii . Scientists have noticed that when bacteria are exposed to lethal drugs , they often react by releasing vesicles , small ‘sacs’ made of pieces of the outer membranes which can contain DNA or enzymes . How this strategy protects the cells against antibiotics such as polymyxin B remains poorly understood . To investigate this question , Park et al . examined different strains of A . baumannii , showing that bacteria resistant to polymyxin B had lower levels of outer membrane proteins but would release more vesicles . Adding vesicles from resistant strains to non-resistant A . baumannii cultures helped cells to survive the drugs . In fact , this protective effect extended to other species , shielding whole communities of bacteria against polymyxin B . In vivo , the vesicles protected bacteria in moth larvae infected with A . baumannii , leading to a higher death rate in the animals . Experiments showed that the negatively-charged vesicles worked as decoys , trapping the positively-charged polymyxin B away from its target . Taken together , the findings by Park et al . highlight a new strategy that allows certain strains of bacteria to protect themselves from antibiotics , while also benefitting the rest of the microbial community .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2021	A novel decoy strategy for polymyxin resistance in Acinetobacter baumannii
The maternal and paternal genomes play different roles in mammalian brains as a result of genomic imprinting , an epigenetic regulation leading to differential expression of the parental alleles of some genes . Here we investigate genomic imprinting in the cerebellum using a newly developed Bayesian statistical model that provides unprecedented transcript-level resolution . We uncover 160 imprinted transcripts , including 41 novel and independently validated imprinted genes . Strikingly , many genes exhibit parentally biased—rather than monoallelic—expression , with different magnitudes according to age , organ , and brain region . Developmental changes in parental bias and overall gene expression are strongly correlated , suggesting combined roles in regulating gene dosage . Finally , brain-specific deletion of the paternal , but not maternal , allele of the paternally-biased Bcl-x , ( Bcl2l1 ) results in loss of specific neuron types , supporting the functional significance of parental biases . These findings reveal the remarkable complexity of genomic imprinting , with important implications for understanding the normal and diseased brain . In mammalian brains , neural computations underlying signal processing and behavioral control are conducted by a large diversity of cell types , each defined by unique but flexible patterns of connectivity , electrical properties , and profiles of gene expression and chromatin states ( Fishell and Heintz , 2013 ) . New experimental and conceptual frameworks have begun to address the nature of long-lasting regulatory changes at the level of the chromatin that may contribute to the establishment of neuronal identities and participate in stable encoding of cellular memories ( Dulac , 2010 ) . Genomic imprinting is a unique and long lasting form of epigenetic inheritance that relies on chromatin modifications or ‘imprints’ established in the parental germ lines and maintained in cells of the developing and adult organism , resulting in the differential expression of the maternally- or the paternally-inherited allele ( Bartolomei and Ferguson-Smith , 2011 ) . Although expression differences between parental alleles are commonly assumed to be all-to-none , a substantial number of imprinted genes have been shown to exhibit a biased expression from one of the parental alleles in at least some tissues ( Dao et al . , 1998; Lewis et al . , 2004; Khatib , 2007; Babak et al . , 2008; Menheniott et al . , 2008; Tierling et al . , 2009; Gregg et al . , 2010; Gregg , 2014 ) . Imprinted genes have been shown to play key roles during embryonic development ( Cleaton et al . , 2014 ) , in the placenta ( Tunster et al . , 2013 ) , and more recently , in the developing and adult brain ( Wilkinson et al . , 2007; Keverne , 2012 ) . A systematic survey of the expression patterns of known imprinted genes in the brain indicates a preferential enrichment of imprinted gene expression in a subset of brain areas , in particular those involved in feeding , social , and motivated behaviors ( Gregg et al . , 2010 ) . Moreover , the prominent role of imprinted genes in brain development and function is evidenced by a number of human disorders and mouse mutant phenotypes ( Wilkinson et al . , 2007 ) . For instance , patients with Prader–Willi syndrome , caused by loss of paternal expression in the q11-13 region of human chromosome 15 , display abnormal development , hyperphagia , mental retardation , and volatile behavior ( Peters , 2014 ) . In contrast , Angelman syndrome , which is caused by loss of maternal expression in the same genomic region , results in mental retardation , impaired speech , and an abnormally joyful demeanor ( Peters , 2014 ) . Similarly , Birk-Barel mental retardation syndrome ( Barel et al . , 2008 ) is caused by mutations in the human maternally expressed KCNK9 gene , and phenotypes associated with loss of the imprinted genes Grb10 , Mest , and Peg3 in the mouse exhibit impairments in specific social behaviors ( Lefebvre et al . , 1998; Li et al . , 1999; Isles et al . , 2006; Garfield et al . , 2011 ) . A key feature of genomic imprinting lies in the transmission of epigenetic marks that remain stable across cell divisions throughout the lifespan of the organism and in different tissues . Surprisingly , subsets of genes have been reported to exhibit tissue-specific imprinting , and the whole brain , neurons and certain brain regions emerged as hot spots for such regulation ( Albrecht et al . , 1997; Gregg et al . , 2010; Sato and Stryker , 2010; Prickett and Oakey , 2012 ) . A genome-wide identification of allelic parental bias throughout the adult and developing brain appears therefore necessary to fully assess the role of genomic imprinting in the nervous system . Such a quest has been noticeably difficult to achieve . Initial methods to uncover imprinted genes based on the differential expression between parthenogenetic ( containing only maternally derived chromosomes ) and androgenetic ( containing only paternally derived chromosomes ) embryos , and subsequent discovery of adjacent imprinted loci within the genome were mainly focused on early developmental stages , and led to the identification of approximately 100 imprinted genes ( Kaneko-Ishino et al . , 1995; Hagiwara et al . , 1997; Morison et al . , 2005; Ruf et al . , 2006 ) . The development of next generation RNA sequencing ( RNA-seq ) allowed for genome-wide screens of parentally biased allelic expression in any tissue of interest using F1s of reciprocal crosses between distantly related mouse strains . An intriguing question was whether or not this new , and presumably more powerful , experimental strategy would uncover novel imprinted genes . The answer to this question was proven challenging and controversial . In pioneer RNA-seq analyses of mouse hybrids , Wang et al . ( 2008 ) and Babak et al . ( 2008 ) used neonatal brains and E9 . 5 embryos , respectively , and determined parental biases by testing if the sum of parentally phased reads along a gene significantly deviates from biallelic expression . This approach , combined with shallow sequencing , only identified a handful of novel imprinted genes and failed to detect genes known to be imprinted in the profiled tissues . Next , Gregg et al . ( 2010 ) conducted an imprinting study at higher resolution by characterizing the preoptic area and prefrontal cortex of adult males and females , and the E15 brain , with an over 10-fold higher sequencing depth compared to the two previous studies . This experimental design , combined with testing for deviation from biallelic expression of parentally phased reads at each single SNP rather than along an entire gene , resulted with a much larger number of novel imprinted gene candidates . However , most of the novel imprinted candidates were not subject to independent experimental validation . In turn , DeVeale et al . ( 2012 ) criticized the use of single SNPs to infer imprinting and the lack of systematic independent validation of Gregg et al . ( 2010 ) as leading to a large and underestimated false discovery rate , and suggested that the vast majority of imprinted genes in mouse were likely already uncovered ( DeVeale et al . , 2012; Kelsey and Bartolomei , 2012 ) . These early studies indicate that several critical considerations should be taken into account to allow for accurate and powerful detection of parental bias using RNA-seq . First , the use of large numbers of biological samples is essential to achieve high statistical power; second , since transcripts are purified and sequenced it is more appropriate to quantify the relative abundances of all allele-specific transcript variants of a given gene rather than rely on the less accurate assessment of allelic expression at the gene level; third , a principled and formal statistical model should be used that explicitly accounts for the biological variability among samples and all factors in the experimental design ( e . g . , cross , sex , and age ) , and finally , a systematic independent validation of all imprinted candidates is necessary . The present study addresses these various challenges and establishes a novel and rigorous framework for detecting imprinting from RNA-seq data . We have processed the RNA-seq data in an allele-aware manner and inferred parent-of-origin biased expression with a newly developed Bayesian regression allelic imbalance model ( BRAIM ) , which accounts for all sources of variability in the experimental design . We further build upon these methodological and experimental advances to gain insights into the spatial and developmental dynamics of parent-of-origin allelic expression . From our study , the apoptotic pathway emerges as an important target of imprinting regulation , and we provide evidence that the paternal and maternal alleles of Bcl-x ( Bcl2l1 ) contribute unequally to brain development . These findings open exciting new avenues for investigating epigenetic mechanisms underlying the normal and pathological brain . We performed a genome-wide screen for imprinted genes in the cerebellum , a brain structure involved in motor control and implicated in autism spectrum disorders ( Amaral et al . , 2008; Tsai et al . , 2012; Wang et al . , 2014b ) . The cerebellum is large enough to enable RNA profiling from tissue of a single animal and contains well-characterized cell types and connectivity . Moreover , unlike other brain structures , cerebellar development occurs largely postnatally , thus providing direct access to key neurodevelopmental processes such as cell proliferation , migration , and synaptogenesis ( Sillitoe and Joyner , 2007; Hashimoto and Hibi , 2012 ) . The high variability of RNA expression across individuals requires the use of large numbers of biological samples to achieve statistical power and accurate detection of parental bias . Tissue was dissected from F1 mouse hybrids of Cast/EiJ and C57Bl/6J reciprocal crosses and RNA transcripts were sequenced from 48 individual cerebella representing two developmental stages and both sexes ( Figure 1A ) . Half of the cerebella were collected at postnatal day 8 ( P8 ) , a period in which newly-born granule cells migrate to the inner granule layer . The remaining samples were collected from adult animals ( P60 ) . Each age group had equal numbers of males and females . Therefore , our experimental design includes six replicates for each cross , sex , and age combination , which enables inference of genomic imprinting across all individuals , while taking into account the effects of the different experimental factors . 10 . 7554/eLife . 07860 . 003Figure 1 . Workflow of transcriptome-wide profiling of allele-specific expression . ( A ) First , F1 hybrids were generated by crossing C57Bl/6J males with Cast/EiJ females ( F1 initial ) and reciprocally crossing Cast/EiJ males with females C57Bl/6J ( F1 reciprocal ) . Second , RNA sequencing ( RNA-seq ) data from each of the F1 samples were mapped to a splice-junction-aware diploid C57Bl/6J , Cast/EiJ genome . Third , we transformed genomic alignments to transcriptomic alignments and filtered alignments that did not map to the transcript set using custom code . Fourth , expression levels and associated errors of all expressed transcripts in the diploid C57Bl/6J , Cast/EiJ transcriptome were estimated using MMSEQ ( Turro et al . , 2011 ) for each mapped RNA-seq sample . Finally , for each heterozygous-expressed-transcript in the diploid C57Bl/6J , Cast/EiJ transcriptome , the parental expression bias and the effects of the mouse cross , the age , and sex were estimated using Bayesian regression allelic imbalance model ( BRAIM ) . ( B ) Histograms of the distributions of the marginal posterior probabilities ( PPs ) of the parental expression biases , and the effects of cross , age , and sex of all 38 , 066 autosomal heterozygous transcripts in the diploid C57Bl/6J , Cast/EiJ transcriptome to which BRAIM was fitted . ( C ) Proportion of previously reported imprinted genes as well as newly identified imprinted genes subjected to pyrosequencing validation . ( D ) Relationship between pyrosequencing and RNA-seq estimates of parental biases as indicated by the percentage of expression contributed by the preferred parental allele . Note: the orange dot at ∼0 . 8 , 0 . 8 is considered a false positive because the preferred allele observed in the RNA-seq data is opposite to the one observed by pyrosequencing . ( E ) Number of genes exhibiting preferential expression of the maternal allele ( red ) , paternal allele ( blue ) , or in which the maternal and paternal alleles preferentially express different isoforms ( purple ) in the cerebellum . ( F ) Distribution of the magnitudes of the parental bias ( % of total expression from the preferred allele ) in the cerebellum . DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 00310 . 7554/eLife . 07860 . 004Figure 1—figure supplement 1 . Determining expression level cutoff in RNA-seq data . ( A ) Distribution of average allele-specific expression levels of all transcripts ( posterior median of MMSEQ estimated natural log of transcripts per million ( TPM ) units ( ln ( TPM ) ) across all 48 RNA-seq cerebellar samples . The solid line represents the expression cutoff below which all expression levels are set to zero . The dashed line represents a more stringent expression level cutoff ( see C ) . ( B ) Each transcript with a significant parental effect is represented by the fraction of expression of the preferred allele ( Y-axis ) ( blue and red for the paternal and maternal allele , respectively ) and its expression level in ln ( TPM ) units ( X-axis ) . The lower concentration of points left to the dashed line ( see C ) indicates that detection of significant parental effects below that expression level is less robust . ( C ) Discrepancy between the estimated percentage of expression contributed by the preferred parental allele according to pyrosequencing and RNA-seq . At low expression levels the discrepancy is pronounced and the fraction of false positives is relatively high . A more stringent expression level cutoff was placed where the discrepancy dramatically drops ( dashed line at ln ( TPM ) = −1 . 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 004 RNA-seq libraries generated from tissues with high cellular heterogeneity and transcriptional complexity , such as found in the brain , are likely to contain genes represented by multiple transcript species ( isoforms ) at various abundances . Thus , accurate transcriptional analysis requires estimating the expression level of each transcript rather than quantifying expression at the gene level . In turn , quantification at the transcript level is also necessary to accurately estimate the expression of a given gene , which is merely the sum of all its isoforms ( Jiang and Wong , 2009; Turro et al . , 2011; Trapnell et al . , 2013 ) . However , since a large fraction of reads do not map uniquely to a single transcript ( e . g . , reads originating from constitutive exons ) or even to unique genomic locations , transcript- and gene-expression levels can only be estimated with some degree of certainty , that is , with estimation error ( Jiang and Wong , 2009; Turro et al . , 2011; Trapnell et al . , 2013 ) . The uncertainty in estimating transcript- and gene-expression levels is even more pronounced when quantifying allele-specific expression , because the high sequence similarity between the two alleles results in a high level of read-mapping ambiguity . An additional concern arises from the mapping of RNA-seq hybrid data to the mouse C57Bl/6J reference genome , as it will favor the mapping of reads originating from the C57Bl/6J allele , and may therefore lead to inaccurate estimates of allele-specific expression levels ( Vijaya Satya et al . , 2012 ) . Finally , accurate inference of genomic imprinting requires a statistical model that explicitly accounts for expression-level uncertainty in each sample , together with the biological and technical variability , and the effects of all factors in the experimental design , such as sex , age , and , in the case of mouse hybrids , the cross of each sequenced subject . To address these challenges , we applied the following steps in the analysis of the cerebellar RNA-seq data ( Figure 1A and ‘Materials and methods’ ) . First , we generated Cast/EiJ and C57Bl/6J diploid genomes and transcriptomes by incorporating Cast/EiJ and C57Bl/6J single nucleotide- and short insertion and deletion polymorphisms ( SNPs and indels , obtained from the Mouse Genome Project: ftp://ftp-mouse . sanger . ac . uk/REL-1303-SNPs_Indels-GRCm38 ) into the Mus musculus GRCm38 reference genome sequence using the AlleleSeq package ( Rozowsky et al . , 2011 ) . We then mapped each of the Cast/EiJ×C57Bl/6J hybrid RNA-seq libraries to the splice-junction aware diploid genome using STAR RNA-seq aligner ( Dobin et al . , 2013 ) . For each mapped RNA-seq library , we then estimated the expression level of each transcript with its respective error , for each allele in a Cast/EiJ×C57Bl/6J diploid transcriptome using MMSEQ ( Turro et al . , 2011 ) . Finally , we developed a statistical model testing for genomic imprinting at the level of each transcript , while taking into account the effects of cross , sex , and age across all samples . Specifically , we developed a BRAIM which is a Bayesian variable selection regression model extending Chipman et al . ( 1997 ) by accounting for the measurement error in the response ( ‘Materials and methods’ ) . In our settings , we define the response as the difference between the paternal and the maternal expression levels for a given transcript in each sample in the experiment , that is , the parental bias . The model therefore computes a posterior distribution of the magnitude of the parental bias , and of the effects that all factors have on the parental bias , along with their marginal posterior probabilities ( PPs ) of being significantly different from zero . We fitted our model to 49 , 464 heterozygous autosomal and X-linked transcripts ( comprised of 32 , 399 protein-coding and 17 , 065 non-coding transcripts , see ‘Materials and methods’ ) from 31 , 547 genes expressed at levels above 0 . 01 transcripts per million ( TPM ) units ( solid line in Figure 1—figure supplement 1A and ‘Materials and methods’ ) . A minimal threshold of 0 . 01 TPM was chosen because it allows for accurate and independently validated detection of parentally biased genes . We note that choosing a more stringent expression level cutoff ( dashed line in Figure 1—figure supplement 1A and see ‘Materials and methods’ ) had a negligible effect on all the analyses described from here on . The distribution of the PPs of the parental effect for autosomal transcripts shows that most are not inferred to be imprinted ( Figure 1B and Supplementary file 1A–C ) . It also clearly identifies a group of autosomal transcripts with PP > 0 . 95 , which we set as our cutoff for calling an effect significant ( ‘Materials and methods’ ) . Our analysis also identifies a clear and substantial effect of strain on expression , as evidenced by the distribution of PPs when we account for the cross ( Figure 1B and Supplementary file 1A–C ) . This is most likely due to the divergence of the Cast/EiJ and C57Bl/6J strains , specifically in cis-regulatory regions of the genes expressed in the cerebellum . We did not identify autosomal imprinted transcripts with a sex effect with PP above the 0 . 95 cutoff , indicating that genomic imprinting is sex invariant in the mouse cerebellum . However , a group of autosomal imprinted transcripts were found to have an age effect with PPs above the 0 . 95 cutoff , indicating age-regulated imprinting ( described in more detail below ) . Finally , among X-linked transcripts , above the PP cutoff , we detect two transcripts of the known imprinted Xlr3b gene ( Davies et al . , 2005 ) , and a single transcript from each of two Xlr3b paralogs , Xlr3a and Xlr3c ( Supplementary file 1D–F ) . The high sequence similarity between these paralogs does not allow us to distinguish which of these transcripts are actually imprinted and we therefore excluded them from the remaining analyses . Using these criteria , we identified 124 genes represented by 169 transcripts as candidate imprinted genes in our data ( Supplementary file 1G ) . Among these candidates , 74 genes were reported and validated as imprinted in previosuly published studies ( Supplementary file 1H ) . The remaining 50 genes have not been previously described as imprinted ( Figure 1C ) . To independently evaluate parental allelic expression bias in all these candidates , we used pyrosequencing , a real-time sequence-by-synthesis approach relying on light emissions after nucleotide incorporation ( Wang and Elbein , 2007 ) . As positive and negative controls , we tested 11 known imprinted genes and 11 randomly selected genes with no significant parental effects according to our RNA-seq analysis , respectively . For each candidate and control gene , we tested an average of two SNPs per gene ( ‘Materials and methods’ ) in 12 individual cerebella dissected from P60 and/or 12 P8 hybrids . All samples used in the pyrosequencing validations are distinct from those used in the RNA-seq experiments . We estimated parental allelic effects in the pyrosequencing data using BRAIM . The pyrosequencing data confirmed significant parental effects for 41 of the candidate novel imprinted genes and the expected significant and non-significant parental effects for all positive and negative controls , respectively ( Figure 1C , D and Supplementary file 1I ) . Among the imprinted cerebellar genes identified , we observed a slightly greater number of genes with a paternal bias ( Figure 1E ) . Interestingly , five genes preferentially express distinct isoforms from both the maternal and paternal alleles ( Figure 1E and see below ) . The distribution of the parental biases in the 115 novel and known imprinted genes spans a wide range , from a weak bias ( just above 50:50% ) up to strong monoallelic expression ( 100:0% ) . The bias distribution follows a bimodal shape at its two extremes ( Figure 1F ) , with 36 genes displaying weak biases ( from slightly above 50:50–60:40% ) and 55 genes showing strong biases ( 90:10–100:0% ) . The remaining 32 genes show a moderate bias , from above 60:40% to below 90:10% . Interestingly , the weak bias mode is enriched with newly identified imprinted genes ( yet includes 10 detected known imprinted genes ) whereas stronger biases are enriched with known imprinted genes ( yet includes 18 newly detected imprinted genes ) . A strong correspondence ( Pearson correlation coefficient = 0 . 91 , p-value < 10−16 ) is found between the biases identified by RNA-seq and pyrosequencing , respectively , thus suggesting high accuracy in their quantification ( Figure 1D ) . Figure 2 shows examples of various levels of allelic parental biases . A newly detected imprinted gene , Fkbp6 , shows almost exclusive expression from the paternal allele , while Ago2 and Nhlrc1 show moderate to weak parental biases , respectively , and Clptm1l is accurately detected as biallelically-expressed . Importantly , these effects appear highly reproducible across cerebellum samples analyzed both with RNA-seq and pyrosequencing . Considering 74 known imprinted genes detected in this analysis as true positives , together with 41 newly validated imprinting genes out of 50 novel candidates ( altogether represented by 160 transcripts ) , the precision of our approach is ∼93% and increases the number of identified mouse imprinted genes by ∼30% ( from 138 to 179 , see Supplementary file 1G , H ) . 10 . 7554/eLife . 07860 . 005Figure 2 . Examples of genes imprinted in the cerebellum and of a biallelic control . Fkbp6 , Ago2 , and Nhlrc1 show parentally-biased expression in the cerebellum as observed by RNA-seq ( left ) and confirmed with pyrosequencing ( right ) . Clptm1l shows biallelic expression in the cerebellum in both RNA-seq and pyrosequencing experiments . For each replicate ( N = 48 ) , red indicates maternal expression while blue indicates paternal expression . The Y-axis shows the RNA-seq expression level in natural log of TPM units ( ln ( TPM ) ) , as derived from the posterior distribution of expression levels reported by MMSEQ . Each box is centered at the posterior mean , extends one posterior standard deviation away , and the bottom and top notches are the minimum and maximum posterior samples , respectively . Effects summary shows the posterior distributions of the effects of all experimental factors: parental , cross , sex , and age ( box centered at the posterior mean , extends one posterior standard deviation away , and bottom and top notches are the minimum and maximum posterior samples , respectively ) with their respective PPs on top . For the parental effect , blue represents paternal and red represents maternal expression . For the cross effect , gray represents F1r and black represents F1i . For the sex effect , pink represents female and cyan represents male . For the age effect , light khaki represents P8 and dark khaki represents P60 . DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 005 We considered the possibility that true imprinted genes may not meet our parental bias PP cutoff ( i . e . , false negatives ) . We therefore handpicked 18 genes below our 0 . 95 cutoff that displayed a trend resembling the allele-specific expression patterns of weakly imprinted genes . Using pyrosequencing we successfully confirmed a significant parental effect in 10 of these 18 genes ( Supplementary file 1I ) . For example , Casd1 , a gene reported to be imprinted in other tissues ( Ono et al . , 2003 ) , only displayed a paternal effect with PP of 0 . 89 in the RNA-seq analysis , but showed a mild paternal bias ( 52:48%; PP = 0 . 95 ) by pyrosequencing . This observation supports the existence of additional parental biases in our data that do not meet the chosen 0 . 95 cutoff . Our approach readily estimates parental bias at the transcript level , as seen in the allelic analysis of the 11 expressed isoforms of the imprinted gene Rian ( Hatada et al . , 2001 ) ( Figure 3A ) . Our data suggest that most imprinted genes , including Rian , show a consistent imprinting pattern in all isoforms . However , isoform-specific imprinting is readily identified by our approach , and can be seen in two different genomic contexts . A few genes harboring a paternally expressed gene inserted within an intron ( often a retrogene ) have been shown to generate different isoforms from the maternal and paternal alleles ( Wood et al . , 2008; Gregg et al . , 2010; Cowley et al . , 2012 ) . Although it is unclear exactly how such regulation arises , transcriptional interference by the intronic paternally expressed gene may play an important role ( McCole and Oakey , 2008 ) . Our analysis readily identified previously reported cases of genes in which isoforms are subject to such regulation ( e . g . , the H13-Mcts2 locus , Figure 3B ) . Moreover , it detected additional imprinted transcripts in the Herc3 gene- either or both of two short transcripts ( indistinguishable by our sequence data ) from a promoter upstream to the large 25 exons-long transcript ( Gencode transcript IDs: ENSMUST00000141600 . 1 and ENSMUST00000122981 . 1 ) , which are preferentially expressed by the maternal allele ( Figure 3C ) . Other known cases of isoform-specific imprinting are due to differential methylation of alternative promoters ( Arnaud et al . , 2003; Choi et al . , 2005; Peters and Williamson , 2007 ) . Interestingly , we detect a novel maternally expressed short transcript ( 791 bp ) ( Gencode transcript ID: ENSMUST00000149496 . 1 ) at the locus of the paternally expressed Mest gene , whose transcription starts at exon 9 , suggesting alternative promoter usage ( Figure 3D ) . This transcript is presumably a non-coding RNA since no open reading frame could be identified . In total , we detected 8 of the 10 known cases of isoform-specific imprinting ( the missing two cases are the short isoform of Cdh15 , which is not expressed in the cerebellum , and the long isoform of Blcap , which is not heterozygous in the Cast/EiJ×C57Bl/6J hybrids ) , and we uncovered additional imprinted isoforms , including a novel example of isoform-specific imprinting in the Mest locus . For a complete list of genes exhibiting isoform-specific imprinting see Supplementary file 1G . 10 . 7554/eLife . 07860 . 006Figure 3 . Assessment of imprinting at the transcript level . ( A ) Imprinting in all identified spliced variants of Rian in the P8 cerebellum . ( B ) Imprinting of Mcts2 and H13 transcripts in the P8 cerebellum . ( C ) Imprinting of Nap1l5 and Herc3 transcripts in the P60 cerebellum . ( D ) Imprinting in transcripts of Mest in the P60 cerebellum . In all figures , gene models are in dark gray and transcript models are colored according to the preferentially expressed parental allele ( maternal in red and paternal in blue ) . Parental-specific expression across all replicates is shown below transcript models in natural log of TPM units ( ln ( TPM ) ) , with the posterior distribution of the parental biases across replicates indicated in the right inset boxes with corresponding PPs on top . Asterisks indicate that the associated transcript is developmentally regulated . DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 006 Comparison of imprinted gene expression in the cerebellum in adulthood and at P8 , a critical milestone of cerebellar development during which granule cells undergo cell division , migration , and synaptogenesis ( Sillitoe and Joyner , 2007; Hashimoto and Hibi , 2012 ) shows that the imprinted status of eight genes expressed at both P8 and P60 changes according to age ( 7 genes are exclusively imprinted at P8 and one is exclusively imprinted at P60 Figure 4A ) . In addition , we observe changes in the magnitude of the parental bias according to age for 29 out of 107 imprinted genes that are expressed at both age groups , with the majority exhibiting a stronger parental bias at P8 than in the adult ( Figure 4B and Supplementary file 1A ) Moreover , analysis of the overall expression level ( sum of paternal and maternal allelic expression ) reveals that 34 of the 107 imprinted genes are differentially expressed between P8 and P60 ( Supplementary file 1A ) , together with 8 imprinted genes that are exclusively expressed at P8 . Similar to the parental bias , the majority of the 34 age-regulated genes show higher expression in the developing cerebellum than in the adult ( Figure 4C ) . Altogether , 59 imprinted genes ( 51% of all imprinted genes expressed in the cerebellum ) are regulated in either parental bias and/or overall expression level according to developmental stage ( age effect PP > 0 . 95 ) . For the majority of these genes , the highest magnitude of the parental bias and the highest level of expression are observed at P8 , suggesting a potential role for these genes in neurodevelopmental processes . 10 . 7554/eLife . 07860 . 007Figure 4 . Age regulation of imprinted genes in the cerebellum . ( A ) Venn diagram of imprinted genes at P8 and at P60 . ( B ) Proportions of imprinted genes in which the parental bias is regulated during cerebellum development . Age regulated genes ( in green ) include genes with significantly higher parental bias at P8 ( dark green ) and genes with significantly higher parental bias at P60 ( light green ) . ( C ) Proportions of imprinted genes with overall expression regulated according to the developmental stage of the cerebellum . Age regulated genes ( in green ) include genes with significantly higher overall expression at P8 ( dark green ) and genes with significantly higher overall expression at P60 ( light green ) . ( D ) Relation between age fold change ( ln ( P8/P60 ) ) of the overall expression level of imprinted genes and their parental bias . Genes located in the top right quadrant exhibit higher overall expression and stronger parental biases at P8 than at P60 . Genes located in the lower left quadrant exhibit lower overall expression and lower parental biases at P8 than at P60 . Genes located in the top left quadrant exhibit lower overall expression but stronger parental biases at P8 than at P60 . Genes located in the bottom right quadrant exhibit stronger expression but lower parental biases at P8 than at P60 . Although the age-invariant genes ( salmon colored circles ) overlap to some extent with the age-regulated genes ( green colored circles ) , their age PPs are lower than the 0 . 95 cutoff . ( E ) Relation between age fold change ( ln ( P8/P60 ) ) of the expression of the preferred ( PA ) and the non-preferred allele ( Non-PA ) of imprinted genes . Genes aligned along the X-axis display age regulated expression of the PA , while genes aligned along the Y-axis show age regulated expression of the non-PA . Points aligned on the diagonal are imprinted genes with age-regulated expression of both alleles . Genes in which a significant age effect on parental bias and/or overall expression requires change in both alleles are displayed by points scattered around the origin of axes . ( F ) Three examples of age-regulated imprinted genes . For each replicate ( N = 48 ) , maternal expression is in red and paternal expression is in blue . Y-axis is the RNA-seq expression level in natural log of TPM units ( ln ( TPM ) ) . Effects summary shows the posterior distributions of the effects of the experimental factors with their respective PPs on top . For the age effect , light khaki represents P8 and dark khaki represents P60 . Expression effect shows the fraction of overall expression at P8 and P60 in magenta and purple , respectively . Paternal-allele effect shows the fraction of expression exclusively form the paternal allele at P8 and P60 in light blue and royal blue , respectively . Maternal-allele effect shows the fraction of expression exclusively form the maternal allele at P8 and P60 in light red and dark red , respectively . ( G ) A novel imprinted cluster at distal chromosome 1 exhibits age-dependent regulation of parental bias ( corresponding PPs of the parental biases are indicated ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 00710 . 7554/eLife . 07860 . 008Figure 4—figure supplement 1 . Detection of changes in both parentally biased expression and overall expression levels . Additional examples of age-regulated imprinted genes . For each replicate ( N = 48 ) , red indicates maternal expression and blue indicates paternal expression . Y-axis is RNA-seq expression level in natural log of TPM units ( ln ( TPM ) ) . Effects summary shows the posterior distributions of the effects of the experimental factors . For the age effect , light khaki represents P8 and dark khaki represents P60 . Expression effect shows the fraction of overall expression at P8 and P60 in magenta and purple , respectively . Paternal Allele effect shows the fraction of expression exclusively form the paternal allele at P8 and P60 in light blue and royal blue , respectively . Maternal Allele effect shows the fraction of expression exclusively form the maternal allele at P8 and P60 in light red and dark red , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 00810 . 7554/eLife . 07860 . 009Figure 4—figure supplement 2 . Developmental regulation of isoform-specific imprinting and/or expression . ( A ) Imprinting of Mcts2 and H13 transcripts at the P8 and P60 cerebellum . ( B ) Overall expression changes of the Mest paternally expressed T1 transcript and maternally biased T2 transcript between the P8 and P60 cerebellum . DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 009 Next , we examined the relationship between changes in the parental bias of a transcript and changes in its expression level . Our analysis shows that the age effects on the magnitude of the parental bias and on the overall expression are positively correlated , such that both the magnitude of the parental bias and the overall expression level of imprinted genes tend to increase or decrease together with age ( Pearson correlation coefficient = 0 . 33; p-value = 4 × 10−4 , Figure 4D ) . Since the ability to accurately detect and estimate parental biases is limited at low expression levels , we repeated this analysis with a higher expression level threshold ( right to the dashed line in Figure 1—figure supplement 1A , B ) but found no significant effect on our results . The relationship between parental bias and overall expression identified in our analysis may result from the fact that the two parental alleles experience differential transcriptional regulation during development , thereby altering both the magnitude of the parental bias as well as the overall level of expression . To test this hypothesis we fitted our model to the data where we defined the response as either the paternal or the maternal expression level . This analysis reveals that age regulated expression or parental bias ( age effect PP > 0 . 95 ) is achieved either by a significant change in the expression level only of the most highly expressed ( preferred ) parental allele ( PA; 19 genes , aligned along the X-axis in Figure 4E ) , a significant change in the expression level only of the non-preferred allele ( 8 genes , aligned along the Y-axis in Figure 4E ) , or a significant change in the expression levels of both alleles ( 12 genes , aligned along the diagonal in Figure 4E ) . Finally , some genes have a significant change in expression and/or parental bias only when both alleles are combined but not when the alleles are analyzed separately ( 11 genes , scattered around the origin of axes in Figure 4E ) . Thus , our data suggest that altering the expression level of the preferred allele is the most common mode through which imprinted genes are regulated according to age ( p-value = 0 . 03; χ2 test ) , and this allelic regulation changes both the parental bias as well as the overall expression level . It is also possible that the enhanced expression and parental bias observed in the P8 cerebellum originates from different cellular compositions of the developing and adult cerebellum . These scenarios are not mutually exclusive . A number of genes for which the parental bias in the cerebellum is affected by age are associated with developmental processes such as cell proliferation , differentiation , and survival . For instance , the Asb4 gene , which regulates embryonic stem-cell differentiation ( Townley-Tilson et al . , 2014 ) , exhibits highly maternally-biased expression during cerebellar development but is biallelically expressed during adulthood , which is achieved by a significant decrease in maternal expression and a slight increase in paternal expression ( Figure 4F ) . The growth suppressor Grb10 gene exhibits biallelic expression at P8 but exclusive paternal expression in the adult resulting from silencing of the maternal allele ( Figure 4F ) . Interestingly , we observed a switch in the parental bias for the transcription factor Zim1 , from maternal during development to paternal in the adult cerebellum due to a reduction in maternal expression level ( PP = 1 . 0 ) , and no change in paternal expression ( PP = 0 . 19 ) ( Figure 4—figure supplement 1 ) . Our analysis uncovered a novel imprinted locus at the distal end of chromosome 1 , which exhibits age-dependent regulation . The genes Ier5 , Mr1 , Stx6 , and the putative BC034090 gene , which we name here Impar ( for Imprinted and Age Regulated ) are located side by side within 136 KB and show a maternal bias during cerebellum development but biallelic expression in the adult ( Figure 4F , G ) . For all genes in this locus , the shift in parental bias is accompanied by a reduction in the expression level of the maternal allele and to a lesser extent in the paternal allele ( Figure 4F and Figure 4—figure supplement 1 ) . Stx6 has been shown to regulate neuronal migration and the formation of neural processes ( Kabayama et al . , 2008; Tiwari et al . , 2011 ) , two events necessary for the integration of granule cells to the cerebellar circuit occurring at the P8 stage . It would therefore be interesting to determine whether the allelic regulation of this gene , which directly affects its expression level , is critical for this process . Finally , we also observe age effects on the parental biases of specific isoforms of Herc3 , Mest , and H13 , which all show isoform-specific imprinting ( asterisks in Figure 3B–D and Figure 4—figure supplement 2 ) . Of the 115 genes found imprinted in the cerebellum , 106 ( 92% ) are located within 1 MB of at least one other imprinted gene . The remaining 9 genes ( 8% ) are isolated from other known or newly identified imprinted genes . Moreover , although most newly identified imprinted genes from this study are located within or around known imprinted clusters , 7 of the 41 ( 17% ) appear isolated in the genome ( Figure 5A , B ) . We noticed that a substantial number of imprinted genes displaying weak to moderate parental bias localize to the vicinity of imprinted genes showing stronger parental bias ( Figure 5A ) . For example , Ankrd34c and Ctsh , two genes that exhibit a weak paternal bias ( note: the parental bias of Ankrd34c is significant according to the pyrosequencing data only ) , are located up- and downstream to Rasgrf1 , a gene exclusively expressed from the paternal allele ( Figure 5C ) . This observation led us to explore whether parental bias consistently decays as a function of the distance from strongly biased genes . We first defined imprinted cluster centers as genes for which the parental bias of the preferentially expressed allele reaches at least 85% , and subsequently assigned adjacent imprinted genes with an intergenic distance of up to 1 MB as members of the same cluster ( ‘Materials and methods’ ) . This step partitioned the imprinted genes in our data to a total of 24 imprinted clusters . Analysis of the magnitude of the parental bias as a function of the distance to the cluster center reveals a statistically significant negative effect ( p-value = 1 . 08 × 10−4 , ‘Materials and methods’ ) , supporting a model of decay of parental bias from imprinted cluster centers ( Figure 5D ) . Importantly , not all of the imprinted clusters follow these patterns . For instance , the developmentally regulated imprinted cluster on chromosome 1 ( described above and in Figure 4G ) does not contain any strongly imprinted gene that could be categorized as a cluster center , nor does it exhibit a decay in the parental bias of adjacent genes . 10 . 7554/eLife . 07860 . 010Figure 5 . Genomic patterns of mild to moderate parental biases . ( A ) Distribution of imprinted genes according to the level of parental bias and genomic location relative to the nearest imprinted gene . ( B ) Chromosomal locations of novel imprinted genes ( green: novel imprinted genes within 1 MB of known ones , purple: isolated novel imprinted genes ) . ( C ) Example of an imprinted cluster on chromosome 9 where the monoallelically expressed imprinted gene Rasgfr1 is flanked by newly identified imprinted genes exhibiting moderate parental biases . ( D ) The decay of parental bias , as a function of the distance from an imprinted cluster center . Magenta: regression line , gray polygon: corresponding standard errors . ( E ) Example of mammalian conservation of the micro-synteny of the Tgfb1i1 , Bag3 , and Inpp5f genes within an imprinted cluster on distal chromosome 7 . ( F ) Distribution of the mammalian conservation of micro-synteny of mouse gene-pair orthologs . The numbers of weakly imprinted gene-pair orthologs are indicated at the bottom and the vertical dashed line indicates the mean conservation across all pairs . ( See ‘Materials and methods’ for further details . ) DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 010 The clustering of imprinted genes is regarded as a hallmark of their genomic organization and is thought to reflect common regional control within a given cluster ( Reik and Walter , 2001 ) . Our observations suggest that genes with weak to moderate parental biases may be indirectly affected by the silencing taking place at adjacent and strongly biased imprinting cluster center genes , perhaps through local chromatin conformation . Alternatively , more directed processes may regulate weak to moderate parental biases as these biases may be functionally important . Moreover , genes with weak biases in some tissue may display robust parental biases in specific cell types or developmental stages . We reasoned that if the imprinting regulation of genes in which we detect weak-to-moderate parental bias at the boundaries of imprinted clusters is functionally important , natural selection may operate against disruption of the clustered organization , and thus the clustered organization , or micro-synteny , of such genes would tend to be conserved during mammalian evolution . Alternatively , if imprinting regulation of genes with weak-to-moderate parental bias is not functionally important , and simply reflects the indirect byproduct of regulation of adjacent genes , the micro-synteny of these genes is not expected to be strongly conserved across mammalian genomes . To test these hypotheses , we derived all pairs of adjacent genes in the mouse genome with orthologs in at least one of 15 other mammalian genomes ( 409 , 874 gene pairs ) and estimated the tendency of their mammalian orthologs to be adjacent as well ( Figure 5E , F , and ‘Materials and methods’ ) , using a probabilistic phylogenetic model analyzing phyletic patterns of presence and absence ( Cohen et al . , 2008 ) . This analysis reveals that the mean propensity of adjacency of mammalian orthologs of weakly imprinted gene pairs in the mouse ( parental biases below 85% ) is significantly higher than the mean propensity of adjacency of orthologs of all mouse adjacent gene pairs ( p-value = 0 . 007 , Figure 5F ) . These findings suggest that the micro-synteny of imprinted clusters is evolutionary conserved , thus supporting the idea that the imprinting regulation of mouse genes with weak-to-moderate parental biases is functionally important . This also suggests that the orthologs of many of these genes may be imprinted in other mammalian species . In addition to newly identified imprinted genes associated with known imprinted clusters , several novel imprinted genes appear isolated from any other known imprinted gene ( >2 MB away ) . Previous studies have reported a differential methylation between the two parental alleles for some of these genes . For example , differential methylation was observed in a region of chromosome 13 immediately downstream of Nhlrc1 ( Xie et al . , 2012 ) , a novel paternally-biased gene in our results ( Figure 2 ) . In humans , mutations in Nhlrc1 cause Lafora progressive myoclonic epilepsy ( Romá-Mateo et al . , 2012 ) , a fatal neurological disorder characterized by the presence of massive intracellular inclusions observed in several neuronal cell types across the brain including the cerebellar granule cells . Differential methylation but not parentally biased expression in the brain was also reported within the Actinin alpha 1 ( Actn1 ) gene at chromosome 12 ( Calaway and Domínguez , 2012 ) , which codes for a protein that regulates cytoskeleton interactions with the membrane . Our results show that this gene is indeed preferentially expressed from the paternal allele in the cerebellum . We also uncovered the paternal-specific expression of the Fkbp6 gene , which is isolated on distal chromosome 5 ( Figure 2 ) . Interestingly , this gene displays maternal allele-specific binding of ZFP57 , a DNA-binding protein that specifically binds to the majority of imprinted genes and protects them from demethylation after fertilization ( Quenneville et al . , 2011 ) . Tissue-dependent regulation of imprinting has been described for a subset of known imprinted genes ( Gregg et al . , 2010; Prickett and Oakey , 2012 ) . We selected 28 imprinted genes ( 20 known and 8 novel representing multiple independent clusters ) based on the most interesting predicted biological functions and performed a systematic pyrosequencing quantification of the parental expression bias in 16 brain macro-regions and seven non-neural peripheral tissues in the adult ( Figure 6A , B and ‘Materials and methods’ ) . This analysis revealed pronounced changes of parental bias across genes and tissues ( Figure 6C , D and Supplementary file 1J , K ) . 10 . 7554/eLife . 07860 . 011Figure 6 . Spatial regulation of imprinted genes . ( A ) Legend of brain regions analyzed , colored according to their broad developmental relatedness . OB: Olfactory Bulb , AC: Anterior Cortex , PC: Posterior Cortex , Hp: Hippocampus , CA: Cortical Amygdala ( which is lateral to the brain midline and hence not captured by this sagittal section ) , CP: Caudate Putamen , NA: Nucleus Accumbens , Pa: Pallidum , SA: Striatum-like Amygdala , Th: Thalamus , Hy: Hypothalamus , DM: Dorsal Midbrain , VM: Ventral Midbrain , Cb: Cerebellum , Po: Pons , My: Medulla . ( B ) Legend of body tissues analyzed . Lg: Lung , Hr: Hearth , Sp: Spleen , Lv: Liver , Kd: Kidney , Sk: Skin , Mu: Muscle . ( C ) Examples of genes whose parental bias is regulated according to organ ( brain vs body tissues ) and of genes whose parental bias is dynamically regulated across the brain . Origins of bar graphs represent biallelic expression . Positive values represent preferential maternal expression ( colored red ) while negative values represent preferential expression of the paternal allele ( colored blue ) . N = 6 in each bar . ( D ) Hierarchical clustering of the heat map representing deviations from biallelic expression ( N = 6 in each square ) in imprinted genes . To the right of the heat map are the ANOVA p-values testing for variability of parental bias across the brain regions ( left columns ) and paired t-test p-values testing for differential parental bias in brain vs body ( right column ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 01110 . 7554/eLife . 07860 . 012Figure 6—figure supplement 1 . Hierarchical clustering of the heat map representing deviations from biallelic expression of imprinted genes in the brain . DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 01210 . 7554/eLife . 07860 . 013Figure 6—figure supplement 2 . Shared patterns of spatial regulation of parental biases for different imprinted genes in the brain . ( A ) Shared patterns in the maternal biases of genes in an imprinted cluster on distal chromosome 15 . The patterns also correlate with the spatial regulation of maternally biased genes in an imprinted cluster on proximal chromosome 6 . ( B ) Example of three imprinted genes in separate genomic locations exhibiting similar patterns in the regulation of parental biases . DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 013 We found that a number of previously known imprinted genes such as Igf2 , Igf2r , Zim1 , Copg2 , and Ago2 display noticeable differences either in their imprinting status and/or direction of parental bias across different brain regions and in non-brain tissues ( Figure 6C , D and Supplementary file 1J , K ) . In particular , the significant differences observed between brain regions and non-neural tissues suggest different contributions from the parental genomes to these regions of the body . Igf2 , previously shown to be maternally biased in the adult cortex and preoptic area ( Gregg et al . , 2010 ) , is indeed confirmed to be robustly maternally biased in all 16 brain regions tested , while it is mainly expressed from the paternal allele in non-brain tissues . This stands in contrast to Grb10 , which is paternally biased in the brain and maternally biased in the body ( Charalambous et al . , 2003 , 2010; Garfield et al . , 2011 ) ( Figure 6D ) . Igf2r , which exerts a function that is antagonistic to that of the paternally expressed Igf2 , has been shown to be maternally expressed during development ( Filson et al . , 1993; Ludwig et al . , 1996 ) and in most adult tissues with the exception of the brain ( Hu et al . , 1999 ) . This pattern is also observed in our data ( Figure 6C , D ) . Finally , the putative transcription factor Zim1 shows an intriguing pattern of both paternal and maternal biases both in the brain and in the body ( Figure 6C , D ) . Hierarchical clustering of the parental biases of the tested genes showed three main clades comprising maternally biased genes , paternally biased genes , and genes that are sporadically biased in the brain . Several genes exhibit sharp contrasts in parental bias between the brain and body . This includes genes that are exclusively or nearly exclusively biased in and throughout the brain , such as the maternally-biased Ube3a ( Rougeulle et al . , 1997 ) , Trappc9 , Bag3 , and B3gnt2 genes and the paternally-biased Bcl-x long isoform ( Bcl-xL ) , Inpp5f_v2 isoform ( Choi et al . , 2005 ) , and Begain gene ( Figure 6C , D ) . Hierarchical clustering also clearly separates tissues into two main clades , non-brain and brain , with parental bias in brain appearing much more consistent and robust . The brain is further subdivided into additional sub-clades , which roughly group developmentally related regions . One sub-clade clusters most of the telencephalon , the thalamus , and cerebellum , another groups mesencephalic and rhombencephalic regions , and a third groups diencephalic and basal ganglia regions ( Figure 6D ) . In order to further identify common patterns of parental bias across the brain , we performed a hierarchical clustering analysis confined to the 16 brain regions surveyed , excluding peripheral tissues ( Figure 6—figure supplement 1 ) . This analysis revealed that Ago2 , Chrac1 , and the long isoform of Trappc9 , which co-localize to an imprinted cluster in the distal end of chromosome 15 , exhibit a similar pattern of maternal bias across the brain , stronger in the cortex and weaker in the olfactory bulb , hippocampus , and cerebellum ( Figure 6C and Figure 6—figure supplement 2A and Supplementary file 1K ) . Interestingly , Copg2 and the short isoform of Mest , which co-localize near the centromeric region of chromosome 6 , exhibit a similar pattern of bias to that of genes on the distal end of chromosome 15 ( Figure 6C and Figure 6—figure supplement 2A and Supplementary file 1K ) . The Zim1 , Asb4 genes , and Herc3 long isoform , which are located in the proximal end of chromosome 7 , the proximal end of chromosome 6 , and near the centromeric region of chromosome 6 , respectively , also exhibit shared patterns of biallelic expression ( or in the case of Zim1 a weak paternal bias ) in telencephalic regions and cerebellum but strong maternal biases in other brain regions ( Figure 6—figure supplement 2B and Supplementary file 1K ) . These results suggest that the brain executes region-specific programs of imprinting involving genes from the same or distinct imprinted clusters . The differences in parental biases observed across the brain raise the question of when during development these specificities are established . To address this issue , we performed a pyrosequencing analysis of the parental biases of 13 genes inferred to be temporally and/or spatially regulated across the brain and in non-brain tissues , at postnatal days 0 , 8 , 15 , and 64 in the cortex , hypothalamus , and cerebellum ( Supplementary file 1I ) . These three brain regions were selected for developmental analysis because they display contrasting imprinting patterns in the adult . In addition , we analyzed the parental bias of these 13 genes in the entire E15 brain . This analysis revealed substantial spatiotemporal dynamics of the parental bias of several genes ( Figure 7A , B ) . For instance , Blcap exhibits a gradual decrease in maternal bias over development consistently across the three analyzed brain regions . In contrast , the switch from maternal to paternal bias of Zim1 during cerebellum development , occurs gradually along development in both the cerebellum and cortex , yet is not mirrored in the hypothalamus where maternal bias is strongly maintained into adulthood . Moreover , this analysis reveals that the sharp contrast between the parental biases of Igf2 and Grb10 , observed across the brain , seems to be temporally co-regulated . The switch in the expressed allele for both genes happens earlier in the cortex and hypothalamus than in the cerebellum , which roughly coincides with the completion of their development ( Levitt et al . , 1997; Sillitoe and Joyner , 2007; Shimogori et al . , 2010 ) . These results demonstrate that imprinting is a remarkably dynamic process , and suggest that the highly coordinated spatiotemporal regulation of parent-of-origin expression may in turn orchestrate development across different brain regions . 10 . 7554/eLife . 07860 . 014Figure 7 . Spatiotemporal regulation of imprinted genes . ( A ) Analysis of parental biases at different time points during development of the whole brain ( WB ) at embryonic day 15 , and of the cortex , hypothalamus , and cerebellum at four post-natal stages . Origins of bar graphs ( N = 6 in each bar ) represent biallelic expression . Positive values represent preferential maternal expression ( colored red ) , and negative values represent preferential expression of the paternal allele ( colored blue ) . ( B ) Hierarchical clustering of the heat map representing deviations from biallelic expression in imprinted genes ( N = 6 in each square ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 014 In order to gain initial insights into the functional implications of the observed patterns of parental bias across the brain , we investigated whether specific biological pathways are enriched among cerebellum-imprinted genes . The category of programmed cell death ( apoptosis ) includes eight genes previously shown to exhibit parent-of-origin monoallelic expression , and five genes that we have newly identified as exhibiting parentally biased expression ( Figure 8A ) . Our data thus uncover the programmed cell death pathway as a prominent target of imprinting . Noticeably , most of the maternally-biased genes ( Cdkn1c , Kcnk9 , Blcap , and Bmf ) promote apoptosis , while most of the paternally-biased genes ( Ndn , Peg3 , Peg10 , Plagl1 , and Magel2 ) inhibit apoptosis . One gene of particular interest is the paternally-biased Bcl-x ( Gregg et al . , 2010 ) , which can produce two distinct protein isoforms: the anti-apoptotic Bcl-xL and the pro-apoptotic Bcl-xS , translated from two distinct mRNA transcripts ( Boise et al . , 1993 ) . Bcl-xL is by far the predominant Bcl-x transcript in the brain ( Krajewska et al . , 2002 ) , where it exhibits widespread paternally-biased expression ( Figure 8B ) . Also of interest is the gene Bag3 , which was newly identified by our study as maternally-biased in the brain and biallelic in peripheral tissues ( Figure 8B ) . The protein products of genes Bcl-xL and Bag3 have multiple functions , including some unrelated to apoptosis ( Roth and D'Sa , 2001; Rosati et al . , 2011 ) . However , a suggested function of these proteins , mediated by their mutual interaction , is to prevent the mitochondrial release of factors controlling activation of caspases and thus the irreversible commitment to undergoing apoptosis ( Jacobs and Marnett , 2009 ) . Our RNA-seq analysis of the cerebellum revealed that the maternal bias of Bag3 substantially increases from P8 to P60 whereas Bcl-xL shows a substantial decrease in paternal bias between these two time points ( Figure 8—figure supplement 1A ) . Further quantification of parental biases across 16 regions of the adult brain ( Figure 8B ) shows that , with the exception of the ventral midbrain and pons , the parental biases of Bag3 and Bcl-xL are significantly anti-correlated ( Pearson correlation coefficient = −0 . 56; p-value = 0 . 04 , Figure 8B ) . This pattern is most evident in the cortex and cerebellum , as well as in central striatal regions , caudate putamen , and hippocampus . Variation in the parental biases of Bcl-xL and Bag3 are also observed in our developmental analysis but an anti-correlation between the two is not as apparent ( Figure 8C ) . 10 . 7554/eLife . 07860 . 015Figure 8 . The role of genomic imprinting in the apoptotic pathway in the brain . ( A ) Imprinted genes ( color coded by parental bias: red for maternal and blue for paternal ) involved in the regulation of apoptosis . Colored boxes denote genes exhibiting weak to moderate parental biases . Lines with arrowheads indicate pro-apoptotic function and notched lines indicate anti-apoptotic function . ( B ) Anti-correlated spatial regulation of Bcl-xL and Bag3 biases in the adult brain and body ( Pearson correlation coefficient = −0 . 56; p-value = 0 . 04 ) . ( C ) Regulation of Bcl-xL and Bag3 biases in the developing brain . ( D ) Body and brain weights of adult ( P80 ) mice bearing nervous system-specific deletions of the maternal ( MD ) , paternal ( PD ) , or neither ( wild type ( WT ) ) allele of Bcl-x . Each data point represents an individual mouse . Data include both males and females , as no significant gender-specific differences were observed . ( E ) Representative images of NISSL-stained coronal sections that were used for measuring cortical areas . Tissue sections shown are from P80 male mice of each genotype . ( F ) Cross-sectional areas of cortex , olfactory bulb ( OB ) , and cerebellum ( Cb ) for each genotype . Each data point represents a coronal section from a total of six P80 male mice . ( G ) Quantification within the cortex of the number of cells per section labeled with specific markers: DAPI ( all cells ) , NEUN ( neurons ) , S100ß ( subset of glia ) , Vglut1 ( Slc17a7 , subset of excitatory neurons ) , and Gad1 ( subset of inhibitory neurons ) . Each data point represents a coronal section from a total of six P80 male mice . DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 01510 . 7554/eLife . 07860 . 016Figure 8—figure supplement 1 . Cell-type specific effects in the cortex of brain-specific deletion of the paternal and maternal Bcl-x alleles . ( A ) Bag3 shows biallelic expression in the developing cerebellum , maternally biased expression in the adult cerebellum , and its overall expression is up regulated from P8 to P60 . Bcl-xL shows paternally biased expression in the developing cerebellum that is significantly reduced in the adult cerebellum , and its overall expression is up regulated from P8 to P60 . ( B ) Quantification within the cortex of the density of cells per section labeled with specific markers: DAPI ( all cells ) , NEUN ( neurons ) , S100ß ( glia ) , Vglut1 ( Slc17a7 , excitatory neurons ) , and Gad1 ( inhibitory neurons ) . Each data point represents a coronal section from a total of six P80 male mice . ( C ) Representative images from two-color immunofluorescence analyses of NEUN ( green ) and S100β ( red ) expression within the cortex . Tissue sections shown are from P80 male mice of each genotype . Magnified views of boxed regions are shown ( right ) . ( D ) Representative images from two-color RNA fluorescent in situ hybridization analyses of Vglut1 and Gad1 expression within the cortex . Tissue sections shown are from P80 male mice of each genotype . Magnified views of boxed regions are shown ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 01610 . 7554/eLife . 07860 . 017Figure 8—figure supplement 2 . Effects of brain-specific deletion of the paternal and maternal Bcl-x alleles . ( A and B ) Representative images of NISSL-stained coronal sections that were used for measuring olfactory bulb ( A ) and cerebellum ( B ) areas . Tissue sections shown are from P80 male mice of each genotype . ( C ) Cross-sectional areas per section for lobules 4–6 ( left ) , the granular layer of lobule 6 ( middle ) , and the molecular layer of lobule 6 ( right ) of cerebella from mice of each genotype . Each data point represents the average of three coronal sections from a single P80 male mouse . ( D ) Quantification of the density of cells ( left ) and average number of cells ( right ) per section within the molecular layer of lobule 6 of the cerebella from mice of each genotype . Each data point represents the average of three coronal sections from a single P80 male mouse . ( E ) Representative image of RNA fluorescent in situ hybridization staining for Gad1 transcripts in a coronal section of the cerebellum of a WT P80 male mouse . A magnified view of the boxed region is shown ( right ) . ( F ) Quantification of the perimeter length ( left ) , density ( middle ) , and average number ( right ) of Purkinje cells per section within lobule 6 of the cerebella from mice of each genotype . Each data point represents the average of three coronal sections from a single P80 male mouse . ( G ) Quantification of the area ( left ) , density ( middle ) , and total number ( right ) of Gad1+ cells per section within the molecular layer ( excluding Purkinje cells ) of lobule 6 of the cerebella from mice of each genotype . Each data point represents the average of three coronal sections from a single P80 male mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 017 We next sought to investigate the functional significance of the parentally biased expression of genes within the programmed cell death pathway by analyzing the Bcl-x ( Bcl2l1 ) gene , which has a well-characterized role in inhibiting neuronal apoptosis during brain development due to production of the Bcl-xL protein isoform ( Roth and D'Sa , 2001 ) . Complete knockout of Bcl-x results in massive embryonic apoptosis and lethality , whereas loss of one allele has been reported to reduce brain size by ∼15% ( Kasai et al . , 2003 ) . We hypothesized that if the paternal and maternal alleles unequally contribute to the expression of Bcl-xL in the brain , and if this differential parental contribution is of functional significance , the differential loss of each parental allele should result in distinct functional outcomes . In particular , loss of the paternal ( more highly expressed ) allele should result in a significantly more severe phenotype than the loss of the maternal ( less expressed ) allele . To test this hypothesis , we generated deletions of the maternal and paternal alleles of Bcl-x specifically in the brain by transmitting a floxed Bcl-x allele from either the mother or the father to offspring along with a Nestin::Cre transgene , which is expressed specifically in the brain . The floxed Bcl-x allele enables CRE-dependent deletion of both the anti-apoptotic Bcl-xL and pro-apoptotic Bcl-xS isoforms . However , since Bcl-xL is substantially more highly expressed in the brain ( Krajewska et al . , 2002 ) , the brain-specific deletion of Bcl-x is expected to affect the Bcl-xL isoform nearly exclusively . As expected , neither the brain-specific maternal deletion nor paternal deletion of Bcl-x had a significant effect on the body weight of adult mice ( week 11–12 , ∼P80 ) . However , we found that mice with a paternal Bcl-x deletion had a ∼15% reduced brain weight compared to wild type ( WT ) littermate controls , which carried the Nestin::Cre transgene and two WT copies of Bcl-x ( Figure 8D ) . In contrast , mice with a maternal Bcl-x deletion exhibited no significant difference in brain weight compared to WT littermates . We next asked whether the paternal Bcl-x deletion has observable effects on brain morphology by measuring the cross-sectional area of three different brain structures: cortex , olfactory bulb , and cerebellum . In all three structures , significant differences in area were observed between mice with a paternal Bcl-x deletion and both mice with a maternal deletion and WT littermates , while no significant differences were found between mice with a maternal Bcl-x deletion and WT littermates ( Figure 8E , F , and Figure 8—figure supplement 2A , B ) . Taken together , these results support the hypothesis that the paternal allele of Bcl-x makes a more substantial contribution to brain development than the maternal allele . We next assessed the cell specificity of the reduced brain weight phenotype observed upon the paternal deletion of Bcl-x . For these analyses , we used histochemical methods to quantify total cortical cells ( DAPI ) as well as subsets of cells that express specific cellular markers: NEUN ( RBFOX3 , neurons ) , S100ß ( subset of glia ) , Vglut1 ( Slc17a7 , subset of excitatory neurons ) , and Gad1 ( subset of inhibitory neurons ) . Our data show that DAPI+ , NEUN+ , and Vglut1+ cells were significantly reduced in paternal Bcl-x deletion cortices compared to their maternal deletion and WT counterparts , while S100ß+ and Gad1+ cells were not significantly affected ( Figure 8G , H , and Figure 8—figure supplement 1B–D ) . Cell densities between the three genotypes differed only modestly for all cell types examined ( Figure 8—figure supplement 1B ) . Similar analyses of the effects of paternal and maternal Bcl-x deletion on lobule 6 of the cerebellum revealed significantly fewer cells in the molecular layer of paternal deletion compared to WT mice ( Figure 8—figure supplement 2C–E ) . Within the molecular and Purkinje layers , Gad1+ neurons were significantly reduced in number in mice with a paternal Bcl-x deletion compared to WT littermates ( Figure 8—figure supplement 2F , G ) . Together , these results reveal that deletion of the paternal allele of Bcl-x causes dramatic reductions in brain size and cell number that are not observed upon deletion of the maternal allele . Moreover , these effects differ according to cell type and brain region , with specific cell types affected in a given region . They therefore suggest a largely unexpected role of genomic imprinting in the regulation of the distribution of specific cell types in distinct areas of the mouse brain . In this study we used the high resolution afforded by RNA-seq to profile genomic imprinting in the mouse cerebellum with high statistical power ( 48 biological samples ) . We analyzed expression of the parental alleles at the individual transcript level , and simultaneously estimated the effects of all factors in the experimental design ( age , sex , and mouse cross ) on parental allelic expression . This methodological approach allows precise quantification of the entire spectrum of parental biases observed in imprinted genes , and accurately reveals the contribution of each parental allele to the overall gene expression . In addition to 74 imprinted genes that had been previously validated in earlier studies ( Morison et al . , 2005; Tunster et al . , 2013 ) , our analysis detects 41 novel imprinted genes that we successfully confirmed by pyrosequencing . Taken together , our approach has a precision of ∼93% and our data increase the total number of reported imprinted genes in the mouse by approximately 30% . Our approach demonstrates reliable inference of genomic imprinting when based on a large number of biological samples and a statistical method such as BRAIM that explicitly estimates the biological variability . Previous surveys of imprinting in the mouse either had only a single pair of F1i and F1r sample ( Gregg et al . , 2010 ) or had multiple such samples yet combined counts of reads across samples of the same cross ( Babak et al . , 2008; Wang et al . , 2008; DeVeale et al . , 2012 ) thereby overlooking biological variability and risking reaching erroneous conclusions ( Simpson's paradox ) . A direct side-by-side comparison between our results and previously published data is not a fair comparison as our method naturally penalizes for the lower number of samples and possibly the weaker signal due to shallower sequencing . In addition , since none of the previously published high throughput screens for imprinting aimed to systematically validate all identified candidate imprinted genes , it is impossible to objectively estimate their associated false positive rates as achieved in this work . Importantly , most of the newly identified imprinted genes , as well as a substantial number of those previously known , display weak-to-moderate parental biases rather than the more conventionally expected monoallelic expression . As the tissue samples analyzed by RNA-seq are comprised of a mixture of cell types it is impossible to distinguish whether the parental biases are present in all cell types of the cerebellum or whether they may result from averaging monoallelic expression of given transcripts in some cell types and biallelic representation in others . Interestingly , cell-type-specific imprinting has been reported for Gnas in the kidney within different segments of nephrons ( Weinstein et al . , 2000 ) , and for Ube3a and Snx14 in neurons but not glial cells ( Yamasaki et al . , 2003; Huang et al . , 2014 ) . Clarifying the origin of parental biases is clearly warranted and will require allelic resolution in defined cell types . Is parentally-biased expression a form of genomic imprinting ? Imprinted genes are often described and even defined as mononallelically expressed in a parent-of-origin specific manner . Such definition , however , is incomplete since a substantial number of known imprinted genes appear preferentially expressed from one of the parental alleles . Indeed , the concurrent detection of both mononallelically-expressed and parentally-biased imprinted genes has been highlighted in multiple genome-wide profiles of genomic imprinting ( Babak et al . , 2008; Wang et al . , 2008; Gregg et al . , 2010; DeVeale et al . , 2012; Zou et al . , 2014 ) . Whether parentally-biased genes correspond to the same biological entity as other imprinted genes has been discussed ( Khatib , 2007; Gregg , 2014 ) , but the regulation and functional significance of this phenomenon remains largely obscure . Our data suggest that monoallelic and parentally biased representations share similar characteristics , and may thus represent different ranges of a common mode of differential regulation of parental allelic expression . First , the majority of imprinted genes with weak-to-moderate parental bias are located nearby or at known imprinted clusters , and we show that this localization is evolutionarily conserved , suggesting it is under purifying selection . Second , genes like Actn1 and Nhlrc1 , which exhibit weak parental biases and are isolated in the genome , have been reported to have differentially methylated regions ( DMRs ) between the parental alleles ( Calaway and Domínguez , 2012; Xie et al . , 2012 ) . These DMRs may thus serve as imprinting control regions as has been shown for monoallelically expressed genes . Third , a substantial proportion of genes with weak-to-moderate biases , as well as strongly imprinted genes , show tissue and developmental-stage specificities in their parental biases . For instance , Copg2 shows a weak maternal bias in the cerebellum ( ∼60% maternal expression to 40% paternal expression ) while its maternal bias in the cortex is close to monoallelic ( 90% maternal expression to 10% paternal expression ) . It is therefore possible that many genes with weak-to-moderate parental bias in the cerebellum are actually strongly imprinted in other tissues or developmental stages . Finally , many of the genes with weak-to-moderate biases are implicated in the same biological processes as genes with strong biases , including cell survival and apoptosis . Altogether , the parental bias and/or overall expression of more than half of the genes imprinted in the cerebellum are regulated according to age , with approximately 25% of cerebellum imprinted genes exhibiting changes in the magnitude of their parental bias between P8 and adult stages . Postnatal day 8 is an important milestone of cerebellar development that includes a peak of granule cell precursor proliferation , the active migration of granule cells to the inner granule layer , and the establishment of connectivity between mossy fibers , granule cells , and Purkinje cells . Interestingly , most of the regulated genes exhibit stronger parental biases and/or higher overall expression during this time point than in the mature cerebellum . This is the case for Cdkn1c , Plagl1 , and Ago2 , which are known to regulate cell survival and differentiation during embryonic development ( Cheloufi et al . , 2010; Tury et al . , 2012; Schmidt-Edelkraut et al . , 2014 ) . A smaller subset of genes , including Stx6 , Adam23 , and Pcdhb12 , have been shown to regulate cell motility and neuronal connectivity , suggesting that genomic imprinting is also involved in these developmental processes ( Owuor et al . , 2009; Tiwari et al . , 2011; Chen and Maniatis , 2013 ) . Previous studies have shown differential regulation in the overall expression and , in some cases , the imprinting status of imprinted genes across body tissues and brain regions ( Albrecht et al . , 1997; Gregg et al . , 2010; Prickett and Oakey , 2012 ) . Our study adds substantial new insights into this phenomenon by investigating the regulation of parental bias of 28 known and novel imprinted genes , across 16 brain regions and 7 somatic non-neural tissues . Three interesting observations emerge from this analysis . First and most noticeably , many genes tested appear uniquely imprinted in the brain . Second , a subset of the genes tested show significant variability in the degree of parental bias across different brain regions . Third , genes within an imprinted cluster , as well as occasional genes located on distinct chromosomes , display interesting similarities in the patterns of paternal bias across brain regions . Additionally , for a subset of imprinted genes , we observed regulation of parental bias according to both the developmental stage and the brain region , suggesting that differences in parental bias observed throughout the adult brain may originate during the lineage specification of distinct brain regions , and may also result from different cellular compositions of the tissues analyzed . The substantial spatial and temporal regulation of parent-of-origin allelic expression observed in our study , particularly between brain and non-brain tissues , raises the question of the specific mechanisms that may control these dynamics . For most imprinted genes , monoallelic or parentally biased expression has been shown to be primarily dependent on DNA methylation marks established in one of the parental germ lines , and maintained post-fertilization in the parental alleles of the embryo and mature organism ( Bartolomei and Ferguson-Smith , 2011 ) . In a smaller subset of imprinted genes , the differential methylation of the parental alleles is established in somatic tissues ( Wang et al . , 2014a ) . Other genes such as Dlk1 and Cdh15 become biallelically expressed in specific cell types after gaining DNA methylation on the normally unmethylated allele ( Ferrón et al . , 2012; Proudhon et al . , 2012 ) . Based on these established mechanisms , it is possible that the observed spatial and temporal regulation of parental bias depends on cell-lineage-specific acquisition of DNA methylation or other epigenetic modifications in one of the parental alleles . It is also possible that changes in parental bias result from loss of methylation or perhaps hydroxymethylation ( Guo et al . , 2011 ) of DMRs . Alternatively , spatial and temporal variations in parental biases may depend on the presence or absence of chromatin regulators directed by DNA methylation , which in turn control necessary steps to achieve preferential expression of one of the parental alleles ( Kulinski et al . , 2013 ) , or may rely on the use of brain- and brain region-specific promoter elements . If the parental bias of a certain gene changes according to cell-type , different cellular compositions may also contribute to the observed variability across tissues . This possibility is particularly intriguing in cases where parental biases change according to brain region . What , if any , is the functional significance of the dynamic regulation of parental biases identified in our study ? We reasoned that the notion of a significant functional contribution of the observed parental bias to normal brain function and development should be supported by at least three elements: first , the parental bias of a given gene must be evolutionary conserved , second , changes in parental bias of a given gene should be consistent with changes in overall expression level , and third , uniparental deletions should lead to different phenotypes . Our data partially support these three notions . First , although we did not investigate parental biases in other species , the evolutionary conserved micro-synteny of biased genes suggests that natural selection may operate against the disruption of the biased expression . Second , an attractive hypothesis underlying parental allelic biases may lie on the requirement for different brain regions , and possibly neuronal types , to tightly regulate the expression of certain genes for proper cellular or network function . Converging evidence from mental disorders associated with slight over- or under-regulation of certain genes ( Dölen et al . , 2007; Ramocki and Zoghbi , 2008 ) suggests that normal brain function may indeed require precise gene titration . Our study reveals that certain genes undergo striking variations in parental bias from one brain region to the next , or from one developmental stage to the other . For example , Ago2 shows a modest 60% maternal bias in the cerebellum and olfactory bulb , but a robust ∼80% maternal expression in the cortex . Similarly , Zim1 shows biallelic expression in cortical regions , 85% maternal bias in the pallidum and hypothalamus , and 60% paternal expression in the cerebellum . These data , however , were collected using pyrosequencing , which does not inform on the overall gene expression level , hence such spatial changes in parental biases may , or may not , be associated with up- or down regulation of genes . The RNA-seq data enabled us to detect a positive correlation between developmental changes in the magnitude of parental biases and overall gene expression for most imprinted genes . Moreover , we find evidence that in the majority of cases , an age-regulated increase in parental bias ( age effect PP > 0 . 95 ) is accompanied by a significant up-regulation in the expression level of the preferred allele and thus , an increase in parental bias correlates with an increase in overall expression level . This indicates that changes in parental bias and overall expression are coordinated according to development stage and brain region , which may effectively modulate gene dosages in individual cells . In addition , or alternatively , changes in parental biases may arise from different cellular distributions across developmental and adult brain areas . Both scenarios commonly point to a highly dynamic regulation of genomic imprinting according to neuronal types . Finally , the distinct phenotypes observed in mice bearing paternal and maternal deletions of Bcl-x argue strongly that , even for genes displaying weak-to-moderate biases , the two parental alleles may differ in their functional significance . Several imprinted genes are known to play important roles in the apoptotic pathway , and we uncovered additional imprinted genes with apoptosis-related functions . Apoptosis plays an important role during brain development where it regulates cell population size and refines neuronal circuits by removing poorly connected or non-functional cells ( Buss et al . , 2006 ) . Accordingly , most of the imprinted apoptosis-related genes observed in our study exhibit stronger overall expression and parental biases in the developing cerebellum than in the adult . We also observe a significant number of pro- and anti-apoptotic genes that are differentially expressed according to brain region . Considering that most neurons in the adult brain are post-mitotic , the functional significance of such spatial regulation is still unclear . Some imprinted genes , however , are known to be highly pleiotropic ( e . g . , Cdkn1c , Tury et al . , 2012 ) , hence alternative functions of these imprinted apoptotic genes in the adult brain cannot be ruled out . The Bcl-x gene , which performs a critical anti-apoptotic function in developing neurons ( Kuan et al . , 2000 ) , exhibits a moderate bias in most adult brain regions and a slightly larger bias in the developing brain . Remarkably , the deletion of the paternal allele of Bcl-x results in a significant reduction in brain size and brain cell number compared to littermate controls , while deletion of the maternal allele has no significant effect . In addition , our histochemical analyses of brain regions from mutant mice and littermate controls reveal intriguing evidence for distinct functional consequences of the paternal Bcl-x deletion on specific cell types within a given region of the brain . In the cortex , we find that the paternal Bcl-x deletion preferentially affects neurons , but not glia , and more specifically , Vglut1-positive excitatory , but not Gad1-positive inhibitory neurons . In contrast , the Gad1-positive Purkinje cells are clearly affected by the paternal deletion in the cerebellum . An unambiguous interpretation of these results will require future investigations to identify the origin of the paternal bias of Bcl-xL . If the observed parental bias of Bcl-xL within a given brain region is shared within all cells of the tissue , the deletion of the paternal allele should result in a larger reduction of Bcl-xL expression than deletion of the maternal allele , and the observed phenotype may reveal a high cellular sensitivity to the dosage of Bcl-xL at the cellular level . In contrast , if the observed parental bias of Bcl-xL in a given brain area originates from a mixture of cells expressing exclusively the paternal allele of Bcl-xL with cells expressing both parental alleles equally , the deletion of the paternal allele should result in the complete absence of Bcl-xL in mononallelically-expressing cells , likely resulting in cell death . Deletion of either parental allele of Bcl-xL in biallelically-expressing cells may only generate a mild phenotype if any . At the level of the organism , our data uncover how paternal and maternal alleles of Bcl-x make vastly different contributions to brain development , a result that has profound implications for the analysis of parentally-inherited polymorphisms in human health . F1 hybrids were generated by reciprocally crossing C57Bl/6J and Cast/EiJ mouse strains , where we denote by F1i an F1 hybrid derived from a C57Bl/6J father and a Cast/EiJ mother , and by F1r an F1 hybrid derived from a Cast/EiJ father and a C57Bl/6J mother . For the RNA-seq data , we used 48 animal subjects covering both crosses , both sexes , and two age groups: developing animals sacrificed at postnatal day 8 , denoted as P8 , and adult animals sacrificed in the range of postnatal days 56–64 , denoted as P60 . Our experimental design is balanced thus having three factors: cross , sex , and age , with six animal replicates in each factor block . We created a C57Bl/6J , Cast/EiJ diploid genome by incorporating C57Bl/6J and Cast/EiJ single nucleotide polymorphisms and indels ( obtained from the Mouse Genome Project: ftp://ftp-mouse . sanger . ac . uk/REL-1303-SNPs_Indels-GRCm38 ) into the M . musculus GRCm38 reference genome sequence . We additionally created a transcriptome annotation set as follows . We first downloaded the Gencode ( Engström et al . , 2013; Steijger et al . , 2013 ) M2 mouse main gene annotation ( gencode . vM2 . annotation ) general transfer format ( GTF ) file and removed from it the following RNA types: Mt_rRNA , Mt_tRNA , miRNA , rRNA , snRNA , snoRNA , Mt_tRNA_pseudogene , tRNA_pseudogene , snoRNA_pseudogene , snRNA_pseudogene , scRNA_pseudogene , rRNA_pseudogene , miRNA_pseudogene , as they are not supposed to be enriched in our RNA libraries . In order to include as an extensive set of transcripts as possible and to specifically cover retroposed genes due to their known involvement in genomic imprinting ( McCole and Oakey , 2008 ) , we followed these steps to add additional annotated transcripts to gencode . vM2 . annotation . First , we downloaded the Gencode Retrotransposed ( gencode . vM2 . 2wayconspseudos ) GTF file and the ucscRetroInfo2 mm10 mouse genome assembly retrogenes annotation file from the University of California Santa Cruz ( UCSC ) genome browser ( Karolchik et al . , 2014 ) . We eliminated any redundancy between the two transcript sets by selecting the longest transcript between any two transcripts represented in both files . Following that , we eliminated any redundancy between that retrogene set and gencode . vM2 . annotation transcript set by selecting the longest transcript between any two transcripts represented in both sets . In order to remove any redundancy between our retrogene set and single-exon protein coding transcripts ( which is a structural prominent feature of retrogenes ) in gencode . vM2 . annotation we kept the longest of any intersecting protein-coding single-exon transcript in gencode . vM2 . annotation and retrogene in our retrogenes set . Subsequently , we added all transcripts from the UCSC knownGene mm10 mouse genome assembly annotation file , which are not indicated to be represented in the gencode . vM2 . annotation set . Finally , we added all functional RNAs from the functional RNA database ( fRNAdb , Mituyama et al . , 2009 ) , which did not intersect with any of the transcripts in the set we generated in the previous steps and is longer than our 59 bp read length . Altogether , this amounted in 148 , 120 transcripts from 92 , 965 genes , comprised of 60 , 978 protein coding transcripts , 87 , 142 non-coding transcripts , among which 16 , 404 are pseudogenes , 15 , 538 are retrogenes , 2518 are lincRNAs , and the remaining 52 , 682 are of other miscellaneous types . We then used the UCSC liftOver utility to generate a C57Bl/6J , Cast/EiJ diploid transcriptome set from our generated transcriptome set . We note that many retrogenes are highly redundant in sequence with their paralogs . In the case of short , single-end read RNA-seq data , the accuracy of their expression level estimates would therefore , by and large , be low . For this reason , retrogenes added from the ucscRetroInfo2 retrogenes annotation , which are of lower certainty than retrogenes in the Gencode annotation , are indicated in Supplementary file 1A–F . RNA was isolated from tissues of interest using the Trizol reagent ( Life Technologies , Carlsbad , CA ) according to the manufacturer instructions and further purified using DNase I digestion and the RNeasy kit ( Qiagen , Netherlands ) . We required samples to have RNA integrity score of above 9 , according to the Agilent ( Santa Clara , CA ) 2100 Bioanalyzer , to be used for RNA-seq library preparation . For each sample we used 3 μg of total RNA to prepare libraries according to the Illumina ( San Diego , CA ) Tru-Seq RNA Kit v2 sample preparation protocol . Sample purity and integrity was confirmed using the Agilent 2100 Bioanalyzer . We selected an average fragment size of 250 bp . Each animal subject was used to prepare a single library and was sequenced on an individual lane generating 59 bp single-end reads . The average read depth across our samples was 168 , 991 , 714 . 5 . Each RNA-seq library was first subjected to quality and adapter trimming using the Trim Galore utility ( http://www . bioinformatics . babraham . ac . uk/projects/trim_galore ) with stringency level 3 . We then mapped each of our C57Bl/6J×Cast/EiJ hybrid RNA-seq sequenced libraries to the C57Bl/6J , Cast/EiJ diploid genome and transcriptome splice junctions using STAR RNA-seq aligner ( Dobin et al . , 2013 ) allowing a maximum of three mismatches . Specifically , we mapped the data twice where after the first mapping step we incorporated valid splice junctions which were reported by STAR to exist in our RNA-seq data . We note that restricting the number of allowed mismatches to one had no apparent effect on the remaining downstream analyses . Subsequent to the second STAR mapping step we transformed our genomic alignments to transcriptomic alignments and thus filtered any alignments that did not map to our transcript set using custom code . We note that by doing so we allow the reads to unbiasedly align to their best locations in the splice-junction aware genome and subsequently keep our alignments of interest as opposed to aligning the read data directly to the transcriptome ( e . g . , Roberts and Pachter , 2013 ) . Following that , we estimated the expression levels with their respective uncertainties of each transcript in our C57Bl/6J , Cast/EiJ diploid transcriptome using MMSEQ ( Turro et al . , 2011 ) . MMSEQ uses a Bayesian model for estimating expression levels and therefore computes a posterior distribution of the expression levels of each transcript in fragment per kilobase per million ( FPKM ) units . We first transformed these posterior FPKM samples to TPM units as TPM units were shown to be less biased and more interpretable ( Wagner et al . , 2012 ) . We set the minimum expression TPM cutoff to 0 . 01 . While this is a very low expression level , it still enables to detect genes with parentally biased expression , albeit with lower accuracy , in the estimated fraction of the preferentially expressed allele ( see ‘Results’ ) . In any RNA-seq sample , any transcript for which its MMSEQ posterior median TPM was lower than 0 . 01 was set to 0 . 01 ( which we thus used as the minimal measureable expression level and therefore avoid taking the logarithm of 0 ) . At a 0 . 01 TPM expression level cutoff we detect and validate genes with parentally biased expression , although at the very low expression levels the parental biases estimated by RNA-seq strongly deviate from those estimated by pyrosequencing and there is a higher fraction of false positives ( Figure 1—figure supplement 1C ) . This indicates that the accuracy of parental biases estimated by RNA-seq is limited at this range of low expression levels . Accordingly , the distribution of parental biases at very low expression levels does not follow the bimodal shape of parental biases of genes expressed at higher levels ( Figure 1—figure supplement 1B ) . We therefore empirically defined a more stringent expression level cutoff corresponding to where the discrepancy between the parental biases estimated by RNA-seq and pyrosequencing drops dramatically ( dashed line in Figure 1—figure supplements 1C and correspondingly in Figure 1—figure supplements 1B ) . Using an extensive transcriptome annotation set has the advantage of estimating the expression levels ( and therefore testing for parentally-biased expression ) of as many known transcripts that are expressed in the tissue from which RNA was purified , as possible . However , it is very likely that highly similar transcripts ( e . g . , NAGNAG alternative splice forms , Bradley et al . , 2012 ) will not be distinguishable by the read data . This , in turn , would increase the expression level uncertainties of such lowly identifiable transcripts and would therefore reduce power when testing whether their expression is parentally biased . Ideally , one would detect such lowly identifiable transcripts and combine their expression level estimates into that of a single merged transcript , indicating that either or all of them are expressed . Such combined transcripts would therefore have lower expression level uncertainty than their constituents , which would therefore increase power when testing whether their expression is parentally biased . To achieve this , we adopted the approach of Turro et al . ( 2014 ) for combining lowly identifiable transcripts based on the posterior correlation of their expression level estimates , tailored for a diploid transcriptome case . In this approach , for any given RNA-seq sample we compute the Pearson correlation coefficient of the posterior TPM samples of any pair of transcripts from the same locus and the same allele . Subsequently , if the mean Pearson correlation coefficient across all RNA-seq samples for a pair of transcripts in both alleles is lower than a defined cutoff ( which we empirically set to −0 . 25 ) , we combine each of these pairs into a single combined transcript . This process continues iteratively until no pair of transcripts ( or pairs of already combined transcripts ) can be further combined . This consistency between the alleles in the combining process ensures that the resulting combined transcripts are identical for the two alleles and can therefore be tested for parentally biased expression . For inferring whether a given transcript is imprinted , we define our estimand of interest as the difference in the expression level between its paternal and maternal alleles , that is , the parental bias . Intuitively , if the parental bias is approximately zero across all samples we would conclude that the transcript is not imprinted . The reality of RNA-seq data , however , is more complicated than that . First , as mentioned above , we do not obtain accurate estimates of expression levels but rather estimates with uncertainty . Next , our experimental design may include inherent factors that can affect allele-specific expression levels to various extents , such as the mouse cross , sex , and age . Whether of interest or not , the effects of these factors need to be explicitly accounted for . In addition , even though our nearly genetically identical animal subjects are treated with similar conditions , thereby minimizing effects of any additional factors to the ones specified above ( such as environment ) , we still expect biological variability in RNA levels across our subjects ( e . g . , due to litter effects ) . Finally , we would expect technical variability across experiments to add to the biological variability , yet unless addressed explicitly ( e . g . , sequencing the same RNA library as technical replicates ) , the two are indistinguishable . To address all of these issues we developed a statistical model for inferring genomic imprinting from our experimental design for every transcript in our annotation set . Specifically , we have chosen to extend the Bayesian variable selection regression model of Chipman et al . ( 1997 ) by accounting for the measurement error in the response , as uncertainties are naturally propagated in a Bayesian framework . In our model , the response of sample j ( y^j , where j = 1 , … , n samples ) for a certain transcript is the mean posterior difference between the paternal and maternal natural log posterior TPM samples , that is , ( 1 ) y^j=∑s=1Slog ( TPMj ( s ) ( paternal ) +c ) −log ( TPMj ( s ) ( maternal ) +c ) S , where TPMj ( s ) ( paternal ) and TPMj ( s ) ( maternal ) denote the paternal and maternal posterior sample s of S posterior samples , respectively for RNA-seq sample j . By c we denote the minimum detectable expression cutoff of that RNA-seq sample ( described above ) that we add to the TPM in order to avoid taking the log of zero . Since regression parameters are sensitive to the scale of inputs ( Gelman , 2008 ) yet we wish to have a common interpretation for all transcripts we fit our model to , we divide the response y^j by the standard deviation of y^j's across all j ∈ n samples and denote this scaled response as y^j′ . We define the uncertainty ( or measurement error ) of the response as: ( 2 ) ε^j=∑s=1S[log ( TPMj ( s ) ( paternal ) +c ) −log ( TPMj ( s ) ( maternal ) +c ) −y^j]2S−1 . This therefore represents the posterior variance of the estimated parental bias for RNA-seq sample j . We thus model: ( 3 ) ( y^1′ , … , y^j′ , … , y^n′ ) =y′\|z∼MVN ( z , E ) . We denote z = ( z1 , … , zj , … , zn ) as the unobserved ( latent ) true value of the response , and E=diag ( ε^ ) as the covariance matrix of the response errors ε^= ( ε^1 , … , ε^j , … , ε^n ) . We denote by n the total number of samples . In our experimental design , each of our k factors , namely , cross , sex , and age has two levels , and each factor is represented by nrep replicates , therefore n corresponds to 2k × nrep observations . We model the , unobserved , true value of the response , z , as: ( 4 ) z\|β , σ2∼MVN ( Xβ , Σ ) . We denote the experimental design matrix by X = ( X1 , … , Xp ) , the regression factor parameters by β = ( β1 , … , βp ) , and across-samples errors as E = diag ( σ2 ) . In our model we define the parental bias as the mean response across all samples , which is the intercept of the regression . Therefore , X1 is a column vector of 1's , β1 is the parameter that quantifies the effect of the parental bias , and p = k + 1 is the number of factors . Since we are interested in testing whether each effect i is significant ( i . e . , whether βi is significantly different from zero ) , we model the β's as a mixture of two normals , such that the first normal is centered around zero with a small variance , representing a non-significant effect , and the second normal is centered around the estimated βi , and considers the effect as significant: ( 5 ) βi\|τi , ci , δi={N ( 0 , τi2 ) , δi=0N ( 0 , ( ciτi ) 2 ) , δi=1 . δi can be interpreted as a random variable that indicates whether factor i has an effect on the response . δ's are therefore modeled as Bernoulli i . i . d . , with probability pi for each δi: ( 6 ) π ( δ ) ∝∏i∈ppi . The τ's and c's are hyper-parameters for scaling the two normals and pi is the prior probability that factor i has a significant effect on the response . Finally , we put a conjugate prior on σ2: ( 7 ) σ2∼IG ( ν/2 , νλ/2 ) , where ν and λ are hyper-parameters for the location and scale of σ2 , respectively . In order to sample each of the parameters we employ a Gibbs sampling strategy where we seek to sample from the joint full posterior distribution . The joint posterior distribution of the observed data , y′ , the covariance matrix of the response errors , E=diag ( ε^ ) , and the parameters and latent variables , θ = ( z , β , σ2 , δ ) is: ( 8 ) f ( y^′ , E;θ ) =f ( y^′\|z , E ) π ( z\|β , σ2 ) π ( β\|δ , σ2 ) π ( δ ) π ( σ2 ) , where , ( 9 ) f ( y^′\|z , E ) ∝\|E\|−1/2 exp{−12 ( y^′−z ) TE ( y^′−z ) } , ( 10 ) π ( z\|β , σ2 ) ∝\|Σ\|−1/2 exp{−12 ( z−Xβ ) TΣ−1 ( z−Xβ ) } , ( 11 ) π ( β\|δ , σ2 ) ∝ ( σ ) −p exp{−12βT∑δ−1β} , where , ( 12 ) ∑δ=diag ( σciδiτi ) , ( 13 ) π ( σ2 ) ∝ ( σ2 ) −ν2+1 exp{−νλ2σ2} , and , ( 14 ) π ( δ ) ∼multinom ( p ) . In a Gibbs sampling strategy each parameter is iteratively sampled from its conditional posterior distribution given all other parameters . The conditional posterior distribution of z is: ( 15 ) f ( z\|σ2 , β , y^′ ) ∝f ( y^′\|z , E ) π ( z\|β , σ2 ) ∝\|E\|−1/2 exp{−12 ( y^′−z ) TE ( y^′−z ) }×\|Σ\|−1/2 exp{−12 ( z−Xβ ) TΣ−1 ( z−Xβ ) } , where by dropping the terms not involving z we get: ( 16 ) f ( z\|σ2 , β , y^′ ) ∝ exp{−12[ ( y^′−z ) TE ( y^′−z ) + ( z−Xβ ) TΣ−1 ( z−Xβ ) ]} . We use the proposition that if: ( 17 ) f ( z\|μ , y , S , Σ ) ∝ exp{−12[ ( z−μ ) TS−1 ( z−μ ) + ( y−z ) TΣ−1 ( y−z ) ]} , then ( see proof in Gelman et al . , 2013 ) , ( 18 ) z∼MVN ( μz , Λz ) , where , ( 19 ) μz=Λz ( S−1μ+Σ−1y ) , and , ( 20 ) Λz= ( S−1+Σ−1 ) −1 . We thus obtain , ( 21 ) f ( z\|σ2 , β , y^′ ) ∝ exp{−12[ ( z−μz ) TΛz−1 ( z−μz ) ]} , where , ( 22 ) μz=Λz ( Σ−1Xβ+E−1y^′ ) , and , ( 23 ) Λz= ( E−1+Σ−1 ) −1 , and therefore , ( 24 ) z\|σ2 , β , y^′∼MVN ( μz , Λz ) . The conditional posterior distribution of β is: ( 25 ) f ( β\|z , σ2 , δ ) ∝π ( z\|β , σ2 ) π ( β\|δ , σ2 ) ∝\|Σ\|−1/2 exp{−12 ( z−Xβ ) TΣ−1 ( z−Xβ ) }×σ−p exp{−12βT∑δ−1β} , where by dropping the terms not involving β we get: ( 26 ) f ( β\|z , σ2 , δ ) ∝ exp{−12[ ( z−Xβ ) TΣ−1 ( z−Xβ ) +βT∑δ−1β]} . We use the proposition that if: ( 27 ) y\|β , Σ∼MVN ( Xβ , Σ ) , and , ( 28 ) β\|D∼MVN ( 0 , D ) , then ( see proof in Lindley and Smith , 1972 ) , ( 29 ) β\|y , Σ , D∼MVN ( μβ , Λβ ) , where , ( 30 ) μβ=ΛβXTΣ−1y , and , ( 31 ) Λβ= ( XTΣ−1X+D−1 ) −1 . We thus obtain , ( 32 ) f ( β\|z , σ2 , δ ) ∝ exp{−12[ ( β−μβ ) TΛβ−1 ( β−μβ ) ]} , where , ( 33 ) μβ=Λβ ( XTΣ−1z ) , and , ( 34 ) Λβ= ( XTΣ−1X+Σδ−1 ) −1 , and therefore , ( 35 ) β\|z , σ2 , δ∼MVN ( μβ , Λβ ) . The conditional posterior distribution of σ2 is: ( 36 ) f ( σ2\|z , β , δ ) ∝π ( z\|β , σ2 ) π ( β\|δ , σ2 ) π ( σ2 ) ∝[σ2]−n2 exp{−12σ2 ( z−Xβ ) T ( z−Xβ ) }×[σ]−p exp{−12σ2βT∑δ−1β}×[σ2]−ν2 exp{−νλ2σ2} , where by dropping the terms not involving σ2 we get: ( 37 ) f ( σ2\|z , β , δ ) ∝[σ2]− ( n+p+ν ) 2−1 exp{−12σ2[νλ+ ( z−Xβ ) T ( z−Xβ ) +βT∑δ−1β]} . Therefore , ( 39 ) σ2\|z , β , δ∼IG ( 12 ( n+p+ν ) , 12[νλ+ ( z−Xβ ) T ( z−Xβ ) +βT∑δ−1β] ) . The conditional posterior distribution of δ after dropping the terms not involving δ is: ( 40 ) f ( δ\|z , β , σ2 ) ∝π ( β\|δ , σ2 ) π ( δ ) . The conditional posterior distribution of δ , however , is unknown , and therefore it is more suitable to sample each δi independently , given the set δ[−i] = {δ1 , . . , δi−1 , δi+1 , … , δp} . Using Equation 40 we get: ( 41 ) f ( δi\|δ[−i] , z , β , σ2 ) ∝π ( δi\|δ[−i]β , σ2 ) π ( δi , δ[−i] ) . Therefore , ( 42 ) p ( δi=1\|δ[−i] , z , β , σ2 ) =π ( β\|δi=1 , δ[−i] , σ2 ) π ( δi , δ[−i] ) π ( β\|δi=1 , δ[−i] , σ2 ) π ( δi=1 , δ[−i] ) +π ( β\|δi=0 , δ[−i] , σ2 ) π ( δi=0 , δ[−i] ) =π ( δi , δ[−i] ) π ( δi=1 , δ[−i] ) +π ( β\|δi=0 , δ[−i] , σ2 ) π ( β\|δi=1 , δ[−i] , σ2 ) π ( δi=0 , δ[−i] ) , where π ( β\|δi=0 , δ[−i] , σ2 ) π ( β\|δi=1 , δ[−i] , σ2 ) is the ratio of the normal mixture for β from Equation 11 . Our Gibbs sampler therefore follows these steps:Initialize the parameters . Iterate until convergence on sampling each parameter from its conditional posterior distribution:a . Sample z\|σ2 , β , y^′∼MVN ( μz , Λz ) ( Equations 18 , 22 , 23 ) . b . Sample β\|y , Σ , D∼MVN ( μβ , Λβ ) ( Equations 29 , 33 , 34 ) . c . Sample σ2\|z , β , δ∼IG ( 12 ( n+p+ν ) , 12[νλ+ ( z−Xβ ) T ( z−Xβ ) +βT∑δ−1β] ) . d . For i = 1 , … , p , sample δi according to Equation 42 . In all our analyses we ran our Gibbs sampler for 10 , 000 iterations discarding the first 1000 as burn-in . In our experimental design all factors have binary levels: parental effect: paternal and maternal; cross effect: F1i and F1r; sex effect: male and female; and , age effect: P8 and P60 . Since in this study we are interested in contrasting the two levels in each factor we set the X matrix cross , sex , and age columns to 1 for F1i's , males , and P8's , and to −1 for F1r's , females , and P60's . Each of the β parameters , which quantifies the effect of the corresponding factor , can therefore be split into the respective levels of the factor . For example , the β which quantifies the parental effect corresponds to the mean effect across samples . Therefore: ( 43 ) β^=∑j=1nyjn=∑j=1n[log ( TPMj ( paternal ) ) −log ( TPMj ( maternal ) ) ]n=log[∏j=1n ( TPMj ( paternal ) TPMj ( maternal ) ) ]n . Therefore: ( 44 ) exp ( β^ ) =∏j=1nTPMj ( paternal ) n∏j=1nTPMj ( maternal ) n . The geometric means ∏j=1nTPMj ( paternal ) n and ∏j=1nTPMj ( maternal ) n approximate the median paternal and maternal expression across samples . Denoting the overall expression as: ( 45 ) ∏j=1nTPMj ( paternal ) n+∏j=1nTPMj ( paternal ) n=1 , allows us to represent the median paternal and maternal fractions in terms of β^: ( 46 ) ∏j=1nTPMj ( paternal ) n=exp ( β^ ) 1−exp ( β^ ) , and ( 47 ) ∏j=1nTPMj ( maternal ) n=11−exp ( β^ ) . For inferring an effect as statistically significant , we require that the mean posterior value of its corresponding δ parameter samples ( which denote as posterior probability , PP ) be higher than 0 . 95 . As described above , our model has several hyper-parameters that need to be specified . Namely , the normal mixture prior distribution of each βi is scaled by respective τi and ci ( Equation 5 ) . The c parameter acts as a multiplying constant that determines how much higher an important effect has to be relative to a negligible effect in order to be considered significant . The τ parameter allows scaling each effect independently . The shape and scale of the inverse gamma prior distribution of σ2 is determined by ν and λ , respectively ( Equation 7 ) . The ν parameter should be set to a relative uninformative value , yet larger than zero in order to avoid obtaining low values of σ2 which will result in often selecting effects as significant . Finally , each δi has prior probability pi of being selected . We follow Chipman et al . ( 1997 ) in setting ν to a value near 2 ( specifically 2 . 5 ) and setting λ=2var ( y^′ ) 51ν ( ν2−1 ) . Since var ( y^′ ) =1 for all transcripts , for ν = 2 . 5 we get λ = 0 . 04 . We chose τ and c of all transcripts to have empirical values of 0 . 1 and 4 . 25 , respectively , in order to provide a good separation between imprinted and non-imprinted transcripts at the selected PP = 0 . 95 cutoff . In addition , we set p of all effects to 0 . 1 , reflecting our prior belief of the effects being significant . Although the choice of these hyper-parameter values is arbitrary to some extent , if the model is robust this choice should not affect the ranking of transcripts according to the PPs . Therefore , in order to evaluate how the inference of genomic imprinting by our model is affected by our choice of empirical hyper-parameter values we performed the following sensitivity analysis . The empirical value that we chose for each of the five hyper-parameters: τ , c , ν , λ , and p , was perturbed by selecting four other values . Specifically , we perturbed the empirical τ = 0 . 1 value with 0 . 005 , 0 . 01 , 1 , and 2 thus both lowering and elevating the posterior probability of an effect being significant; we perturbed the empirical c = 4 . 25 value only with the higher values: 1 . 7 , 3 . 4 , 5 . 3125 , and 10 . 625 as 4 . 25 was found to be approximately the minimal value for detecting significant effects; we perturbed the empirical ν = 2 . 5 value with 5 , 12 . 5 , 25 , and 50 as ν cannot assume values lower than 2; we perturbed the empirical λ = 0 . 04 value with 0 . 002 , 0 . 004 , 0 . 4 , and 0 . 8 shifting the σ2 prior distribution from informative to uninformative; and finally , we perturbed the empirical p = 0 . 1 value with higher prior probabilities of 0 . 2 , 0 . 3 , 0 . 4 , and 0 . 5 . In each such perturbation we re-fitted our model to the data where all other hyper-parameter values are held fixed at their original empirical unperturbed values ( thereby achieving a one-at-a-time sensitivity analysis ) . We assessed the perturbed results using a receiver operating characteristic ( ROC ) analysis in which the imprinted transcripts obtained by the unperturbed inferences were used as the ground-truth positives , and all other non-imprinted transcripts were used as ground-truth negatives . For practical computational considerations , for all perturbations we used a random sample of 10% of the 47 , 676 transcripts in our data set . In all perturbations the area under the ROC curve ( AUC ) was found to be 1 , except for the τ = 1 and τ = 2 perturbations which obtained AUCs of 0 . 99 and 0 . 95 , respectively ( Figure 9 ) . These results therefore indicate that the ranking of transcripts according to the PP of the parental effect , obtained by our model , is robust to the choice of hyper-parameter empirical values . Therefore , the practice employed in this study of selecting empirical values , setting a parental effect PP cutoff , and experimentally confirming all transcripts with a PP of parental effect above that cutoff , is a reasonable choice for obtaining reliable inference of genomic imprinting from RNA-seq data . 10 . 7554/eLife . 07860 . 018Figure 9 . Sensitivity analysis of BRAIM to the choice of hyper-parameter values . Receiver operating characteristic curves describing the performance of BRAIM in detecting imprinted genes using perturbed values of τ ( A ) , c ( B ) , ν ( C ) , λ ( D ) , and the prior of δ , p ( E ) hyper-parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 07860 . 018 In order to detect all imprinted transcripts in our data we fitted our model to all 48 P8 and P60 samples , thus estimating cross , sex , and age effects on the parental bias . This , however , requires that a given transcript is expressed in both age groups . In addition , a significant age effect may mean that a transcript has a strong parental bias in one age group but none in the other . In that case , the PP of the parental bias may not be significant . For these reasons we additionally fitted our model to the 24 P8 and P60 samples , independently , thus estimating only cross and sex effects on the parental bias for each of the age groups . We thus report all transcripts with a parental effect PP > 0 . 95 in the combined P8 , P60 dataset , or exclusively expressed in one of the age group datasets with a parental effect PP > 0 . 95 as imprinted . Transcripts with an age effect PP > 0 . 95 in the P8 , P60 dataset and with a parental effect PP > 0 . 95 in either of the individual P8 and P60 datasets were additionally reported as imprinted . BRAIM is implemented in R—see Source code 1 . To quantify the effect that physical chromosomal distance of imprinted genes from their corresponding imprinted cluster center has on the magnitude of their parentally-biased expression we performed the following analysis . We started off with the list of transcripts which were either validated to be imprinted in this study or their corresponding genes were validated to be imprinted in previous studies ( Supplementary file 1H ) . Retrogenes were excluded from this list as the context of their genomic location is not strongly relevant for the question addressed by this analysis . For each transcript in this list we computed the parental bias as: ( 48 ) max ( exp ( β^ ) 1−exp ( β^ ) , 11−exp ( β^ ) ) . In words , the median parental expression fraction across all samples ( see Equations 43–47 ) . In order to represent a gene by a single transcript , for each gene we chose the isoform with the most significant parental bias ( i . e . , the isoform with the maximal parental effect PP ) , derived from the model fitted to both age groups . The same criterion was also used to select between host and resident genes as the representative of their genomic location . In order to assign genes to clusters we linearly scanned each chromosome from its 5′ to 3′ end . The most upstream gene on a chromosome ( where we represent the start and end sites of genes encoded on the antisense strand as their end and start sites on the sense strand , respectively ) , was assigned to the first imprinted cluster of that chromosome . Then , if the start site of the next downstream gene is within 1 MB of the end site of the previous upstream gene it was assigned to the same cluster , else it was assigned to a new cluster . Thus , the maximal distance between any two genes within a cluster can be 1 MB . We then defined the genes with a parental bias larger than 85:15% as candidate centers of their respective clusters . Since in some imprinted clusters more than a single gene met this condition , we grouped all physically consecutive candidate cluster center genes to a single gene which start site was defined as the start site of the most upstream candidate center gene and end site was defined as the end site of the most downstream candidate center gene . If an imprinted cluster resulted with more than one such group of candidate center genes flanked by genes with parental biases lower than 85:15% it was broken down to sub-clusters centered on each of these candidate centers . We defined the boundaries of each sub-cluster as half the distance between the end site of its center gene and the start site of the center gene of the adjacent sub-cluster ( or up to 1 MB in the cases of the most up- and downstream sub-clusters ) and thus flanking genes were assigned to the sub-cluster which boundary was downstream to their start site location . We finally removed any gene which start site was more than 1 MB away from its respective cluster center . We then fitted a general linear fixed effects model to non-cluster center genes , where we defined the response as the parental bias and the fixed effect as the intergenic distance from the cluster center . To test whether the clustered organization of mouse imprinted genes with weak parental-expression biases ( lower than 85:15% ) is conserved in mammalian evolution , which therefore suggests functional importance , we carried out the following analysis of conservation of micro-synteny . We defined a pair of adjacent genes in the mouse reference genome as any two genes for which intergenic distance was below 1 MB ( as 92% of the imprinted genes in our data are within 1 MB of another imprinted gene ) . Using this set of adjacent pairs of genes as reference we then constructed a phyletic pattern of orthologous gene pairs as follows . We downloaded the gene orthology assignment between the mouse reference genome and all mammalian reference genomes available in the Ensembl genome browser ( Flicek et al . , 2014 ) , which are assembled at the chromosome level . This included the genomes of the following species: opossum , pig , sheep , cow , horse , cat , dog , rat , marmoset , rhesus , orangutan , gorilla , chimpanzee , and human ( Figure 5E ) . For each comparative genome , if both orthologs of a mouse adjacent gene pair exist in that genome , we labeled this orthologous pair by ‘1’ if their intergenic distance is lower than 1 MB ( i . e . , they are also adjacent in the comparative genome ) and ‘0’ otherwise ( i . e . , they are not adjacent in the comparative genome ) . If either or both orthologs do not exist in the comparative genome we labeled this pair as ‘ ? ’ ( i . e . , unknown ) . Ensembl tables of orthology assignment provide two types of homology between a query and search genomes: a one-to-one and a one-to-many homology . In addition , these tables also provide a binary confidence score for the orthology assignment ( 0 for low and 1 for high ) . As a conservative approach , we filtered out all one-to-many and low confidence orthologs unless the genes orthologous to the mouse pair were found to be within 1 MB of each other . As a result we obtained a phyletic pattern , which is a type of a multiple sequence alignment , where in our case each site in the phyletic pattern is a gene pair orthologous to an adjacent gene pair in the mouse genome with the characters of ‘1’ , ‘0’ , and ‘ ? ’ . All phyletic pattern columns in which all sites were ‘ ? ’ , were removed thus obtaining a phyletic pattern of 409 , 874 sites . We then fitted to this phyletic pattern and corresponding phylogenetic species tree ( Figure 5E ) a probabilistic phylogenetic model estimating the substitution rates between presence and absence ( ‘1’ and ‘0’ , respectively ) and vice-versa ( Cohen et al . , 2008 ) . We specifically allowed the presence/absence ratio to vary across sites and the presence and absence probabilities at the root of the phylogenetic tree to be independently estimated . As a result , we obtained a posterior expectation of the absence-to-presence substitution rate ( i . e . , rate of gene-pair gain: ‘0’ to ‘1’ ) and a posterior expectation of the presence-to-absence substitution rate ( i . e . , rate of gene-pair loss: ‘1’ to ‘0’ ) for each site in the phyletic pattern . The propensity of each orthologous gene pair to be adjacent throughout the mammalian evolution represented by the species used in this analysis is therefore the posterior expectation of its rate of gene-pair gain divided by the posterior expectation of its rate of gene-pair loss . In order to test whether the propensity of adjacency of orthologs of mouse weakly parentally-biased imprinted gene pairs ( with a bias lower than 85:15% ) , is significantly higher than what would be expected by chance , we compared the mean propensity of adjacency of these gene pairs with the mean propensity of adjacency of all gene pairs in the phyletic pattern using a one-sided z-test . We note that not applying any distance cutoff for defining a pair of adjacent genes means that genes with much longer intergenic distances will be included for which the micro-synteny is expected to be much less evolutionary conserved thus potentially biasing our analysis . Pyrosequencing validations were performed in cerebella derived from a different batch of F1 hybrids from that used in the RNA-seq experiments . The same age of animals was used for the pyrosequencing validations whenever the RNA-seq data showed an age-specific parental bias . Otherwise , either age group was used . RNA purification and quality control followed the same procedures described for the RNA-seq data . An average of three SNPs suitable for pyrosequencing analyses were identified for each candidate gene . Previously described three-primer strategy ( Royo et al . , 2007 ) , including a 3′-biotinylated primer common to all reactions , was employed for the amplification of each targeted SNP . All primers were designed using Pyromark Assay Design 2 . 0 . Pyromark One Step RT-PCR kit ( Qiagen ) was used for the amplification of each targeted region , followed by purification of single stranded biotinylated DNA according to the manufacturer instructions . Pyrosequencing was performed on the Pyromark Q96 MD ( Qiagen ) . For each SNP at least 12 replicates were separately analyzed . Statistical analyses of the pyrosequencing data for each SNP were performed using BRAIM . For each tissue we collected six samples from F1i and F1r females . The coordinates of the Allen Reference Atlas guided dissections of all brain regions interrogated . RNA purification , pyrosequencing and analyses of parental biases were conducted as described for pyrosequencing validations . For the heat maps in Figure 6D , Figure 6—figure supplement 1 , and Figure 7B we binned the parental biases with PP > 0 . 95 to five bins: 50:50–60:40 , 60:40–70:30 , 70:30–80:20 , 80:20–90:10 , and 90:10–100:0 , and assigned the parental biases with PP ≤ 0 . 95 to the 50:50 category . The bars in Figure 6C , Figure 6—figure supplement 2A , B , Figure 7A , and Figure 8B , C , correspond to the actual parental biases , regardless of their PPs , along with an error notch corresponding to the 95% confidence interval ( Supplementary file 1H , I ) . To quantify the statistical significance of the difference in the magnitude of the parental bias of each gene between the 16 brain regions and 7 non-neural somatic tissues we used a t-test ( assuming unequal variances ) comparing its brain and non-neural somatic tissues samples . To quantify the statistical significance of the variability of the parental biases across the different brain regions , for each gene , we used a one-way ANOVA test , where the response is the parental bias computed for all samples and the independent variable is the brain region . Mice with a nervous system-specific paternal deletion of Bcl-x were generated by crossing males heterozygous for both a Nestin::Cre transgene ( Tronche et al . , 1999 ) and a floxed allele of Bcl-x ( Rucker et al . , 2000 ) with WT females ( carrying two WT alleles of Bcl-x and lacking the Nestin::Cre transgene ) that had a strain background that was a 50% mixture of each of the Nestin::Cre ( +/+ ) and Bcl-x ( fl/fl ) background strains . Mice with a nervous system-specific maternal deletion of Bcl-x were generated by crossing females bearing a floxed allele of Bcl-x with males bearing a Nestin::Cre transgene . From the same crosses , mice bearing the Nestin::Cre transgene and carrying two WT alleles of Bcl-x were used as WT littermate controls . This is expected to knock out the function of both Bcl-xL and Bcl-xS , as both proteins are encoded primarily by exon 2 , however , Bcl-xL is believed to be substantially more expressed in the brain ( Krajewska et al . , 2002 ) , and thus primarily affected by the Nestin::Cre driven recombination . To minimize any possibility of background strain effects , all mice analyzed had a strain background that was a 50% mixture of each of the Nestin::Cre ( +/+ ) and Bcl-x ( fl/fl ) background strains . Genotyping was performed by PCR , as described ( Rucker et al . , 2000 ) . A non-experimenter performed the genotyping . This person assigned unique animal numbers to each of the animals such that from here-on-out , researchers were completely blinded to the genotypes . Researchers performed all of the tissue-processing and image quantifications fully-blinded to the genotypes of the animals . The data were then given to another researcher who unblinded the animals' genotype , performed statistics , and created figures . Mice aged P78-85 were weighed , transcardially perfused with PBS followed by 4% paraformaldehyde ( PFA ) in PBS . Brains were removed , weighed , postfixed in 4% PFA ( in PBS ) for 4 hr at 4°C , immersed in 30% sucrose ( in PBS ) overnight at 4°C , and then frozen at −80°C until being sectioned . Coronal sections ( 14 µm ) of brain from male mice were prepared using a sliding microtome ( Leica , Germany ) and mounted in series on eight slides , which were subsequently stored at −80°C . Area quantifications were performed on brain sections stained using the NISSL method . In brief , sections were hydrated in a graded series of alcohol , stained with cresyl violet , dehydrated in alcohol , cleared with xylenes , and coverslipped with DPX . Slide-mounted sections were warmed ( 37°C , 5 min ) , equilibrated in PBS ( 5 min , RT ) , fixed in PFA ( 4% in PBS; 10 min , RT ) , washed in PBS ( 3 min , RT ) , permeabilized with Triton X-100 ( 0 . 5% in PBS; 30 min , RT ) , washed in TNT ( 3 × 5 min , RT ) , blocked in fetal bovine serum ( FBS; 10% in TN; 30 min , RT ) , incubated with mouse anti-NEUN ( EMD Millipore , Billerica , MA , MAB377; 1:1000 ) and rabbit anti-S100ß ( Abcam , UK , ab41548; 1:500 ) antibodies ( in 10% FBS; 12 hr , 4°C ) , washed in TNT ( 3 × 5 min , RT ) , incubated with secondary antibodies ( Alexa488- and Alexa647-labeled; Invitrogen , Carlsbad , CA; 1:1000 in 10% FBS; 12 hr , 4°C ) , and washed in TNT ( 3 × 15 min , RT ) . Slides were mounted using Vectashield containing DAPI ( 5 μg/ml ) . Probe target sequences for Vglut1 ( Slc17a7 ) and Gad1 were amplified by PCR and inserted into pCRII-TOPO vector ( Life Technologies ) . Sequences were identical to those used by the Allen Brain Atlas ( http://www . brain-map . org/ ) . Labeled antisense RNA probes were generated from linearized plasmid template using T7 or Sp6 polymerases ( Promega , Madison , WI ) with a digoxigenin RNA labeling mix ( Roche , Switzerland; for Gad1 ) or a dinitrophenyl RNA labeling mix ( Perkin Elmer , Waltham , MA; for Vglut1 ) , treated with DNaseI ( Promega ) , ethanol precipitated , and dissolved in a 30-μl volume of water . Slide-mounted sections were warmed ( 37°C , 5 min ) , equilibrated in phosphate-buffered saline ( PBS; pH 7 . 2; 5 min , room temperature [RT] ) , fixed in PFA ( 4% in PBS; 10 min , RT ) , washed in PBS ( 3 min , RT ) , permeabilized with Triton-X-100 ( 0 . 5% in PBS; 10 min , RT ) followed by Proteinase K ( 20 μg/ml in 50 mM Tris , 5 mM EDTA; 15 min , RT with gentle shaking ) , washed in PBS ( 3 × 3 min , RT ) , fixed in PFA ( 4% in PBS; 10 min , RT ) , washed in PBS ( 3 × 3 min , RT ) , incubated in acetylation solution ( triethanolamine [0 . 1 M; pH 7 . 5] , acetic anhydride [0 . 25%]; 10 min , RT ) , washed in PBS ( 3 × 3 min , RT ) , incubated in hybridization solution ( formamide [50%] , SSC [5×] , Denhardts [5×] , yeast tRNA [250 μg/ml] , herring sperm DNA [200 μg/ml]; 30 min , RT ) , hybridized simultaneously with both Vglut1 and Gad1 antisense RNA probes ( 1:300 each in hybridization solution; 16 hr , 68°C ) , washed with SSC ( 2×; 5 min , 68°C ) , washed with SSC ( 0 . 2×; 3 × 30 min , 68°C ) , incubated in H2O2 ( 3% in TN [Tris-HCl ( 0 . 1 M; pH 7 . 5 ) , 0 . 15 M NaCl]; 30 min , RT ) , washed in TNT ( Tween-20 [0 . 05%] in TN; 3 × 3 min , RT ) , incubated in TNB ( Blocking Reagent [Perkin Elmer; 0 . 05% in TN]; 30 min , RT ) , incubated with anti-digoxigenin-POD antibody ( 1:1000 in TNB; 12 hr , 4°C ) , and washed in TNT ( 3 × 20 min , RT ) . Fluorescent signals corresponding to the Gad1 probe were generated using the Tyramide Signal Amplification ( TSA ) Plus Fluorescein Kit ( Perkin Elmer ) according to the manufacturer's instructions , after which sections were washed in TNT ( 2 × 3 min , RT ) , incubated in H2O2 ( 3% in TN; 1 hr , RT ) , washed in TNT ( 3 × 3 min , RT ) , incubated with anti-dinitrophenyl-HRP antibody ( Perkin Elmer; 1:500 in TNB; 12 hr at 4°C ) , and washed in TNT ( 3 × 20 min , RT ) . Fluorescent signals corresponding to the Vglut1 probe were generated using the TSA Plus Cyanine5 Kit ( Perkin Elmer ) according to the manufacturer's instructions . Slides were mounted using Vectashield ( Vector Laboratories ) containing DAPI ( 5 μg/ml ) . Sections were imaged using a Zeiss ( Germany ) Axioscan . Z1 microscope with a 10× objective . Areas of brain regions were measured from images of NISSL-stained sections using Zen software ( Zeiss ) . Cell densities and numbers were quantified from two-color IF and in situ hybridization images using ImageJ software . Five anterior cortical sections that include the somatomotor cortex ( corresponding to Figures 29–33 in Paxinos and Franklin , 2007 ) were analyzed . Three coronal cerebellar sections that included lobules 4–6 of the cerebellar vermis ( corresponding to Figures 86–88 in Paxinos and Franklin , 2007 ) were analyzed . Four olfactory bulb sections immediately rostral to the external plexiform layer of the accessory olfactory bulb ( including and rostral to Figure 2 in Paxinos and Franklin , 2007 ) were analyzed . The RNA-seq raw and processed data as well as Supplementary file 1 have been deposited to GEO database under the accession number GSE67556 .	Most cells in the human body contain two copies of each chromosome—one that was inherited from the individual's mother and one from the father—and therefore contain two copies of every gene . While both copies are usually used equally and simultaneously to produce proteins , in a minority of cases the gene from one parent is silenced in a process known as genomic imprinting . This is generally achieved via the addition of chemical marks onto the DNA , which prevent the molecular machinery that activates genes from accessing the genetic material . Previous efforts to map imprinting in the brain throughout the mouse genome have yielded inconsistent results , due in part to the large number of factors that can affect gene expression . Perez , Rubinstein , Fernandez et al . have now addressed this issue by applying a combined approach , which includes developing a powerful statistical model that takes into account variation in age , sex , and mouse strain and extensively validating each imprinted gene candidate using an independent experimental technique . Perez , Rubinstein , Fernandez et al . analyzed genomic imprinting initially in a part of the brain called the cerebellum in both young and adult mice . This analysis confirmed the occurrence of imprinting in 74 genes identified in previous studies , and revealed imprinting for the first time in a further 41 genes . The degree of imprinting varied between genes . In some genes only one copy was expressed and the other was completely silenced whereas others only deviated from the two copies being expressed equally . For individual genes , imprinting varied with age , tending to be more pronounced in young animals than in adults . It also varied between brain regions and typically genes were imprinted more in the brain compared to elsewhere in the body . Mapping the activities of the imprinted genes revealed that many are involved in regulating the process of controlled cell death , or ‘apoptosis’ . For one particular test gene , selectively deleting either the maternal or paternal copy had different effects on the mice , thereby confirming that imprinting can affect brain development and activity . With this in mind , the potential impact of imprinting should also be considered when evaluating the effects of inherited mutations on human health .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Quantitative and functional interrogation of parent-of-origin allelic expression biases in the brain
Deciphering effective ways to suppress tumor progression and to overcome acquired apoptosis resistance of tumor cells are major challenges in the tumor therapy field . We propose a new concept by which tumor progression can be suppressed by manipulating tumor cell identity . In this study , we examined the effect of ER stress on apoptosis resistant tumorous cells in a Caenorhabditis elegans germline tumor model . We discovered that ER stress suppressed the progression of the lethal germline tumor by activating the ER stress sensor IRE-1 . This suppression was associated with the induction of germ cell transdifferentiation into ectopic somatic cells . Strikingly , transdifferentiation of the tumorous germ cells restored their ability to execute apoptosis and enabled their subsequent removal from the gonad . Our results indicate that tumor cell transdifferentiation has the potential to combat cancer and overcome the escape of tumor cells from the cell death machinery . A major challenge in the tumor therapy field is the development of new strategies to eliminate tumors and cancer cells . Whereas most of the current therapeutic strategies are based on apoptosis induction in the tumor cells , the effectiveness of these approaches is limited due to acquired apoptosis resistance ( Hanahan and Weinberg , 2000 , 2011 ) . Thus , deciphering ways to restore apoptosis sensitivity to tumorous cells that acquired apoptosis resistance may revive ‘old’ tools with therapeutic potential to eliminate tumor cells . The gld-1 ( GermLine Development defective ) gene encodes a germline-specific QUAKING-like RNA binding protein , which represses the translation of a variety of germline transcripts ( Jungkamp et al . , 2011; Wright et al . , 2011 ) . Consequently , GLD-1 regulates many aspects of germ cell biology ( Francis et al . , 1995a , 1995b; Kadyk and Kimble , 1998; Jan et al . , 1999; Hansen et al . , 2004; Ciosk et al . , 2006 ) . One of the striking consequence of a deficiency in gld-1 is the formation of a proximal germline tumor that fills the gonad ( Francis et al . , 1995a ) . This germline tumor is the result of re-entry of meiotic germ cells into the mitotic cell cycle instead of maturing into oocytes ( Francis et al . , 1995a ) . Importantly , some aspects of tumorigenesis are exhibited in the gld-1 germline tumor model . These include the ability of the tumorous germ cells to proliferate in a growth factor–independent manner ( Francis et al . , 1995a ) and their regulation by genes homologous to known human oncogenes or human tumor suppressor genes ( Pinkston-Gosse and Kenyon , 2007 ) . Notably , these tumorous germ cells acquired resistance to apoptosis ( Gumienny et al . , 1999 ) . In addition , some precocious germ cell transdifferentiation into ectopic somatic cells has been reported to occur at a low frequency in gld-1-deficient animal ( Ciosk et al . , 2006 ) . This transdifferentiation of the germ cells can be further enhanced by manipulation of RNA binding proteins , P granule components , transcription factors and histone modifiers; all of which regulate gene expression within the transdifferentiating germ cells themselves ( Ciosk et al . , 2006; Tursun et al . , 2011; Patel et al . , 2012; Updike et al . , 2014 ) . In this study , we investigated whether cellular and organismal stress can affect germ cell fate and tumorigenicity in the gld-1 tumor model . In Caenorhabditis elegans , chemically or genetically-induced ER stress promote germ cell apoptosis in normal ( i . e . , non-tumorous ) germlines ( Levi-Ferber et al . , 2014 ) . Nevertheless , the germ cells in gld-1-deficient animals were reported to be resistant to physiological and DNA damage-induced apoptosis ( Gumienny et al . , 1999 ) , suggesting that they may have acquired global resistance to apoptosis promoting signals , a common phenomenon of transformed cells . To directly test whether ER stress succeeds or fails to induce apoptosis in gld-1-deficient animals , we exposed the animals to ER stress , and assessed the presence of apoptotic corpses within the gonad using the SYTO12 apoptotic dye . ER stress was induced either by treating the animals with tunicamycin ( a specific inhibitor of N-glycosylation ) or by treating the animals with tfg-1 RNAi ( tfg-1 encodes a component of COPII-coated vesicles required for the export of cargo from the ER [Witte et al . , 2011] ) . Both treatments specifically induce ER stress ( Levi-Ferber et al . , 2014 ) . As previously reported ( Gumienny et al . , 1999 ) , no apoptotic corpses representing physiological germ cell apoptosis were detected in the tumorous gonads in the absence of ER stress ( Figure 1A , B and Figure 1—figure supplement 1 ) . However , we consistently detected SYTO12-labeled corpses in tumorous gonads of gld-1 RNAi-treated animals exposed to ER stress induced either by genetic means ( i . e . , tfg-1 RNAi ) or by chemical means ( i . e . , tunicamycin ) ( Figure 1A , B and Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 08005 . 003Figure 1 . Apoptotic corpses are detected in the gonads of gld-1-deficient animals upon induction of ER stress . ( A–C ) Day-3 animals treated with the indicated RNAi were stained with SYTO12 to detect apoptotic cell corpses . The average number of SYTO12-labeled apoptotic corpses per gonad is shown in B . The relative size of the SYTO12-labeled nuclei is shown in C . See Figure 1—figure supplement 1 for SYTO12-labeling of tunicamycin treated animals . ( D , E ) CED-1::GFP expressed in the gonadal sheath cells was used to follow engulfment of apoptotic cells within the gonad of day-3 animals . The relative average area of the engulfed cells is shown in E . Note that in non-tumorous animals the apoptotic cells are detected in the distal gonad zone ( DZ ) , whereas in the ER stressed-tumorous animals they are detected in the proximal gonad zone ( PZ ) . Asterisk marks Student's T-test values of p < 0 . 001 compared to animals treated with a mixture of control and tfg-1 RNAi . gld-1 RNAi knocked down GLD-1 protein levels to a similar extent upon treatment with control or tfg-1 RNAi ( see Figure 1—figure supplement 2 ) . At least 40 gonads of each genotype were analyzed . DOI: http://dx . doi . org/10 . 7554/eLife . 08005 . 00310 . 7554/eLife . 08005 . 004Figure 1—figure supplement 1 . Apoptotic cell corpses are detected in the gonads of tunicamycin-treated tumorous animals . Representative micrographs showing gonads ( x400 ) of day-3 gld-1 RNAi-treated animals treated with either 45 μg/ml tunicamycin or DMSO as of L4 and stained with SYTO12 to detect apoptotic cell corpses . The average number of SYTO12-labeled apoptotic corpses per gonad is shown . DOI: http://dx . doi . org/10 . 7554/eLife . 08005 . 00410 . 7554/eLife . 08005 . 005Figure 1—figure supplement 2 . GLD-1 protein levels are efficiently reduced by gld-1 RNAi in the single , double and triple RNAi mixtures . ( A ) Representative western blot of GLD-1 and tubulin in day 3 wild-type animals treated with the following RNAi combinations: 1 = control RNAi . 2 = gld-1 RNAi . 3 = gld-1 and control RNAi double mix . 4 = gld-1and tfg-1 RNAi double mix . 5 = gld-1 , control and ced-3 RNAi triple mix . 6 = gld-1 , tfg-1and ced-3 triple mix . Dashed line indicates removal of irrelevant lanes . ( B ) Bar graph shows the mean ratio of GLD-1 protein levels normalized to tubulin levels ±SEM in 3 independent experiments . Note that GLD-1levels were efficiently reduced in all RNAi conditions compared to lane 1 . Note that GLD-1 levels were similarly reduced upon both conditions of double RNAi treatment ( compare lanes 3–4 ) as well as upon both conditions of triple RNAi treatment ( compare lanes 5–6 ) . Furthermore , under all conditions , the gonads of gld-1 RNAi-treated animals appeared tumorous , and lacked oocytes and embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 08005 . 005 We further confirmed the presence of engulfed apoptotic corpses in the ER stressed tumorous gonads using CED-1::GFP expressed in the engulfing pseudopodia of the gonadal sheath cells ( Figure 1D ) . CED-1::GFP-engulfed apoptotic corpses were only detected in tumorous gonads upon induction of ER stress . These differences were not the result of restoration of gld-1 expression in the stressed animals as GLD-1 protein levels were similarly reduced in animals treated with gld-1 RNAi along with control RNAi or tfg-1 RNAi ( Figure 1—figure supplement 2 ) . Remarkably , the apoptotic corpses in the tumorous gonads of ER-stressed animals were distinct from those detected in non-tumorous gonads in several ways . First , in terms of location within the gonad—whereas in non-tumorous gonads apoptotic germ cell corpses were usually detected at the turn of the gonad , where oogenesis occurs , those detected in ER-stressed tumorous gonads were located in the proximal region of the gonad ( Figure 1A , D ) . In addition , in terms of size , both the SYTO12-labeled nuclei and the CED-1::GFP-engulfed cells in the tumorous gonads exposed to ER stress were significantly larger than those detected in non-tumorous animals ( Figure 1A , C , D , E ) . Furthermore , unlike engulfed germ cell corpses that are typically round , the CED-1::GFP engulfed cells in the ER-stressed tumorous gonads displayed a variety of shapes ( Figure 1D ) . Altogether , the differences in size , shape and location suggest that the apoptotic cell corpses in ER-stressed tumorous gonads are distinct from the germ cell corpses observed in non-tumorous gonads . We hypothesized that if cells undergoing apoptosis were continuously engulfed and removed from the stressed tumorous gonads , then blocking apoptosis should result in their accumulation in the gonads . To this end , ced-3 expression was inactivated to prevent apoptosis and the pattern of the nuclei in the gonad was assessed by DAPI ( 4′ , 6-diamidino-2-phenylindole‎ ) staining . Strikingly , blocking apoptosis resulted in the accumulation of cells with large and misshaped nuclei occupying nearly 40% of the stressed gonad ( Figure 2A , B ) . Even without exposing the animals to ER stress , blocking apoptosis resulted in the accumulation of ectopic cells with large and misshaped nuclei in the gonads of the gld-1 RNAi-treated animals . However , in the absence of ER stress , the abnormally large nuclei occupied only 10% of the gonad ( Figure 2A , B ) , indicating that ER stress promotes their induction . Interestingly , blocking apoptosis enabled the detection of ectopic cells within the gonads of nearly 90% of the gld-1 RNAi-treated animals examined , both under normal growth conditions as well as under ER stress , albeit to different extents ( Figure 2—figure supplement 1 and Figure 2A , B ) . The enhanced accumulation of ectopic cells in the ER stressed gonads was not due to altered efficiency of the gld-1 RNAi as the GLD-1 protein levels were similarly reduced upon gld-1 RNAi treatment in combination with control/tfg-1 and ced-3 RNAi ( Figure 1—figure supplement 2 ) . Furthermore , treatment of gld-1 ( q485 ) ; ced-3 ( n1286 ) double mutants with ER-stress inducing tfg-1 RNAi also increased the accumulation of cells with large nuclei in the gonads ( Figure 2C ) , recapitulating the phenomena observed in gld-1 , tfg-1 and ced-3 RNAi-treated animals . Altogether , these findings suggest that these cells with large nuclei in the gonads of gld-1-deficient animals are the ones that are normally cleared from the gonad by apoptosis , and not the typical germ cells . 10 . 7554/eLife . 08005 . 006Figure 2 . Ectopic cells with large nuclei accumulate in the gonads of gld-1; ced-3 animals . ( A ) Representative micrographs ( x400 ) of DAPI-stained gonads of day-4 animals . Animals were treated with the indicated RNAi . gld-1 RNAi was used to induce a germline tumor . ced-3 RNAi served to block apoptosis . tfg-1 RNAi was used to induce ER stress . Treatment with tfg-1 RNAi increased the levels of ectopic cells with large misshaped nuclei at the proximal zone of the gonad of gld-1 deficient animals , especially upon apoptosis inactivation . DZ marks the distal zone of the gonad . PZ marks the proximal zone of gonad . ( B ) Bar graph presents percentage of gonad area occupied by large nuclei in the indicated genotypes ( n = at least 60 gonads per genotype ) . Asterisks mark Student's T-test values of p < 0 . 001 of tfg-1 RNAi-treated animals compared to their non-stressed controls . Note that ectopic cells with large nuclei were detected to different extents in most of the animals examined ( see Figure 2—figure supplement 1 ) . ( C ) The induction of ectopic cells in the gonad by tfg-1-induced ER stress was recapitulated in gld-1 ( q485 ) ; ced-3 ( n1286 ) double mutants . Arrows point at axon-like structures detected within the gonads of gld-1-deficient animals upon ER stress . DOI: http://dx . doi . org/10 . 7554/eLife . 08005 . 00610 . 7554/eLife . 08005 . 007Figure 2—figure supplement 1 . Ectopic somatic cells are detected in most of the gonads of gld-1-deficient animals . Bar graphs present the percentage of animals treated with a mixture of gld-1 and ced-3 RNAi which contained ectopic cells in their gonads . ( A ) Ectopic cell were detcted by abnormally large DAPI-stained nuclei in the gonads ( n = 50–60 gonads per genotype ) ( B–D ) Ectopic cell were detcted by the expression of transgenic fluorescent somatic markers ( n = 50–60 gonads per genotype ) . Pelt-2::gfp is an intestinal marker . Pmyo-2::gfp is a pharyngal muscle marker . Punc-119::gfp is a neuronal marker . DOI: http://dx . doi . org/10 . 7554/eLife . 08005 . 007 A low frequency of precocious activation of embryonic-like differentiation of the germ cells in gld-1-deficient animals has been reported before ( Ciosk et al . , 2006 ) . Hence , we suspected that the abnormal nuclei within the gonad of gld-1-deficient animals , which were further induced upon ER stress , could be differentiated somatic cells as well . Thus , we examined the expression of transgenic somatic markers within the gonad of apoptosis-defective animals . In accordance with the detection of ectopic cells in the tumorous gonads upon blockage of apoptosis , ced-3 inactivation allowed the detection of differentiation markers in 80–90% of the gonads of gld-1-RNAi treated animals ( Figure 2—figure supplement 1 ) . Genetically-induced ER stress resulted in a 3–6 fold increase in the fluorescence of each of the three germ layers transgenic markers in the tumorous gonads ( Figure 3A ) . Similarly , treatment with tunicamycin increased the fluorescence of the neuronal marker Punc-119::gfp in the tumorous gonads of gld-1; ced-3 deficient animals ( Figure 3B ) . 10 . 7554/eLife . 08005 . 008Figure 3 . The ectopic somatic cells in the ER-stressed gonad of gld-1-deficient animals are germ cell-derived differentiated somatic cells . ( A ) Representative fluorescence micrographs ( x400 ) of somatic differentiation markers expressed in the gonads of gld-1; ced-3-deficient animals on day-4 of adulthood . gld-1 RNAi was used to sensitize the germline for transdifferentiation . ced-3 RNAi was used to prevent the clearance of the ectopic somatic cells from the gonad . Punc-119::gfp is a neuronal marker , Pmyo-2::gfp is a pharyngeal muscle marker . Pelt-2::NLS::gfp is an intestinal marker . ER stress was induced by tfg-1 RNAi or by an xbp-1 mutation . Asterisks mark Student's T-test values of p < 0 . 001 compared to non-stressed conditions . At least 40 animals were analyzed per genotype . ( B ) Representative fluorescence micrographs ( x400 ) of Punc-119::gfp in the gonads of ced-3 ( n1286 ) day-4 animals treated with gld-1 RNAi . ER stress was induced chemically with tunicamycin and compared to DMSO treatment . ( C ) Representative fluorescence micrographs of DAPI-stained nuclei of ER-stressed day 4 adults treated with a mixture of gld-1 , ced-3 and tfg-1 RNAi . Accumulation of abnormal somatic-like nuclei was detected in germ cell ( + ) animals and not in germ cell ( − ) glp-1 ( − ) mutants . At least 50 gonads were analyzed per genotype . Asterisk marks Student's T-test values of p < 0 . 001 . See Figure 3—figure supplement 1 for co-localization of the somatic marker expressing cells and the cells with the large nuclei and/or the cell under engulfment . DOI: http://dx . doi . org/10 . 7554/eLife . 08005 . 00810 . 7554/eLife . 08005 . 009Figure 3—figure supplement 1 . The ectopic somatic cells in the tumorous gonad have large nuclei and are engulfed by the surrounding gonad . ( A ) Day 4 Punc-119::gfp transgenic animals treated with a mixture of gld-1 , ced-3 and tfg-1 RNAi were stained with the nuclear dye Hoechst . Hoecst staining and GFP expression were individually captured and used to assess co-localization between the pattern of the GFP-expressing cells in the gonad and the cells harboring ectopically large nuclei . ( B ) Day 3 transgenic animals co-expressing the somatic marker Pelt-2::NLS::GFP and the engulfment marker Plim-7::ced-1::gfp were treated with a mixture of gld-1 and tfg-1 RNAi . Solid arrow indicates an engulfed cell expressing the somatic marker , demonstrating that the ectopic somatic cells in the tumorous gonad are engulfed and removed by the surrounding cells . Dashed arrow indicates an engulfed cell which does not express the somatic marker . DOI: http://dx . doi . org/10 . 7554/eLife . 08005 . 009 Do the cells with abnormal large nuclei also express the somatic markers , or are these two separate abnormal cell populations that arise in the tumorous gonad upon ER stress ? Analysis of the fluorescence pattern of a Pelt-2::NLS::gfp intestinal somatic marker , which by virtue of its NLS signal specifically labeled the nuclei of the corresponding somatic cells , revealed that these nuclei were larger than those of typical germ cells ( Figure 3A ) . In addition , Hoechst-nuclei staining of a strain expressing the neuronal Punc-119::gfp somatic marker was used to assess whether the pattern of the ectopic somatic cells in the gonad overlapped with the cells harboring small germline-like nuclei or whether it overlapped with the ectopically large nuclei . We found that under ER stress conditions , the cells that expressed the neuronal marker co-localized with the cells in the gonad that had atypically large nuclei . No expression of the Punc-119::gfp somatic marker was detected in regions of the gonad devoid of large-nucleated cells ( Figure 3—figure supplement 1A ) . In addition , we examined whether the cells that undergo apoptosis in the stressed tumorous gonad are the ones that express the somatic markers . To this end , we examined a strain that co-expresses ced-1::gfp in the sheath cells of the gonad as well as the intestinal marker Pelt-2::gfp::NLS . Indeed , some of the engulfed cells also expressed the somatic marker ( Figure 3—figure supplement 1B ) . These findings are consistent with the interpretation that the ectopic somatic cells that arise in the stressed tumorous gonad normally undergo apoptosis and are consequently engulfed and cleared from the gonad . Why was the expression of the intestinal somatic marker apparent only in some of the ced-1::gfp labeled phagosomes and not in all of them ? Likewise , why weren't all the large-nucleated cells co-localized with the cells in the gonad that expressed the neuronal marker ? These phenomena are likely the consequence of the fact that the teratoma contains somatic cells of the different germ-layers , implying that only a fraction of the teratoma cells can be detected using somatic markers of specific tissues or germ layers . The somatic cells in gld-1-deficient animals arise by precocious differentiation and loss of pluripotency of the germ cells ( Ciosk et al . , 2006 ) . Thus , we explored whether the ectopic somatic cells in the gonad of ER stressed-animals are derived from germ cells as well . To this end , we treated animals with a normal germline or glp-1 ( lof ) mutants , whose primary germ cells do not proliferate at non permissive temperatures , with a mixture of gld-1 , tfg-1 and ced-3 RNAi , in order to generate optimal conditions for the detection of ectopic somatic cells in the gonad ( RNAi efficacy was confirmed as described in ‘Materials and methods’ ) . Day-4 adults were analyzed for the presence of DAPI-stained ectopic large nuclei . Ectopic somatic cells were not detected in the gonad of germ cell-deficient glp-1 animals , although they were detected in control germ cell ( + ) animals upon induction of ER stress ( Figure 3C ) . These findings are consistent with the possibility that the ectopic somatic cells induced by ER stress in the tumorous gonad are derived from the germ cells themselves , uncovering ER stress as a potent regulator of germ cell pluripotency . Importantly , unlike previously identified regulators of germ cell pluripotency , all of which act at the final steps of the transdifferentiation process by directly regulating gene expression ( Ciosk et al . , 2006; Tursun et al . , 2011; Patel et al . , 2012; Updike et al . , 2014 ) ; ER stress and ER homeostasis likely act further upstream , linking between cellular and organismal physiology and germ cell fate . After establishing that ER stress promotes germ cell transdifferentiation in gld-1-deficient animals we examined whether this induction is mediated by one of the ER stress sensor proteins . To this end , wild-type animals and mutant animals deficient in one of the ER stress sensors ire-1 , pek-1 or atf-6 were treated with a mixture of gld-1 , ced-3 and tfg-1 RNAi . At day 4 of adulthood , the presence of ectopic cells in the gonad was assessed by staining of the animals' nuclei with DAPI . We detected significantly increased levels of ectopic nuclei in the gonads of wild-type , pek-1-deficient and atf-6-deficient animals , in the presence of tfg-1 RNAi ( Figure 4A ) . In contrast , tfg-1 RNAi failed to increase germ cell transdifferentiation in ire-1-deficient animals ( Figure 4A ) . Nevertheless , a low level of ectopic somatic cells , similar to that observed in gld-1-deficient animals that were not exposed to ER stress , was still detected in gld-1; ire-1-deficient animals ( Figure 4A ) . This basal level of germ cell transdifferentiation was not increased by tfg-1 RNAi-induced ER stress or by the ER stress associated with the ire-1 deficiency . We conclude that ire-1 is not required for germ cell transdifferentiation per se . However , it is required for germ cell transdifferentiation in response to ER stress . Furthermore , these findings indicate that the regulation of germ cell pluripotency by ER stress in gld-1-deficient animals requires the activation of the ER stress sensor IRE-1 , and is not simply the result of interference with ER homeostasis and function . 10 . 7554/eLife . 08005 . 010Figure 4 . ER stress induces germline transdifferentiation in an ire-1-dependent but xbp-1-independent manner . ( A ) Representative micrographs ( x400 ) of DAPI-stained gonads of day-4 animals treated with either a mixture of control , gld-1and ced-3 RNAi or with a mixture of tfg-1 , gld-1and ced-3 RNAi . Treatment with tfg-1 , gld-1and ced-3 RNAi failed to induce germ cell transdifferentiation in ire-1 mutants . ( B ) Representative micrographs of whole body ( x100 ) and gonads ( x400 ) of DAPI-stained day-4 animals of the indicated genotypes treated with gld-1 and ced-3 RNAi . Solid arrows indicate mitotic germ cells . Dashed arrows indicate somatic nuclei . Bar graphs present percentage of gonad area occupied by ectopic cells . Asterisk marks Student's T-test of p < 0 . 001 relative to wild-type animals . Bar graphs present percentage of gonad area occupied by ectopic cells of the indicated genotypes ( n = at least 70 gonads per genotype ) . Asterisks mark Student's T-test of p < 0 . 001 relative to the same animals treated with control , gld-1and ced-3 RNAi . Note that both alleles of xbp-1 similarly increased the percentage of gonad area occupied by ectopic somatic cells ( p = 0 . 23 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08005 . 010 Although one of the main downstream targets of ire-1 is the ER stress-regulated transcription factor xbp-1 ( Shen et al . , 2001; Calfon et al . , 2002 ) , ire-1 can also directly activate signaling cascades by virtue of its oligomerization propensity ( Urano et al . , 2000; Yoneda et al . , 2001 ) , or regulate mRNA stability by virtue of its RNase activity via the RIDD pathway ( Hollien and Weissman , 2006; Han et al . , 2009 ) . We hypothesized that if ire-1 promotes germ cell transdifferentiation by activating the transcription factor xbp-1 , then ER stress would fail to induce germ cell transdifferentiation in xbp-1-deficient animals . Initially , we used DAPI staining to assess the levels of germ cell transdifferentiation in two strains deficient in xbp-1- xbp-1 ( zc12 ) and xbp-1 ( tm2457 ) . Strikingly , we detected significantly increased levels of ectopic nuclei in the gonads of both xbp-1 mutants upon treatment with a mixture of gld-1 and ced-3 RNAi on day 4 of adulthood ( Figure 4B ) . Previous studies in mice and in C . elegans xbp-1 ( zc12 ) mutants have demonstrated that the mere deficiency in XBP1 leads to activation of the ER stress sensor IRE1 ( Hu et al . , 2009; Richardson et al . , 2011; Hur et al . , 2012; Safra et al . , 2013 ) Thus , we hypothesized that this high basal level of germ cell transdifferentiation in xbp-1 mutants may be due to the increased activity of IRE-1 in these mutants . Consistent with this interpretation , the high level of germ cell transdifferentiation in xbp-1 mutants was completely dependent on ire-1 ( Figure 4B ) . Thus , although ER stress-induced germ cell transdifferentiation is completely dependent on the ER stress sensor gene ire-1 , it is not dependent on its downstream target xbp-1 , implying that it is mediated by an ire-1-dependent xbp-1-independent signal . Since we have found that ER stress-induced germ cell transdifferentiation renders apoptosis resistant cells into apoptosis-sensitive cells ( Figure 1 ) , we hypothesized that it may allow the removal of cells from the tumorous gonad and thus may improve the health of the animals . To this end , we analyzed the progression of the germline tumor in gld-1-deficient animals in the presence or absence of ER stress . We found that xbp-1 inactivation , not only induced germ cell transdifferentiation but also suppressed the germline tumor; demonstrating that the two phenomena are tightly correlated . The physiological improvement was manifested in several ways: ( a ) the density of the germ cells in the germline tumor was reduced in the tumorous animals exposed to ER stress such that the germline tumor that occupied the proximal gonad did not fill the entire gonad ( Figure 5A and Figure 6—figure supplement 1 ) . ( b ) as the germline tumor progresses , the gonad becomes over-packed with proliferating germ cells , resulting in increased rigidity of the gonad . Consequently , the motion of animals with a packed gonad germline tumor is limited resulting in paralyzed animals that can move their heads but cannot move their body . We find that ER stress significantly delayed the paralysis of the tumorous animals and allowed the animals to move freely at time-points where most of the non-stressed tumorous animals were paralyzed ( Figure 5B , Videos 1 , 2 and Figure 5—Figure supplement 1A ) . This implies that the gonad of the animals subjected to ER stress were not as packed with germ cells as the non-stressed animals . ( c ) as the germline tumor progresses it ultimately kills the animal . We found that in wild-type animals , treatment with gld-1 RNAi shortened the lifespan by more than 30% , whereas it shortened the lifespan of ER stressed-xbp-1 mutants by only 13% ( Figure 5C ) . Similarly , treatment with gld-1 RNAi shortened wild-type animals lifespan by more than 30% , whereas it shortened the lifespan of tfg-1 RNAi-treated animals by only 15% ( Figure 5—figure supplement 1B ) . Importantly , no physiological improvement in terms of lifespan , movement or germline density were observed in ire-1 deficient animals , in which ER stress-induced germ cell transdifferentiation does not occur , even upon xbp-1 inactivation ( Figure 5 and Figure 5—figure supplement 1 ) . Thus , the genetic requirements for suppression of the germline tumor and for induction of germ cell transdifferentiation converge on ire-1 . 10 . 7554/eLife . 08005 . 011Figure 5 . ER stress suppresses the germline tumor in an ire-1-dependent manner . ( A ) Representative micrographs showing whole body ( x100 ) and gonads ( x400 ) of day-4 animals stained with DAPI . ( B ) Paralysis assay in wild type , xbp-1 ( tm2457 ) , ire-1 ( ok799 ) and xbp-1 ( tm2457 ) ; ire-1 ( ok799 ) animals . At least 90 synchronized adult animals per genotype were placed on gld-1RNAi plates and their paralysis was scored on days 5 , 6 , 8 and 9 . Bar graphs present percentage of paralyzed animals . At all timepoints the xbp-1 mutation significantly decreased the paralysis of the tumorous animals in an ire-1-dependent manner . Asterisks mark Student's T-test of p < 0 . 001 for reduced paralysis relative to wild-type animals . ( C ) Lifespan analysis of wild type , xbp-1 ( tm2457 ) , ire-1 ( ok799 ) and xbp-1 ( tm2457 ) ; ire-1 ( ok799 ) animals treated with either gld-1 ( RNAi ) to induce tumor formation or with control RNAi . Lifespan shortening was significantly suppressed in xbp-1 mutants in an ire-1-dependent manner . Mean lifespan and p-values are indicated within each graph . tfg-1 RNAi similarly supressed gld-1-RNAi-induced paralysis and lifespan shortening ( see Figure 5—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08005 . 01110 . 7554/eLife . 08005 . 012Figure 5—figure supplement 1 . tfg-1 RNAi suppresses the germline tumor in an ire-1-dependent manner . ( A ) Paralysis assay in wild type and ire-1 ( ok799 ) animals . At least 90 synchronized adult animals per genotype were placed on gld-1/control RNAi plates or gld-1/tfg-1 RNAi plates and their paralysis was scored on days 5 , 6 , 8 and 9 . Bar graphs present percentage of paralyzed animals . At all timepoints the tfg-1 RNAi treatment significantly decreased the paralysis of the tumorous animals in an ire-1-dependent manner . Asterisks mark Student's T-test of p < 0 . 001 for reduced paralysis relative to wild-type animals . ( B ) Lifespan analysis of wild type and ire-1 ( ok799 ) animals treated with the indicated RNAi combinations . Lifespan shortening by gld-1 RNAi was significantly suppressed in animals co-treated with tfg-1 RNAI in an ire-1-dependent manner . Mean lifespan and p-values are indicated within each graph . DOI: http://dx . doi . org/10 . 7554/eLife . 08005 . 01210 . 7554/eLife . 08005 . 013Video 1 . Impaired motility of tumorous gld-1 ( q485 ) animals on day-4 of adulthood . DOI: http://dx . doi . org/10 . 7554/eLife . 08005 . 01310 . 7554/eLife . 08005 . 014Video 2 . ER stress induced by tfg-1 RNAi treatment improved the motility of tumorous gld-1 ( q485 ) animals on day 4 of adulthood . DOI: http://dx . doi . org/10 . 7554/eLife . 08005 . 014 DNA damage , pathogens , oxidative , osmotic , heat shock , starvation and ER stress are all known inducers of germ cell apoptosis in normal ( i . e . , non-tumorous ) germlines ( Gartner et al . , 2000; Salinas et al . , 2006; Levi-Ferber et al . , 2014 ) . We wondered whether such stresses that induce germ cell apoptosis in non-tumorous gonads would induce germ cell transdifferentiation in gld-1-deficient animals , similarly to ER stress . To address this , wild-type or gld-1-deficient animals were exposed to ER stress , genotoxic stress , mitochondrial stress or osmotic stress . These stresses were induced by manipulating genes whose inactivation is known to induce only one specific stress response , without globally stressing the animals . tfg-1 RNAi was used to activate the ER stress response ( Levi-Ferber et al . , 2014 ) . rad-51 RNAi , which targets a DNA recombinase required for the repair of dsDNA breaks was used for the induction of genotoxic stress ( Gartner et al . , 2000 ) . Mitochondrial stress was induced by RNAi targeting ddl-3 , which encodes a kinesin light chain whose inactivation specifically activates the mitochondrial stress response ( Shore et al . , 2012 ) . Osmotic stress was induced by inactivation of a negative regulator of the osmotic stress response , osm-8 ( Rohlfing et al . , 2011 ) . We confirmed that all of these treatments significantly increased the amount of apoptotic corpses in the gonads of animals with a non-tumorous germline ( Figure 6A and white bar graph ) . We found that although some treatments induced more germ cell apoptosis than tfg-1 RNAi , only treatment with tfg-1 RNAi led to the detection of cells with large nuclei in the gonads of gld-1; ced-3-deficient animals ( Figure 6B and black bar graphs ) . Thus , although ER stress induces germ cell apoptosis in normal gonads and hinders germ cell pluripotency in gld-1-deficient animals , the latter does not occur upon exposure to other severe cellular stresses , including genotoxic stress . This indicates that the transdifferentiation of the germ cells is not simply a side-effect or a default fate of apoptosis-resistant tumorous cells under proapoptotic stress which they fail to execute . Moreover , since only treatment with tfg-1 RNAi suppressed the progression of the germline tumor ( Figure 6—figure supplement 1 ) , this lends further support to the notion that induction of germ cell transdifferentiation is a prerequisite for the suppression of the germline tumor . 10 . 7554/eLife . 08005 . 015Figure 6 . Loss of germ cell pluripotency in gld-1-deficient animals does not occur in response to all stresses . ( A ) Representative micrographs ( x400 ) of gonads of ced-3 ( + ) day 2 animals which were stained with SYTO12 to detect apoptotic cell corpses . White bar graphs present number of apoptotic nuclei per gonad arm ( n = 60 gonads per genotype ) . ( B ) Representative micrographs ( x400 ) of gonads of gld-1 ( RNAi ) ; ced-3 ( RNAi ) day-4 animals which were stained with DAPI to detect germline and somatic nuclei within the animals' gonads . Black bar graph presents the percentage of gonad area occupied by large nuclei ( n = 60 gonads per genotype ) . ER stress was induced by tfg-1 RNAi . Osmotic stress was induced by osm-8 inactivation . Mitochondrial stress was induced by RNAi targeting ddl-3 . Genotoxic stress was induced by rad-51 RNAi . Asterisks mark Student's T-test of p < 0 . 001 relative to control RNAi-treated animals . Note that the stresses that did not induce germ cell transdifferentiation in gld-1 ( − ) ; ced-3 ( − ) animals also failed to suppress the germline tumor in gld-1 ( − ) ; ced-3 ( + ) animals ( see Figure 6—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08005 . 01510 . 7554/eLife . 08005 . 016Figure 6—figure supplement 1 . Not all stresses suppress the germline tumor in gld-1-deficient animals . Representative micrographs ( x400 ) of DAPI-stained gonads of gld-1 ( RNAi ) day-4 animals ( n = at least 50 gonads per genotype ) . Note that the ability to execute apoptosis has not been manipulated in these animals , in contrast to the animals in Figure 6B . ER stress was induced by tfg-1 RNAi . Osmotic stress was induced by osm-8 inactivation . Mitochondrial stress was induced by RNAi targeting ddl-3 . Genotoxic stress was induced by rad-51 RNAi . Only tfg-1 RNAi suppressed tumor progression . DOI: http://dx . doi . org/10 . 7554/eLife . 08005 . 016 Understanding the molecular events that regulate germ cell fate in a normal germline , and even more so in a tumorous germline , is of fundamental importance in reproductive biology , stem cell research , and tumorigenesis and cancer therapy . In this study , we gained new and fascinating insights into the complex coupling between ER stress and germ cell fate in a C . elegans germline tumor model . We find that while severe stress limits organismal survival and well-being under most circumstances , ER stress is beneficial for the animals' healthspan in gld-1-deficient animals , as it suppresses their lethal germline tumors . This is in accordance with previous findings that a stress-inducing interference such as knockout of hsf1 in mice , which has deleterious impact on organismal survival under normal growth conditions as well as under stress , actually benefits the same organism in the case of cancer ( Dai et al . , 2007 ) . How does ER stress suppress the tumor in the gld-1-germline tumor model ? Our findings indicate that the suppression of the germline tumor is the combined effect of germ cell transdifferentiation coupled with a limited half-life of the transdifferentiated germ cells ( see model in Figure 7 ) . The latter is achieved by restoring the responsiveness of the tumorous cells to apoptosis once they transdifferentiate into ectopic somatic cells and the subsequent removal of their ectopic somatic corpses from the gonad . 10 . 7554/eLife . 08005 . 017Figure 7 . Model—tumor progression and germline fate in ER stressed tumorous gonads . In animals with a normal germline , ER stress induces germ cell apoptosis . Inactivation of the gld-1 genes alters many aspects of germ cell fate: differentiation of germ cells into oocytes is abrogated , germ cell proliferation is enhanced , a germline tumor is formed and the germ cells lose their responsiveness to execute physiological and stress-induced apoptosis . Furthermore , in gld-1 deficient animals the germ cells are prone to generate teratoma as they become sensitized to precociously transdifferentiate into somatic cells . Under these conditions ER stress can suppress and limit the germline tumor . This suppression is achieved by enhancing germline transdifferentiation into ectopic somatic cells . Soon after the transdifferentiation , these ectopic cells undergo apoptosis , and are removed from the gonad , suppressing the germline tumor . DOI: http://dx . doi . org/10 . 7554/eLife . 08005 . 017 Importantly , although germ cell transdifferentiation normally occurs in gld-1-deficient animals ( Ciosk et al . , 2006 ) , its extent is not sufficient to effectively suppress the tumor by itself . Effective suppression is only achieved under conditions that further enhance germ cell transdifferentiation beyond its basal level , as in the case of ER stress . All in all , our results indicate that tumor cell transdifferentiation may serve as a protective anti-tumor mechanism . This implies that transdifferentiation-promoting genes may effectively suppress tumors . Many cellular stresses that induce germ cell apoptosis in normal gonads failed to affect germ cell pluripotency and tumor progression in gld-1-deficient animals . This suggests that the regulation of these processes in gld-1-deficient animals is specifically associated with ER stress , and is not simply a side-effect of apoptosis-resistant cells under stress . How might ER stress promote germ cell transdifferentiation ? A major consequence of ER stress is interference with ER functions , which may directly impinge upon the production of a variety of signaling molecules such as secreted peptides , hormones , cholesterol and lipids which are metabolized in the ER . However , it is unlikely that this directly accounts for ER stress-induced germline transdifferentiation and apoptosis of the germ cells as these do not occur in animals under severe ER stress in the absence of ire-1 . This suggests that it is the activation of the ER stress sensor IRE-1 , rather than a non-specific consequence of ER stress per se , that regulate germ cell pro-differentiation and tumor progression in gld-1-deficient animals . Furthermore , since germ-cell transdifferentiation is promoted in animal deficient in xbp-1 , this indicates that it is mediated by an ire-1-dependent xbp-1 independent signal . What are the signaling molecules produced ? Is it a single molecule or an arsenal of signaling molecules that regulate germ cell fate ? These are all open questions to be addressed in future studies . Interestingly , in humans , mutations that compromise the activity or the expression of RNase L , an IRE1-related endoribonuclease , have been implicated with increased susceptibility to prostate cancer ( Carpten et al . , 2002; Casey et al . , 2002 ) . Unlike IRE1 , RNase L has lost specificity to xbp-1 mRNA . Nevertheless , it shares with IRE-1 its promiscuous RNase activity . This similarity may be sufficient as the tumor-suppressive properties of ire-1 described herein are independent of xbp-1 . This implies that tumor-suppressive properties of IRE-1-related endoribonucleases may be evolutionarily conserved , and may not be specifically associated with one specific type of prostate cancer . One hallmark of aggressive tumors is their adaptation to their natural primary niche ( Hanahan and Weinberg , 2000 , 2011 ) . Thus , transformation into an ectopic kind of cell , rather than differentiation into a type of cell which can normally be found in the same primary niche , may have an added value in suppressing aggressive tumors . Strikingly , in 95% of human ovarian germline tumors , the totipotent germ cells precociously differentiate into a variety of ectopic somatic cells generating teratoma ( Koonings et al . , 1989; Ulbright , 2005 ) . These naturally occurring teratoma are usually benign whereas the remaining 5% are usually malignant ( Ulbright , 2005 ) . This suggests that teratoma may be a preferential germline tumor in terms of survival and fitness of the organism . As in other tumor-suppressive mechanisms , cancer cells may evolve to bypass this line of defense . Accordingly , some of the ovarian teratoma undergo a malignant transformation which occurs after the development of the teratoma . Nevertheless , this occurs only in 1 . 5% of teratoma ( Comerci et al . , 1994; Ayhan et al . , 2000; Ulbright , 2005 ) . This suggests that the conclusions derived from our studies in the C . elegans tumor model system are already naturally applied for human germline tumor biology , as if they were selected to do so by evolution . The potential to suppress tumors by forcing their transdifferentiation requires the tumor cells to have differentiation potential . This raises the question whether transdifferentiation-mediated tumor suppression is relevant only for germline tumors , or is it relevant to a wider range of cell types ? In mammals , in addition to the totipotent germ cells , subpopulations of stem cells , which maintain differentiation potential ( i . e . , pluri or multi-potent ) exist in many organs and tissues , albeit at low numbers ( Reya et al . , 2001 ) . Furthermore , within tumors and hematological cancers , one can find a subpopulation of cancer stem cells that possess stem cells characteristics , including the ability to give rise to a variety of cell types ( Reya et al . , 2001 ) . Accordingly , differentiation therapy has been used in the clinic and is proven to be effective in limiting some kinds of tumors ( Sell , 2004 ) . Interestingly , whereas these treatments induce differentiation of cells within the same lineage , induction of transdifferentiation into a completely different cellular lineage may be even more effective by creating a discrepancy and maladaptiveness between the tumorous cells and their surrounding niche . Thus , induction of tumor cell transdifferentiation has the potential to be applied as a novel approach to combat cancer and overcome the escape of tumor cells from the cell death machinery in a wide variety of tissues . The number of apoptotic cells in the gonads of day-2 or day-3 animals was assessed by scoring the number of SYTO12/CED-1::GFP labeled cells in the gonad . SYTO12 ( Molecular Probes , Eugene , Oregon ) staining was performed as previously described ( Gumienny et al . , 1999 ) . Day-0 animals were placed on plates containing 45 μg/ml tunicamycin ( Calbiochem , Billerica , MA ) . SYTO12 staining and expression of somatic transgenes were analyzed on day 3 or day 4 of adulthood . Bacteria expressing dsRNA were cultured overnight in LB containing tetracycline and ampicillin . Bacteria were seeded on NGM plates containing IPTG and carbenicillin . RNAi clone identity was verified by sequencing . Eggs were placed on plates and synchronized from day-0 ( L4 ) . The efficacy of the tfg-1 RNAi was confirmed by the animals' reduced body size ( Witte et al . , 2011 ) . The efficacy of the ced-3 RNAi was confirmed by the lack of apoptotic corpses in the gonads . The efficacy of the gld-1 RNAi was confirmed by the tumorigenicity of the gonads , by the absence of oocytes and embryos and by western blotting . Some experiments involved double or triple RNAi mixtures , in which the relative amount of each RNAi bacteria was kept equal between samples by supplementing with control RNAi as needed . To follow expression of fluorescent transgenic markers , transgenic animals were anaesthetized on 2% agarose pads containing 2 mM levamisol . Images were taken with a CCD digital camera using a Nikon 90i fluorescence microscope . For each trial , exposure time was calibrated to minimize the number of saturated pixels and was kept constant through the experiment . The NIS element software was used to quantify mean fluorescence intensity as measured by intensity of each pixel in the selected area within the gonad . To determine the fraction of the gonad area occupied by ectopic cells day-4 animals were fixed and stained with DAPI . The NIS element software was used to manually select and quantify the gonad area as well as the area within the gonad that was occupied by abnormal DAPI-stained nuclei in the tumorous animals . A similar number of animals were boiled in protein sample buffer containing 2% SDS . Proteins were separated using standard PAGE separation , transferred to a nitrocellulose membrane and detected by western-blotting using anti-GLD-1 ( 1:1000 , kindly provided by Prof Anton Gartner [Rutkowski et al . , 2011] ) and anti-tubulin ( DHSB , 1:5000 ) . RNAi treatments were performed continuously from the time of hatching . Eggs were placed on plates seeded with the RNAi bacteria of interest . Paralysis and lifespan were scored every 1–2 days . Related lifespans were performed concurrently to minimize variability . In all experiments , lifespan was scored as of the L4 stage which was set as t = 0 . Animals that ruptured or crawled off the plates were included in the lifespan analysis as censored worms . SPSS program was used to determine the means and the p values . p values were calculated using the log-rank ( i . e . , Mantel–Cox ) method . Error bars represent the standard error of the mean ( SEM ) of at least 3 independent experiments . Except for lifespan analysis , p values were calculated using the unpaired Student's t test . The following lines were used in this study: N2 , CF2012: pek-1 ( ok275 ) X , CF2988: atf-6 ( ok551 ) X , CF2473: ire-1 ( ok799 ) II , CF3208: xbp-1 ( tm2457 ) III , CF2472: xbp-1 ( zc12 ) III , SHK62: ire-1 ( ok799 ) II; xbp-1 ( tm2457 ) III , MD701: Plim-7:: ced-1::gfp V , CF2185: ced-3 ( n1286 ) IV , DP132: edIs6 ( punc-119::GFP ) IV , SHK75: irIS25 ( pJM86; pelt-2::NLS::GFP::LacZ + rol-6 ) V , SHK40: glp-1 ( e2141 ) III; irIS25 ( pJM86; pelt-2::NLS::GFP::LacZ + rol-6 ) V , MT3571: osm-8 ( n1518 ) II , SHK118: gld-1 ( q485 ) /unc-13 ( e51 ) I; ced-3 ( n1286 ) IV , SHK152: ced-3 ( n1286 ) edIs6 ( Punc-119::gfp ) IV .	If a cell in the body becomes damaged or stops working properly , it can trigger its own destruction . This helps to prevent the accumulation of damaged cells . However , cancer cells can often tolerate much greater damage than normal cells . Toxic chemotherapies , which are often used to treat cancer , work by severely damaging the cells to help trigger their self-destruction . Unfortunately , chemotherapy does not work on all cancer cells , and the remaining treatment-resistant cells may continue to grow and spread in more aggressive ways . Now , Levi-Ferber et al . have found a way to change the identity of cancer cells , which makes them more likely to self-destruct . The experiments used roundworms called Caenorhabditis elegans that had a genetic mutation that causes them to develop tumors in their reproductive organs . Normally , the cells in these tumors do not self-destruct . Levi-Ferber et al . exposed tumor cells from the worms to chemicals or to genetic modifications that cause unfolded proteins to accumulate inside the cell . This build-up of proteins stresses a structure in the cell called the endoplasmic reticulum . Normally , if endoplasmic reticulum stress gets too high , the cell activates various pathways to relieve the stress , and if these fail , the cell self-destructs . Levi-Ferber et al . showed that a protein called IRE-1 , which senses endoplasmic reticulum stress , caused the tumor cells to change into a type of non-cancerous cell . After the change , the cells were also more sensitive to self-destruction . This meant that tumors grew more slowly and ended up smaller , allowing the animals to survive longer . Together , the experiments suggest that treatments that force cancer cells to become a different cell type might be one way to prevent the emergence of treatment-resistant tumor cells . Future research will be needed to investigate exactly how IRE-1 causes the identity of the cell to change , and to see if this process could treat other kinds of cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2015	Transdifferentiation mediated tumor suppression by the endoplasmic reticulum stress sensor IRE-1 in C. elegans
The vascular wall is a source of progenitor cells that are able to induce skeletal repair , primarily by paracrine mechanisms . Here , the paracrine role of extracellular vesicles ( EVs ) in bone healing was investigated . First , purified human perivascular stem cells ( PSCs ) were observed to induce mitogenic , pro-migratory , and pro-osteogenic effects on osteoprogenitor cells while in non-contact co-culture via elaboration of EVs . PSC-derived EVs shared mitogenic , pro-migratory , and pro-osteogenic properties of their parent cell . PSC-EV effects were dependent on surface-associated tetraspanins , as demonstrated by EV trypsinization , or neutralizing antibodies for CD9 or CD81 . Moreover , shRNA knockdown in recipient cells demonstrated requirement for the CD9/CD81 binding partners IGSF8 and PTGFRN for EV bioactivity . Finally , PSC-EVs stimulated bone repair , and did so via stimulation of skeletal cell proliferation , migration , and osteodifferentiation . In sum , PSC-EVs mediate the same tissue repair effects of perivascular stem cells , and represent an ‘off-the-shelf’ alternative for bone tissue regeneration . Stromal progenitor cells within vessel walls have multipotent properties ( Cathery et al . , 2018; Corselli et al . , 2012; Covas et al . , 2008; Crisan et al . , 2008; Dellavalle et al . , 2007; Farrington-Rock et al . , 2004 ) , are native forerunners of mesenchymal stem cells ( MSCs ) , and participate in endogenous bone repair ( Diaz-Flores et al . , 1991; Diaz-Flores et al . , 1992; Grcevic et al . , 2012 ) . When purified based on expression of CD146 ( Mel-CAM ) and CD34 , human perivascular stem cells ( PSCs ) from adipose tissue or other tissue compartments speed bone repair ( Askarinam et al . , 2013; Chung et al . , 2014; James et al . , 2017a; James et al . , 2012a; James et al . , 2012b; Tawonsawatruk et al . , 2016 ) . Although direct incorporation of human PSCs into chondroblasts , osteoblasts and osteocytes occurs ( James et al . , 2012b ) , PSCs induce bone healing either primarily or exclusively via paracrine stimulation of the resident host cells within the defect niche ( Chung et al . , 2014; Tawonsawatruk et al . , 2016 ) . High expression and secretion of osteoinductive proteins has been observed among freshly sorted or culture-expanded PSCs , including bone morphogenetic proteins ( BMPs ) and vascular endothelial growth factor ( VEGF ) among others ( Chen et al . , 2009; Hardy et al . , 2017; James et al . , 2017b ) . However , the exact paracrine intermediaries of PSC-induced bone defect healing are not known . Extracellular vesicles ( EVs ) , including exosomes and microvesicles , carry a repertoire of bioactive molecules: proteins , nucleic acids , lipids and carbohydrates ( Colombo et al . , 2014; Cvjetkovic et al . , 2016; Mateescu et al . , 2017 ) . The heterogenous nature of EVs is well described , with EVs of a diameter <100–150 nm and multivesicular body ( MVB ) origin termed exosomes ( Gould and Raposo , 2013; Mateescu et al . , 2017 ) . Stromal progenitor cells elaborate EVs , and EVs have been observed to harbor many of the regenerative properties of their parent cell ( Lakhter and Sims , 2015 ) . In bone repair , CD9 knockout mice , which have impaired EV elaboration , have delayed appendicular bone healing , that may be rescued by mesenchymal progenitor cell-derived EVs ( Furuta et al . , 2016 ) . As well , induced pluripotent stem cell-derived EVs have been shown to incite osteogenesis and vasculogenesis in vivo ( Qin et al . , 2016; Todorova et al . , 2017 ) . The potential role of miRNA cargo in EVs has been implicated in these effects , however the mechanisms that underlie EV-mediated reparative effects in bone are essentially unknown . Here , we observe that in similarity to their parent cell , PSC-derived EVs incite pleiotropic effects on skeletal progenitor cells , including mitogenic , pro-migratory , pro-osteogenic effects , and broad ranging changes in the cellular transcriptome of the recipient cell . These bioactive effects require EV tetraspanin activity , as well as expression of their binding partners IGSF8 and PTGFRN by the recipient progenitor cell . These cellular effects of PSC-EVs on skeletal progenitor cells converge to incite intramembranous bone repair in a mouse model . Perivascular stem/stromal cells ( PSCs ) were derived from human white adipose tissue using previously reported methods , by FACS selection of the related populations of CD34+CD146-CD45-CD31- adventitial progenitor cells and CD146+CD34-CD45-CD31- pericytes ( Corselli et al . , 2012; Crisan et al . , 2008; James et al . , 2017a; Meyers et al . , 2018aMeyers et al . , 2018b; Meyers et al . , 2018b ) ( Figure 1—figure supplement 1 ) . Frequencies of human PSCs within adipose tissue are summarized in Supplementary file 2 , and were within our previously reported ranges ( James et al . , 2012a; West et al . , 2016 ) . In order to better understand the paracrine effects of PSCs in bone repair , we first defined a set of replicable in vitro paracrine effects of human PSCs on recipient human BMSCs ( Figure 1 ) . Prior to all experiments , cell surface antigens of human PSCs and BMSCs were typified by flow cytometry , and the multilineage differentiation potential of BMSCs was confirmed ( Figure 1—figure supplement 2 and Figure 1—figure supplement 3 ) . Co-culture results showed that PSCs induced significant mitogenic ( Figure 1A , MTS assay ) , pro-migratory ( Figure 1B , scratch wound healing assay ) , and pro-osteogenic effects ( Figure 1C , alkaline phosphatase activity ) on BMSCs when placed in non-contact co-culture conditions . When a lipophilic dye ( PKH26 ) was used to label the cell membrane of PSCs , this dye was incorporated into the BMSCs , suggesting the exchange of membrane from parent PSCs to recipient cells under non-contact culture conditions ( Figure 1D ) . As primary candidates for mediating non-contact dependent cell-cell communication , PSC-derived extracellular vesicles ( EVs ) were examined ( Figure 1E–H ) . PSC-EVs were enriched by ultracentrifugation of serum-free supernatant of PSCs . Guidelines from the International Society of Extracellular Vesicles ( ISEV ) for EV characterization were observed ( Lötvall et al . , 2014 ) . Spherical PSC-EV morphology was observed by transmission EM ( Figure 1E ) , with enrichment of tetraspanin molecule expression ( CD9 , CD63 , and CD81 ) , but without expression of the endoplasmic reticulum protein calnexin ( Figure 1F ) . Size distribution of PSC-EVs was most commonly ~100 nm , as shown by either TEM quantification ( Figure 1G ) or nanoparticle tracking analysis ( NanoSight ) ( Figure 1H ) . EV yield per cell per day as assessed by protein content was 1 . 11 ± 0 . 18 pg , as determined by the Bradford method ( Supplementary file 3 ) . Thus , PSCs elaborate EVs in culture which then are received by stromal progenitor cells , representing a primary candidate for PSC-mediated paracrine activity . EVs represent a mixture of protein , lipid , and RNA that may affect cellular processes of the recipient cell ( Bidarimath et al . , 2017 ) . To begin to examine this , total RNA sequencing was performed on three PSC-EV preparations and compared to the RNA content of their parent PSC cells ( Figure 1I–K ) . To determine which genes were most expressed , transcripts were normalized by fragments per kilobasepair per million mapped ( FPKM ) , and those with Log2 FPKM >−0 . 8 underwent further analysis . Among these , 10 , 256 annotated genes were expressed in all samples of 54 , 136 total RNA transcripts ( 19% of total , including 10 , 256 protein coding RNA; six non-coding RNA; and four pseudo RNA ) . Clear separation between gene expression profiles were observed when comparing PSC-EV RNA content to their parent cells , as observed by unsupervised hierarchical clustering ( Figure 1I ) and principal component analysis ( Figure 1J ) . Putative gene markers of human perivascular MSC were cross-referenced within PSCs and PSC-EVs ( Cho et al . , 2017 ) . Higher expression of most transcripts were seen in each PSC isolate ( such as CD44 , NT5E , ENG , LEPR , PDGFRA , and PDGFRB ) , with comparatively less expression among PSC-EVs ( Figure 1K ) . The highest expressing 100 genes within PSC-EVs are listed in Supplementary file 4 . All 100 were protein coding genes , 57 encoded for ribosomal proteins ( e . g . RPS18 , RPL37A , and RPL41 ) and 68 were included within Gene Ontology term extracellular exosome ( GO term: 0070062 ) . Reflecting their adipose tissue origin , these highly expressed PSC-EV transcripts included several genes associated with adipocytes or adipogenesis , such as VIM , RACK1 , and LGALS1 . Highly expressed PSC-EV genes also included transcripts involved in the regulation of cellular proliferation ( e . g . FTL , RAB13 , MT2A , and TMSB10 ) or cellular migration ( e . g . RAB13 , TMSB10 , S100A6 , and EEF1A1 ) . Next , PSC-EV RNA content was cross-referenced to a previously published RNA Seq dataset within unpurified adipose stromal cell-derived EVs from porcine tissue ( Eirin et al . , 2014 ) . As expected , some overlap existed in EV content between these two adipose stromal cell derivatives ( Supplementary file 5 ) . For example , of 39 transcription factors previously found to be enriched in adipose tissue stem/stromal cell ( ASC ) -EVs ( Eirin et al . , 2014 ) , 14 genes ( 35 . 9% ) were also enriched in human PSC-EVs . Of these , known positive regulators of cellular proliferation ( e . g . JMJD1C , NRIP1 , and TRPS1 ) and cellular migration ( e . g . JMJD1C , TCF4 , and KLF7 ) were identified . Of note , several transcriptional repressors were found within the previously published ASC-EVs ( e . g . ZNF568 , ZHX1 , ZBTB1 , and RUNX1T1 ) , which were not found at high levels in human PSC-EVs . Thus , PSC-EVs demonstrate both commonalities and distinct differences in RNA content when compared to their either parent cells , as well as to known expression profiles among unpurified adipose stromal EVs . The direct effects of PSC-EVs on recipient BMSCs were next defined ( Figure 2 ) . To demonstrate interaction of EVs with the recipient cells , PSC-EVs were first labeled with PKH26 dye , and fluorescently labeled PSC-EVs were directly applied to human BMSCs . After 48 hr , fluorescence was observed on the recipient BMSCs in a cytosolic distribution ( Figure 2A ) . PSC-EVs were next labeled with a pH dependent dye ( pHrodo Red Maleimide ) , which only fluoresces after exposure to an acidic environment such as the cellular interior ( Figure 2B ) . After 48 hr , fluorescence was again observed suggesting EV internalization by the recipient BMSCs . PSC-EVs exerted a dose-dependent mitogenic effect on recipient BMSCs ( Figures 2C and 1 , 2 . 5 , and 5 μg/mL , MTS assays ) . Using a scratch wound healing assay , PSC-EVs represented a significant pro-migratory stimulus , with the lowest PSC-EV concentration demonstrating the largest effect ( Figure 2D ) . Under osteogenic differentiation conditions , PSC-EVs induced a dose dependent increase in ALP activity ( Figure 2E ) and bone nodule deposition ( Figure 2F ) . Quantitative PCR analysis of characteristic osteoblastic gene markers confirmed a dose-dependent positive regulatory effect by PSC-EVs ( Figure 2G ) , including RUNX2 ( Runt related transcription factor 2 ) and SP7 ( Osterix ) . We next queried as to whether recipient stromal cells responded to PSC-EVs in a tissue-specific manner . Here , human culture-derived ASCs were exposed to the identical PSC-EV treatment conditions with broad similarities and minor differences found ( Figure 2—figure supplement 1 ) . Unlike BMSCs , the mitogenic effect of PSC-EVs was not observed among recipient ASCs . The pro-migratory and pro-osteogenic effects of PSC-EVs were conserved findings across both recipient stromal cells , although the maximum efficacious dose for each bioactivity assay was different when comparing BM- and AT-derived recipient cells . Finally , we compared the effects of EVs from unsorted ASCs and homogeneous PSCs on recipient BMSCs . Like PSC-EVs , ASC-EVs exerted mitogenic , pro-migratory , and pro-osteogenic effects on recipient BMSCs ( Figure 2—figure supplement 2 ) . However , the magnitude of change was significantly different between EV preparations , and PSC-EVs demonstrated a stronger induction of cell proliferation and osteogenic differentiation , while ASC-EVs enhanced cell migration to a greater degree . In sum , perivascular EVs retain bioactive effects of their parent perivascular cell type and exert overall similar effects across different recipient multipotent mesenchymal cell types . Next , the changes in the transcriptome of the recipient BMSCs were examined . Here , the Affymetrix Clariom D microarray assayed the BMSC transcriptome at 48 hr post-PSC-EV treatment ( Figure 2H–K ) . 38 , 416 annotated genes were expressed in all samples among 135 , 750 total probesets ( 20 , 014 protein coding RNA; 5054 non-coding RNA; 2795 pseudo RNA; 207 sno RNA; 55 snRNA; 26 rRNA; two scRNA ) . Clear differences between control- and PSC-EV-treated BMSC gene expression were observed by unsupervised hierarchical clustering ( Figure 2H ) and principal component analysis ( Figure 2I ) . As shown by volcano plot ( Figure 2J ) , 4129 transcripts showed >2 SD increase in expression with PSC-EV treatment ( 3 . 04% of total , red dots ) , while 2589 transcripts showed >2 SD reduction in expression with PSC-EV treatment ( 1 . 91% of total , blue dots ) . QIAGEN Ingenuity Pathway Analysis ( IPA ) showed that the majority of the activated pathways were associated with the positive regulation of cell proliferation , migration , and/or osteogenesis , including for example SAPK/JNK signaling ( Zha et al . , 2016 ) , HGF signaling ( Forte et al . , 2006 ) , Sirtuin signaling ( Simic et al . , 2013 ) , FGF Signaling ( Gharibi and Hughes , 2012 ) , and PDGF Signaling ( Li et al . , 2014 ) , among recipient BMSCs ( Figure 2K , see Supplementary file 6 for a complete list ) . Conversely , downregulated signaling pathways among recipient BMSCs included those known to negatively regulate cell proliferation , such as AMPK Signaling ( de Meester et al . , 2014 ) . PTEN signaling was also downregulated , which negatively regulates BMSC proliferation ( Shen et al . , 2018 ) , migration ( Comer and Parent , 2002 ) and osteogenic differentiation ( Liu et al . , 2017 ) ( Figure 2K , see Supplementary file 7 for a complete list ) . Next , the transcriptome of PSC-EVs ( as previously determined by RNA-Seq ) was cross-referenced to changes within the BMSC transcriptome after EV treatment ( Figure 2—figure supplement 3 ) . Here , of 7789 PSC-EV transcripts with a mean FPKM >0 , approximately half of these transcripts demonstrated were increased among PSC-EV-treated BMSCs ( 3678 transcripts , 47 . 22% of total ) . Thus , global changes in the recipient BMSC transcriptome were not well explained by simple transfer of EV RNA cargo . Instead , PSC-EVs demonstrate prominent regulation of recipient cell gene transcription resulting in significant mitogenic , pro-migratory , and pro-osteogenic effects in vitro . EV uptake and downstream bioactive effects may be dependent on membrane-bound protein interaction with the recipient cell ( Escrevente et al . , 2011; Janas et al . , 2016; Morelli et al . , 2004 ) . To assess this , trypsinization was performed to digest EV membrane-bound proteins and the cellular effects of trypsinized PSC-EVs were again assessed ( Figure 3—figure supplements 1 and 2 . 5 μg/mL PSC-EVs used ) . Pretreatment with trypsin completely abrogated the mitogenic ( Figure 3—figure supplement 1A ) , pro-migratory ( Figure 3—figure supplement 1B ) , and pro-osteogenic effects of PSC-EVs ( Figure 3—figure supplement 1C ) . Thus , surface-associated vesicular proteins are integral for PSC-EV bioactivity . Tetraspanins are enriched among exosomes including PSC-EVs ( see again Figure 1F ) , and both CD9 and CD81 are fusogenic , involved in spermatozoa and phagocyte fusion ( Rubinstein et al . , 2006a; Rubinstein et al . , 2006b; Takeda et al . , 2003; van Dongen et al . , 2016; Zhu et al . , 2002 ) . In contexts outside mesenchymal progenitor cell biology , antibodies against tetraspanins have mitigated EV interaction with the recipient cell ( Morelli et al . , 2004 ) . To test the requirement of surface-associated tetraspanins , neutralizing antibodies directed against CD9 or CD81 were pre-incubated with PSC-EVs before application to the recipient cell ( Figure 3 ) . PSC-EVs were labeled with a pH dependent dye and treated with or without neutralizing antibodies ( Figure 3 ) . After 48 hr and in contrast to control conditions , minimal fluorescence was seen with either neutralizing antibody , suggesting that CD9 and CD81 were required for EV internalization ( Figure 3A ) . Results demonstrated that both anti-CD9 and anti-CD81 reversed the mitogenic effects of PSC-EVs ( Figure 3B , C ) . Likewise , the pro-migratory effects of PSC-EVs were completely abrogated by either neutralizing antibody ( Figure 3D–G ) . The pro-osteogenic effects of PSC-EVs were partially reversed by anti-CD9 ( Figure 3H , I ) , but not anti-CD81 neutralizing antibodies ( Figure 3J , K ) . To explore other EV-associated proteins that may be important for PSC-EV bioactivity , we analyzed other CD markers that were enriched in PSC-EVs ( mean FPKM >0; see Supplementary file 8 ) . Several additional fusogenic proteins were identified , including CD46 and CD63 ( Anderson et al . , 2004; Raaben et al . , 2017 ) . In addition , several other CD markers were enriched among PSC-EV that have described pro-osteogenic effects , including CD44 , CD82 , and CD99 ( Bergsma et al . , 2018; Oranger et al . , 2015; Yeh et al . , 2014 ) . In sum , intact activity of CD9 or CD81 are essential for the majority of bioactive effects of PSC-EVs on recipient osteoprogenitor cells . CD9 and CD81 are known to interact with several cell surface proteins on the recipient cell , including immunoglobulin superfamily , member 8 ( IGSF8 ) ( Glazar and Evans , 2009 ) and prostaglandin F2 receptor inhibitor ( PTGFRN ) ( Charrin et al . , 2001 ) . In a candidate fashion , shRNA-mediated knockdown of IGSF8 and PTGFRN was performed in human recipient BMSCs ( Figure 4A , 52 . 9% and 58 . 4% knockdown of IGSF8 and PTGFRN gene expression , respectively ) . PSC-EVs were labeled with a pH-dependent dye and incubated with BMSCs with or without knockdown . After 48 hr and unlike vector control conditions , no fluorescence was seen among BMSCs with IGSF8 or PTGFRN knockdown ( Figure 4B ) . In comparison to vector control , knockdown of either IGSF8 or PTGFRN abrogated the mitogenic effects of PSC-EVs ( Figure 4C , 2 . 5 μg/mL PSC-EVs used ) . Likewise , IGSF8 or PTGFRN KD nullified the pro-migratory effect of PSC-EVs ( Figure 4D ) . Finally , PTGFRN KD inhibited the pro-osteogenic effect of PSC-EVs ( Figure 4E ) . In contrast , IGSF8 KD paradoxically increased BMSC osteogenic differentiation; however , this was observed in both control and PSC-EV treatment conditions and therefore appeared to be an EV independent phenomenon . Thus and in aggregate , EV-associated CD9 and CD81 along with their known binding partners on the recipient cell are required for the majority of bioactive effects of perivascular EVs . Next , we investigated whether PSC-EV treatment would improve calvarial defect repair ( Figure 5 ) . The effects of the implanted parent PSC themselves on calvarial bone repair have been previously documented ( James et al . , 2012a ) . In order to confirm that human PSC-EVs would demonstrate bioactivity to mouse recipient osteoprogenitor cells , additional in vitro studies were first performed . Either mouse ASCs or neonatal mouse calvarial cells ( NMCCs ) were isolated and treated with or without PSC-EVs ( 1–5 μg/mL ) . In similarity to human recipient cells , PSC-EVs treatment resulted in similar mitogenic , pro-migratory , and pro-osteogenic effects on both mouse ASCs and NMCCs ( Figure 5—figure supplement 1 ) . To assess the ability of PSC-EVs to speed bone repair , we chose a bone injury model that shows some modest healing overtime ( a 1 . 8 mm diameter , full thickness , circular frontal bone defect in the calvaria ) ( Zhang et al . , 2018 ) . Percutaneous injection of PSC-EVs ( 1 and 2 . 5 μg ) was performed over the defect site twice weekly , and analyses were performed at 4 weeks postoperative . A summary of the animal treatment protocol is provided as Figure 5—figure supplement 2 . A summary of animal numbers and treatment allocation is shown in Supplementary file 9 . As in our in vitro studies , the effects of PSC-EVs on skeletal cell proliferation , migration , and osteodifferentiation were sequentially addressed . PSC-EV-treated defects showed an increase in osteoblastic proliferation at the bone defect edge , as shown by Ki67 immunostaining and quantitation of immunoreactive cells at the bone defect edge ( Figure 5A , B , appearing red with arrowheads ) . In order to assess progenitor cell migration into the defect , a lineage tracing strategy was employed using Pdgfrα-CreER;eGFP animals ( Figure 5C ) . Reporter activity within Pdgfrα-CreER;eGFP highlights bone-lining stromal cells which then migrate into populate the bone injury site . PSC-EV-treated defects showed a prominent increase in GFP+ stromal progenitor cell migration into the defect mid-substance ( Figure 5C , D , quantified right as the mean fluorescence intensity within the middle of the defect site ) . Microcomputed tomography ( μCT ) reconstructions revealed an increase in defect re-ossification in comparison to PBS control ( Figure 5E ) . Quantitative indices confirmed an increase in bone healing across all metrics , including bone volume ( BV , Figure 5F ) , bone formation area ( BFA , Figure 5G ) , and semi-quantitative healing score ( Figure 5H ) ( Spicer et al . , 2012 ) . Routine H&E staining confirmed a significant narrowing of the gap between bony fronts ( Figure 5I , black arrowheads ) and an enrichment of osteocalcin ( OCN ) + cells at the leading edges of the defect site with PSC-EV treatment ( Figure 5J , K ) . In sum , the mitogenic , pro-migratory , and osteoinductive effects of PSC-EVs converge in vivo to speed bone defect healing . In sum , perivascular EVs induce proliferation , migration and osteogenic differentiation of osteoprogenitor cells , and the confluence of these effects positively regulates bone defect repair . Perivascular EVs require surface-associated tetraspanins for bioactivity , and recipient skeletal cells require their binding partners to respond to a perivascular EV stimulus . These data solidify the pleiotropic paracrine effects of perivascular stem cells on bone repair , and suggest the importance of perivascular EVs in both endogenous bone repair as well as in skeletal tissue engineering . Future studies must consider the issues of process optimization , including the optimum method for PSC-EV isolation , storage , and sustained delivery . PSCs were isolated from human subcutaneous adipose tissue via fluorescence activated cell sorting ( FACS ) based on prior protocols ( Hardy et al . , 2017; Meyers et al . , 2018b ) . Human lipoaspirate was obtained from healthy adult donors ( four different donors ) under IRB approval at JHU with a waiver of informed consent , and was stored for less than 48 hr at 4°C before processing . The SVF ( stromal vascular fraction ) of human lipoaspirate was obtained by type II collagenase digestion . Briefly , lipoaspirate was diluted with an equal volume of phosphate-buffered saline ( PBS ) and digested with Dulbecco's modified Eagle's medium ( DMEM ) containing 0 . 5% bovine serum albumin ( Sigma-Aldrich , St . Louis , MO ) and 1 mg/ml type II collagenase ( Worthington Biochemical , Freehold , NJ ) for 1 hr under agitation at 37°C . Adipocytes were separated and removed by centrifugation . The cell pellet was resuspended in red blood cell lysis buffer ( 155 mM NH4Cl , 10 mM KHCO3 , and 0 . 1 mM EDTA ) and incubated at room temperature for 5 min . After centrifugation , cells were resuspended in PBS and filtered at 70 μm . The resuspended SVF cells were cultured in T75 flask and expanded as ASCs . For PSC isolation , the resulting SVF was further processed for FACS sorting , using a mixture of the following directly conjugated antibodies: anti-CD34-allophycocyanin ( 1:100 , RRID:AB_398614; BD Pharmingen , San Diego , CA ) , anti-CD45-allophycocyanin-cyanin 7 ( 1:30 , RRID:AB_396891; BD Pharmingen ) , anti-CD146-fluorescein isothiocyanate ( 1:100 , RRID:AB_324069; Bio-Rad , Hercules , CA ) , and anti-CD31-allophycocyanin-cyanin 7 ( 1:100 , RRID:AB_10643590; Bio Legend , San Diego , CA ) . A summary of antibodies used is presented in Supplementary file 10 . All incubations were performed at 4°C for 20 min . The solution was then passed through a 40 μm cell filter and then run on a FACS Diva 8 . 0 . 1 cell sorter ( BD Biosciences ) . FlowJo software ( version 7 . 6 , RRID:SCR_008520 ) was used for the analysis of flow cytometry data . In this manner , microvessel pericytes ( CD146+CD34-CD45-CD31- ) and adventitial cells ( CD34+CD146-CD45-CD31- ) were isolated and combined to constitute the PSC population . Each patient sample of cells was cultured independently . For in vitro experiments , cells were cultured at 37°C in a humidified atmosphere containing 95% air and 5% CO2 . ASCs and PSCs were expanded in growth medium ( GM ) consisting of DMEM , 15% fetal bovine serum ( FBS ) ( Gibco , Grand Island , NY ) , 1% penicillin/streptomycin ( Gibco ) . Medium was changed every 3 day unless otherwise noted . Bone marrow mesenchymal cells ( BMSCs ) of de-identified arthroplasty specimens of the human femur and tibia were flushed with PBS . All samples were obtained under IRB approval at JHU with a waiver of informed consent . Marrow cells were passed through a 70 μm cell strainer ( BD Bioscience ) to obtain a single-cell suspension of all nucleated cells . BMSCs were expanded in growth medium as above . Non-adherent cells were removed by washing the cultures with PBS twice and replacing the medium after 5 days . Multilineage differentiation capacity of human BMSCs was confirmed using previously reported methods , including osteogenic differentiation and Alizarin red staining ( Xu et al . , 2016 ) , adipogenic differentiation and Oil red O staining ( Barlian et al . , 2018 ) , and chondrogenic differentiation in high density micromass and Alcian blue staining ( Huang et al . , 2014 ) . Briefly , differentiation protocols were as follows: For osteogenic or adipogenic differentiation , after reaching 90% confluency in monolayer , growth medium was replaced with either osteogenic induction medium ( 100 nM dexamethasone , 50 μM ascorbic acid , and 10 mM β-glycerophosphate ( Sigma-Aldrich ) ) or adipogenic induction medium ( MesenCult Adipogenic Differentiation Medium ( Human; Stemcell Technologies , Vancouver , Canada ) ) . In high-density micromass , chondrogenic induction medium was used ( MesenCult-ACF Chondrogenic Differentiation Medium ( Stemcell Technologies ) ) . Human BMSCs were cultured and used for experiments at passage three unless otherwise noted . Neonatal mouse calvarial cells ( NMCCs ) were collected from C57BL/6J mice at postnatal day 1 . Dissected frontal and parietal bones were subjected to six sequential enzymatic digestions with collagenase type I ( Worthington Biochemical; 1 mg/mL ) and collagenase type II ( Worthington Biochemical; 1 mg/mL ) . The dissociated cells ( from sequential digestions 3–5 ) were then filtered through a 40 μm strainer and cultured in α-MEM supplemented with 15% FBS and 1% penicillin/streptomycin . For isolation of murine adipose-derived stromal/stem cells , subcutaneous fat pads were excised separately from C57BL/6J mice , finely minced , and digested using collagenase type II ( Worthington Biochemical; 0 . 75 mg/mL ) for 30 min at 37°C . The cell suspension was filtered through a 70 μm strainer and centrifuged at 1000 rpm for 5 min . The cells were plated in T75 flask and cultured in α-MEM supplemented with 15% FBS and 1% penicillin/streptomycin . Co-culture experiments were performed using 24-well 0 . 4 um transwell inserts ( Millipore , Darmstadt , Germany ) with human PSCs ( passage 8 ) placed in the upper insert and human BMSCs ( passage 3 ) in the lower well . Proliferation was measured after 72 hr co-culture in GM using the CellTiter96 AQueous One Solution Cell Proliferation Assay kit ( MTS , G358A; Promega , Madison , WI ) , where 8 × 103 BMSCs were cultured with or without 3 × 105 PSCs . Briefly , 20 μl of MTS solution was added to each well . After incubation for 1 hr at 37°C , the absorbance was measured at 490 nm using Epoch microspectrophotometer ( Bio-Tek , Winooski , VT ) . Cell migration was measured with Ibidi inserts ( Ibidi , Planegg/Martinsried , Germany ) , where 1 . 4 × 104 BMSCs were seeded and grown to confluency followed by co-culture with or without 3 × 105 PSCs in reduced FBS conditions ( GM with 1% FBS ) . Inserts were removed and cell migration into the empty area was monitored by brightfield microscopy at 0 , 8 , or 24 hr . The equilibrium width of the gap was calculated using the ImageJ software ( Version 1 . 49 v , RRID:SCR_003070; NIH , Bethesda , MD ) . Here , gap closure = ( scratch width at hour 0 − scratch width at hour 8 or 24 ) /scratch width at hour 0 × 100% . Osteogenic differentiation in co-culture was performed using osteogenic differentiation medium ( OM ) consisting of GM with 50 μM ascorbic acid , 10 mM β-glycerophosphate , and 100 nM dexamethasone ( Sigma-Aldrich ) . 4 × 104 BMSCs were cultured with or without of 5 × 105 PSCs , and alkaline phosphatase ( ALP ) staining was performed at 72 hr using an ALP staining kit ( Sigma-Aldrich ) . Relative staining intensity was quantified using ImageJ software and normalized to the control group . In select studies , PSCs were labeled with PKH26 Red Fluorescent Cell Linker Kit ( Sigma-Aldrich ) . After 48 hr co-culture of PKH26-labeled PSCs ( 5 × 105 ) with unlabeled BMSCs ( 2 × 104 ) , images were acquired with an Olympus IX71 inverted microscope ( Olympus , Cypress , CA ) . Unless otherwise stated , each experimental study was done in three technical replicates using two different PSC preparations . EVs were derived from passage 3–9 cells using ultracentrifugation based on previously validated protocols ( Sung et al . , 2015; Théry et al . , 2006 ) . Each preparation of EVs was isolated from independent cells and were not mixed . Briefly , ASCs or PSCs were expanded in growth medium . Upon reaching subconfluency and after triplicate washes in PBS , cells were cultured in DMEM only for 48 hr . EVs were collected by serial centrifugation at 300 × g for 10 min , 2 000 × g for 30 min , 10 000 × g for 30 min , and 120 000 × g for 4 hr at 4°C . The supernatant was discarded and the pellets were resuspended in 1X phosphate-buffered saline ( PBS ) . The protein concentration of EVs were quantified using the Pierce BCA Protein Assay Kit ( Thermo Scientific , Waltham , MA ) , according to the manufacturer’s instruction in similarity to prior reports ( Li et al . , 2016 ) . For transmission electron microscopy ( TEM ) , 10 μl of sample was adsorbed to glow-discharged 400 mesh ultra-thin carbon coated grids ( EMS CF400-CU-UL ) for two min , followed by three quick rinses of TBS and staining with 1% UAT ( uranyl acetate with 0 . 05 Tylose ) . Grids were immediately observed with a Philips CM120 at 80 kV and images captured with an AMT XR80 high-resolution ( 16-bit ) 8 Mpixel camera . The size distribution of EVs was examined by analysis of serial TEM images ( n = 57 images analyzed ) or using nanoparticle tracking analysis ( NTA ) with NanoSight NS500 ( Malvern , Worcestershire , UK ) with a 405 nm laser . For Western blotting , cells were lysed in RIPA buffer ( Thermo Scientific ) with protease inhibitor cocktail ( Cell Signaling Technology , Danvers , MA ) . Proteins were separated by SDS–polyacrylamide gel electrophoresis and transferred onto a nitrocellulose membrane . Protein extraction was blocked with 5% bovine serum albumin and incubated with primary antibodies at 4°C overnight . Finally , membranes were incubated with a horseradish-peroxidase ( HRP ) -conjugated secondary antibody and detected with ChemiDoc XRS+ System ( Bio-rad ) . EVs ( 1 , 2 . 5 , or 5 μg/mL protein content ) were added to culture medium during proliferation , migration and osteogenic differentiation as described above . The ratio of particles to protein was 1 . 18109 particles/μg . In select experiments , PSC-EVs were labeled using PKH26 lipophilic dye or a pH sensitive dye ( pHrodo Red Maleimide , Thermo Scientific ) prior to use . Upon completion of the reaction with the PSC-EVs , an excess of glutathione was added to consume the excess thiol-reactive reagent . PSC-EVs and dye were separated by a gel filtration column ( Sephadex G-25 column; GE Healthcare , Marlborough , MA ) . Proliferation assays were performed in 96-well plates ( 2 × 103 BMSCs/well ) and assayed at 48 , 72 , and 96 hr . Migration was performed as above with or without EVs supplementation ( 1 , 2 . 5 , or 5 μg/mL ) , with endpoints at 8 and 24 hr . Osteogenic differentiation was performed with OM with or without EVs ( 1 , 2 . 5 , or 5 μg/mL ) , with medium and EVs replenished every 3 days . The degree of mineralization was assessed by Alizarin Red S staining after 7 days ( Sigma-Aldrich ) , followed by incubation with 0 . 1 N sodium hydroxide and photometric quantification using Epoch microspectrophotometer ( 548 nm absorbance ) . Total RNA was extracted from the cultured cells using TRIzol Reagent ( Invitrogen , Carlsbad , CA ) according to the manufacturer’s instructions . 0 . 8 μg of total RNA was used for reverse transcription with iScript cDNA synthesis kit ( Bio-Rad ) following manufacturer’s instructions . Real-time PCR was performed using SYBR Green PCR Master Mix ( Thermo Scientific ) according to the manufacturer’s protocol . Relative gene expression was calculated using a 2-ΔΔCt method by normalization with GAPDH . Primer sequences are presented in Supplementary file 11 . In select experiments , PSC-EVs were incubated with trypsin ( 1 mg/ml , Sigma ) for 1 hr at 37°C and re-isolated by ultracentrifugation prior to application . In select experiments , PSC-EVs were pre-incubated with anti-CD9 ( RRID:AB_302894 ) or anti-CD81 ( RRID:AB_2811127 ) neutralizing antibodies , or their appropriate IgG isotype control prior to application . For neutralizing antibody experiments , 2 . 5 μg PSC-EVs was incubated with 10 μl neutralizing antibody or isotype control ( Anti-CD9: 1 mg/mL; Anti-CD81: 0 . 88 mg/mL ) for 4 hr at 4°C prior to use . In select experiments , shRNA-mediated knockdown of the CD9/CD81 receptors PTGFRN and IGSF8 was performed among primary human BMSCs prior to PSC-EVs application . ShRNA was transfected using TransIT-LT1 Transfection Reagent ( Mirus Bio , Madison , WI ) as described by the manufacturer . The target sequence of PTGFRN mRNA was 5′- GCCTTTGATGTGTCCTGGTTT-3′ and that of IGSF8 mRNA was 5′-GCTGCTGCTAATGCTAGGAAT-3′ . The medium was changed after 4 hr . Validation by regular PCR and semi-quantification was performed in ImageJ software . The RNA content of PSC-EVs and parent PSCs was detected by total RNA sequencing . Briefly , total RNA was extract from PSCs by Trizol ( Life technologies corporation ) . PSC-EV-derived RNA was isolated using exoRNeasy Serum Plasma Kits ( Qiagen , Hilden , Germany ) in accordance with the manufacturer’s instructions . The RNA samples were sent to the JHMI Deep Sequencing and Microarray core ( JHU ) and quantified by deep sequencing with the Illumina NextSeq 500 platform ( Illumina , San Diego , CA ) . Data analyses were performed using software packages including CLC Genomics Server and Workbench ( RRID:SCR_017396 and RRID:SCR_011853 ) , Partek Genomics Suite ( RRID:SCR_011860 ) , Spotfire DecisopnSite with Functional Genomics ( RRID:SCR_008858 ) , and QIAGEN Ingenuity Pathway Analysis ( RRID:SCR_008653 ) . The transcriptome of BMSCs treated with PSC-EVs or PBS control for 48 hr was examined by expression microarray . Briefly , total RNA was extracted from BMSCs by Trizol ( Life technologies corporation , Gaitherburg , MD , USA ) . The RNA samples were sent to the JHMI Deep Sequencing and Microarray core ( JHU , Baltimore , MD ) and quantified by microarray analyses on the Affymetrix Clariom_D Array ( Affymetrix , Santa Clara , CA ) . Data analyses were performed using software packages including CLC Genomics Server and Workbench ( RRID:SCR_017396 and RRID:SCR_011853 ) , Partek Genomics Suite ( RRID:SCR_011860 ) , Spotfire DecisionSite with Functional Genomics ( RRID:SCR_008858 ) , and QIAGEN Ingenuity Pathway Analysis ( RRID:SCR_008653 ) . All animal experiments were performed according to the approved protocol of the Animal Care and Use Committee ( ACUC ) at Johns Hopkins University ( Approval No . MO16M226 ) . 10-week-old , male C57BL/6J mice were purchased from the Jackson Laboratory ( strain #000664 , RRID:IMSR_JAX:000664 , Bar Harbor , ME ) . To assess progenitor cell migration , 10-week-old , male Pdgfrα-CreER;eGFP transgenic reporter mice were a kind gift from the Dwight Bergles laboratory ( Kang et al . , 2010 ) . Validation of the high specificity of Pdgfrα reporter activity has been previously confirmed ( Kang et al . , 2010 ) . In Pdgfrα-CreER;eGFP transgenic reporter mice , tamoxifen ( TM ) administration was performed by i . p . injection as per published protocols 14 days prior to defect creation ( Kang et al . , 2010 ) . TM free base ( MP Biomedicals , Solon , OH ) in 98% sunflower seed oil and 2% ethanol was administered at a concentration of 0 . 07 mg/kg daily for 5 days by i . p . injection as per previously validated protocols ( Kang et al . , 2010 ) . For calvarial defect creation , anesthesia was performed with 2–3% isoflurane in 100% oxygen at a flow rate of 1 L/min and animals were operated upon on a warm , small animal surgery station . Post-operative monitoring was performed in accordance with institutional policy . Analgesia was administered using buprenorphine ( 0 . 1 ml/25 g body weight ) via intraperitoneal injection after surgery , and a repeat injection was performed after 48 hr . A 4 mm skin incision was made over the right frontal bone . Next , a 1 . 8 mm diameter full thickness circular defect was created in the non-suture associated frontal bone using an Ideal Micro-Drill and a burr ( Xemax Surgical , Napa Valley , CA ) . Meticulous care was taken not to injure the underlying dura mater . Finally , the skin was sutured and the animal was monitored per established postoperative protocols . PSC-EVs ( 1 or 2 . 5 μg total dose ) or vehicle control were percutaneously injected into the tissue directly overlying the defect every 3 days ( 27G × 5/8’ Insulin Syringes , BD , Franklin Lakes , NJ ) . Mice were euthanized after 4 weeks for postmortem analysis . Samples were fixed in 4% PFA ( paraformaldehyde ) for 24 hr and imaged using a high-resolution microcomputed tomography ( microCT ) imaging system ( SkyScan 1294; Bruker MicroCT N . V , Kontich , Belgium ) . Scans were obtained at an image resolution of 10 μm , with the following settings: 1 mm of aluminum filter , X-ray voltage of 65 kVP , anode current of 153 uA , exposure time of 65 ms , frame averaging of 4 , and rotation step of 0 . 3 degrees . Three-dimensional images were then reconstructed from the 2D X-ray projections by implementing the Feldkamp algorithm using a commercial software package NRecon software ( 2 . 0 . 4 . 0 SkyScan , Bruker ) . For the 3D morphometric analyses of images , CTVox and CTAn software were used ( 1 . 13 SkyScan , Bruker ) . For calvarial defect analysis , a cylindrical volume of interest centered around each defect site was defined as the 1 . 8 mm in diameter and 1 mm in height with a threshold value of 80 . The amount of bone formation was analyzed and quantified in three different ways . Firstly , bone volume ( BV ) was calculated from binary x-ray images . Second , bone fractional area ( BFA ) was calculated by using CTVox to create a 3D rendering of calvarial defect and measuring pixels of bone in defect divided by total defect area using Adobe Photoshop ( RRID:SCR_014199; Adobe , San Jose , CA ) . Lastly , a bone healing score from 0 to 4 was assigned by three blinded observers according to previous published grading scales for calvarial defect healing ( Spicer et al . , 2012 ) . Briefly , the grading system was as follows: 0–no bone formation , 1–few bony spicules dispersed through defect , 2–bony bridging only at defect borders , 3–bony bridging over partial length of defect , and 4–bony bridging entire span of defect at longest point . After radiographic imaging , samples were transferred to 14% EDTA for decalcification for 14–21 days . Samples were then embedded in optimal cutting temperature compound ( OCT ) and sectioned in a coronal plane at 10 μm thickness . H&E staining was performed on serial sections . For immunofluorescent staining , additional sections were incubated with the following primary antibodies: anti-Ki67 ( 1:200 , RRID:AB_302459 ) , and anti-Osteocalcin ( 1:100 , RRID:AB_10675660 ) . Sections were washed with phosphate buffered saline ( PBS ) three times , 10 min each . All sections were blocked with 5% goat serum in PBS for 1 hr at 25°C; antigen retrieval was by trypsin enzymatic antigen retrieval solution for 10 mins at 37°C ( ab970; Abcam , Cambridge , MA ) . Primary antibodies were added to each section at their respective dilutions and incubated at 37°C for 1 hr and then overnight at 4°C . Next , a Dylight 594 goat anti-rabbit IgG ( H+L ) polyclonal ( 1:200 , RRID:AB_2336413 ) was used as the secondary antibody . Sections were counterstained with DAPI mounting medium ( H-1500 , Vector laboratories , Burlingame , CA ) . For studies in Pdgfrα-CreER;eGFP mice , histologic preparations were examined at 7 days post-injury , and total reporter activity within 3–4 random high-powered images per defect site were quantified using ImageJ software . All histological sections were examined under a Zeiss 700 confocal microscope ( Zeiss , Thornwood , NY ) . Quantitative data are expressed at mean ± SD . Statistical analyses were performed using the SPSS16 . 0 software ( RRID:SCR_002865 ) . All data were normally distributed . Student’s t test was used for two-group comparisons , and one-way ANOVA test was used for comparisons of three or more groups , followed by Tukey’s post hoc test . Differences were considered significant when p<0 . 05 , p<0 . 01 , and *p<0 . 001 . Expression data that support the findings of this study have been deposited in Gene Expression Omnibus ( GEO ) with the accession codes GSE118961 and GSE130086 .	Throughout our lives , our bodies need to heal after injury . Blood vessels are found throughout the body’s tissues and are a source of cells that guide the process of repair . These cells , called perivascular stem cells ( PSCs ) , are a type of stem cell found in the lining of blood vessels . Stem cells are cells that can become one of several different types of mature cells , depending on what the body needs . Extracellular vesicles are bundles of chemical signals that cells send into their external environment . Just like an address or a tag on a parcel , specific molecules mark the exterior surface of these bundles to deliver the message to the right recipient . Stem cells often use extracellular vesicles to communicate with surrounding cells . One role of PSCs is repairing damage to bones . Unusually , they do not turn into new bone cells and so do not directly contribute to the re-growing tissue . Instead , PSCs act indirectly , by stimulating the cells around them . How PSCs send these ‘repair instructions’ has , however , remained unclear . Xu et al . wanted to determine if PSCs used extracellular vesicles to direct bone repair , and if so , what ‘tags’ needed to be on the vesicles and on the receiving cells for this to happen . Experiments using PSCs and immature bone cells grown in the laboratory allowed the PSCs’ effect on bone cells to be simulated in a Petri dish . The two types of cells were grown on either side of a barrier , which separated them physically but allowed chemical signals through . In response to the PSCs , the immature bone cells multiplied , started to move ( which is something they need to do to heal damaged tissue ) , and began to resemble mature bone cells . Analysis of the signals released by the PSCs revealed that these were indeed extracellular vesicles , and that they were tagged by specific proteins called tetraspanins . Genetic manipulation of the immature bone cells later showed that these cells needed specific ‘receiver’ molecules to respond to the PSCs . Adding only extracellular vesicles to the bone cells , without any PSCs , confirmed that it was indeed the vesicles that triggered the healing response . Finally , giving the vesicles to mice with bone damage helped them to heal faster than untreated animals . These results have uncovered a key mechanism by which stem cells control the repair of bone tissue . This could one day lead to better treatments for patients recovering from fractures or needing bone surgery .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "cell", "biology" ]	2019	Human perivascular stem cell-derived extracellular vesicles mediate bone repair
Neurons in developing sensory pathways exhibit spontaneous bursts of electrical activity that are critical for survival , maturation and circuit refinement . In the auditory system , intrinsically generated activity arises within the cochlea , but the molecular mechanisms that initiate this activity remain poorly understood . We show that burst firing of mouse inner hair cells prior to hearing onset requires P2RY1 autoreceptors expressed by inner supporting cells . P2RY1 activation triggers K+ efflux and depolarization of hair cells , as well as osmotic shrinkage of supporting cells that dramatically increased the extracellular space and speed of K+ redistribution . Pharmacological inhibition or genetic disruption of P2RY1 suppressed neuronal burst firing by reducing K+ release , but unexpectedly enhanced their tonic firing , as water resorption by supporting cells reduced the extracellular space , leading to K+ accumulation . These studies indicate that purinergic signaling in supporting cells regulates hair cell excitability by controlling the volume of the extracellular space . The developing nervous system must generate , organize , and refine billions of neurons and their connections . While molecular guidance cues forge globally precise neuronal connections between distant brain areas ( Dickson , 2002; Stoeckli , 2018 ) , the organization of local connections is initially coarse and imprecise ( Dhande et al . , 2011; Kirkby et al . , 2013; Sretavan and Shatz , 1986 ) . Coincident with the refinement of topographic maps , nascent circuits experience bursts of intrinsically generated activity that emerge before sensory systems are fully functional ( Kirkby et al . , 2013 ) . This intrinsically generated activity consists of periodic bursts of high frequency firing that promotes the survival and maturation of neurons in sensory pathways ( Blankenship and Feller , 2010; Moody and Bosma , 2005 ) . The precise patterning of this electrical activity appears crucial for refinement of local connections , as its disruption results in improper formation of topographic maps ( Antón-Bolaños et al . , 2019; Burbridge et al . , 2014; Xu et al . , 2011 ) and impaired maturation and specification of sensory neurons ( Shrestha et al . , 2018; Sun et al . , 2018 ) . In all sensory systems that have been examined , spontaneous burst firing arises within their respective developing sensory organs ( e . g . retina , olfactory neuroepithelium , and cochlea ) ( Blankenship and Feller , 2010; Yu et al . , 2004 ) . Although the mechanisms that induce spontaneous activity in the developing retina have been extensively explored , much less is known about the key steps involved in triggering auditory neuron burst firing in the developing cochlea . Understanding these processes may provide novel insights into the causes of developmental disorders of hearing , such as hypersensitivity to sounds and auditory processing disorders that prevent children from communicating and learning effectively . The mechanisms responsible for initiating spontaneous activity appear to be unique to each sensory system , reflecting adaptations to the structure and cellular composition of the sensory organs . In the cochlea , two distinct models have been proposed to initiate burst firing of inner hair cells ( IHCs ) . One model proposes that the initiation of burst firing results from intermittent hyperpolarization of tonically active IHCs by cholinergic efferents ( Johnson et al . , 2011; Wang and Bergles , 2015 ) , which provide prominent inhibitory input to IHCs prior to hearing onset ( Glowatzki and Fuchs , 2000 ) . While transient activation of acetylcholine receptors in acutely isolated cochleae caused IHCs to switch from sustained to burst firing ( Johnson et al . , 2011 ) , in vivo recordings from auditory brainstem revealed that neuronal burst firing remains , with altered temporal structure , in α9 acetylcholine receptor knockout ( KO ) mice ( Clause et al . , 2014 ) that lack functional efferent signaling in IHCs ( Clause et al . , 2014; Johnson et al . , 2013 ) . Burst firing also persists in IHCs and auditory neurons in cochleae maintained in vitro without functional efferents ( Johnson et al . , 2013; Tritsch et al . , 2007 ) . These findings suggest that cholinergic efferents modulate the temporal characteristics of bursts , but are not essential to initiate each event . An alternative model proposes that IHCs are induced to fire bursts of action potentials by the release of K+ from nearby inner supporting cells ( ISCs ) , which together form a transient structure known as Köllikers organ ( Greater Epithelial Ridge ) that is prominent in the cochlea prior to hearing onset . K+ release from ISCs occurs following a cascade of events that begins with the spontaneous release of ATP and activation of purinergic autoreceptors ( Tritsch et al . , 2010a ) . Purinergic receptor activation induces an increase in intracellular Ca2+ in ISCs , opening of Ca2+-activated Cl– channels ( TMEM16A ) , efflux of Cl– and subsequently K+ to balance charge ( Tritsch et al . , 2007; Wang et al . , 2015 ) . The loss of ions during each event draws water out of ISCs through osmosis , leading to pronounced shrinkage ( crenation ) of ISCs . While these pathways have been extensively studied in vitro , the molecular identity of the purinergic receptors has remained elusive and few manipulations of this pathway have been performed in vivo , limiting our understanding of how spontaneous activity in the cochlea influences patterns of neuronal activity in auditory centers at this critical stage of development . Here , we show that the key initial step in generation of spontaneous activity in the auditory system involves activation of P2RY1 autoreceptors in ISCs . These metabotropic receptors induce Ca2+ release from intracellular stores that allow TMEM16A channels to open . Pharmacological inhibition of P2RY1 or genetic deletion of P2ry1 dramatically reduced burst firing in spiral ganglion neurons ( SGNs ) and blocked the coordinated , spatially restricted activation of ISCs , IHCs , and SGNs in the cochlea . Unexpectedly , P2RY1 activation also promoted the dissipation of K+ away from IHCs by increasing the volume of extracellular space . Conversely , inhibition of P2RY1 reduced the extracellular space and restricted the redistribution of K+ within the cochlear epithelium , causing IHCs to depolarize and fire tonically , demonstrating an important role for purinergic receptor-mediated extracellular space changes in controlling IHC excitability . Using in vivo widefield epifluorescence imaging of the auditory midbrain in unanesthetized mice , we show that acute inhibition of P2Y1 dramatically reduced burst firing of auditory neurons in isofrequency domains . Together , these data indicate P2RY1 autoreceptors in non-sensory supporting cells in the cochlea play a crucial role in generating bursts of activity among neurons that will ultimately process similar frequencies of sound , providing the means to initiate the maturation of auditory pathways before hearing onset . Periodic release of ATP from ISCs in the developing cochlea initiates a signaling cascade in these cells that increases intracellular calcium ( Ca2+ ) , opens Ca2+-activated Cl– channels ( TMEM16A ) , and ultimately results in efflux of chloride and K+ into the extracellular space . Although the increase in intracellular Ca2+ following activation of purinergic autoreceptors is sufficient to induce both depolarization and osmotic shrinkage ( Wang et al . , 2015 ) , the relative contributions of Ca2+ influx ( e . g . through Ca2+-permeable , ionotropic P2X receptors ) and release from intracellular stores ( e . g . following metabotropic P2Y receptor activation ) to these cytosolic Ca2+ transients is unclear . To define the signaling pathways engaged by purinergic receptor activation , we examined the sensitivity of spontaneous ISC whole-cell currents and crenations to inhibitors of intracellular Ca2+ release pathways ( Figure 1A ) . Spontaneous inward currents and crenations were abolished following a 15 min incubation of excised cochlea in BAPTA-AM ( 100 μM ) , a cell permeant Ca2+ chelator ( Figure 1B–F ) , and after depleting intracellular Ca2+ stores with thapsigargin ( 2 μM ) , an inhibitor of endoplasmic reticulum Ca2+-ATPase ( Figure 1B–F ) . These data suggest that Ca2+ release from intracellular stores is necessary for spontaneous electrical activity in ISCs . Metabotropic Gq-coupled receptors typically induce PLC-mediated cleavage of phosphatidylinositol 4 , 5-bisphosphate ( PIP2 ) and subsequent binding of inositol trisphosphate ( IP3 ) to IP3 receptor-channels on the endoplasmic reticulum to release Ca2+ into the cytoplasm . To determine if PLC signaling is necessary to generate spontaneous activity in ISCs , we recorded spontaneous currents and crenations from ISCs in the presence of U73122 ( 10 μM ) , a PLC inhibitor , and U73343 ( 10 μM ) , an inactive succinimide analog . The frequency of spontaneous currents and crenations were significantly reduced by U73122 , but not by U73343 ( Figure 1B–F ) ; the amplitudes and charge transfer of residual activity also trended lower during PLC inhibition , but this did not reach significance due to high variance in the sizes of the spontaneous responses ( Figure 1B–F ) . Intracellular Ca2+ stores can also be mobilized through calcium-induced calcium release ( CICR ) involving ryanodine receptors . However , neither spontaneous currents nor crenations were affected by ryanodine ( 10 μM ) ( Figure 1—figure supplement 1 ) . Together , these results suggest that engagement of a Gq-coupled purinergic autoreceptor is a critical first step in initiating PLC-mediated Ca2+ release from intracellular stores and subsequent activation of TMEM16A channels . There are eight members of the metabotropic purinergic receptor family in mouse , four of which are Gq-coupled ( P2RY1 , P2RY2 , P2RY4 , and P2RY6 ) . Gene expression studies in the developing mouse cochlea revealed that non-sensory cells express P2RY1 mRNA at high levels , >100 fold higher than any other P2RY ( Figure 2A; Scheffer et al . , 2015 ) and that expression of this receptor progressively increases during early postnatal development ( Figure 2A , inset ) concurrent with increases in spontaneous activity ( Tritsch and Bergles , 2010 ) . To determine which cells in the sensory epithelium express P2RY1 , we isolated cochleae from P2ry1-LacZ reporter mice and performed X-gal staining . Intense blue labeling was present along the entire length of the cochlea within Kölliker’s organ ( Greater Epithelial Ridge; Figure 2B ) , and cross-sections of cochlea revealed that staining was present within ISCs , but not IHCs ( Myosin VIIA , Figure 2C ) , indicating that P2RY1 is appropriately localized to sense ATP release from ISCs prior to hearing onset . To determine if P2RY1 is responsible for spontaneous ATP-mediated currents in ISCs , we examined the sensitivity of these responses and associated crenations to the P2RY1 selective antagonist MRS2500 ( Figure 3A , B ) . Acute inhibition of P2RY1 with MRS2500 ( 1 µM ) markedly reduced both spontaneous ISC currents ( Figure 3B , C ) and crenations ( Figure 3D , E ) ; near complete inhibition occurred within minutes at both room temperature ( Figure 3B , C ) and near physiological temperature ( Figure 3—figure supplement 1A–G ) , with only sporadic , small amplitude events remaining that were not mediated by purinergic receptors ( Figure 3—figure supplement 1B–E ) . Consistent with the involvement of P2RY1 , the amplitude and total charge transfer of ISC events ( Figure 3—figure supplement 2A , B ) and size of spontaneous crenations ( Figure 3—figure supplement 2C , D ) were smaller in cochleae in P2ry1 KO mice relative to controls . However , supporting cells in P2ry1 KO mice exhibited some aberrant , gain-of-function activity consisting of more frequent , small amplitude currents ( Figure 3—figure supplement 2A , B ) , that were not blocked by MRS2500 or broad-spectrum P2 receptor antagonists ( Figure 3—figure supplement 2E , F ) . ATP-mediated signaling in ISCs activates TMEM16A , triggering K+ efflux that depolarizes nearby IHCs . To assess whether P2RY1 signaling is also required for periodic excitation of IHCs prior to hearing onset , we assessed the sensitivity of spontaneous IHC inward currents to MRS2500 ( Figure 3F ) . Consistent with the supporting cell origin of IHC activity , application of MRS2500 ( 1 µM ) also abolished spontaneous currents in IHCs ( Figure 3F , G ) . Together , these data suggest that P2RY1 is the primary purinergic autoreceptor on ISCs responsible for inducing periodic excitation of IHCs prior to hearing onset . Although P2RY1 inhibition abolished most transient inward currents in both ISCs and IHCs , a progressively increasing inward current ( downward shift in baseline ) appeared in both cell types with prolonged application of MRS2500 ( Figure 3B , F ) . Prior studies in CNS brain slices indicated that Gq-coupled purinergic receptors in astrocytes regulate extracellular K+ concentration and neuronal excitability ( Wang et al . , 2012 ) . The slowly progressing nature of the response in IHCs and ISCs suggest that it may arise from accumulation of K+ released from cells in the organ of Corti . If this hypothesis is correct , then inhibiting the main sources of extracellular K+ should diminish this inward current . Indeed , when IHC and SGN excitation was inhibited with cadmium ( CdCl2 , 100 µM ) and tetrodotoxin ( TTX , 1 µM ) , and K+ transporters Na , K-ATPase and NKCC were inhibited with ouabain ( 10 µM ) and bumetanide ( 50 µM ) , no inward current was induced in ISCs upon blocking P2RY1 ( Figure 3H , I ) . Similarly , if K+ accumulation is responsible for the current in IHCs , it should be abolished when the ability of IHCs to detect changes in K+ is reduced . When whole cell recordings were performed from IHCs using an internal solution containing Cs+ and TEA , which blocks most IHC K+ channels ( Kros et al . , 1998; Marcotti et al . , 2003 ) , MRS2500 also did not induce an inward current ( Figure 3J , K ) . Together , these results suggest that P2RY1 has two distinct effects in the cochlea; it induces the transient inward currents that trigger IHC burst firing ( Figure 3F; see also Figure 6D , G ) and it accelerates clearance of K+ away from IHCs within the organ of Corti . To directly assess the relationship between P2RY1 activity and extracellular K+ accumulation near IHCs , we monitored K+ levels in the extracellular space using IHC K+ channels . Focal P2RY1 stimulation with a selective agonist ( MRS2365 , 10 μM ) , which mimics the effect of endogenous ATP by eliciting an inward current and crenations in ISCs in control but not P2ry1 KO mice ( Figure 4A–C ) , was combined with assessments of the reversal potential of K+ currents in IHCs using a voltage protocol similar to that used to assess extracellular K+ buildup at vestibular calyceal synapses ( Lim et al . , 2011 ) ( Figure 4D–F ) . This protocol consisted of: ( 1 ) a hyperpolarizing step to –110 mV to relieve K+ channel inactivation , ( 2 ) a depolarizing step to +30 mV to activate outward K+ currents , and ( 3 ) a step to –70 mV to obtain a ‘tail’ current . Because the conductance during this last step is largely mediated by K+ channels , it is highly sensitive to shifts in K+ driving force induced by changes in extracellular K+ ( Contini et al . , 2017; Lim et al . , 2011 ) . Following transient stimulation of P2RY1 , these K+ tail currents immediately shifted inward , as would be expected if extracellular K+ increases ( Figure 4G , J ) , consistent with the effects metabotropic purinergic receptor stimulation on synaptically-evoked K+ currents in IHCs ( Wang et al . , 2015 ) . However , after a few seconds these K+ currents shifted outward relative to baseline , indicative of a gradual decrease in extracellular K+ below that present prior to P2RY1 stimulation , before gradually returning to pre-stimulation levels after several minutes ( Figure 4G , J ) . The outward shift in K+ tail current closely followed the time course of supporting cell crenation ( Figure 4G ) , suggesting that the shrinkage of ISCs induced by P2RY1 activation results in a prolonged increase in extracellular space around IHCs that allows greater dilution and more rapid redistribution of K+ in the organ of Corti . Alternatively , buildup of extracellular K+ alone may stimulate greater uptake . To determine if rapid increases in extracellular K+ and Cl– were sufficient to stimulate K+ redistribution in the absence of crenation , we focally applied KCl ( 130 mM ) into the supporting cell syncytium near IHCs in the presence of P2RY1 antagonists ( Figure 4H , J ) . As expected , this transient increase in extracellular K+ induced an inward shift in K+ tail currents and a brief optical change induced by fluid delivery; however , K+ tail currents rapidly returned to baseline and did not shift outward , suggesting that extracellular K+ ( and Cl– ) elevation are not sufficient to enhance K+ redistribution rates . In addition , we transiently stimulated P2RY1 in Tecta-Cre;TMEM16Afl/fl mice , in which purinergic receptor activation is preserved , but crenations are abolished ( Wang et al . , 2015 ) . In these mice , ISCs failed to crenate , IHCs did not depolarize , and K+ tail currents remained stable throughout the duration of the recording ( Figure 4I , J ) . These results suggest that purinergic autoreceptors on ISCs influence extracellular K+ levels by triggering K+ release and by altering K+ redistribution by controlling the size of the extracellular space . To evaluate the role of P2RY1 in initiating coordinated cellular activity in the cochlea , we monitored large-scale activity patterns in excised cochleae from Pax2-Cre;R26-lsl-GCaMP3 mice , which express GCaMP3 in nearly all cells of the inner ear . Time lapse imaging revealed that the spontaneous Ca2+ elevations that occur simultaneously within groups of ISCs , IHCs , and SGNs ( Eckrich et al . , 2018; Tritsch and Bergles , 2010; Zhang-Hooks et al . , 2016 ) ( Figure 5A ) were abolished following inhibition of P2RY1 with MRS2500 ( Figure 5B , C ) and were dramatically reduced in P2ry1 KO mice ( Pax2-Cre;R26-lsl-GCaMP3;P2ry1–/– ) ( Figure 5—figure supplement 1A , B ) . Moreover , in accordance with the progressive increase in extracellular K+ that follows P2RY1 inhibition , there was a gradual increase in spontaneous , uncoordinated Ca2+ transients in IHCs in the presence of MRS2500 ( Figure 5D–F ) , suggesting that this K+ accumulation increases IHC firing . Similarly , IHCs in P2ry1 KO mice displayed a higher level of uncorrelated Ca2+ transients in hair cells ( Figure 5—figure supplement 1C–E ) , indicative of enhanced excitability . Together , these results indicate that P2RY1 is required for coordinated activation of ISCs , IHCs , and SGNs before hearing onset and that P2RY1 inhibition leads to higher rates of uncorrelated activity . IHCs in the developing cochlea exhibit regenerative Ca2+ spikes that strongly activate post-synaptic SGNs , resulting in bursts of action potentials that propagate to the CNS . To determine if P2RY1 initiates burst firing in SGNs , we recorded spontaneous activity from SGNs using juxtacellular recordings from their somata ( Figure 6A ) . Application of MRS2500 resulted in a dramatic reduction of high frequency burst firing in SGNs , visible as a decrease in burst frequency and action potentials per burst ( Figure 6E , F ) . All SGN spiking was abolished by the AMPA receptor antagonist NBQX ( 50 µM ) ( Figure 6D ) , indicating that their activity requires synaptic excitation by IHCs . The precise patterning of action potentials within bursts was also disrupted by P2RY1 inhibition , as there were fewer interspike intervals in the 75–125 ms range ( Figure 6C , F ) , which correspond to the maximum rate of Ca2+ spike generation by IHCs during ATP-mediated excitation ( Tritsch et al . , 2010b ) . Additionally , the coefficient of variation measured for interspike intervals was significantly lower following P2RY1 inhibition , suggesting SGNs fire more randomly ( Figure 6E ) . However , the average frequency of action potentials remained unchanged during P2RY1 inhibition ( Figure 6E ) due to increases in non-burst firing , consistent with depolarizing inward currents observed in IHCs with P2RY1 inhibition ( Figure 3F , K ) . SGNs in P2ry1 KO cochleae exhibited activity similar to wildtype SGNs in the presence of MRS2500 , with a lower burst firing rate , fewer interspike intervals in the 75–125 ms range , and a lower coefficient of variation of interspike intervals relative to controls ( Figure 6G–I ) . However , despite the profound contribution of P2RY1 to ISC and IHC activity , some burst-like behavior ( clustered , but lacking the same density of action potentials in control conditions ) was still observed in SGNs ( Figure 6C , D , G ) suggesting that other forms of excitation emerge in the absence of P2RY1 , perhaps due to the increase in overall excitability or compensatory developmental changes . Together , these data indicate that P2RY1 is required to generate discrete bursts of action potentials in SGNs and that loss of these receptors enhances uncorrelated firing . The highly synchronized electrical activity exhibited by IHCs prior to hearing onset propagates through the entire developing auditory system to induce correlated firing of auditory neurons within isofrequency zones ( Babola et al . , 2018; Tritsch et al . , 2010b ) . To determine if P2RY1 is required to produce this form of correlated activity , we used in vivo wide-field epifluorescence microscopy of the inferior colliculus ( IC ) in mice that express GCaMP6s in all neurons ( Snap25-T2A-GCaMP6s and Snap25-T2A-GCaMP6s;P2ry1–/– mice ) . Time lapse imaging revealed that both control and P2ry1 KO mice exhibited correlated neuronal activity confined to stationary bands oriented along the tonotopic axis ( Figure 7A–C ) . Spontaneous events were less frequent in P2ry1 KO mice ( 9 . 7 ± 0 . 8 events per minute compared to 13 . 4 ± 0 . 7 events per minute in control; two-tailed Student’s t test , p=0 . 002 ) , although the events were similar in amplitude and duration ( half-width ) ( Figure 7D ) , suggesting that some compensatory amplification of events occurs in these mice . Spontaneous activity in P2ry1 KO mice differed from controls in three other ways . First , the contralateral bias exhibited for each event was higher , with the weaker relative to stronger side amplitude decreasing from 0 . 61 ± 0 . 02 to 0 . 44 ± 0 . 02 ( two-tailed Student’s t test , p=3 . 0e-6 ) ( Figure 7D ) . Second , the coefficient of variation ( ratio of standard deviation to the mean ) of event amplitudes was 40% higher relative to controls ( Figure 7D ) . Third , a detailed examination of the spatial location of events across the tonotopic axis ( Figure 7E ) revealed that activity in brain areas later responsible for processing higher frequency tones ( ~8–16 kHz ) exhibited the greatest reduction ( 68% ) in P2ry1 KO mice , while activity in low frequency areas was unaltered ( Figure 7F–H ) . In P2ry1 KO mice , bilateral removal of both cochleae abolished activity in the IC , demonstrating that activity in these mice still originates in the periphery ( Figure 7—figure supplement 1A–C ) . Although P2ry1 KO mice mimic some aspects of acute P2RY1 inhibition , the absence of P2RY1 signaling throughout life may have led to compensatory changes , such as the increase in non-purinergic ISC activity ( see Figure 3—figure supplement 2E ) . Therefore , to better assess the role of P2RY1 in generating spontaneous activity in vivo , we acutely inhibited these receptors by administering MRS2500 into the intraperitoneal cavity of mice while imaging activity in the IC . Compared to mice injected with control solution ( 5% mannitol ) , mice injected with MRS2500 exhibited dramatic reductions in IC event frequency ( from 13 . 3 ± 0 . 8 to 3 . 9 ± 1 . 1 events per minute; two-tailed Student’s t test , p=0 . 0005 ) and amplitude ( from 9 . 9 ± 0 . 5 to 4 . 9 ± 0 . 8% ΔF/Fo; two-tailed Student’s t test , p=0 . 002 ) ~5 min after administration ( Figure 8A–D ) . This decrease was specific to the IC , as SC retinal wave activity ( Ackman et al . , 2012 ) was unaffected by acute MRS2500 administration ( Figure 8B , C , E ) , suggesting that the locus of action is likely within the cochlea , which has been shown to have a permeable blood-tissue barrier at this age ( Suzuki et al . , 1998 ) . Spatial analysis revealed that unlike the selective deficit observed in higher frequency zones in P2ry1 KO mice , the inhibition with MRS2500 administration was not limited to certain tonotopic regions , but rather occurred evenly across all frequency zones ( Figure 8F , G ) . Together , these data indicate that P2RY1 autoreceptors on ISCs within the cochlea play a critical role in initiating spontaneous bursts of neural activity in auditory centers within the brain prior to hearing onset . Before the onset of hearing , neurons in the auditory system that will process similar sound frequencies exhibit periodic bursts of highly correlated activity , an entrainment that is initiated within the cochlea by the release of ATP ( Babola et al . , 2018; Clause et al . , 2014; Sonntag et al . , 2009; Tritsch et al . , 2010b ) . The mechanism of ATP release from ISCs within the developing cochlear epithelium remains undefined , but may involve gap-junction hemichannels , as gap-junction antagonists profoundly inhibit spontaneous activity and lowering extracellular calcium , a manipulation that increases the open probability of hemichannels ( Peracchia , 2004 ) , markedly enhances the frequency of spontaneous activity ( Tritsch et al . , 2007 ) . Extruded ATP then activates ISC purinergic receptors , triggering a rapid increase of intracellular Ca2+ , gating of TMEM16A Ca2+-activated Cl– channels , and subsequent Cl– and K+ efflux into the extracellular space ( Tritsch et al . , 2007; Wang et al . , 2015 ) . This transient K+ efflux is sufficient to depolarize nearby IHCs , resulting in a burst of Ca2+action potentials , release of glutamate , and suprathreshold activation of postsynaptic SGNs via AMPA and NMDA receptors ( Tritsch et al . , 2010b; Zhang-Hooks et al . , 2016 ) ( Figure 9A ) . These bursts propagate throughout the entire developing auditory system ( Babola et al . , 2018 ) and likely activate the olivocochlear efferent reflex arc , providing bursts of inhibition to IHCs that may modulate the envelope of excitation ( Clause et al . , 2014 ) and/or the number of IHCs activated during each event . Our results show that activation of metabotropic P2RY1 autoreceptors is a key first step in this transduction pathway . P2RY1 is highly expressed by ISCs at a time when spontaneous activity is prominent in the cochlea ( Scheffer et al . , 2015; Tritsch and Bergles , 2010 ) ( Figure 2A ) , and spontaneous activity was reduced when intracellular Ca2+ stores were depleted or PLC was inhibited ( Figure 1B–F ) , manipulations that disrupt canonical Gq-coupled GPCR signaling pathways ( Erb and Weisman , 2012; Fabre et al . , 1999 ) . Moreover , our pharmacological studies indicate that P2RY1 is both necessary and sufficient for spontaneous current generation in supporting cells ( Figures 3B and 4B ) , and acute inhibition of P2RY1 in vivo profoundly decreased cochlea-generated activity in the auditory midbrain ( Figure 8C ) . This reliance on P2RY1 is somewhat unexpected , as ionotropic P2X and other metabotroic P2Y receptors are also widely expressed in the developing cochlea ( Brändle et al . , 1999; Eckrich et al . , 2018; Huang et al . , 2010; Lahne and Gale , 2008; Liu et al . , 2015; Nikolic et al . , 2003; Scheffer et al . , 2015; Tritsch et al . , 2007 ) . The lack of P2X or other Gq-coupled P2Y receptor engagement may reflect the particular spatial-temporal characteristics of ATP release by ISCs , which may occur in locations enriched in P2RY1 or yield ATP concentration transients that favor P2RY1 activation . Exogenous ATP can induce all of the phenomenon associated with spontaneous events ( ISC currents , crenation , IHC depolarization , SGN burst firing ) ; however , it is possible that other nucleotides are released that have greater affinity for P2RY1 ( e . g . ADP ) , or that extracellular nucleotidases rapidly convert ATP to ADP that favor activation of native metabotropic receptors ( von Kügelgen , 2006; Vlajkovic et al . , 1998; Vlajkovic et al . , 2002 ) . Pharmacological inhibition of P2RY1 unexpectedly induced IHCs to gradually depolarize and begin tonic , uncorrelated firing , a phenotype also observed in P2ry1 KO mice ( Figure 5—figure supplement 1C ) . Our studies indicate that this phenomenon occurs because P2RY1 controls the volume of the extracellular space in the organ of Corti . Activation of P2RY1 induces ISCs to shrink osmotically ( crenate; Figure 9A ) , a consequence of ion and water efflux that is triggered by opening of TMEM16A channels ( Figure 4G ) . The resulting increase in extracellular space lasts for many seconds and enhances dissipation of extracellular K+ ( Figure 9A ) , visible through the time-dependent shift in the reversal potential of K+-mediated tail currents ( Figure 4G , I ) . Conversely , inhibition of P2RY1 increased the size of ISCs , a swelling-induced ‘relaxation’ that concomitantly decreased extracellular space around IHCs ( Figure 9B ) . K+ accumulation and depolarization of IHCs followed , an effect absent when IHC K+ channels were inhibited ( Figure 3J , K ) or K+ release from cells in the organ of Corti was reduced ( Figure 3H–K ) , leading to tonic firing of IHCs and post-synaptic SGNs ( Figure 6D , G and Figure 9B ) . This phenomenon is consistent with the depolarizing shift in resting membrane potential of IHCs observed in Tmem16A cKO mice ( Wang et al . , 2015 ) , which similarly blocks ISC crenation . Of note , epileptiform activity can be elicited in the CNS by inducing cell swelling with hypoosmotic solutions or by impairing K+ buffering ( Larson et al . , 2018; Murphy et al . , 2017; Thrane et al . , 2013 ) . Basal P2RY1 activation in supporting cells hyperpolarizes nearby IHCs by expanding the extracellular space and lowering local K+ concentrations . These changes increase the dynamic range of IHCs , allowing them to respond to a wider range of ATP release events and enabling finer control of their excitability through transient ATP-mediated signaling events . The tonic inward current that develops in ISCs in response to P2RY1 block was abolished when homeostatic K+ release pathways ( Na+ channels , Ca2+ channels , Na+-K+-Cl–cotransporters , and Na , K-ATPase ) were inhibited ( Figure 3H , I ) , suggesting that K+ redistribution mechanisms , in the absence of ISC crenation , are weak at this stage of development . Indeed , although the membrane potential of ISCs is close to EK , their membrane conductance is dominated by intercellular gap junction channels; when uncoupled from their neighbors , they exhibit very high ( 1–2 GΩ ) input resistance ( Jagger and Forge , 2015; Wang et al . , 2015 ) , suggesting that few K+ leak channels are expressed . The presence of tight junctions at the apical surface of the cochlear epithelium and the limited K+ conductance of ISCs may restrict passive diffusion and dilution of K+ , similar to what has been described in the vestibular epithelium ( Contini et al . , 2017 ) , thus necessitating uptake via alternative mechanisms . Both inner phalangeal and Deiters’ cells ( which envelop the inner and outer hair cells , respectively ) express K+-Cl– symporters , Na , K-ATPase pumps , and inwardly-rectifying K+ channels that may siphon K+ into the supporting cell syncytium after extrusion from hair cells . However , the apparently low capacity of these systems places a greater dependence on diffusion within the extracellular volume fraction controlled by the supporting cells . Similar to ISCs , astrocytes in the CNS facilitate rapid dissipation of extracellular K+ by K+ uptake and redistribution through the glial syncytium via gap junctions , a mechanism termed spatial buffering ( Kofuji and Newman , 2004 ) . Astrocytes are efficient K+ sinks due to their highly negative resting potential ( ~–85 mV ) and large resting K+ conductance dominated by inward rectifying K+ channels ( Olsen , 2012 ) and two-pore leak K+ channels ( Ryoo and Park , 2016 ) . While uptake of K+ through these channels is passive , recent studies suggest that K+ buffering in astrocytes is actively regulated by purinergic receptors . Following stimulation of native astrocyte purinergic receptors or foreign Gq-coupled receptors ( MrgA1 ) and release of Ca2+ from intracellular stores , Na , K-ATPase activity increased , resulting in a transient decrease in extracellular K+ , hyperpolarization of nearby neurons , and reduction in their spontaneous activity ( Wang et al . , 2012 ) . Although P2RY1 is expressed by some astrocytes and can trigger Ca2+ waves ( Gallagher and Salter , 2003 ) , this mechanism does not appear to regulate IHC excitability in the cochlea , as stimulation of P2RY1 in Tmem16a cKO mice , which have intact metabotropic receptor signaling but no crenations ( Wang et al . , 2015 ) , did not hyperpolarize IHCs ( Figure 4I , J ) . Thus , astrocytes and cochlear ISCs use purinergic signaling in distinct ways to maintain the ionic stability of the extracellular environment and control the excitability of nearby cells . Our understanding of how non-sensory cells contribute to spontaneous activity has been limited by a lack of in vivo mechanistic studies . Recent advances in visualizing cochlea-induced spontaneous activity in central auditory centers in vivo using genetically-encoded calcium indicators ( Babola et al . , 2018 ) allowed us to assess whether supporting cell purinergic receptors are involved in generating this activity . Prior to hearing onset , the blood-labyrinth barrier within the inner ear is not fully formed ( Suzuki et al . , 1998 ) , permitting pharmacological access to the cochlea at this age . Infusion of a P2RY1 inhibitor into the intraperitoneal space dramatically decreased the frequency and amplitude of transient , spatially restricted bands of activity in the inferior colliculus within minutes , while retina-induced activity in the superior colliculus ( Ackman et al . , 2012 ) was unaffected ( Figure 8 ) , suggesting that inhibition is not due to activation of astrocyte P2RY1 receptors or broad disruption of neural activity; as noted above , inhibition of P2RY1 in astrocytes would be expected to enhance , rather than inhibit neuronal activity ( Wang et al . , 2012 ) . Within the CNS , P2ry1 mRNA is expressed in the cochlear nucleus and P2RY1-mediated calcium transients can be induced in spherical bushy cells ( Milenkovic et al . , 2009 ) . Although the consequences of P2RY1 inhibition on neuronal activity in auditory centers is not yet known , it is possible that some effects of MRS2500 also result from actions on P2RY1 in the CNS . In vivo imaging in P2ry1 KO mice recapitulated many aspects of changes seen when P2RY1 was acutely inhibited , with significantly reduced neuronal activity observed in lateral regions of the IC ( later active to 8–16 kHz tones; Figure 7 ) . However , neuronal burst firing persisted within central regions of the IC , regions that will ultimately process lower frequency sounds ( 3–8 kHz ) . At present , we do not know the reason for the differences between P2RY1 inhibition and constitutive P2ry1 deletion , but the absence of P2RY1 throughout development may have induced widespread compensatory changes as a result of the new patterns of activity experienced by these developing neurons ( Figure 3—figure supplement 2 ) . Indeed , developing sensory systems exhibit a remarkable ability to preserve spontaneous activity . In the visual system , cholinergic antagonists injected directly into the eye blocks retinal waves in vivo ( Ackman et al . , 2012 ) , but genetic removal of the β2 acetylcholine receptor subunit alters , but does not abolish , peripherally-generated activity ( Zhang et al . , 2012 ) . In the auditory system , in vivo spontaneous activity can be blocked by acute inhibition of cochlear AMPARs , but deaf mice in which hair cell glutamate release is abolished ( Vglut3 KO mice ) exhibit activity patterns remarkably similar to control mice ( Babola et al . , 2018 ) . These robust homeostatic mechanisms allow spontaneous activity to persist despite disruption of key transduction components . Local purinergic signaling within the cochlea may still initiate tonotopic activity in central auditory circuits of P2ry1 KO mice , perhaps by engaging other subtypes of metabotropic purinergic receptors , as events in the IC exhibited spatial and temporal characteristics similar to controls . IHCs and SGNs are more depolarized in these mice , reducing the threshold for activation by other purinergic receptors . The role of P2RY1 in the developing CNS remains unexplored , but potential compensatory effects may be compounded in central circuits , due to loss of P2RY1 in astrocytes and neurons . Although such gain-of-function changes in the developing nervous system present challenges for interpretation of genetic manipulations , the preservation of early , patterned activity in children that carry deafness mutations may improve the outcome of later therapeutic interventions to restore hearing . In the adult inner ear , members of all purinergic receptors subtypes ( ionotropic P2X receptors , metabotropic P2Y , and adenosine P1 receptors ) are expressed by cells throughout the sensory epithelium , Reisner’s membrane , stria vascularis , and SGNs ( Housley et al . , 2009; Huang et al . , 2010 ) . The widespread expression of these receptors coupled with observations of increased endolymphatic ATP concentrations following trauma ( Muñoz et al . , 1995 ) have led to the hypothesis that these receptors serve a neuroprotective role . Indeed , infusion of ATP into the inner ear profoundly reduces sound-evoked compound action potentials in the auditory nerve ( Bobbin and Thompson , 1978; Muñoz et al . , 1995 ) , presumably due to decreased endolymphatic potential following shunting inhibition through P2RX2 ( Housley et al . , 2013 ) or excitotoxic damage ( Cisneros-Mejorado et al . , 2015 ) to IHCs or SGNs . Consistent with these observations , P2RX2 null mice and humans with a P2RX2 variant ( c . 178G > T ) experience progressive sensorineural hearing loss ( Yan et al . , 2013 ) . Ca2+ imaging and recordings from adult cochleae have also revealed robust responses to UTP in the inner sulcus , pillar cells , and Deiters' cells ( Sirko et al . , 2019; Zhu and Zhao , 2010 ) , suggesting that metabotropic purinergic receptors continue to be expressed . Following traumatic noise damage , ATP release could activate K+ buffering mechanisms in supporting cells , enhance K+ redistribution , reduce IHC depolarization and prevent excitotoxic damage . Purinergic receptors may also contribute to IHC gain control by influencing their membrane potential , as ATP circulates in the endolypmph at low nanomolar concentrations ( Muñoz et al . , 1995 ) . Further studies involving conditional deletion of P2ry1 from ISCs in the adult cochlea may help to define the role of this receptor in both normal hearing and injury contexts . For inner supporting cell recordings , apical segments of the cochlea were acutely isolated from P6-P8 rat ( Figure 1 ) and mouse pups ( all other figures ) and used within 2 hr of the dissection . Cochleae were moved into a recording chamber and continuously superfused with bicarbonate-buffered artificial cerebrospinal fluid ( 1 . 5–2 mL/min ) consisting of the following ( in mM ) : 119 NaCl , 2 . 5 KCl , 1 . 3 MgCl2 , 1 . 3 CaCl2 , 1 NaH2PO4 , 26 . 2 NaHCO3 , 11 D-glucose and saturated with 95% O2/5% CO2 to maintain a pH of 7 . 4 . Solutions were superfused at either room temperature or near physiological temperature ( 32–34°C ) using a feedback-controlled in-line heater ( Warner Instruments ) , as indicated in figure legends . Whole-cell recordings of inner supporting cells ( ISCs ) were made under visual control using differential interference contrast microscopy ( DIC ) . Electrodes had tip resistances between 3 . 5–4 . 5 MΩ when filled with internal consisting of ( in mM ) : 134 KCH3SO3 , 20 HEPES , 10 EGTA , 1 MgCl2 , 0 . 2 Na-GTP , pH 7 . 3 . Spontaneous currents were recorded with ISCs held at −80 mV . For inner hair cell recordings , apical segments of the cochlea were acutely isolated from P6-P8 mouse pups and used within 2 hr of the dissection . Cochleae were moved into a recording chamber and continuously superfused with bicarbonate-buffered artificial cerebrospinal fluid ( 1 . 5–2 mL/min ) consisting of the following ( in mM ) : 115 NaCl , 6 KCl , 1 . 3 MgCl2 , 1 . 3 CaCl2 , 1 NaH2PO4 , 26 . 2 NaHCO3 , 11 D-glucose . Solutions were saturated with 95% O2/5% CO2 to maintain a pH of 7 . 4 . Solutions were superfused at room temperature . Electrodes had tip resistances between 4 . 5–6 MΩ when filled with internal consisting of ( in mM ) : 134 KCH3SO3 , 20 HEPES , 10 EGTA , 1 MgCl2 , 0 . 2 Na-GTP , pH 7 . 3 . For hair cell recordings with K+ channels inhibited with cesium and TEA , the internal solution consisted of ( in mM ) : 100 cesium methanesulfonate , 20 TEA-Cl , 10 EGTA in CsOH , 20 HEPES , 1 MgCl2 , 0 . 2 Na-GTP , pH 7 . 3 with CsOH . Spontaneous currents were recorded with IHCs held at near their resting membrane potential ( –75 to –80 mV ) . Errors due to the voltage drop across the series resistance and the liquid junction potential were left uncompensated for recordings of spontaneous activity . For IHC recordings with K+ accumulation voltage protocols ( Figure 4 ) , the amplifier compensation circuit was used to compensate 70% of the access resistance . Recordings that displayed more than a 10% increase in access resistance or access resistances > 30 MΩ were discarded . ISC and IHC spontaneous currents were recorded with pClamp 10 software using a Multiclamp 700B amplifier , low pass filtered at 2 kHz , and digitized at 5 kHz with a Digidata 1322A analog-to-digital converter ( Axon Instruments ) . For SGN juxtacellular recordings , cochleae were dissected and cultured for 2 days ( see Cochlear Explant Culture section below ) . Cochleae were then transferred to a recording chamber and continuously superfused with bicarbonate-buffered aCSF ( same as ISCs ) at 1 . 5–2 mL/min . Recordings were performed at room temperature . Electrodes for SGN recordings had tip resistances between 1 . 5–2 . 5 MΩ when filled with artificial cerebrospinal fluid . Extracellular potentials were recorded for 10 min with pClamp10 software using a Multiclamp 700B amplifier , low pass filtered at 20 kHz , and digitized at 50 kHz with a Digidata 1322A analog-to-digital converter ( Axon Instruments ) . For MRS2500 experiments , spikes were analyzed in five-minute windows; five minutes of baseline preceded ten minutes of superfusion of aCSF containing 1 μM MRS2500 . Firing behavior in the latter five minutes of MRS2500 was used for measurement . Action potentials were analyzed offline using custom routines written in Matlab 2017b ( Mathworks ) . Briefly , raw traces were high-pass filtered to remove baseline drift and spikes were identified using an amplitude threshold criterion . As described previously ( Tritsch and Bergles , 2010 ) , bursts were identified by classifying interspike intervals into non-bursting intervals ( >1 s ) , burst intervals ( 30 ms - 1 s ) , and mini-burst intervals ( <30 ms ) . Bursts were defined as clusters of at least 10 consecutive burst intervals ( with mini-burst intervals being ignored in the context of burst detection ) . Spikes within mini-bursts were included when calculating the number of spikes within a burst . Colored raster plots were generated by grouping spikes into one-second bins and applying a color map to the resulting data ( modified ‘hot’ colormap; Matlab ) . Cochleae were dissected from postnatal day 5–6 control ( P2ry1+/+ or Pax2-Cre;R26-lsl-GCaMP3 ) and P2ry1 KO ( P2ry1–/–or Pax2-Cre;R26-lsl-GCaMP3;P2ry1–/– ) mice in ice-cold , sterile-filtered HEPES-buffered artificial cerebrospinal fluid ( aCSF ) consisting of the following ( in mM ) : 130 NaCl , 2 . 5 KCl , 10 HEPES , 1 NaH2PO4 , 1 . 3 MgCl2 , 2 . 5 CaCl2 , and 11 D-Glucose . Explants were mounted onto Cell-Tak ( Corning ) treated coverslips and incubated at 37° C for 24–48 hr in Dulbecco’s modified Eagle’s medium ( F-12/DMEM; Invitrogen ) supplemented with 1% fetal bovine serum ( FBS ) and 10 U/mL penicillin ( Sigma ) prior to recording or imaging . Cochlear segments were imaged with a Olympus 40x water immersion objective ( LUMPlanFl/IR ) and recorded using MATLAB and a USB capture card ( EZ Cap ) . Difference movies were generated by subtracting frames at time tn and tn+5 seconds using ImageJ software to generate an index of transmittance change over time . To quantify transmittance changes , a threshold of three standard deviations above the mean was applied to the values . To calculate the frequency of these events , the whole field was taken as an ROI and peaks were detected using MATLAB ( findpeaks function ) . To calculate area of these events , a Gaussian filter ( sigma = 2 . 0 ) was applied to the image after thresholding and the borders detected using MATLAB ( bwlabel function ) . The area was then calculated as the number of pixels within the border multiplied by the area scaling factor ( µm/pixel ) 2 measured with a stage micrometer . Mice were deeply anesthetized with isoflurane and perfused with freshly prepared paraformaldehyde ( 4% ) in 0 . 1 M phosphate buffer . Cochleae were post-fixed for 45 min at room temperature and stored at 4°C until processing . For X-gal reactions , P6-P8 cochleae were removed from the temporal bone and washed 3 × 5 min with PBS . Tissue was then incubated for 24 hr in the dark at 37°C in X-gal working solution consisting of ( in mM ) : five potassium ferricyanide crystalline , five potassium ferricyanide trihydrate , two magnesium chloride , and 0 . 1% X-gal ( GoldBio ) dissolved in DMSO . After washing 3 × 5 min with PBS , images of cochleae were acquired on a dissecting microscope ( Zeiss Stemi 305 ) . For immunohistochemistry , fixed tissue was washed 3 × 5 min in PBS , placed in 30% sucrose solution overnight , and incubated in OCT mounting medium overnight at 4°C . Ten micron thick cross-sections of the cochlea were made on a cryostat and mounted on Superfrost Plus slides ( Fisher ) , which were then allowed to dry for 1 hr before processing . Cross-sections were incubated overnight with primary antibodies against β-gal ( anti-Chicken; 1:4000 , Aves ) and Myosin-VIIa ( anti-Rabbit; 1:500 , Proteus BioSciences ) for detection of β-gal and with Myosin-VIIa only for qualitative analysis of the Tecta-Cre;TdT reporter mouse line . Sections were then rinsed three times with PBS and incubated for two hours at room temperature with secondary antibodies raised in donkey ( Alexa-488 and Alexa-546; 1:2000 , Life Technologies ) . Slides were washed three times in PBS ( second with PBS + 1:10 , 000 DAPI ) , allowed to dry , and sealed using Aqua Polymount ( Polysciences , Inc ) . Images were captured using a laser scanning confocal microscope ( LSM 510 or 880 , Zeiss ) . After one day in vitro , cochleae were moved into a recording chamber and continuously superfused with bicarbonate-buffered artificial cerebrospinal fluid ( 1 . 5–2 mL/min ) consisting of the following ( in mM ) : 119 NaCl , 2 . 5 KCl , 1 . 3 MgCl2 , 1 . 3 CaCl2 , 1 NaH2PO4 , 26 . 2 NaHCO3 , 11 D-glucose , and saturated with 95% O2/5% CO2 to maintain a pH of 7 . 4 . A piezo-mounted objective was used to rapidly alternate between SGN cell bodies and ISCs/IHCs . Images were captured at one frame per second using a Zeiss laser scanning confocal microscope ( LSM 710 , Zeiss ) through a 20X objective ( Plan APOCHROMAT 20x/1 . 0 NA ) at 512 × 512 pixel ( 354 × 354 µm; 16-bit depth ) resolution . Sections were illuminated with a 488 nm laser ( maximum 25 mW power ) . MRS2500 ( 1 µM , Tocris ) was applied by addition to the superfusing ACSF . Images were imported into ImageJ and image registration ( MultiStackReg ) was used to correct for drifts in the imaging field . Since images were obtained at two different z-planes , images were combined into one stack for analysis . This was done by eliminating the empty bottom half of the imaging field containing ISCs and IHCs and the empty top half of the field containing SGN cell bodies and merging the two images . For analysis of coordinated activity throughout the cochlea , regions of interest were drawn around the entirety of ISCs , IHCs , and SGNs . Fluorescence changes were normalized as ΔF/Fo values , where ΔF = F – Fo and Fo was defined as the fifth percentile value for each pixel . Peaks in the signals were detected in MATLAB using the built-in peak detection function ( findpeaks ) with a fixed value threshold criterion ( mean + three standard deviations for each cell ) . To quantify frequency and areas of Ca2+ transients , a threshold of three standard deviations above the mean was applied to each pixel within the ROI . To calculate the frequency of these events , the whole field was taken as an ROI and peaks were detected using MATLAB ( findpeaks function ) on the number of thresholded pixels per frame . To calculate area of these events , a Gaussian filter ( sigma = 2 . 0 ) was applied to the image after thresholding and the borders detected using MATLAB ( bwlabel function ) . The area was then calculated as the number of pixels within the border multiplied by an area scaling factor ( 1 µm/pixel ) 2 measured with a stage micrometer . For correlation analysis , ROIs were drawn around every IHCs in the field of view . Pairwise correlation coefficients were performed between every hair cell pair and represented as correlation matrices . Inhalation anesthesia was induced with vaporized isoflurane ( 4% for 5 min , or until mice are non-responsive to toe-pinch ) and surgical plane maintained during the procedure ( with 1–2% isoflurane ) with a stable respiration rate of 80 breaths per minute . A midline incision beginning posterior to the ears and ending just anterior to the eyes was made . Two subsequent cuts were made to remove the dorsal surface of the scalp . A headbar was secured to the head using super glue ( Krazy Glue ) . Fascia and neck muscles overlying the interparietal bone were resected and the area bathed in sterile , HEPES-buffered artificial cerebrospinal fluid that was replaced as necessary throughout the surgery . Using a 28G needle and microblade , the sutures circumscribing the interparietal bone were cut and removed to expose the midbrain . The dura mater was removed using fine scissors and forceps , exposing the colliculi and extensive vasculature . A 5 mm coverslip ( CS-5R; Warner Instruments ) was then placed over the craniotomy , the surrounding bone was dried using a Kimwipe , and super glue was placed along the outer edges of the coverslip for adhesion to the skull . Replacement 0 . 9% NaCl solution was injected IP and a local injection of lidocaine was given to the back of the neck . Animals were weaned off isoflurane , placed under a warming lamp , and allowed to recover for a minimum of 1 hr prior to imaging . Spontaneous activity was not seen in deeply anesthetized animals and emerged ~30 min after recovery from isoflurane exposure , as reported previously ( Ackman et al . , 2012 ) . After 1 hr of post-surgical recovery from anesthesia , pups were moved into a swaddling 15 mL conical centrifuge tube . The top half of this tube was removed to allow access to the headbar and visualization of the midbrain . Pups were head-fixed and maintained at 37°C using a heating pad and temperature controller ( TC-1000; CWE ) . During the experiments , pups were generally immobile; however , occasional limb and tail twitching did occur . For wide field epifluorescence imaging , images were captured at 10 Hz using a Hamamatsu ORCA-Flash4 . 0 LT digital CMOS camera attached to a Zeiss Axio Zoom . V16 stereo zoom microscope . A 4 × 4 mm field of view was illuminated continuously with a metal halide lamp ( Zeiss Illuminator HXP 200C ) and visualized through a 1X PlanNeoFluar Z 1 . 0x objective at 17x zoom . Images were captured at a resolution of 512 × 512 pixels ( 16-bit pixel depth ) after 2 × 2 binning to increase sensitivity . Each recording consisted of uninterrupted acquisition over 10 min or 20 min if injected with pharmacological agents . After induction of anesthesia and before installing the cranial window , a catheter was placed in the intraperitoneal ( IP ) space of neonatal mouse pups . A 24G needle was used to puncture the peritoneum and a small-diameter catheter ( SAI Infusion Technologies , MIT-01 ) was placed . A drop of Vetbond secured the catheter to the pup’s belly . Installation of cranial window proceeded as described above . Imaging sessions consisted of 5 min of baseline activity measurements , followed by a slow push of either 50 µL of sham ( 5% mannitol solution ) or MRS2500 solution ( 500 µM in 5% mannitol solution ) . Imaging was continuous throughout and 20 min of activity total were collected . No discernable diminishment of activity was observed in sham animals . For wide field imaging , raw images were imported into the ImageJ environment and corrected for photobleaching by fitting a single exponential to the fluorescence decay and subtracting this component from the signal ( Bleach Correct function , exponential fit ) . Images were then imported into MATLAB ( Mathworks ) and intensities were normalized as ΔF/Fo values , where ΔF = F – Fo and Fo was defined as the fifth percentile value for each pixel . Ovoid regions of interest ( ROIs ) encompassing the entire left and right inferior colliculi were drawn . Across all conditions , the size of the ROIs was invariant , however , due to small differences in the imaging field between animals , the ROIs were placed manually for each imaging session . Peaks in the signals were detected in MATLAB using the built-in peak detection function ( findpeaks ) using a fixed value threshold criterion; because fluorescence values were normalized , this threshold was fixed across conditions ( 2% ΔF/Fo ) . Occasionally , large events in the cortex or superior colliculus would result in detectable fluorescence increases in the IC . These events broadly activated the entire surface of the IC and did not exhibit the same spatially-confined characteristics as events driven by the periphery . These events were not included in the analysis . As shown in Figure 6D , a rectangle of size 125 × 50 pixels was placed perpendicular to the tonotopic axis of the IC ( ±55° rotation , respectively ) . The columns of the resulting matrix were averaged together to create a line scan ( 125 pixels x one pixel ) for the entire time series . Peaks were detected using MATLAB’s imregionalmax function with a constant threshold of 3% ΔF/Fo across all animals . Histograms of events along the tonotopic axis were generated by summing the number of events in ~25 µm bins . Lateral and medial designations were assigned by splitting the area evenly between the lateral edge and the location of defined single-band events in the medial portion of the IC . Events detected on the medial edge of single-band events , reflective of the bifurcation of this information was not included in the medial/lateral analysis . ROIs ( 200 × 150 pixels ) were placed over each lobe of the superior colliculus and downsampled by a factor of five . Signals were normalized as ΔF/Fo values , where ΔF = F – Fo and Fo was defined as the fifth percentile value for each pixel . In order to eliminate periodic whole-sample increases in fluorescence , the mean intensity of all pixels was subtracted from each individual pixel . Following this , pixels were considered active if they exceeded the mean + three standard deviations . For each point in time , the number of active pixels was summed . Retinal waves were defined as prolonged periods ( >1 s ) , where more than five pixels were active simultaneously . Retinal wave durations were defined as the total continuous amount of time that more than five pixels were active . Frequencies and durations are similar to earlier reports ( Ackman et al . , 2012 ) . A crRNA ( TAATGATGAATAATTCATCC ) targeted near exon 2 of the Tecta gene , tracrRNA , Cas9 recombinase , and a donor plasmid containing an iCre-WPRE-polyA sequence ( 500 base pair homology arms ) were injected into single-cell embryos that were then transferred to pseudopregnant recipient mothers . After birth , mouse pups were screened for insertion of the gene at the correct locus with two pairs of primers: one pair amplified DNA beginning 5’ of the 5’ homology arm and ending within the Cre sequence and the other amplified DNA within the polyA sequence and ending 3’ of the 3’ homology arm . These primers were then used to sequence the junctions . Of these , all mice used for experiments were derived from a single founder that was positive for both sets of primers and had 100% sequence validation . Mice were crossed to a TdTomato reporter line to examine cell-specific recombination . All statistics were performed in the MATLAB ( Mathworks ) programming environment . All statistical details , including the exact value of n , what n represents , and which statistical test was performed , can be found in the figure legends . To achieve statistical power of 0 . 8 with a 30% effect size with means and standard deviations similar to those observed in previous studies ( Figure 1E of Tritsch et al . , 2007 and Figures 1B and 3D in Wang and Bergles , 2015 ) , power calculations indicated that seven animals in each condition are necessary ( µ1 = 10 , µ2 = 7 , σ = 2 , sampling ratio = 1 ) . While this number was used as a guide , power calculations were not explicitly performed before each experiment; many experiments had much larger effect sizes and sample sizes were adjusted accordingly . For transparency , all individual measurements are included in the figures . Unless otherwise noted , data are presented as mean ± standard error of the mean ( SEM ) . Because the main comparison between conditions was the mean , the SEM is displayed to highlight the dispersion of sample means around the population mean . All datasets were tested for Gaussian normality using the D’Agostino’s K2 test . For single comparisons , significance was defined as p<=0 . 05 . When multiple comparisons were made , the Bonferroni correction was used to adjust p-values accordingly to lower the probability of type I errors . For multiple condition datasets , one-way ANOVAs were used , followed by Tukey’s multiple comparison tests .	As the brain develops , billions of cells respond to genetic and environmental cues to form the trillions of connections that make up its neural networks . However , before these brain circuits can respond to real life stimuli , their connections are refined by bursts of electrical activity . For example , sensory cells in the ear produce bursts of spontaneous electrical activity that mimic those made by sounds . This activity allows the neural network in the hearing system to ‘practice’ responding to sounds . However , the origin of these electrical bursts is unusual as they do not start in the sensory cells themselves , but are initiated by the non-sensory cells around them . Past research has shown that as the ear develops these non-sensory cells , or supporting cells , release regular doses of a molecule called ATP . The supporting cells then detect their own ATP release using specialized receptor proteins on their surface . This self-stimulation causes the supporting cells to release potassium ions that interact with the sensory cells and trigger bursts of electrical activity . However , the identity of this ATP-detecting receptor was not known , and without this information it was unclear how the electrical activity starts and why it happens in rhythmic bursts . To fill this knowledge gap , Babola et al . measured electrical activity in ear cells isolated from mice , and examined nerve cell activity in live mice during this critical stage of development . This revealed that the bursts of activity in the ear depend on a receptor called P2RY1 which can be found on the supporting cells located next to sensory cells . When P2RY1 is activated it triggers the release of calcium ions inside the supporting cells . This opens channels in the cell membrane , allowing the potassium ions to flow out and electrically activate the sensory cells . But , when the potassium ions leave the supporting cells , water is drawn out with them , causing the cells to shrink and the space around the cells to get bigger . As a result , the released potassium ions disperse more quickly , moving away from the sensory cells and stopping the burst in electrical activity . Conversely , when P2RY1 is inhibited , this causes the supporting cells to swell , trapping potassium ions near the sensory cells and making them fire continuously . This indicates that bursts in electrical activity are controlled by the rhythmic swelling and shrinking of supporting cells . Although supporting cells cannot detect sound themselves , they seem to play a crucial role in developing the hearing system . A better understanding of these cells could therefore aid research into hearing problems without a known cause such as hypersensitivity to sound , tinnitus , and complex auditory processing disorders in children .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2020	Purinergic signaling in cochlear supporting cells reduces hair cell excitability by increasing the extracellular space
External and internal morphological characters of extant and fossil organisms are crucial to establishing their systematic position , ecological role and evolutionary trends . The lack of internal characters and soft-tissue preservation in many arthropod fossils , however , impedes comprehensive phylogenetic analyses and species descriptions according to taxonomic standards for Recent organisms . We found well-preserved three-dimensional anatomy in mineralized arthropods from Paleogene fissure fillings and demonstrate the value of these fossils by utilizing digitally reconstructed anatomical structure of a hister beetle . The new anatomical data facilitate a refinement of the species diagnosis and allowed us to reject a previous hypothesis of close phylogenetic relationship to an extant congeneric species . Our findings suggest that mineralized fossils , even those of macroscopically poor preservation , constitute a rich but yet largely unexploited source of anatomical data for fossil arthropods . An organism’s morphology represents a complex solution to myriad ecological and environmental challenges it and its ancestors have confronted over evolutionary time . Inferring a comprehensive evolutionary history of a lineage requires consideration of a wide range of morphological features , and how they may have been shaped by selection , drift , and developmental constraints . While external characters predominate in ecomorphological and systematic studies , internal characters also play critical roles ( Perreau and Tafforeau , 2011 ) . In fossil specimens , however , these characters are usually not preserved or difficult to access ( Siveter et al . , 2007 ) . While combined phylogenetic analyses of extant species frequently utilize internal anatomy , analyses including fossil taxa are generally limited to external characters . Moreover , it is often difficult to distinguish whether unobserved morphological characters were originally absent or lost due to taphonomic processes , potentially leading to misinterpretations of character evolution and erroneous phylogenetic placements ( Sansom , 2015 ) . Several types of preservation or certain combinations of them are known for arthropod fossils . These are adpressions ( compressions or impressions ) ( Wedmann et al . , 2007; 2011 ) , casts , voids , embeddings , mineral replications , charcoalified remains , or inclusions in amber ( Grimaldi et al . , 1994; Martı́nez-Delclòs et al . , 2004; Grimaldi and Engel , 2005; Dunlop and Garwood 2014; Penney and Jepson , 2014 ) . Amber inclusions are famous for exquisitely preserving three-dimensional external shape and sometimes internal characters ( Perreau and Tafforeau , 2011 ) . Three-dimensional arthropods may also be preserved within concretions ( e . g . in siderite nodules [Nitecki , 1979; Garwood et al . , 2009] ) , calcareous incrustations ( e . g . in travertine [Rosendahl et al . , 2013] ) , encapsulations in minerals ( e . g . in onyx-marble [Pierce , 1951] , chert [Anderson and Trewin , 2003] , or gypsum crystals [Schlüter et al . , 2003] ) , and mineral replications ( e . g . as calcite [McCobb et al . , 1998] , silica [Miller and Lubkin , 2001] , goethite [Grimaldi , 2009; Barling et al . , 2014] , pyrite [Grimaldi and Engel , 2005] , or phosphate [Duncan and Briggs , 1996; Hellmund and Hellmund , 1996; Waloszek , 2003] ) . Some of these preservation types have revealed surprisingly detailed insights into the internal and soft tissue anatomy of several arthropods , for instance from several Paleozoic marine deposits ( e . g . Siveter et al . , 2007; 2013; 2014; Ma et al . , 2014; Cong et al . , 2014; Edgecombe et al . , 2015 ) . For insects , e . g . eyes ( Duncan and Briggs , 1996 ) and muscle fibers ( Grimaldi , 2009 ) have been reported . Abundant arthropod fossils preserved by mineralization of calcium phosphate are known from the Oligocene fissure fillings of Ronheim ( Hellmund and Hellmund , 1996 ) , the Late Oligocene/Early Miocene limestones of Riversleigh ( QLD , Australia ) ( Duncan and Briggs , 1996 ) and from Paleogene deposits at Quercy ( south-central France ) ( Filhol , 1877; Gervais , 1877; Flach , 1890; Thévenin , 1903; Handschin , 1944 ) . These localities have long been famous for their rich vertebrate fossils as well ( e . g . Legendre et al . , 1997; Laloy et al . , 2013 ) . The arthropod fossils of Quercy were documented by Swiss entomologist Eduard Handschin ( 1944 ) . He described the hister beetle Onthophilus intermedius ( Coleoptera: Histeridae ) from eight specimens , and considered it distinct but closely related to the extant European species O . striatus ( Forster , 1771 ) . The description , however , was vague and based mainly on the external morphology of the two best-preserved specimens ( Handschin , 1944 ) . X-ray microtomography has become established for the detailed examination of both extant ( e . g . Betz et al . , 2007; Bosselaers et al . , 2010; van de Kamp et al . , 2011; 2014; 2015; Brehm et al . , 2015; Sombke et al . , 2015 ) and extinct ( Sutton , 2008; Sutton et al . , 2014 ) arthropods , including fossils preserved in amber ( Lak et al . , 2009; Pohl et al . , 2010; Soriano et al . , 2010; Perreau and Tafforeau , 2011; Riedel et al . , 2012 ) . We explored the application of this technique to mineralized fossils by re-examination of Handschin's specimens of Onthophilus intermedius . To ensure a direct morphological comparison , we performed tomographic scans ( Figure 1 ) of ethanol-fixed and air-dried O . striatus using the same experimental setup . Furthermore we tested the hypothesis that the two are closely related with a global phylogenetic analysis of Onthophilus Leach , 1817 . 10 . 7554/eLife . 12129 . 003Figure 1 . Comparison between the fossil Onthophilus intermedius ( A , D , G ) and EtOH-fixed ( B , E , H ) and air-dried ( C , F , I ) specimens of O . striatus . Slices of tomographic volumes showing head region ( A–C ) , thorax ( D–F ) and abdomen ( G–I ) . ae = aedeagus; ag = accessory gland; bpae = basal part of aedeagus; hg = hindgut; m = musculature; ml = median lobe; mr = muscles remnants; mscx = mesocoxa; msf = mesofemur; mst = mesotibia; mt = muscle tissue; mtcx = metacoxa; mtf = metafemur; mtt = metatibia; pcx = procoxa; sph = spherical particle; sm = stony matrix; t8 = 8th abdominal tergite; t9 = 9th abdominal tergite; t10 = 10th abdominal tergite; te = tentorium; tr = trachea . DOI: http://dx . doi . org/10 . 7554/eLife . 12129 . 003 We found internal characters in all fossils ( Table 1 ) . Three specimens show remains of inner organs , especially of the sclerotized genitalia , allowing their identification as two males and one female . The outer surfaces of most specimens appear smooth ( Figure 2 ) ; the distinct punctuation found in extant Onthophilus species ( Kovarik and Caterino , 2005 ) is faint . 10 . 7554/eLife . 12129 . 004Table 1 . Notes on the fossil Onthophilus intermedius specimens from Quercy and their preservation . DOI: http://dx . doi . org/10 . 7554/eLife . 12129 . 004IDInternal structures preservedNotesF1951some sclerites ( incl . coxa-trochanteral joints ) and tracheaethe only specimen depicted by Handschin ( 1944 ) ; but not explicitly designated as holotypeF1992some sclerites and small tracheaehead , prothorax missingF1993some sclerites ( incl . coxa-trochanteral joints ) head , pygidia missing; elytra partly abradedF1994most sclerites , muscle parts , tracheae , parts of alimentary system , large parts of male genitalsthe only specimen of the collection that is ventrally encrusted by a stone matrixF1995some sclerites , parts of male genitalshead present; abdomen deeply abraded dorsallyF1996some scleriteshead , prothorax missingF1997some sclerites , remains of muscles below the elytrahead , prothorax partly abradedF1998some sclerites ( incl . coxa-trochanteral joints ) , parts of female genitaliahead , prothorax partly abraded10 . 7554/eLife . 12129 . 005Figure 2 . Surface renderings of the eight Onthophilus intermedius specimens . Note the unique encrustation of F1994 . DOI: http://dx . doi . org/10 . 7554/eLife . 12129 . 005 The specimen F1994 ( Figures 1A , D , G , 2 , 3 , Supplementary file 1 ) differs from all other samples by the presence of a stony matrix , covering the ventral part of the beetle . Its dorsal part and head are exposed; the elytra are missing and were probably detached before embedding . The exposed surface is partly eroded , especially in the anterior region of the head , and no appendages are visible from the outside . The matrix , however , concealed the best-preserved fossil from the collection , which we examine here in detail . 10 . 7554/eLife . 12129 . 006Figure 3 . Digital reconstruction of the fossil . ( A ) Photograph of Onthophilus intermedius ( F1994 ) ventrally embedded in a stony matrix . ( B ) Digital reconstruction showing fossilized beetle ( green ) and matrix ( brown ) . ( C ) Beetle digitally isolated from the stone , revealing well-preserved morphology hidden by the matrix . ( D ) Perspective view of the fossil showing parts of exoskeleton , tracheal network , alimentary canal and genitals . ( E , F ) Comparison of the male genitals of the extant O . striatus ( E ) and the fossil O . intermedius ( F ) ; outer sclerites cut to reveal internal anatomy . See Supplementary file 1 for an interactive version of the 3D reconstruction . DOI: http://dx . doi . org/10 . 7554/eLife . 12129 . 006 The ventral portion of the beetle covered by the matrix reveals an extraordinary preservation of exoskeletal fine structures and internal anatomy ( Figures 3 , 4 and 5; Supplementary file 1 ) . While some fractions of the cuticle appear to be mineralized , the latter is mostly represented by air-filled spaces in the fossil ( Figure 1A , D , G ) . The surface of the exoskeleton is preserved as a three-dimensional imprint of remarkable detail; the body sclerites show characteristic punctuation of the genus . The right foreleg is not preserved; the left one is truncated from the trochanter; distal parts of the leg were lost prior to fossilization . The right mid and hind legs are eroded at the edge of the matrix , but their encrusted left counterparts appear complete except for the most distal part of the metafemur of the hind leg that would protrude from the matrix . Moreover , many anatomical characters can be recognized inside the fossil ( Figure 3D ) . Apart from internal invaginations of the exoskeleton ( e . g . tentorium , furcal arms and metendosternite ) , large parts of the alimentary canal and tracheal system are visible . The oesophagus appears to be shrunken and is connected to the crop , which is truncated posteriorly . The anterior part of the hindgut is hollow , while the middle part is apparently filled with mineral matrix but well-defined . Conspicuous spherical particles may constitute remnants of gut content ( Figure 1G ) . The hindmost part of the gut can be roughly retraced by aggregations of tiny holes inside the mineral matrix . Like in the alimentary canal , some large tracheae appear to be filled with matrix , while others are hollow . Except for the musculature connecting the right pro- and mesofurcal arms ( Figure 1D ) , most muscles can only be recognized by remnants at the insertion areas ( Figure 1G ) . The genitals are extraordinary well-preserved ( Figure 3F ) . While testes and Ductus ejaculatorius could not be recognized , other soft tissues such as the spiral accessory glands and parts of the gland ducts are conspicuous . The genital sclerites , including aedeagus , median lobe , gonopore , tergites 8-10 and sternites 8 & 9 are almost perfectly preserved as imprints . 10 . 7554/eLife . 12129 . 007Figure 4 . Coxa-trochanteral joints . Comparison of the joints ( cut ) of the left mid- ( A , B ) and hind leg ( C , D ) of Onthophilus striatus ( A , C ) and O . intermedius ( B , D ) , showing coxae ( green ) and trochanters ( yellow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12129 . 00710 . 7554/eLife . 12129 . 008Figure 5 . Digital endocast of Onthophilus intermedius ( specimen F1994 ) . A digital endocast ( A , B ) artificially created from tomography data resembles the shape of the other fossils ( Figure 2 ) much closer than the original surface of the beetle ( C , D ) hidden by the stony matrix . DOI: http://dx . doi . org/10 . 7554/eLife . 12129 . 008 The remarkable preservation state of the fossil is emphasized when its morphological characters are compared to those of an extant ethanol-fixed specimen of the same genus ( Figures 1 , 3E , F and 4 ) . The new anatomical data from this specimen facilitated an extended description of the species according to modern taxonomic standards ( Appendix 1 ) . Handschin ( 1944 ) hypothesized a close relationship ( ‘particularly striking similarity’ ) between Onthophilus intermedius and O . striatus based on then-observable external morphology . However , phylogenetic analysis ( Material and methods ) of the more diverse character set now accessible places these species in distinct clades . The analysis resulted in 72 most parsimonious trees of length 185 ( CI 0 . 27 , RI 0 . 61 ) . The strict consensus of these trees ( Figure 6 ) is well resolved apart from a few rearrangements of some outgroup taxa and within a relatively derived group related to O . niponensis Lewis , 1907 . O . intermedius is part of a trichotomy involving O . silvae Lewis , 1884 and a large group of species descended from the common ancestor of O . giganteus Heleva , 1978 and O . niponensis , though in reweighted trees it is resolved as sister to O . silvae alone . In all analyses O . striatus is nested within a lineage of Nearctic and far-eastern Palaearctic species , including O . flavicornis Lewis , 1884 , O . flohri Lewis , 1888 and others . 10 . 7554/eLife . 12129 . 009Figure 6 . Strict consensus tree . The analysis places Onthophilus striatus within a lineage of Nearctic and far-eastern Palaearctic species ( red ) , while O . intermedius is a member of a separate Holarctic lineage ( blue ) . Four internal ( purple ) and three external ( orange ) unambiguous synapomorphies supporting their respective placements are mapped onto the cladogram - Onthophilus striatus group: Character 22:2 , mesoventrite wide and short; 30:1 , pygidial median carina absent; 35:2 , tegmen of aedeagus abruptly downturned apically . O . intermedius group: 29:2 , pygidium laterally impunctate; 36:2 , tegmen of aedeagus abruptly narrowing apically; 40:2 , lateral halves of eighth sternite large and nearly meeting at midline; 41:2 , stem of spiculum gastrale broad throughout its length . DOI: http://dx . doi . org/10 . 7554/eLife . 12129 . 009 Inclusion of diverse characters revealed by microtomography of Onthophilus intermedius yields a well-supported topology and a more comprehensive picture of the biogeographic and morphological history of the group . Of the characters scored for both O . intermedius and O . striatus , there are seven by which their states differ , three external and four internal . Of these , two external ( chars . 29 & 30 ) and one internal ( char . 36 ) are reconstructed as autapomorphies ( Figure 6 ) . Only one external synapomorphy ( char . 22 ) separates them , while three of the four genitalic differences ( chars . 35 , 40 , and 41 ) represent synapomorphies of their respective lineages . Exclusion of internal characters for O . intermedius did not affect the topology , but did prevent genitalic characters from supporting its larger containing clade . Critical diagnostic differences in external morphology , such as mesoventral proportions and pygidial sculpturing , were also revealed by visualization of features previously obscured by matrix . Based on our examinations we can reconstruct the probable fossilization process of the Quercy Onthophilus specimens , which culminates in a partial mineralization of inner organs in combination with the cuticle preserved as voids . An accurate three-dimensional conservation of soft tissues does not occur if the specimens are dried in air ( Figure 1C , F , I ) . Therefore , the fixation process must have occurred fast , possibly due to the animal being immediately penetrated and enclosed by phosphate rich water . In arthropods , this type of fossilization is only known from a handful of localities , which are better known for a rich vertebrate fauna ( Riversleigh: Duncan and Briggs , 1996; Ronheim: Hellmund and Hellmund , 1996; Quercy: Handschin , 1944 ) . Replication of soft tissues by phosphatization may be accomplished over a period of weeks ( Martı́nez-Delclòs et al . , 2004 ) . Possible sources for high phosphorous concentrations in water circulating through the fissure fill are rocks or abundant phosphate-rich vertebrate bones , which may have been deposited along with them ( Handschin , 1944; Hellmund and Hellmund , 1996 ) . After encrustation and internal mineralization , the cuticle largely decayed , leaving air-filled spaces . Erosion processes probably removed the outer stony matrix of most specimens , including fragile appendages and the imprint of the outer surface of the exoskeleton , leaving a mineralized endocast . Thus , the exterior of the fossils merely represents the inner surface of the exoskeleton – the deep grooves ( Figure 2 ) actually being inner folds or apophyses . While the smooth dorsal part of F1994 resembles the other fossils in appearance , its ventral surface covered by the mineral matrix shows a distinct surface sculpturing as present in extant species of the genus . In contrast , an artificial ‘digital endocast’ created from the tomographic data of F1994 ( Material and methods ) bears a striking resemblance to the other fossils ( Figure 5 ) , on which Handschin based his original description . Summing up , the Quercy hister beetles represent three-dimensional ‘hybrid’ fossils , comprising cuticle imprints and mineralized soft tissue , combining to preserve both exoskeletal fine structure and internal anatomical characters . Fissure filling fossils preserving three-dimensional internal anatomy will help to overcome taphonomic biases in available fossil data ( Allison and Bottjer , 2011 ) . To date , fossilized insect internal character information has mainly been obtained from well-preserved amber inclusions ( e . g . Pohl , et al . , 2010; Perreau and Tafforeau , 2011 ) . However , the origination of amber as tree resin causes a representational bias toward generally arboreal taxa ( Martı́nez-Delclòs et al . , 2004 ) . The fossil arthropods of Quercy represent an assemblage of taxa more typically associated with forest floor communities ( Handschin , 1944 ) , as exemplified by Onthophilus , typically a predator in various decaying organic materials ( Kovarik and Caterino , 2005; Bajerlein et al . , 2011 ) . Such communities are less commonly preserved than those of many other environments ( Kidwell and Flessa , 1996 ) . Beyond anatomical data on these species , clearer interpretations of evolutionary relationships of these fossils will improve inferences about the evolution of these ecological communities . Thus , reexamination of the Quercy fossils , and likely also of similar mineralized fossils from other localities ( which may represent different ecosystems and/or time periods ) , may provide a highly complementary source of information on the evolutionary history of arthropods . With regard to the methods employed here , we can offer some guidance on improving future imaging attempts on similar materials . Based on our experience , a fast tomography setup combining filtered polychromatic radiation and an optimized detector system ( dos Santos Rolo et al . , 2014 ) is well-suited to achieve sufficient image quality in most fossil specimens . Thus , scan duration per tomogram may be reduced to a couple of seconds ( Material and methods ) , facilitating high-throughput screening of large sample numbers in short time . Our results demonstrate that mineralized arthropod fossils from a sedimentary context may three-dimensionally preserve soft tissue and other internal anatomical characters in remarkable detail , which allows determinations and phylogenetic analyses according to the standards for Recent organisms . Reevaluation of relationships with modern taxa in this extended morphological context will substantially improve estimates of rates and modes of arthropod evolution . This exceptionally detailed preservation may be aided by the presence of a surrounding stony matrix , hinting that encrusted specimens , which therefore were originally considered to be of poor quality , could contain particularly well-preserved external and internal characters . Our findings may trigger the reinvestigation of numerous similar fossils from various localities . 3D X-ray micro-computed tomography scans with synchrotron radiation ( µCT ) were performed at the TOPO-TOMO beamline ( Rack et al . , 2009 ) of the ANKA Synchrotron Radiation Facility at Karlsruhe Institute of Technology ( KIT ) . The measurements consisted of the acquisition of 2500 equiangularly spaced radiographic projections of the sample in a range of 180° . The frame rate was set to 150 images per second , resulting in an overall scan duration of 16 . 67 seconds per sample . The parallel polychromatic X-ray beam produced by a 1 . 5 T bending magnet was spectrally filtered by 0 . 2 mm aluminum to obtain a peak at about 15 keV . The sample was placed 20 cm upstream of the detector , which in turn was located about 33 m from the source . The detector consists of a thin , plan-parallel lutetium aluminum garnet single crystal scintillator doped with cerium ( LuAG:Ce ) , optically coupled via a Nikon Nikkor 85/1 . 4 photo-lens to a pco . dimax camera with a pixel matrix of 2008x2008 pixels . The lens was stopped down to F/4 to remove optical aberrations and to increase its depth of focus , permitting the use of a thicker scintillator to collect a higher fraction of the incident X-ray photons . The magnification of the optical system was adjusted to 3X , yielding an effective X-ray pixel size of 3 . 66 µm ( dos Santos Rolo et al . , 2014 ) . Tomographic reconstruction was performed with the GPU-accelerated filtered back projection algorithm implemented in the software framework UFO ( Vogelgesang et al . , 2012 ) . Microtomographic image data are deposited in Morph·D·Base ( www . morphdbase . de; accession numbers T_vandeKamp_20151216-M-12 . 1 to T_vandeKamp_20151216-M-22 . 1 ) . 3D reconstruction followed the protocol described by Ruthensteiner and Heß ( 2008 ) and van de Kamp et al . , ( 2014 ) ; using Amira ( versions 5 . 5 , 6 , FEI ) and Avizo ( version 8 . 1 , FEI ) for segmentation of the tomographic volumes and CINEMA 4D R15 ( Maxon Computer GmbH ) for assembly of components and rendering of figures . The ‘digital endocast’ ( Figure 5 ) was created from the tomographic stack of specimen F1994 by segmenting solely the dorsal stony matrix , ventrally confined by the inner impression of the beetle’s cuticle . The number of surface polygons was reduced to 10% of its original value in CINEMA 4D: the raw mesh of F1994 contains approx . 30 million polygons , the reduced version ( Figure 3D ) ca . 3 million . Segmentation artifacts were carefully removed using the sculpting tools of the software . For the interactive 3D model ( Supplementary file 1 ) , the polygon count was further reduced to 800 , 000 ( without the stony matrix ) ; the digital mesh was imported into Deep Exploration ( version 6; Right Hemisphere ) , saved as Universal 3D file ( U3D ) and embedded into a PDF document with Adobe® Acrobat® 9 Pro Extended . Our phylogenetic analysis was performed to test Handschin’s ( 1944 ) hypothesis of a close relationship of Onthophilus intermedius to the extant and sympatric O . striatus . Although his hypothesis was not presented in strictly phylogenetic terms ( ‘particularly striking similarity’; our translation ) , the suggestion is of a direct lineal relationship between these heterochronic species . This would be revealed in a cladistic analysis as a sister group relationship between them . Thus , the hypothesis would be rejected by any resolution in which O . intermedius and O . striatus were not found to be sister species . We compiled a character set comprising 41 characters ( Source code 1 ) of internal and external morphology visible in one or more specimens of O . intermedius , as visualized following X-ray microtomography . We scored these characters for a set of 29 of the 39 currently described species in the genus Onthophilus ( Mazur , 2011 ) , as well as seven outgroup Onthophilinae ( including the recently described Cretaceous Cretonthophilus tuberculatus ( Caterino et al . , 2015 ) . Most were scored from direct examination of specimens . However , some taxa were scored from illustrations and descriptions in the literature ( Reichardt , 1941; Helava and Howden , 1977; Helava , 1978; Ôhara and Nakane , 1986; Ôhara , 1989; Howden and Laplante , 2003 ) . Data were analyzed under parsimony using PAUP* 4 . 0a144 ( Swofford , 2002 ) , using a heuristic search with 1000 random addition sequence replicates . Characters were all treated as unordered . We examined the effects of character reweighting ( by rescaled consistency indices ) , and exclusion of various character subsets ( internal vs . external ) . Character transitions were mapped using Mesquite v . 3 . 03 ( Maddison and Maddison , 2015 ) . The tree was rooted with either Anapleus ( Dendrophilinae: Anapleini ) , considered to exhibit plesiomorphic states in many higher level histerid characters ( Caterino and Vogler , 2002 ) , or Cretonthophilus , a recently described taxon from Cretaceous Burmese amber representing the oldest known Onthophiline histerid ( Caterino et al . , 2015 ) ( Source code 1 ) .	Fossils are the preserved remains of animals , plants or other organisms . The most highly prized fossils are those that retain their original three-dimensional shape and provide details needed to identify what species it represents , and what its closest living relatives might be . However , even fossils with the most beautifully preserved external anatomy can lack the internal structures that also help to identify its evolutionary history . “Mineralized” fossils are particularly useful for researchers as they form in a process that helps to preserve the internal anatomy of the organism . This type of fossil forms when mineral-laden water surrounds an organism’s body so that the minerals are deposited in its cells and turn soft tissues to stone . In the 1940s , Swiss scientist Eduard Handschin used eight mineralized fossil specimens to describe a 25-40 million-year-old beetle species called Onthophilus intermedius . On the basis of the external anatomy of the two best-preserved specimens , Handschin claimed this species was distinct from , but closely related to a beetle species called O . striatus that is found in Europe today . Since then , fossil examination methods have greatly advanced and include three-dimensional X-ray based imaging techniques that reveal the internal structures of a fossil while leaving it intact . One such technique is called X-ray computed tomography , in which numerous X-ray images of a solid object are taken from different angles . These images are then reassembled using computer software to create a virtual three-dimensional model of the object . Here , Schwermann et al . used this X-ray technique to re-examine the beetle fossils originally reported by Handschin . This analysis revealed many new details of these specimens’ external and internal anatomies , including their gut , genitals and airways . These new insights place Onthophilus intermedius into a different evolutionary lineage to O . striatus . They also suggest that mineralized fossils could provide a rich source of data for studies on fossil insects and other arthropods , even if they appear to be poorly preserved on the outside .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "tools", "and", "resources" ]	2016	Preservation of three-dimensional anatomy in phosphatized fossil arthropods enriches evolutionary inference
We report that bacterial RNA polymerase ( RNAP ) is the functional cellular target of the depsipeptide antibiotic salinamide A ( Sal ) , and we report that Sal inhibits RNAP through a novel binding site and mechanism . We show that Sal inhibits RNA synthesis in cells and that mutations that confer Sal-resistance map to RNAP genes . We show that Sal interacts with the RNAP active-center ‘bridge-helix cap’ comprising the ‘bridge-helix N-terminal hinge’ , ‘F-loop’ , and ‘link region’ . We show that Sal inhibits nucleotide addition in transcription initiation and elongation . We present a crystal structure that defines interactions between Sal and RNAP and effects of Sal on RNAP conformation . We propose that Sal functions by binding to the RNAP bridge-helix cap and preventing conformational changes of the bridge-helix N-terminal hinge necessary for nucleotide addition . The results provide a target for antibacterial drug discovery and a reagent to probe conformation and function of the bridge-helix N-terminal hinge . Salinamide A ( Sal; SalA ) and salinamide B ( SalB ) are structurally related bicyclic depsipeptide antibiotics , each consisting of seven amino acids and two non-amino-acid residues ( Trischman et al . , 1994; Moore et al . , 1999; Figure 1A ) . SalA and SalB are produced by Streptomyces sp . CNB-091 , a marine bacterium isolated from the surface of the jellyfish Cassiopeia xamachana ( Trischman et al . , 1994; Moore and Seng , 1998; Moore et al . , 1999 ) , and SalA also is produced by Streptomyces sp . NRRL 21611 , a soil bacterium ( Miao et al . , 1997 ) . SalA and SalB exhibit antibacterial activity against both Gram-positive and Gram-negative bacterial pathogens , particularly Enterobacter cloacae and Haemophilus influenzae , but do not exhibit cytotoxicity against mammalian cells ( Trischman et al . , 1994; Moore et al . , 1999; Figure 1B ) . SalA and SalB inhibit both Gram-positive and Gram-negative bacterial RNA polymerase ( RNAP ) in vitro , but do not inhibit human RNAP I , II , or III in vitro ( Miao et al . , 1997; Figure 1C ) . A total synthesis of SalA has been reported ( Tan and Ma , 2008 ) . 10 . 7554/eLife . 02451 . 003Figure 1 . Sal . ( A ) Structures of SalA and SalB ( Moore et al . , 1999 ) . ( B ) Growth-inhibitory activity of SalA and SalB . ( C ) RNAP-inhibitory activity of SalA and SalB . DOI: http://dx . doi . org/10 . 7554/eLife . 02451 . 003 Although previous work had established that Sal exhibits RNAP-inhibitory activity in a purified system in vitro and antibacterial activity in culture ( Trischman et al . , 1994; Miao et al . , 1997; Moore et al . , 1999 ) , previous work had not established a causal relationship between the RNAP-inhibitory activity of Sal and the antibacterial activity of Sal ( i . e . , had not established that RNAP is the functional cellular target of Sal ) . In addition , previous work had not provided information regarding the binding site , mechanism , and structural basis of inhibition of RNAP by Sal . In this work , we show that RNAP is the functional cellular target of Sal , we show that Sal inhibits RNAP through a novel binding site and novel mechanism , we determine crystal structures that define RNAP–Sal interactions , and we set the stage for structure-based design and semi-synthesis of Sal analogs with improved properties . As a first step to determine whether the RNAP-inhibitory activity of Sal is responsible for the antibacterial activity of Sal in culture , we assessed whether Sal inhibits RNAP in bacterial cells in culture . To do this , we assayed macromolecular synthesis by bacterial cells in culture , monitoring incorporation of [14C]-thymidine into DNA , [14C]-uracil into RNA , and [14C]-amino acids into protein . The results in Figure 2A shows that addition of Sal to cultures inhibits RNA synthesis at the first time point following addition and inhibits protein synthesis at later time points . Addition of Sal has no effect on DNA synthesis . The pattern observed for Sal matches the pattern observed for the reference RNAP inhibitor rifampin ( Rif; compare red lines and blue lines in Figure 2A; Lancini and Sartori , 1968; Lancini et al . , 1969 ) , and matches the pattern expected from first principles for an RNAP inhibitor ( i . e . , immediate inhibition of RNAP-dependent RNA synthesis and later inhibition of RNA-dependent protein synthesis; Sergio et al . , 1975; Irschik et al . , 1983 , 1985 , 1995 ) . We conclude that Sal inhibits RNA synthesis in bacterial cells in culture , and we infer that Sal inhibits RNAP in bacterial cells in culture . 10 . 7554/eLife . 02451 . 004Figure 2 . The RNAP-inhibitory activity of Sal accounts for the antibacterial activity of Sal . ( A ) Sal inhibits RNAP in cells . Black , no inhibitor . Red , Sal ( 2 x MIC ) . Blue , Rif ( 2 x MIC ) . Asterisks , statistically significant differences between no-inhibitor data and Sal data ( t test; p<0 . 01 ) . ( B and C ) Sal-resistant mutations occur in RNAP subunit genes . MICwild-type , SalA = 0 . 049 µg/ml; MICwild−type , SalB = 0 . 20 µg/ml . DOI: http://dx . doi . org/10 . 7554/eLife . 02451 . 004 As a second step to determine whether the RNAP-inhibitory activity of Sal is responsible for the antibacterial activity of Sal , we assessed whether Sal-resistant mutations occur in RNAP subunit genes . To do this , we isolated spontaneous Sal-resistant mutants and then PCR-amplified and sequenced genes for RNAP subunits ( Figure 2B , C ) . Spontaneous Sal-resistant mutants were isolated by plating E . coli strain , D21f2tolC—a strain with cell-envelope defects resulting in increased uptake and decreased efflux of small molecules , including Sal ( Fralick and Burns-Keliher , 1994; DD and RHE , unpublished ) —on agar containing Sal and identifying Sal-resistant colonies . For each Sal-resistant isolate , genomic DNA was prepared and the genes for the largest and second-largest RNAP subunits , rpoC encoding RNAP β′ subunit and rpoB encoding RNAP β subunit , were PCR-amplified and sequenced . Spontaneous Sal-resistant mutants were isolated with a frequency of ∼1 × 10−9 ( Figure 2B ) . A total of 47 independent Sal-resistant mutants were isolated , PCR-amplified , and sequenced ( Figure 2B ) . Strikingly , 100% ( 47/47 ) of the analyzed Sal-resistant mutants were found to contain mutations in genes for RNAP subunits: 36 were found to contain mutations in rpoC and 11 were found to contain mutations in rpoB ( Figure 2B ) . A total of 21 different substitutions conferring Sal-resistance were identified ( Figure 2C ) . Substitutions were obtained at 11 sites in RNAP β′ subunit ( residues 690 , 697 , 738 , 748 , 758 , 763 , 775 , 779 , 780 , 782 , and 783 ) and three sites in RNAP β subunit ( residues 569 , 675 , and 677 ) ( Figure 2C ) . Quantitation of resistance levels indicated that all mutants exhibited at least moderate-level ( ≥16-fold ) resistance to SalA and SalB , and that nine mutants exhibited high-level ( ≥128-fold ) resistance to SalA ( Figure 2C ) . In parallel work , we isolated and sequenced induced Sal-resistant mutants ( Supplementary file 1 ) . Random mutagenesis of plasmid-borne rpoC and rpoB genes was performed using error-prone PCR , mutagenized plasmid DNA was introduced into E . coli strain D21f2tolC by transformation , transformants were plated on media containing Sal , and Sal-resistant clones were isolated . The plasmid-borne , induced Sal-resistant mutants were found to contain mutations in the same rpoC and rpoB segments as the spontaneous Sal-resistant mutants ( compare Supplementary file 1 and Figure 2C ) . Transfer of plasmids carrying plasmid-borne , induced Sal-resistant mutants was found to transfer the Sal-resistant phenotype , indicating that no mutation outside of rpoC or rpoB is required for Sal-resistance . From the analysis of spontaneous and induced Sal-resistant mutants , we conclude that a single substitution in an RNAP subunit gene , either rpoC or rpoB , is sufficient to confer Sal-resistance , and we infer that RNAP is the functional cellular target for Sal . In the three-dimensional structure of RNAP , the sites of substitutions conferring Sal-resistance form a tight cluster ( ‘the Sal target’; green surface in Figure 3A ) . The dimensions of the Sal target are ∼35 Å × ∼18 Å × ∼12 Å . The Sal target is sufficiently large to be able to encompass Sal ( ∼16 Å × ∼12 Å × ∼10 Å ) . Based on the observation that substitutions of the Sal target result in Sal-resistance ( Figure 3A ) , we infer that the Sal target is the binding site for Sal on RNAP . 10 . 7554/eLife . 02451 . 005Figure 3 . Target of transcription inhibition by Sal . ( A ) The Sal target overlaps the RNAP active-center region . Structure of bacterial RNAP ( gray ribbons; black circle for active-center region; violet sphere for active-center Mg2+; β' non-conserved region and σ omitted for clarity; PDB 1IW7 ) , showing the sites of Sal-resistant substitutions ( green surface; sequences from Figure 2C; ‘Sal target’ ) . Two orthogonal views . ( B ) The Sal target overlaps the RNAP active-center module designated as the ‘bridge-helix cap’ ( i . e . , the module comprising the N-terminal half of the bridge helix , the F loop , and the link region; Weinzierl , 2010; Hein and Landick , 2010 ) . Sequence alignments of the largest subunits of bacterial RNAP ( top 20 sequences ) and human RNAP I , RNAP II , and RNAP III ( bottom three sequences ) , showing locations of Sal-resistant substitutions ( black rectangles; sequences from Figure 2C; ‘Sal target’ ) , and locations of the RNAP active-center bridge helix , bridge-helix N-terminal hinge ( BH-HC ) , bridge-helix C-terminal hinge ( BH-HC ) , F loop , and link region ( black bars; boundaries from Weinzierl , 2010 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02451 . 005 The Sal target is located adjacent to , and partly overlaps , the RNAP active-center region ( Figure 3A ) . We infer that Sal likely inhibits RNAP by inhibiting RNAP active-center function . Mapping of substitutions conferring Sal-resistance onto the three-dimensional structure of a transcription elongation complex comprising RNAP , DNA , RNA , and an NTP ( Vassylyev et al . , 2007b ) indicates that the Sal target does not overlap the RNAP active-center catalytic Mg2+ ion and does not overlap the RNAP residues that interact with the DNA template , the RNA product , and the NTP substrate . We infer that Sal likely inhibits RNAP active-center function allosterically , through effects on RNAP conformation , and not through direct , steric interactions with the RNAP residues that mediate bond formation , template binding , product binding , or substrate binding . The Sal target overlaps an RNAP active-center module referred to as the ‘bridge-helix cap’ , which , in turn , comprises three active-center subregions: the ‘bridge-helix N-terminal hinge’ ( BH-HN ) , the ‘F-loop’ , and the ‘link region’ ( Figure 3B; active-center subregion nomenclature as in Weinzierl 2010 and Hein and Landick 2010 ) . 18 of the 21 identified substitutions conferring Sal-resistance , and all substitutions conferring high-level Sal-resistance , map to these RNAP active-center subregions ( Figure 3B ) . It recently has been proposed that the BH-HN undergoes conformational changes coupled to , and essential for , the nucleotide-addition cycle in transcription initiation and transcription elongation , and that the F-loop , and possibly the link region , control these conformational changes ( Hein and Landick , 2010; Weinzierl , 2010; Kireeva et al . , 2012; Nedialkov et al . , 2013 ) . Specifically , it has been proposed that the BH-HN segment comprising β′ residues 779–783—a segment that includes the sites of 6 of the 21 identified substitutions conferring Sal-resistance , and 4 of the 9 substitutions conferring high-level Sal-resistance—undergoes a hinge-opening/hinge-closing conformational cycle coupled to the nucleotide-addition cycle ( Hein and Landick , 2010; Weinzierl , 2010; Kireeva et al . , 2012; Nedialkov et al . , 2013 ) . ( These proposals are supported by results of mutagenesis studies and molecular-dynamics simulations . However , these proposals have not been definitively established . Crystal structures showing an ‘open’ ( unbent ) BH-HN conformational state have been reported , but a crystal structure showing a ‘closed’ ( bent ) BH-HN conformational state has not been reported . ) Based on the strong , nearly one-for-one , correspondence between the Sal target and the active-center subregions proposed to mediate and control the BH-HN hinge-opening/hinge-closing conformational cycle , we suggest that Sal inhibits RNAP active-center function by inhibiting the proposed BH-HN hinge-opening and/or hinge-closing . The Sal target does not overlap the targets of the previously characterized RNAP inhibitors Rif ( Ovchinnikov et al . , 1981 , 1983; Lisitsyn et al . , 1984; Jin and Gross , 1988; Severinov et al . , 1993 , 1994; Campbell et al . , 2001; Garibyan et al . , 2003 ) , streptolydigin ( Stl; Lisitsyn et al . , 1985; Heisler et al . , 1993; Severinov et al . , 1993 , 1995; Tuske et al . , 2005; Temiakov et al . , 2005 ) , CBR703 ( Artsimovitch et al . , 2003; X Wang and RHE , unpublished ) , myxopyronin ( Myx; Mukhopadhyay et al . , 2008; Belogurov et al . , 2009; Srivastava et al . , 2011 ) , and lipiarmycin ( Lpm; Ebright , 2005; Tupin et al . , 2010; Srivastava et al . , 2011 ) ( Figure 4A ) . The Sal target is located adjacent to , but does not overlap , the Rif , Stl , and CBR703 targets . The Sal target is distant from the Myx and Lpm targets . 10 . 7554/eLife . 02451 . 006Figure 4 . Relationship between the Sal target and the targets of other RNAP inhibitors . ( A ) The Sal target does not overlap the targets of Rif , Stl , CBR703 , Myx , and Lpm . Structure of bacterial RNAP ( gray ribbons; violet sphere for active-center Mg2+; Mukhopadhyay et al . , 2008 ) , showing sites of substitutions that confer resistance to Sal ( green; Figures 2 , 3 ) , Rif ( red; Ovchinnikov et al . , 1981 , 1983; Lisitsyn et al . , 1984; Jin and Gross , 1988; Severinov et al . , 1993 , 1994; Campbell et al . , 2001; Garibyan et al . , 2003 ) , Stl ( yellow; Lisitsyn et al . , 1985; Heisler et al . , 1993; Severinov et al . , 1993 , 1995; Tuske et al . , 2005 ) , CBR703 ( blue; Artsimovitch et al . , 2003; X Wang and RHE , unpublished ) , Myx ( magenta; Mukhopadhyay et al . , 2008; Srivastava et al . , 2011 ) , and Lpm ( cyan; Ebright , 2005; Srivastava et al . , 2011 ) . Views as in Figure 3 . ( B ) Sal-resistant mutants ( Figure 2C ) do not exhibit cross-resistance to Rif , Stl , CBR703 , Myx , and Lpm . Blue , high-level hypersusceptibility ( MIC ratio ≤0 . 25 ) . MICwild-type , Sal = 0 . 049 μg/ml; MICwild-type , Rif = 0 . 20 μg/ml; MICwild-type , Stl = 3 . 13 μg/ml; MICwild-type , CBR703 = 6 . 25 μg/ml; MICwild-type , Myx = 0 . 20 μg/ml; MICwild-type , Lpm = 1 . 56 μg/ml . ( C ) Rif-resistant mutants ( Jin and Gross , 1988; Garibyan et al . , 2003; DD and RHE , unpublished ) , Stl-resistant mutants ( Tuske et al . , 2005 ) , CBR703-resistant mutants ( Artsimovitch et al . , 2003; X Wang and RHE , unpublished ) , Myx-resistant mutants ( Mukhopadhyay et al . , 2008 ) , and Lpm-resistant mutants ( Ebright , 2005 ) do not exhibit cross-resistance to Sal . MICwild-type , Sal = 0 . 049 μg/ml; MICwild-type , Rif = 0 . 20 μg/ml; MICwild-type , Stl = 1 . 56 μg/ml; MICwild-type , CBR703 = 6 . 25 μg/ml; MICwild-type , Myx = 0 . 20 μg/ml; MICwild-type , Lpm = 1 . 56 μg/ml . ( D ) Co-administration of Sal ( 2 × MIC ) with Rif ( 2 × MIC ) or Myx ( 2 × MIC ) suppresses the emergence of resistance . DOI: http://dx . doi . org/10 . 7554/eLife . 02451 . 006 Consistent with the absence of overlap between the Sal target and the Rif , Stl , CBR703 , Myx , and Lpm targets , Sal-resistant mutants do not exhibit cross-resistance with Rif , Stl , CBR703 , Myx , and Lpm ( Figure 4B ) . Conversely , Rif-resistant , Stl-resistant , CBR703-resistant , Myx-resistant , and Lpm-resistant mutants do not exhibit cross-resistance with Sal ( Figure 4C ) . For approximately one-quarter to one-half of Sal-resistant substitutions , not only is there no cross-resistance to Stl and CBR703 , but also there is significant ( ≥ fourfold ) hyper-susceptibility to Stl and CBR703 ( data in blue in Figure 4B ) . Resistance to a first inhibitor of an enzyme and hyper-susceptibility to a second inhibitor of the enzyme generally is understood to indicate that the two inhibitors affect different reaction steps of the enzyme and/or bind to and stabilize different conformational states of the enzyme ( Tachedjian et al . , 1996; Selmi et al . , 2003 ) . We infer that Sal may inhibit a different RNAP reaction step than Stl and CBR703 and/or may bind to and stabilize a different RNAP conformational state than Stl and CBR703 . The absence of overlap between the Sal target and other RNAP inhibitor targets , and the absence of cross-resistance between Sal and other RNAP inhibitors , suggests that the co-administration of Sal and another RNAP inhibitor may result in an extremely low , effectively undetectable , spontaneous resistance rate , representing the product of the spontaneous resistance rate for Sal and the spontaneous resistance rate for the other RNAP inhibitor . ( For many pairs of antibacterial agents having different targets and no cross-resistance , co-administration potentially results in a spontaneous resistance rate comparable to the product of the individual spontaneous resistance rates [Fischbach , 2011] . This is true even for pairs of antibacterial agents that function through the same pathway and same target protein [Fischbach , 2011] . ) The results in Figure 4D support this hypothesis . Thus , co-administration of Sal ( resistance rate = 2 × 10−9 per generation ) and Rif ( resistance rate = 1 × 10−9 per generation ) results in a resistance rate below the limit of detection ( <2 × 10−12 per generation ) . In the same manner , co-administration of Sal ( resistance rate 2 × 10−9 per generation ) and Myx ( resistance rate 3 × 10−10 per generation ) results in a resistance rate below the limit of detection ( <2 × 10−12 per generation ) . The observed suppression of the emergence of spontaneous resistance has practical implications , in view of the fact that susceptibility to spontaneous resistance is the main limiting factor in clinical use of Rif ( Floss and Yu , 2005 ) and has been cited as a potential barrier to clinical use of Myx ( Moy et al . , 2011 ) . To define the mechanistic basis of transcription inhibition by Sal , we assessed the effects of Sal on individual reaction steps in transcription initiation and transcription elongation ( Figure 5 ) . 10 . 7554/eLife . 02451 . 007Figure 5 . Mechanistic basis of transcription inhibition by Sal . ( A ) Sal does not inhibit formation of a transcription initiation complex ( RPo ) . ( B ) Sal inhibits nucleotide addition in transcription initiation . ( C ) Sal inhibits nucleotide addition in transcription elongation . ( D ) Sal inhibits nucleotide addition noncompetitively . ( E ) Transcription inhibition by Sal does not require the RNAP trigger loop . DOI: http://dx . doi . org/10 . 7554/eLife . 02451 . 00710 . 7554/eLife . 02451 . 008Figure 5—figure supplement 1 . Sal inhibits nucleotide addition in de novo initiation . DOI: http://dx . doi . org/10 . 7554/eLife . 02451 . 00810 . 7554/eLife . 02451 . 009Figure 5—figure supplement 2 . Sal enhances Type-I and Type-II transcriptional pausing . ( A ) Effects of Sal on Type-I pausing at the his pause site . ( B ) Effects of Sal on Type-II pausing at the ops pause site . A29 , 29 nt RNA product in halted elongation complex; c , ‘chase’ reaction; RO , run-off RNA product; T , terminated RNA product; p , ops or his pause-site RNA product; p* , additional pause-site RNA product; t0 . 5 , pause half-life . DOI: http://dx . doi . org/10 . 7554/eLife . 02451 . 00910 . 7554/eLife . 02451 . 010Figure 5—figure supplement 3 . Sal inhibits pyrophosphorolysis . ( A ) Pyrophosphorolysis assay using nucleic-acid scaffold containing G:C as first base pair of downstream duplex ( relatively high pyrophosphorolysis rate in absence of inhibitor ) . ( B ) Pyrophosphorolysis assay using nucleic-acid scaffold containing T:A as first base pair of downstream duplex ( relatively low pyrophosphorolysis rate in absence of inhibitor ) . Gel images show pyrophosphorolysis from 0 to 30 min . 9 nt , nucleic-acid scaffold; 8 nt , product of pyrophosphorolysis; 10 nt , product of ‘chase’ reaction with GTP ( left ) or UTP ( right ) ; nt-DNA , DNA nontemplate strand; t-DNA , DNA template strand . DOI: http://dx . doi . org/10 . 7554/eLife . 02451 . 010 The results in Figure 5A show that Sal does not inhibit formation of a heparin-resistant RNAP-promoter open complex . The results indicate that the mechanism of transcription inhibition by Sal differs from the mechanisms of transcription inhibition by Myx and Lpm , both of which inhibit the formation of the RNAP-promoter open complex ( Mukhopadhyay et al . , 2008; Belogurov et al . , 2009; Tupin et al . , 2010; Srivastava et al . , 2011 ) . The results in Figure 5B show that Sal inhibits nucleotide addition in transcription initiation . Sal inhibits nucleotide addition in both primer-dependent transcription initiation ( Figure 5B ) and de novo transcription initiation ( Figure 5—figure supplement 1 ) . In primer-dependent transcription initiation , Sal inhibits all nucleotide-addition steps , including the first nucleotide-addition step to form a 3-nt RNA product from a 2-nt RNA primer and an NTP ( Figure 5B ) . In de novo transcription initiation , Sal inhibits all nucleotide-addition steps , including the first nucleotide-addition step to form a 2-nt RNA product from NTPs ( Figure 5—figure supplement 1 ) . The results indicate that the mechanism of transcription inhibition by Sal differs from the mechanism of transcription inhibition by Rif , which does not inhibit the first nucleotide-addition step in transcription initiation ( Figure 5B; McClure and Cech , 1978 ) . The results in Figure 5C show that Sal also inhibits nucleotide addition in transcription elongation . In transcription elongation , Sal inhibits nucleotide addition both at non-pause sites ( Figure 5C ) and at type-I and type-II pause sites ( hairpin-stabilized pauses and backtracking-stabilized pauses; Figure 5—figure supplement 2 ) and inhibits not only the forward reaction of nucleotide addition but also the reverse reaction , pyrophosphorolysis ( Figure 5—figure supplement 3 ) . The results confirm that the mechanism of transcription inhibition by Sal differs from the mechanisms of transcription inhibition by Rif , Myx , and Lpm , which do not inhibit transcription elongation ( Figure 5C; McClure and Cech , 1978; Mukhopadhyay et al . , 2008; Belogurov et al . , 2009; Tupin et al . , 2010; Srivastava et al . , 2011 ) . The results in Figure 5D show that inhibition by Sal is noncompetitive with respect to NTP substrate . The Ki for inhibition is 0 . 2 µM , which is equal to the IC50 for inhibition of transcription ( compare Figures 5C and 1C ) . The results indicate that Sal does not inhibit the NTP binding sub-reaction of the nucleotide-addition cycle , but instead inhibits one or more of the bond-formation , pyrophosphate-release , and translocation sub-reactions of the nucleotide-addition cycle . The results in Figure 5E show that transcription inhibition by Sal does not require the RNAP active-center subregion referred to as the ‘trigger loop’ . Thus , Sal inhibits wild-type RNAP and an RNAP-derivative having a deletion of the trigger loop to the same extent and with nearly the same concentration-dependence . These results indicate that the mechanism of transcription inhibition by Sal differs from the mechanism of transcription inhibition by Stl , which absolutely requires the RNAP trigger loop ( Temiakov et al . , 2005 ) . Taken together , the results in Figure 5 establish that Sal inhibits RNAP through a mechanism different from the mechanisms of the previously characterized RNAP inhibitors Rif , Stl , Myx , and Lpm . The observation that Sal-resistant mutants are hyper-susceptible to CBR703 ( Figure 4B ) suggests , but does not prove , that Sal inhibits RNAP through a mechanism that is also different from the mechanism of the previously characterized RNAP inhibitor CBR703 . Based on the data presented to this point , we suggest that Sal inhibits RNAP through a novel binding site and a novel mechanism . Specifically , we suggest that Sal interacts with a binding site in the bridge-helix cap and allosterically interferes with the conformational dynamics of the BH-HN required for one or more of bond formation , pyrophosphate release , and translocation in the nucleotide-addition cycle of transcription initiation and transcription elongation . To define the structural basis of transcription inhibition by Sal , we determined crystal structures of E . coli RNAP holoenzyme and E . coli RNAP holoenzyme in complex with Sal ( Figure 6; Figure 6—figure supplement 1; Supplementary file 2 ) . [At the time this work was performed , all published crystal structures of bacterial RNAP and bacterial RNAP complexes had employed RNAP from the genus Thermus . However , it was found that Sal did not inhibit RNAP from the genus Thermus ( Figure 1C ) . Therefore , it was necessary to determine both a reference crystal structure of a Sal-susceptible bacterial RNAP and a crystal structure of the Sal-susceptible RNAP in complex with Sal . ]10 . 7554/eLife . 02451 . 011Figure 6 . Structural basis of transcription inhibition by Sal: crystal structures of E . coli RNAP holoenzyme and E . coli RNAP holoenzyme in complex with Sal . ( A ) Structure of E . coli RNAP holoenzyme ( two orthogonal views ) . Gray ribbon , RNAP core . Yellow ribbon , σ70 . Violet sphere , active-center Mg2+ . ( B ) Structure of E . coli RNAP holoenzyme in complex with Sal ( two orthogonal views ) . Green , Sal . Other colors as in A . ( C ) Electron density and atomic model for Sal ( two orthogonal views ) . Blue mesh , NCS-averaged Fo-Fc omit map for Sal ( contoured at 3 . 2σ ) . Green , red , and blue , Sal carbon , oxygen , and nitrogen atoms . Gray ribbons , RNAP . BH , FL , and LR , bridge helix , fork loop , and link region . DOI: http://dx . doi . org/10 . 7554/eLife . 02451 . 01110 . 7554/eLife . 02451 . 012Figure 6—figure supplement 1 . Structures of E . coli RNAP holoenzyme: αCTDI and αCTDII . ( A ) Structure of E . coli RNAP holoenzyme ( two orthogonal views ) . Gray , β' , β and ω . Dark green and dark blue , αI subunit N-terminal and C-terminal domains ( αNTDI and αCTDII ) . Light green and light blue , αII subunit N-terminal and C-terminal domains ( αNTDII and αCTDII ) . Yellow , σ70 . Violet sphere , active-center catalytic Mg2+ . ( B ) Closeup view of αCTDI and αCTDII ( stereoview ) . Gray , β flap and β dispensable region 2 ( βDR2 ) . Yellow , σ70 region 4 ( σR4 ) . Other colors as in A . DOI: http://dx . doi . org/10 . 7554/eLife . 02451 . 012 Figure 6A shows the resulting crystal structure of E . coli RNAP holoenzyme at 3 . 9 Å resolution . In the structure , the conformations and interactions of RNAP β′ subunit , β subunit , αI subunit N-terminal domain ( αNTDI ) , αII subunit N-terminal domain ( αNTDII ) , ω subunit , and σ70 regions 1 . 2–4 in our structure match those in recently published structures of E . coli RNAP holoenzyme ( Murakami , 2013; Zuo et al . , 2013; Bae et al . , 2013 ) . Our structure also includes the αI subunit C-terminal domain ( αCTDI ) , with a conformation and interactions matching those in the structure of Murakami , 2013 ( Figure 6—figure supplement 1 ) . ( αCTDI was not present in the RNAP derivatives used for crystallization in the structures of Zuo et al . , 2013 and Bae et al . , 2013 . ) The structure also includes the αII subunit C-terminal domain ( αCTDII ) , positioned adjacent to , and in contact with , αNTDI , the β flap , and β dispensable region 2 ( βDR2 ) ( Figure 6—figure supplement 1 ) . ( αCTDII was not ordered in the structure of Murakami , 2013 , and was not present in the RNAP derivatives used for crystallization in Zuo et al . , 2013 and Bae et al . , 2013 . ) Figures 6B and C show the corresponding structure of E . coli RNAP holoenzyme in complex with Sal at 3 . 9 Å resolution . The structure shows unambiguous experimental electron density for Sal in the genetically-defined Sal target , confirming the hypothesis that the genetically-defined Sal target represents the binding site for Sal on RNAP ( Figure 6B , C ) . To confirm the binding position and binding orientation of Sal shown in Figure 6B , C , we prepared a bromine-containing Sal derivative , and collected X-ray diffraction data for E . coli RNAP holoenzyme in complex with the bromine-containing Sal derivative ( Figure 7; Supplementary file 3 ) . The bromine-containing Sal-derivative ( ‘Sal–Br’ ) contained a residue-9 bromohydrin moiety structurally related to the residue-9 chlorohydrin moiety of SalB ( compare Figures 7A and 1A ) . Sal–Br was prepared by semi-synthesis from SalA , exploiting the unique chemical reactivity of the residue-9 epoxide moiety of SalA ( Figure 7A ) . Sal–Br was found to exhibit essentially full RNAP-inhibitory activity and essentially full antibacterial activity ( Figure 7B ) . 10 . 7554/eLife . 02451 . 013Figure 7 . Structural basis of transcription inhibition by Sal: crystal structure of E . coli RNAP holoenzyme in complex with a bromine-containing Sal derivative . ( A ) Synthesis of Sal–Br . ( B ) Growth-inhibitory activity and RNAP-inhibitory activity of Sal–Br . ( C ) Electron density , Br anomalous difference density , and atomic model for Sal–Br . Blue mesh , NCS-averaged Fo-Fc omit map for Sal ( contoured at 3 . 2σ ) . Pink mesh , Br anomalous difference density for Sal–Br ( contoured at 7 . 0σ ) . Other colors and labels as in Figure 6C . DOI: http://dx . doi . org/10 . 7554/eLife . 02451 . 013 The RNAP–Sal–Br complex exhibited electron density for Sal–Br matching the electron density in the RNAP–Sal complex for Sal ( blue mesh in Figures 6C and 7C ) and exhibited a single peak of Br anomalous difference density immediately adjacent to the electron density for Sal–Br , in the position expected for a Br atom covalently bonded to a carbon atom of the Sal–Br residue-9 bromohydrin ( pink mesh in Figure 7C ) . The results unequivocally confirm the ligand binding position and ligand binding orientation . The structural information shows that Sal binds within the RNAP bridge-helix cap , making direct interactions with the BH-HN , the fork loop , and the link region ( Figures 6C , 7C , and 8 ) . Sal makes direct interactions with all five residues at which substitutions conferring high-level ( ≥128-fold ) Sal-resistance are obtained ( β′ residues Arg738 , Ala779 , and Gly782 , and β residues Asp675 and Asn677; red in Figure 8A ) . Substitution of β′ residue Arg738 would be expected to disrupt an H-bond between RNAP and Sal ( Figure 8B , C ) . Substitution of β′ residue Ala779 or Gly782 by any residue having a larger sidechain would be expected to introduce severe steric clash between RNAP and Sal ( Figure 8B , C ) . Substitution of β residues Asp675 and Asn677 would be expected to disrupt both H-bonds and van der Waals interactions between RNAP and Sal ( Figure 8B , C ) . ( Based on the resolution of the structure and the quality of electron density maps for residues of Sal and residues of RNAP close to Sal , the inferred proximities of individual residues of Sal to individual residues of RNAP are secure , but the inferred details of H-bonds and van der Waals interactions are , at least in part , provisional . ) 10 . 7554/eLife . 02451 . 014Figure 8 . Structural basis of transcription inhibition by Sal: Sal makes direct interactions with the RNAP bridge-helix cap . ( A ) Relationship between Sal ( green ) and sites of substitutions that confer high-level Sal-resistance ( red ) . Views and labels as in Figures 6C and 7C . ( B ) Contacts between RNAP and Sal ( stereoview ) . Gray , RNAP backbone ( ribbon representation ) and RNAP sidechain carbon atoms ( stick representation ) . Green , Sal carbon atoms . Red , oxygen atoms . Blue , nitrogen atoms . Dashed lines , H-bonds . ( C ) Schematic summary of inferred contacts between RNAP and Sal . Red dashed lines , H-bonds . Blue arcs , van der Waals interactions . DOI: http://dx . doi . org/10 . 7554/eLife . 02451 . 01410 . 7554/eLife . 02451 . 015Figure 8—figure supplement 1 . The structurally and chemically accessible epoxide moiety of SalA enables semi-synthesis of novel Sal analogs . Yellow circle , Sal residue-9 epoxide moiety . Other colors as in Figure 8B , C . DOI: http://dx . doi . org/10 . 7554/eLife . 02451 . 015 Six of the RNAP residues that make direct contact with Sal are conserved across Gram-positive bacterial RNAP , Gram-negative bacterial RNAP , and human RNAP I , II , and III ( β′ residues 739 , 745 , 778 , 779 , 782 , and 785; Figures 3B and 8B , C ) . Nine RNAP residues that contact Sal are conserved in Gram-positive bacterial RNAP and Gram-negative bacterial RNAP , but are not conserved , and indeed are radically different , in human RNAP I , II , and III ( β′ residues 738 , 744 , 746 , 747 , 748 , 775 , and 781 , and β residues 675 and 677; Figures 3B and 8B , C ) . The observed interactions account for , and explain , the observation that Sal inhibits Gram-positive and Gram-negative bacterial RNAP , but does not inhibit human RNAP I , II , and III ( Figure 1C ) . Four of the five Sal-contacting residues in the RNAP BH-HN are conserved from bacterial RNAP to human RNAP ( β′ residues 778 , 779 , 782 , and 785 ) , presumably reflecting constraints on sequence variation imposed by the functionally essential , conformationally dynamic , BH-HN . In contrast , only two of the nine Sal-contacting residues in the RNAP fork loop are conserved from bacterial RNAP to human RNAP ( β′ residues 739 and 745 ) , and no Sal-contacting residues in the RNAP link region are conserved from bacterial RNAP to human RNAP , presumably reflecting lower constraints on sequence variation in these RNAP regions . The pattern of residue conservation observed for Sal is reminiscent of the pattern of residue conservation observed for the RNAP inhibitor Myx ( Mukhopadhyay et al . , 2008 ) . In each case , inhibitor-contacting residues within a functionally essential , conformationally dynamic , secondary-structure element—BH-HN for Sal and ‘switch 2’ for Myx—are conserved from bacterial RNAP to human RNAP , but inhibitor-contacting residues in adjacent secondary-structure elements are not , allowing for selective inhibition of bacterial RNAP but not human RNAP . Sal binds within a ∼2000 Å3 pocket formed by the RNAP BH-HN , the RNAP fork loop , and the RNAP link region ( Figure 8B , C ) . Backbone atoms of residues that form the pocket have superimposible conformations in RNAP holoenzyme in the absence of Sal and in RNAP holoenzyme in complex with Sal , indicating that the pocket pre-exists in RNAP holoenzyme in the absence of Sal . The pocket opens at one end onto the RNAP secondary channel and the RNAP active-center ‘i+1’ nucleotide binding site ( Figure 8B , C ) . It seems likely that Sal enters the pocket from the RNAP secondary channel and/or the active-center i+1 nucleotide site . Within the binding pocket , Sal residues 4 , 5 , 7 , and 8 interact with the RNAP BH-HN , Sal residues 1–3 and 6–7 interact with the RNAP fork loop , and Sal residues 8 and 9 interact with the RNAP link region ( Figure 8B , C ) . Sal residue 9 is at the end of the pocket that opens onto the RNAP secondary channel and the active-center i+1 nucleotide binding site ( Figure 8B , C ) . The Sal residue-9 epoxide and methyl moieties extend into this opening and make no or limited interactions with RNAP ( Figure 8B , C ) . The interactions observed in the structure suggest an opportunity for preparation of novel Sal analogs with improved potencies by semi-synthesis . The Sal residue-9 epoxide moiety is chemically reactive ( Figure 7A ) , can be altered without loss of activity ( Figure 7B ) , makes no or limited interactions with RNAP ( Figure 8B , C ) , and is directed toward the RNAP secondary channel and active-center i+1 nucleotide binding site ( Figure 8B , C ) . Accordingly , it should be possible to prepare novel Sal derivatives by semi-synthesis , introducing sidechains at the Sal residue-9 epoxide moiety that make additional interactions with RNAP , thereby potentially increasing RNAP-inhibitory activity and antibacterial activity ( Figure 8—figure supplement 1 ) . By way of example , sidechains that carry a negative charge would be positioned to make favorable electrostatic interactions with a cluster of positively-charged residues located in the RNAP secondary channel ( the ‘basic rim’; Vassylyev et al . , 2007b; Zhang and Landick , 2009 ) . By further way of example , a sidechain carrying a nucleotide , a nucleoside , or a nucleoside analog would be positioned to make highly favorable additional interactions with the RNAP active-center i+1 nucleotide binding site , potentially enabling highly potent RNAP-inhibitory activity and antibacterial activity . The crystal structure of the RNAP-Sal complex also defines effects of Sal on RNAP conformation ( Figure 9 ) . The crystal structure shows that Sal interacts with the RNAP BH-HN in an open ( unbent ) state ( Figure 9A ) , the same state that has been observed in previous crystal structures of RNAP and RNAP complexes ( Zhang et al . , 1999 , 2012; Campbell et al . , 2001; Vassylyev et al . , 2002 , 2007a , 2007b; Temiakov et al . , 2005; Tuske et al . , 2005; Mukhopadhyay et al . , 2008; Belogurov et al . , 2009; Murakami , 2013; Zuo et al . , 2013; Bae et al . , 2013; Figure 9B ) . This conformation is different from the closed ( bent ) BH-HN conformation that has been observed in molecular dynamics simulations of nucleotide-addition reactions in transcription elongation complexes ( Weinzierl , 2010; Kireeva et al . , 2012; Nedialkov et al . , 2013 ) , and that has been postulated to serve as a critical intermediate in the bond-formation , pyrophosphate-release , and/or translocation reactions of the nucleotide-addition cycle ( Hein and Landick , 2010; Weinzierl , 2010; Kireeva et al . , 2012; Nedialkov et al . , 2013 ) . We conclude that Sal interacts with an open ( unbent ) BH-HN conformational state , and we propose that , through its interactions with that state , it stabilizes that state and prevents conformational dynamics required for nucleotide addition . 10 . 7554/eLife . 02451 . 016Figure 9 . Structural basis of transcription inhibition by Sal: Sal interacts with an ‘open’ ( unbent ) state of the bridge-helix N-terminal hinge and an ‘open’ ( unfolded ) state of the trigger loop . ( A ) Electron density and model for bridge helix in crystal structure of RNAP-Sal . Blue mesh , Fo-Fc omit map for bridge helix ( contoured at 2 . 5σ ) . Black ribbon , bridge-helix backbone . Green , red , and blue , Sal carbon , oxygen , and nitrogen atoms . BH-HN , bridge-helix N-terminal hinge . BH-HC , bridge-helix C-terminal hinge . ( B ) Superimposition of bridge helices of E . coli RNAP-Sal ( black ) , E . coli RNAP ( green; unbent BH-HN and BH-HC ) , T . thermophilus RNAP ( cyan; PDB 1IW7 ) , T . thermophilus RPo ( yellow; PDB 4G7H ) , T . thermophilus transcription elongation complex ( pink; PDB 2O5J ) , and paused T . thermophilus transcription elongation complex ( violet; PDB 4GZY ) . ( C ) Predicted absence of steric clash between Sal ( colors as in A ) and trigger loop in open conformational state ( blue; PDB 1ZYR ) and predicted presence of steric clash between Sal and trigger loop in closed conformational state ( pink; PDB 2O5J ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02451 . 016 In the crystal structure of the RNAP-Sal complex , the RNAP trigger loop is disordered . Molecular modelling indicates that the structure of RNAP-Sal is compatible with the open ( unfolded ) trigger-loop conformations observed in crystal structures of RNAP and the transcription elongation complex without a bound NTP substrate ( Zhang et al . , 1999; Campbell et al . , 2001; Vassylyev et al . , 2002 , 2007a; Temiakov et al . , 2005; Tuske et al . , 2005; Mukhopadhyay et al . , 2008; Belogurov et al . , 2009; Murakami , 2013; Zuo et al . , 2013; Bae et al . , 2013 ) , but would be incompatible with the closed ( folded ) trigger loop conformation observed in the crystal structure of the transcription elongation complex with a bound NTP substrate ( Vassylyev et al . , 2007b; Figure 9C ) . We infer that Sal interacts with an open ( unfolded ) trigger-loop conformational state , and likely would prevent the formation of the closed ( folded ) trigger-loop conformational state . It is possible that effects of Sal on trigger-loop conformation may contribute to the mechanism of transcription inhibition by Sal . However , the results in Figure 5E show that the trigger loop is not essential for transcription inhibition by Sal , and therefore , although effects of Sal on trigger loop conformation may contribute to transcription inhibition by Sal , they cannot be essential for transcription inhibition by Sal . The results in Figure 2 show that Sal inhibits RNAP in bacterial cells in culture , and that Sal-resistant mutations occur in RNAP subunit genes . The results establish that the RNAP is the functional cellular target of Sal , confirming the hypothesis that the RNAP-inhibitory activity of Sal is responsible for the antibacterial activity of Sal . The results in Figure 3 establish that transcription inhibition by Sal requires a determinant located within the RNAP active-center bridge-helix cap and comprising residues of the RNAP BH-HN , the RNAP F-loop , and the RNAP link region ( ‘Sal target’ ) . The results in Figure 4 establish that the Sal target is different from , and does not overlap , the targets of the previously characterized RNAP inhibitors Rif , Stl , CBR703 , Myx , and Lpm . Consistent with the absence of overlap , mutants resistant to Sal are not cross-resistant with these other RNAP inhibitors , and , reciprocally , mutants resistant to these other RNAP inhibitors are not cross-resistant with Sal . Consistent with the absence of cross-resistance , co-administration of Sal and Rif , or of Sal and Myx , suppresses the emergence of spontaneous resistance , a finding that is significant since emergence of resistance limits the clinical application of Rif ( Floss and Yu , 2005 ) and has been cited as a potential obstacle to the clinical development of Myx ( Moy et al . , 2011 ) . The results in Figure 5 establish that Sal inhibits nucleotide addition in both transcription initiation and transcription elongation , interfering with one or more of the bond-formation , pyrophosphate-release , or translocation sub-reactions of the nucleotide-addition cycle . The results in Figure 5 show that the mechanism of inhibition by Sal is different from the mechanisms of inhibition by the previously characterized RNAP inhibitors Rif , Stl , Myx , and Lpm; and further results in Figure 4B suggest , although do not prove , that the mechanism of Sal also is different from the mechanism of the previously characterized RNAP inhibitor CBR703 . The crystal structures of RNAP–Sal and RNAP–Sal–Br complexes in Figures 6–8 confirm that Sal binds within the RNAP bridge–helix cap , making interactions with residues of the BH-HN , the F-loop , and the link region . The structures establish that Sal does not contact , or clash with , the RNAP active-center catalytic Mg2+ ion or the RNAP residues that interact with the DNA template , the RNA product , or the NTP substrate , indicating that Sal interferes with nucleotide addition allosterically . The structures further reveal that Sal interacts with an open ( unbent ) state of the BH-HN ( Figure 8A ) . We propose that Sal allosterically inhibits nucleotide addition by interacting with and stabilizing the open ( unbent ) state of the BH-HN . Sal is the first RNAP inhibitor that has been proposed to function through effects on conformational dynamics of the BH-HN . We suggest that Sal will find use as a research tool for dissection of mechanistic and structural aspects of BH-HN conformational dynamics . The semi-synthesis of Sal–Br from SalA ( Figure 7A ) shows that the SalA epoxide moiety provides a chemical reactivity that can be exploited for semi-synthesis of novel Sal analogs . The retention of RNAP inhibitory activity and antibacterial activity by Sal–Br ( Figure 7B ) shows that semi-synthetic modifications at the SalA epoxide moiety can be tolerated without loss of potency . The crystal structures of RNAP–Sal and RNAP–Sal–Br ( Figures 6–8 ) show that the SalA epoxide moiety makes no or limited interactions with RNAP and is located at the entrance to the Sal-binding pocket , directed towards the RNAP secondary channel and RNAP active-center i+1 nucleotide binding site ( Figure 8B , C; Figure 8—figure supplement 1 ) . These findings , together with the published total synthesis of SalA ( Tan and Ma , 2008 ) , set the stage for rational , structure-based design of novel semi-synthetic and fully synthetic Sal analogs with increased potency . Introduction at the SalA epoxide moiety of a sidechain with negatively-charged functionality should enable new , energetically favorable , electrostatic interactions with positively-charged ‘basic-rim’ residues in the RNAP secondary channel . Introduction at the Sal epoxide moiety of a nucleotide or nucleoside analog , should enable new , energetically favorable , interactions with the RNAP active-center i+1 nucleotide binding site . Covalently linking Sal to a nucleotide or nucleoside analog is expected to yield a bipartite inhibitor that interacts simultaneously with the Sal binding pocket and the active-center i+1 nucleotide binding site , and therefore , that potentially exhibits a very high affinity of binding and a very high potency of inhibition . Reciprocally , equipping a nucleoside-analog RNAP inhibitor with chemical functionality able to interact with the Sal pocket should provide a means both to increase potency of the nucleoside-analog inhibitor and to introduce selectivity for inhibition of bacterial RNAP vs inhibition of human RNAP . SalA and SalB were prepared from cultures of Streptomyces sp . CNB-091 as in Moore et al . ( 1999 ) . Sal–Br was prepared by adding 48% HBr ( 10 μl; 89 μmol; Sigma–Aldrich , St . Louis , MO ) to SalA ( 5 mg; 4 . 9 μmol ) in 250 μl chloroform and stirring 15 min at 24°C . The reaction mixture was washed with 100 μl saturated sodium bicarbonate , and the organic layer was separated , washed with 100 ml water , dried with anhydrous Na2SO4 , and evaporated to a white solid . The resulting solid was purified via silica flash chromatography ( 0–10% methanol in chloroform ) . Yield: 5 mg , 93% . MS ( MALDI ) : calculated: m/z 1122 . 456 , 1124 . 454 ( M+Na+ ) ; found: 1122 . 481 , 1123 . 486 , 1124 . 480 , 1125 . 492 . E . coli strain XE54 ( Tang et al . , 1994 ) was transformed with plasmid pREII-NHα ( encodes N-terminally-hexahistidine-tagged E . coli RNAP α subunit under control of tandem lpp and 'lacUV5 promoters; Niu et al . , 1996 ) . A a single colony of the resulting transformant strain was used to inoculate 50 ml fermentation broth ( FB; 32 . 5 mM Na2HPO4 , 17 . 4 mM KH2PO4 , 5 mM MgSO4 , 12 g/l tryptone , 24 g/l yeast extract , and 5 g/l glucose; pH 7 . 1 ) containing 200 mg/l ampicillin in a 200 ml flask , and the culture was incubated 16 hr at 37°C with shaking . A 50 ml aliquot of the culture was used to inoculate 2 . 8 L FB containing 200 mg/l ampicillin and 0 . 5 ml of polypropylene glycol 2000 as antifoam in a Minifors 5 L bioreactor ( INFORS HT , Bottmingen , Switzerland ) , and fermentation was carried out at 37°C with stirring ( 800 rpm ) and with maintenance of O2 ( fresh air inlet and exhaust air outlet; air flow rate governed by O2 sensor ) , pH ( 10 M NaOH supply; flow rate governed by pH sensor ) , and nutrient ( 50% glycerol supply; flow rate equal to flow rate of 10 M NaOH ) ( procedures essentially as in Riek et al . , 2008 ) . When the culture reached OD600 = 10 , the culture was induced by addition of IPTG to 1 mM , and fermentation was continued for 3 hr at 37°C . Cells were harvested by centrifugation ( 5000×g; 30 min at 4°C ) and stored at −80°C . Cells were lysed , and RNAP holoenzyme was purified using procedures in Niu et al . ( 1996 ) . Following the ammonium-sulfate-precipitation step , the pellet was dissolved in 100 ml buffer A ( 10 mM Tris–HCl , pH 7 . 9 , 500 mM NaCl , 10 mM β-mercaptoethanol , and 5% glycerol ) , loaded onto 8 ml Ni-NTA Agarose ( Qiagen , Venlo , Netherlands ) , washed with 50 ml buffer A containing 5 mM imidazole , washed with 50 ml buffer A containing 10 mM imidazole , and eluted with 50 ml buffer A containing 150 mM imidazole . The eluate was diluted with equal volume of buffer B ( 10 mM Tris–HCl , pH 7 . 9 , 1 mM EDTA , 1 mM DTT , and 5% glycerol ) and purified by anion-exchange chromatography on a 16/10 Mono Q column ( GE Healthcare , Piscataway , NJ; 160 ml linear gradient of 300–500 mM NaCl in buffer B; flow rate = 1 ml/min ) . Fractions containing RNAP holoenzyme were pooled , concentrated to 2 ml using Amicon Ultra-15 centrifugal filters ( Millipore , Billerica , MA ) , loaded onto a HiLoad 16/600 Superdex 200 column ( GE Healthcare ) pre-equilibrated in buffer C ( 10 mM Tris–HCl , pH 7 . 9 , 100 mM NaCl , and 1% glycerol ) , and eluted with 120 ml of buffer C . Fractions containing RNAP holoenzyme were pooled , concentrated to 10 mg/ml using Amicon Ultra-15 centrifugal filter ( Millipore ) and stored at −80°C . Yields were 5 mg/l , and purities were >95% . RNAP core was prepared in the same manner , but using E . coli strain BL21 ( DE3 ) ( Invitrogen , Carlsbad , CA ) transformed with plasmids pEcABC-H6 ( encodes RNAP α subunit , β subunit , and N-terminally hexahistidine-tagged β' subunit under control of the bacteriophage T7 gene 10 promoter; Hudson et al . , 2009 ) and pCDFω ( encodes RNAP ω subunit; under control of the bacteriophage T7 gene 10 promoter; Vrentas et al . , 2005 ) . ΔTL RNAP core was prepared in the same manner , but using BL21 ( DE3 ) transformed with plasmids pRL4455-β'Δ ( 931-1137 ) ΩAla3 ( encodes RNAP α subunit , β subunit , C-terminally hexahistidine-tagged β' subunit with residues 931–1137 replaced by Ala-Ala-Ala , and ω subunit , under control of the bacteriophage T7 gene 10 promoter; Toulokhonov et al . , 2007 ) and pCDFω . Yields were 10 mg/l , and purities were >95% . Minimum inhibitory concentrations ( MICs ) were quantified using broth microdilution assays as in Clinical and Laboratory Standards Institute ( 2009 ) , using a starting cell density of 2–5 × 105 cfu/ml , Mueller Hinton II cation adjusted broth ( BD Biosciences , San Jose , CA ) , and an air atmosphere for Enterobacter cloacae , Pseudomonas aeruginosa , Bacillus anthracis , Burkholderia mallei , and Yersinia pestis , and using a starting cell density of 2–5 × 105 cfu/ml , Haemophilus Test Medium broth ( Barry et al . , 1993 ) and a 7% CO2 , 6% O2 , 4% H2 , 83% N2 atmosphere for Haemophilus influenzae , Neisseria gonorrhoeae , and Moraxella catarrhalis . MICs for mammalian cells ( Vero E6 ) in culture were quantified using CellTiter96 assay ( Promega , Madison , WI; procedures as specified by the manufacturer ) . Reaction mixtures contained ( 10 μl ) : 0-100 μM test compound , bacterial RNAP holoenzyme ( 75 nM E . coli RNAP holoenzyme , 75 nM S . aureus RNAP core enzyme and 300 nM S . aureus σA [prepared as in Srivastava et al . , 2011] , or 75 nM T . thermophilus RNAP holoenzyme [prepared as in Zhang et al . , 2012] , 20 nM DNA fragment N25-lacUV5-14 ( positions −100 to −1 of the bacteriophage T5 N25 promoter [Gentz and Bujard , 1985] followed by positions +1 to +29 of the lacUV5 ( +10A;+15C ) promoter; prepared by PCR amplification of a synthetic nontemplate-strand oligodeoxyribonucleotide ) , 0 . 5 mM ApA , 100 μM [α32P]UTP ( 0 . 2 Bq/fmol ) , 100 μM ATP , and 100 μM GTP in TB ( 50 mM Tris–HCl , pH 8 . 0 , 100 mM KCl , 10 mM MgCl2 , 1 mM DTT , 10 μg/ml acetylated bovine serum albumin , 5% methanol , and 5% glycerol ) . Reaction components except DNA , ApA , and NTPs were pre-incubated 10 min at 24°C; DNA was added and reaction mixtures were incubated 10 min at 37°C; ApA , 0 . 15 μl 7 μM [α32P]UTP ( 200 Bq/fmol ) , ATP , and GTP were added and reaction mixtures were incubated 5 min at 37°C; and 0 . 5 μl 2 mM UTP was added and reaction mixtures were incubated 5 min at 37°C . Reactions were terminated by adding 10 μl loading buffer ( 80% formamide , 10 mM EDTA , 0 . 02% bromophenol blue , and 0 . 02% xylene cyanol ) and heating 2 min at 95°C . Products were applied to 7 M urea 15% polyacrylamide ( 19:1 acrylamide:bisacrylamide ) slab gels ( Bio-Rad , Hercules , CA ) , electrophoresed in TBE ( 90 mM Tris-borate , pH 8 . 0 , and 2 mM EDTA ) , and analyzed by storage-phosphor scanning ( Typhoon; GE Healthcare ) . Data shown are means of at least two determinations . Radiochemical assays with human RNAP I/II/III were performed essentially as in Sawadogo and Roeder ( 1985 ) . Reaction mixtures contained ( 20 µl ) : 0-100 µM test compound , 8 U HeLaScribe Nuclear Extract ( Promega ) , 1 µg human placental DNA ( Sigma-Aldrich ) , 400 μM ATP , 400 μM [α32P]UTP ( 0 . 11 Bq/fmol ) , 400 μM CTP , 400 μM GTP , 50 mM Tris–HCl , pH 8 . 0 , 7 mM HEPES-NaOH , 70 mM ( NH4 ) 2SO4 , 50 mM KCl , 12 mM MgCl2 , 5 mM DTT , 0 . 1 mM EDTA , 0 . 08 mM phenylmethylsulfonyl fluoride , 5% methanol , and 16% glycerol . Reaction components other than DNA and NTPs were pre-incubated 10 min at 30°C , DNA was added and reaction mixtures were incubated 15 min at 30°C , NTPs were added and reaction mixtures were incubated 60 min at 30°C . Reaction mixtures were spotted on DE81 filter discs ( Whatman , Kent , UK; pre-wetted with water ) and incubated 1 min at room temperature . Filters were washed with 3 × 3 ml Na2HPO4 , 2 × 3 ml water , and 3 ml ethanol , using a filter manifold ( Hoefer , Holliston , MA ) . Filters were placed in scintillation vials containing 10 ml Scintiverse BD Cocktail ( Thermo Fisher , Waltham , MA ) , and radioactivity was quantified by scintillation counting ( LS6500; Beckman–Coulter , Brea , CA ) . Half-maximal inhibitory concentrations ( IC50s ) were calculated by non-linear regression in SigmaPlot ( SPSS , Chicago , IL ) . Macromolecular synthesis assays were performed essentially as in Cotsonas King and Wu , 2009 . E . coli D21f2tolC ( Fralick and Burns-Keliher , 1994 ) was cultured in 10 ml LB broth ( Sambrook and Russell , 2001 ) at 37°C with shaking until OD600 = 0 . 4–0 . 8 , and cultures were diluted with LB broth to OD600 = 0 . 167 . Aliquots ( 90 µl ) were dispensed into wells of a 96-well plate; were supplemented with 7 µl of pre-warmed 4 µCi/ml [14C]-thymidine , 10 µCi/ml [14C]-uracil , or 30 µCi/ml [14C]-amino acid mix ( PerkinElmer , Waltham , MA ) ; were incubated 10 min at 37°C with shaking; were supplemented with 3 µl 3 . 3 µg/ml SalA in methanol ( final concentration = 2 × MIC ) , 3 µl 13 µg/ml Rif in methanol ( final concentration = 2 × MIC ) , or 3 µl solvent blank; and incubated at 37°C with shaking . At time points 0 , 10 , 20 , and 30 min after the addition of SalA , Rif , or solvent blank , rows of samples were transferred to a second 96-well plate , containing 100 µl ice-cold 10% trichloroacetic acid ( TCA ) in each well , and the second plate was incubated on ice . 1 hr after the final time point , TCA precipitates were collected by filtration onto glass-fiber filters ( Filtermat A; PerkinElmer; pre-rinsed twice with 5% TCA ) , washed with 2 × ∼300 µl 5% TCA , washed with 3 × ∼300 µl water , and washed with 2 × ∼300 µl 10% ethanol , using a Packard FilterMate 196 Cell Harvester with Matrix Filter upper head assembly ( PerkinElmer ) . Filters were dried under a heat lamp , wrapped in a single layer of plastic wrap , and exposed to a storage phosphor screen for 18–19 hr , and analyzed by storage-phosphor scanning ( Typhoon; GE Healthcare ) . E . coli D21f2tolC was cultured to saturation in 5 ml LB broth at 37°C , cultures were centrifuged , and cell pellets ( ∼2 × 109 cells ) were re-suspended in 50 μl LB broth and plated on LB agar ( Sambrook and Russell , 2001 ) containing 0 . 6 or 1 . 2 μg/ml SalA ( 2 × MIC or 4 × MIC under these conditions ) , and incubated 24–48 hr at 37°C . Sal-resistant mutants were identified by the ability to form colonies on this medium and were confirmed by re-streaking on the same medium . Genomic DNA was isolated using the Wizard Genomic DNA Purification Kit ( Promega; procedures as specified by the manufacturer ) and was quantified by measurement of UV-absorbance ( procedures as in Sambrook and Russell , 2001 ) . The rpoC gene and the rpoB gene were PCR-amplified in reactions containing 0 . 2 µg genomic DNA , 0 . 4 µM forward and reverse oligodeoxyribonucleotide primers ( 5′-AGGTCACTGCTGTCGGGTTAAAACC-3′ and 5′-TGACAAATGCTCTTTCCCTAAACTCC-3′ for rpoC; 5′-GTTGCACAAACTGTCCGCTCAATGG-3′ and 5′-TCGGAGTTAGCACAATCCGCTGC-3′ for rpoB ) , 5 U Taq DNA polymerase ( Genscript , Piscataway , NJ ) , and 800 µM dNTP mix ( 200 µM each dNTP; Agilent/Stratagene , La Jolla , CA ) ( initial denaturation step of 3 min at 94°C; 30 cycles of 30 s at 94°C , 45 s at 52°C , and 4 . 5 min at 68°C; final extension step of 10 min at 68°C ) . PCR products containing the rpoC gene ( 4 . 2 kB ) or the rpoB gene ( 4 . 0 kB ) were isolated by electrophoresis on 0 . 8% agarose ( procedures as in Sambrook and Russell , 2001 ) , extracted from gel slices using a Gel/PCR DNA Fragments Extraction Kit ( IBI Scientific , Peosta , IA; procedures as specified by the manufacturer ) , and submitted to the High Throughput Genomics Center ( Seattle , WA ) for sequencing ( Sanger sequencing; eight sequencing primers per gene ) . Induced Sal-resistant mutants were isolated using procedures analogous to those used for isolation of induced Myx-resistant mutants in Mukhopadhyay et al . ( 2008 ) . Random mutagenesis of rpoB plasmid pRL706 ( Severinov et al . , 1997 ) and rpoC plasmid pRL663 ( Wang et al . , 1995 ) was performed by use of PCR amplification , exploiting the baseline error rate of PCR amplification . Mutagenesis reactions were performed using the QuikChange Site-Directed Mutagenesis Kit ( Agilent/Stratagene ) , with pRL706 as template and oligodeoxyribonucleotide forward and reverse primers corresponding to nucleotides 427-446 of lacI ( 5′-GTTCCGGCGTTATTTCTTGA-3′ and 5′-TCAAGAAATAACGCCGGAAC-3′ ) , or with pRL663 as template and oligodeoxyribonucleotide forward and reverse primers corresponding to nucleotides 217-246 of lacI ( 5′-CTGCACGCGCCGTCGAAAATTGTCGCGGCG-3′ and 5′-CGCCGCGACAATTTTCGACGGCGCGTGCAG-3′ ) ( primers at 160 nM; all other components at concentrations as specified by the manufacturer ) . Mutagenized plasmid DNA was introduced by transformation into E . coli XL1-Blue ( Agilent/Stratagene ) . Transformants ( ∼5 × 103 cells ) were applied to LB-agar plates containing 200 μg/ml ampicillin , plates were incubated 16 hr at 37°C , and plasmid DNA was prepared from the pooled resulting colonies . The resulting passaged random-mutagenesis library was pooled in a 1/1 ( wt/wt ) ratio with pooled passaged saturation-mutagenesis libraries of Mukhopadhyay et al . ( 2004 ) , Tuske et al . ( 2005 ) , and Mukhopadhyay et al . ( 2008 ) , and the resulting pooled mutagenized plasmid DNA was introduced by transformation into E . coli D21f2tolC . Transformants ( ∼103 cells ) were applied to LB-agar plates containing 0 . 4 μg/ml SalA ( for pRL706 ) or 1 μg/ml SalA ( for pRL663 ) , 200 μg/ml ampicillin , and 1 mM IPTG , and plates were incubated 24–48 hr at 37°C . Sal-resistant mutants were identified by the ability to form colonies on this medium , were confirmed by re-streaking on the same medium , and were demonstrated to contain plasmid-linked Sal-resistant mutations by preparing plasmid DNA , transforming E . coli D21f2tolC with plasmid DNA , and plating transformants on the same medium . Nucleotide sequences of rpoB and rpoC were determined by Sanger sequencing ( eight primers per gene ) . For complementation assays with rpoC derivatives , temperature-sensitive E . coli strain 397c ( rpoCts397 argG thi lac [λcI857h80St68dlac+]; Christie et al . , 1996 ) was transformed with pRL663 or a pRL663 derivative , transformants ( 103–104 cells ) were applied to LB-agar plates containing 200 μg/ml ampicillin and 1 mM IPTG , plates were incubated 22 hr at 43°C , and bacterial growth was scored . For complementation assays with rpoB derivatives , temperature-sensitive E . coli strain RL585 ( rpoBamcI supDts43 , 74 Δ ( recA-srl ) 306 lacZam2110 galEKam leuam trpam sueA rpsL tsx srl301::Tn10-84; Landick et al . , 1990 ) was transformed with pRL706 or a pRL706 derivative , transformants ( 103–104 cells ) were applied to LB-agar plates containing 200 μg/ml ampicillin , 1 mM IPTG , and 10 μg/ml tetracycline , plates were incubated 22 hr at 43°C , and bacterial growth was scored . Resistance levels of Sal-resistant mutants were quantified by performing broth microdilution assays . Single colonies were inoculated into 5 ml LB broth ( LB broth containing 200 μg/ml ampicillin for induced mutants and wild-type controls for induced mutants ) and incubated at 37°C with shaking until OD600 = 0 . 4–0 . 8 . ( At this point , IPTG was added to a final concentration of 1 mM for induced mutants and wild-type controls for induced mutants , and the cultures were grown for an additional 1 hr at 37°C with shaking . ) Diluted aliquots ( ∼5 × 104 cells in 97 μl LB broth; LB broth containing 200 μg/ml ampicillin and 1 mM IPTG for induced mutants and wild-type controls for induced mutants ) were dispensed into wells of a 96-well plate , were supplemented with 3 μl of a twofold dilution series of SalA or SalB in methanol ( final concentrations = 0 . 0015–50 μg/ml ) , or 3 μl of a solvent blank , and were incubated 16 hr at 37°C with shaking . The MIC was defined as the lowest tested concentration of SalA that inhibited bacterial growth by ≥90% . Cross-resistance levels of Sal-resistant mutants were determined analogously to resistance levels , using 0 . 003–200 μg/ml Rif ( Sigma–Aldrich ) , Stl ( Sourcon-Padena , Tübingen , Germany ) , CBR703 ( Maybridge , Tintagel , UK ) , MyxB ( synthesized as in Ebright and Ebright , 2013 ) , and Lpm ( BioAustralis , Smithfield , Australia ) . Cross-resistance levels of Rif-resistant mutants , Myx-resistant mutants , and Lpm-resistant mutants ( mutations transferred from pRL706 or pRL663 derivatives [Ebright , 2005; Mukhopadhyay et al . , 2008; DD , S Ismail and RHE , unpublished] to the chromosome of E . coli D21f2tolC by λ-Red-mediated recombineering [procedures essentially as in Datsenko and Wanner , 2000 , but using transformation rather than electroporation] ) were determined analogously to resistance levels of spontaneous Sal-resistant mutants . Cross-resistance levels of Stl-resistant mutants and CBR703-resistant mutants ( mutations on pRL706 and pRL663 derivatives; Tuske et al . , 2005; X Wang and RHE , unpublished ) were determined analogously to resistance levels of induced Sal-resistant mutants . Resistance rates were determined using fluctuation assays essentially as in Srivastava et al . ( 2012 ) . Defined numbers of cells of E . coli D21f2tolC ( 108–1011 cfu/plate ) were plated on LB agar containing 0 . 6 μg/ml ( 2 × MIC ) SalA , 1 μg/ml ( 2 × MIC ) Rif , 6 μg/ml ( 2 × MIC ) MyxB , both 0 . 6 μg/ml SalA and 1 μg/ml Rif , or both 0 . 6 μg/ml SalA and 6 μg/ml MyxB , and numbers of colonies were counted after 24 hr at 37°C ( at least five independent determinations each ) . Resistance rates and 95% confidence intervals were calculated using the Ma-Sandri-Sarkar Maximum Likelihood Estimator ( Ma et al . , 1992; Sarkar et al . , 1992 ) as implemented on the Fluctuation Analysis Calculator ( http://www . keshavsingh . org/protocols/FALCOR . html ) ( Hall et al . , 2009 ) . Reaction mixtures contained ( 20 μl ) : test compound ( 0 or 10 μM SalA , 0 . 2 μM Rif , 400 μM Stl , 100 μM CBR703 , 20 μM MyxB , or 100 μM Lpm ) , 40 nM E . coli RNAP holoenzyme , 10 nM DNA fragment containing positions −42 to +426 of the lacUV5 ( ICAP ) promoter ( Naryshkin et al . , 2001 ) , and 100 μg/ml heparin in TB . Reaction components other than DNA and heparin were pre-incubated 5 min at 24°C , DNA was added and reaction mixtures were incubated 15 min at 37°C; heparin was added and reactions were incubated 2 min at 37°C to disrupt non-specific RNAP-promoter complexes and RNAP-promoter closed complexes ( Cech and McClure , 1980 ) . Products were applied to 5% TBE polyacrylamide slab gels ( Bio-Rad ) , gels were electrophoresed in TBE , and gels were stained with SYBR Gold Nucleic Acid Gel Stain ( Life Technologies , Grand Island , NY ) . Reaction mixtures contained ( 10 μl ) : test compound ( 0 or 10 μM SalA , 0 . 2 μM Rif , 400 μM Stl , 100 μM CBR703 , 20 μM MyxB , or 100 μM Lpm ) , 40 nM E . coli RNAP holoenzyme , 10 nM DNA fragment containing positions −42 to +426 of the lacUV5 ( ICAP ) promoter ( Naryshkin et al . , 2001 ) , 0 . 5 mM ApA , and 10 μM [α32P]UTP ( 1 . 5 Bq/fmol ) in TB . Reaction components except DNA , ApA , and [α-32P]UTP were pre-incubated 10 min at 37°C , DNA was added and reaction mixtures were incubated 10 min at 37°C , ApA and [α32P]UTP were added and reaction mixtures were incubated 10 min at 37°C . Reactions were terminated by adding 10 μl loading buffer and heating 4 min at 95°C . Products were applied to 7 M urea 15% polyacrylamide ( 19:1 acrylamide:bisacrylamide ) slab gels , electrophoresed in TBE , and analyzed by storage-phosphor scanning ( Typhoon; GE Healthcare ) . Reaction mixtures contained ( 10 μl ) : 0 or 10 μM SalA , 400 nM E . coli RNAP holoenzyme , 100 nM DNA fragment containing positions −42 to +426 of the lacUV5 ( ICAP ) promoter ( Naryshkin et al . , 2001 ) , 100 μM [α32P]ATP ( 0 . 07 Bq/fmol ) in TB . Reaction components other than DNA and ATP were pre-incubated 10 min at 37°C; DNA was added and reaction mixtures were incubated 10 min at 37°C; and ATP was added and reaction mixtures were incubated 10 min at 37°C . Reactions were terminated by adding 5 μl loading buffer and heating 3 min at 95°C . Products were applied to 7 M urea 24% polyacrylamide ( 19:1 acrylamide:bisacrylamide ) slab gels , electrophoresed in TBE , and analyzed by storage-phosphor scanning ( Typhoon; GE Healthcare ) . Halted transcription elongation complexes ( halted at position +29 ) were prepared essentially as in Revyakin et al . ( 2006 ) . Reaction mixtures ( 13 . 5 μl ) contained: 40 nM E . coli RNAP holoenzyme , 10 nM DNA fragment N25-100-tR2 ( Revyakin et al . , 2006 ) , 100 μg/ml heparin , 5 μM [γ32P]ATP ( 5 . 5 Bq/fmol ) , 5 μM UTP , and 5 μM GTP in TB . Reaction components except heparin and NTPs were pre-incubated 10 min at 37°C , heparin was added and reaction mixtures were incubated 3 min at 37°C , and NTPs were added and reaction mixtures were incubated 5 min at 37°C . The resulting halted transcription elongation complexes were exposed to test compounds by addition of 0 . 75 μl 200 μM SalA , 0 . 75 μl 4 μM Rif , 0 . 75 μl 8 mM Stl , 0 . 75 μl 2 mM CBR703 , 0 . 75 μl 400 μM MyxB , or 0 . 75 μl 2 mM Lpm and incubation 5 min at 37°C , and were re-started by addition of 0 . 75 μl 1 mM CTP and incubation 5 min at 37°C . Reactions were terminated by adding 10 μl loading buffer and heating 4 min at 95°C . Products were applied to 7 M urea 15% polyacrylamide ( 19:1 acrylamide:bisacrylamide ) slab gels , electrophoresed in TBE , and analyzed by storage-phosphor scanning ( Typhoon; GE Healthcare ) . Reaction mixtures contained ( 20 µl ) : 0–0 . 4 µM SalA , 10 nM E . coli RNAP holoenzyme ( Epicentre , Madison , WI ) , 5 nM DNA fragment T7A1 ( −65;+35 ) ( positions −65 to +35 of the bacteriophage T7 A1 promoter [Stackhouse et al . , 1989]; prepared by PCR amplification of a synthetic nontemplate-strand oligodeoxyribonucleotide ) , 25 μg/ml heparin , 6 mM ATP , 0–1 . 6 mM UTP , and 25 μM [α32P]CTP ( 0 . 44 Bq/fmol ) in TB . Reaction components other than DNA , heparin , and NTPs were pre-incubated 30 min at 37°C . DNA was added and reaction mixtures were incubated 15 min at 37°C; heparin was added and reaction mixtures were incubated 2 min at 37°C; and NTPs were added and reactions mixtures were incubated 10 min at 37°C . Reactions were terminated by addition of 10 µl 80% formamide , 10 mM EDTA , 0 . 04% bromophenol blue , 0 . 04% xylene cyanol , and 0 . 08% amaranth red . Products were heated 5 min at 90°C , cooled 5 min on ice , resolved by urea-PAGE ( Sambrook and Russell , 2001 ) , and analyzed by storage-phosphor scanning ( Typhoon; GE Healthcare ) . Data for synthesis of the trinucleotide product pppApUp[α32P]C were fitted to full-competitive , partial-competitive , full-noncompetitive , partial-noncompetitive , full-uncompetitive , partial-uncompetitive , full-mixed , and partial-mixed models of inhibition using the Fit-to-Model feature of the SigmaPlot Enzyme Kinetics Module v1 . 1 ( SPSS ) . Fits were ranked based on the AICc statistic ( Akaike Information Criterion corrected; low values best; −432 . 5 for full-noncompetitive model; −430 . 7 for partial-noncompetitive model; −429 . 8 for next-best model ) , the Sy . x statistic ( standard error of the estimate; low values best; 6 . 647 × 10−4 for full-noncompetitive model; 6 . 651 × 10−4 for partial-noncompetitive model; 6 . 770 × 10−4 for next-best model ) , and the number of parameters ( low values best; 3 for full-noncompetitive model; 4 for partial-noncompetitive model ) . Nucleic-acid scaffolds for assays were prepared as follows: nontemplate-strand oligodeoxyribonucleotide ( 5′-TCGCCAGACAGGG-3′; 0 . 1 μM ) , template-strand oligodeoxyribonucleotide ( 5′-CCCTGTCTGGCGATGGCGCGCCG-3′; 0 . 1 μM ) , and 32P-5′-end-labeled oligoribonucleotide ( 5′-32P-CGGCGCGCC-3′; 0 . 1 μM; 200 Bq/fmol ) in 25 μl 5 mM Tris–HCl , pH 7 . 7 , 200 mM NaCl , and 10 mM MgCl2 , were heated 5 min at 95°C and cooled to 4°C in 2°C steps with 1 min per step using a thermal cycler ( Applied Biosystems , Foster City , CA ) and then were stored at −20°C . Reaction mixtures for assays contained ( 10 μl ) : 0–64 μM SalA 40 nM wild-type or ΔTL E . coli RNAP core enzyme , 10 nM 32P-labeled nucleic-acid scaffold ( 200 Bq/fmol ) , and 20 μM ATP in TB . Reaction components except SalA and ATP were pre-incubated 5 min at 37°C , SalA was added and reaction mixtures were incubated 5 min at 37°C , and ATP was added and reaction mixtures were incubated 0 . 4 min ( wild-type RNAP ) or 10 min ( ΔTL RNAP ) at 37°C . Reactions were terminated by adding 10 μl loading buffer and heating 2 min at 95°C . Products were applied to 7 M urea 15% polyacrylamide ( 19:1 acrylamide:bisacrylamide ) slab gels , electrophoresed in TBE , and analyzed by storage-phosphor scanning ( Typhoon; GE Healthcare ) . Pausing assays were performed essentially as in Ederth et al . ( 2002 ) . Reaction mixtures for formation of halted transcription elongation complexes ( halted at position +29 ) contained ( 74 μl ) : 50 nM E . coli RNA polymerase holoenzyme , 40 nM DNA fragment PT7A1-his-ThisL or PT7A1-opspheP-ThisL ( prepared by PCR using plasmid pIA171 or pIA251 [Artsimovitch and Landick , 2000] as template , and 5′-GGAGAGACAACTTAAAGAG-3′ and 5′-CAGTTCCCTACTCTCGCATG-3′ as primers] , 150 μM ApU , 1 μM [α32P]CTP ( 4 Bq/fmol ) , 2 . 5 μM ATP , 2 . 5 μM GTP , 20 mM Tris–HCl , pH 7 . 9 , 20 mM NaCl , 3 mM MgCl2 , 14 mM 2-mercaptoethanol , and 0 . 1 mM EDTA . Reaction components except ApU and NTPs were pre-incubated 5 min at 37°C , and ApU and NTPs were added and the reaction mixture was incubated 15 min at 37°C . The resulting halted transcription elongation complexes were exposed to SalA ( or buffer blank ) by addition of 4 μl 40 μM SalA ( or buffer blank ) and incubation for 5 min at 37°C , and were re-started by addition of 1 . 2 μl 10 mM ATP , 1 . 2 μl 10 mM CTP , 1 . 2 μM 10 mM UTP , 0 . 8 μl 10 mM GTP , and 2 μl 2 mg/ml heparin . Aliquots ( 10 μl ) were removed at time points ( 0 , 15 , 30 , 60 , 120 , 240 , and 480 s ) and after a subsequent ‘chase’ ( addition of 0 . 1 μl 10 mM ATP , 0 . 1 μl 10 mM CTP , 0 . 15 μl 10 mM GTP , and 0 . 1 μl 10 mM UTP , followed by incubation 5 min at 37°C ) , and aliquots were combined with 10 μl loading buffer , and heated 4 min at 95°C . Products were applied to 7 M urea 15% polyacrylamide gels ( 19:1 acrylamide:bisacrylamide ) slab gels , electrophoresed in TBE , and analyzed by storage-phosphor scanning ( Typhoon; GE Healthcare ) . Pause half-lives and efficiencies were calculated as in Landick et al . ( 1996 ) . Pyrophosphorolysis assays were performed essentially as in Hein et al . ( 2011 ) . Nucleic-acid scaffolds were prepared by mixing 1 μM nontemplate-strand oligodeoxyribonucleotide , 1 μM template-strand oligodeoxyribonucleotide , and 0 . 5 μM 32P-5′-end-labeled oligoribonucleotide ( sequences in Figure 5—figure supplement 3; RNA 32P-5′-end-labeled using T4 polynucleotide kinase [New England Biolabs , Ipswich , MA] and [γ32P]-ATP [PerkinElmer] ) in 50 μl 10 mM Tris–HCl , pH 7 . 9 , 40 mM KCl , 5 mM MgCl2 , and then heating 2 min at 95°C , cooling to 45°C in 30 s , and cooling to 25°C in 2°C steps with 120 s per step , in a thermal cycler ( Applied Biosystems ) . Transcription elongation complexes were reconstituted by mixing 0 or 10 μM Sal and 100 nM E . coli RNAP core enzyme in 90 μl 25 mM HEPES-KOH , pH 8 . 0 , 130 mM KCl , 5 mM MgCl2 , 1 mM DTT , 0 . 15 mM EDTA , 5% glycerol , and 25 μg/ml acetylated bovine serum albumin; incubating 10 min at 24°C; adding 10 µl 500 nM nucleic-acid scaffold; and incubating 15 min at 37°C . Pyrophosphorolysis was initiated by addition of 1 μl 0 . 05 U/μl apyrase ( New England Biolabs ) and 1 μl 50 mM sodium pyrophosphate; reaction mixtures were incubated at 37°C , and 10 μl aliquots were withdrawn after 0 , 1 , 2 . 5 , 5 , 10 , 20 , and 30 min and quenched by mixing with 10 μl 98% formamide , 10 mM EDTA , 0 . 02% bromophenol blue , and 0 . 02% xylene cyanol . To confirm that transcription elongation complexes were catalytically active , a ‘chase’ reaction was performed after the last time point , adding 11 . 4 μl reaction mixture to 0 . 6 μl 20 mM GTP ( scaffold of panel A of Figure 5—figure supplement 3 ) or 0 . 6 μl 20 mM UTP ( scaffold of panel B of Figure 5—figure supplement 3 ) , incubating 5 min at 37°C , and withdrawing and quenching an aliquot as above . Products were applied to 7 M urea 20% polyacrylamide gels ( 19:1 acrylamide:bisacrylamide ) slab gels , electrophoresed in TBE , and analyzed by storage-phosphor scanning ( Typhoon; GE Healthcare ) . Crystallization trials were performed using Crystal Former microfluidic chips ( Microlytic , Burlington , MA ) and SmartScreen solutions ( Microlytic ) ( precipitant inlet: 1 . 5 μl screening solution; sample inlet: 1 . 5 μl 10 mg/ml E . coli RNAP holoenzyme in 10 mM Tris-HCl , pH 7 . 9 , 100 mM NaCl , 1% glycerol; 22°C ) . Under one condition , small crystals appeared within 2 days . Conditions were optimized using the hanging-drop vapor-diffusion technique at 22°C . The optimized conditions ( reservoir: 500 μl 0 . 1 M HEPES , pH 7 . 0 , 0 . 2 M CaCl2 , and 18% PEG400; drop: 1 μl 10 mg/ml E . coli RNAP holoenzyme in 10 mM Tris-HCl , pH 7 . 9 , 100 mM NaCl , 1% glycerol plus 1 μl reservoir solution; 22°C ) yielded large crystals with dimensions of 0 . 2 mm × 0 . 2 mm × 0 . 2 mm in one week . SalA and Sal–Br were soaked into RNAP crystals , yielding RNAP–Sal and RNAP–Sal–Br crystals , by addition of 0 . 2 μl 20 mM SalA or Sal–Br in ( ± ) -2-methyl-2 , 4-pentanediol ( Hampton Research , Aliso Viejo , CA ) to the crystallization drop and incubation 30 min at 22°C . RNAP , RNAP-Sal , and RNAP–Sal–Br crystals were transferred to reservoir solutions containing 15% ( vol/vol ) ( 2R , 3R ) - ( − ) -2 , 3-butanediol ( Sigma–Aldrich ) and then flash-cooled with liquid nitrogen . Diffraction data were collected from cryo-cooled crystals at Cornell High Energy Synchrotron Source beamline F1 and at Brookhaven National Laboratory beamline X25 . Data were processed using HKL2000 ( Otwinowski and Minor , 1997 ) . The structure of E . coli RNAP holoenzyme was solved by molecular replacement using AutoMR ( McCoy et al . , 2007; Adams et al . , 2010 ) . The search model was generated by starting with the crystal structure of T . thermophilus RNAP-promoter open complex ( PDB 4G7H; Zhang et al . , 2012 ) , deleting DNA and non-conserved protein domains , modelling E . coli αI and αII subunit N-terminal domains by superimposing the crystal structure of E . coli α N-terminal domain dimer ( PDB 1BDF; Zhang and Darst , 1998 ) , and modelling E . coli β , β' , ω , and σ70 subunits using Sculptor ( Bunkóczi and Read , 2011; backbone and sidechain atoms for identical residues; backbone and Cβ atoms for non-identical residues ) . Two RNAP molecules are present in the asymmetric unit . Crystal structures of E . coli α subunit C-terminal domain ( PDB 3K4G; Lara-González et al . , 2010 ) , the E . coli β subunit β2-βi4 and βflap-βi9 domains ( PDB 3LTI and PDB 3LU0; Opalka et al . , 2010 ) , and E . coli σ70 region 2 ( PDB 1SIG; Malhotra et al . , 1996 ) were fitted manually to the ( Fo-Fc ) difference electron density map . Early-stage refinement of the structure was performed using Phenix ( Adams et al . , 2010 ) and included rigid-body refinement of each RNAP molecule in the asymmetric unit , followed by rigid-body refinement of each subunit of each RNAP molecule , followed by rigid-body refinement of 216 segments of each RNAP molecule , followed by group B-factor refinement with one B-factor group per residue , using Phenix ( methods as in Zhang et al . , 2012 ) . Density modification , including non-crystallographic-symmetry averaging and solvent flattening using a locally modified version of DM ( Collaborative Computational Project , 1994 ) , was performed to remove model bias and to improve phases . The resulting maps allowed segments that were not present in the search model to be built manually using Coot ( Emsley et al . , 2010 ) . Cycles of iterative model building with Coot and refinement with Phenix improved the model . The final E . coli RNAP holoenzyme model , refined to Rwork and Rfree of 0 . 276 and 0 . 325 , respectively , has been deposited in the PDB with accession code 4MEY ( Supplementary file 2 ) . The structures of the E . coli RNAP–SalA and RNAP–Sal–Br complexes were solved by molecular replacement in AutoMR , using the above crystal structure of E . coli RNAP holoenzyme as the search model . For each structure , after rigid-body refinement with 216 domains , an electron density feature corresponding to one molecule of SalA per holoenzyme was clearly visible in the ( Fo-Fc ) difference map . A structural model of SalA derived from the crystal structure of SalB ( CSD 50962; Trischman et al . , 1994; enantiomorph corrected based on Moore et al . , 1999 ) was fitted to the ( Fo-Fc ) difference map with minor adjustments of SalA conformation , and the fit was confirmed by the position of the peak of Br anomalous difference density for the RNAP–Sal–Br complex . The final E . coli RNAP–SalA complex model , refined to Rwork and Rfree of 0 . 286 and 0 . 325 , respectively , has been deposited in the PDB with accession code 4MEX ( Supplementary file 2 ) .	The need for new antibiotics is becoming increasingly critical , as more and more bacteria become resistant to existing drugs . To develop new treatments , researchers need to understand how antibiotics work . One way antibiotics can kill bacteria is by targeting an enzyme called bacterial RNA polymerase . This enzyme builds chains of RNA that bacteria need to survive . Sal is an antibiotic produced by a marine bacterium found on the surface of a species of jellyfish . Degen , Feng et al . show that Sal kills bacteria by inhibiting bacterial RNA polymerase and explain how Sal inhibits RNA polymerase . Sal binds to a rod-like structural element within RNA polymerase known as the ‘bridge helix’ . The bridge helix has been proposed by others to contain two ‘hinges’ that open and close—allowing the bridge helix to bend and unbend—at specific steps in the cycle through which RNA polymerase builds an RNA chain . Degen , Feng et al . show that Sal binds directly to one of the two hinges and show that Sal binds to the hinge in the unbent state . Therefore , Degen , Feng et al . propose that Sal inhibits the enzyme by preventing the hinge from bending . The binding site on RNA polymerase for Sal is different from , and does not overlap , the binding sites of current antibacterial drugs . As a result , Sal is able to kill bacteria that are resistant to current antibacterial drugs . When Degen , Feng et al . administered Sal in combination with a current antibacterial drug that targets RNA polymerase , bacteria did not detectably develop resistance to either Sal or the current antibacterial drug . The structure of the complex between Sal and RNA polymerase suggests several ways that Sal could be modified to improve its ability to interact with RNA polymerase , thereby potentially increasing Sal's antibacterial activity . Future research could develop a range of new drugs based on Sal that could kill bacteria more effectively .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "microbiology", "and", "infectious", "disease" ]	2014	Transcription inhibition by the depsipeptide antibiotic salinamide A
Prader-Willi syndrome ( PWS ) is a genetic neurodevelopmental disorder that presents with hypotonia and respiratory distress in neonates . The Necdin-deficient mouse is the only model that reproduces the respiratory phenotype of PWS ( central apnea and blunted response to respiratory challenges ) . Here , we report that Necdin deletion disturbs the migration of serotonin ( 5-HT ) neuronal precursors , leading to altered global serotonergic neuroarchitecture and increased spontaneous firing of 5-HT neurons . We show an increased expression and activity of 5-HT Transporter ( SERT/Slc6a4 ) in 5-HT neurons leading to an increase of 5-HT uptake . In Necdin-KO pups , the genetic deletion of Slc6a4 or treatment with Fluoxetine , a 5-HT reuptake inhibitor , restored normal breathing . Unexpectedly , Fluoxetine administration was associated with respiratory side effects in wild-type animals . Overall , our results demonstrate that an increase of SERT activity is sufficient to cause the apneas in Necdin-KO pups , and that fluoxetine may offer therapeutic benefits to PWS patients with respiratory complications . Respiration is a complex function controlled in large part by raphe serotonergic ( 5-HT ) neurons ( Teran et al . , 2014 ) . Central 5-HT depletion induces severe apneas during the early postnatal period ( Barrett et al . , 2016; Trowbridge et al . , 2011 ) and serotonopathy is implicated in the genesis of breathing disorders in human pathologies including neurodevelopmental diseases such as Sudden Infant Death Syndrome ( Duncan et al . , 2010; Hilaire et al . , 2010; Kinney et al . , 2011; Paterson et al . , 2009 ) , Rett syndrome ( Abdala et al . , 2010; Toward et al . , 2013 ) and Prader-Willi Syndrome ( PWS ) ( Zanella et al . , 2008 ) . However , the cellular and molecular events that underlie serotonopathy , and the causal link between serotonopathy and respiratory dysfunction in these pathologies are poorly understood . PWS ( prevalence 1/20000 ) is characterized by a combination of endocrine , metabolic , cognitive and behavioural/psychiatric symptoms ( OMIM #176270 ) . Its associated respiratory disturbances ( Miller and Wagner , 2013; Nixon and Brouillette , 2002; Tan and Urquhart , 2017 ) are highly disruptive to the daily life of patients and represent the most common cause of death ( 73% of infants and 26% of adults ) ( Butler et al . , 2017 ) . They include both obstructive ( Festen et al . , 2006; Pavone et al . , 2015 ) and central sleep apneas {Festen et al . , 2006 #1495; Sedky et al . , 2014 ) , and blunted responses to hypercapnia/hypoxia possibly due to a lack of chemoreceptor sensitivity ( Arens et al . , 1996; Gozal et al . , 1994; Schlüter et al . , 1997; Gillett and Perez , 2016 ) . Central apneas are present at birth ( Zanella et al . , 2008 ) and are prevalent throughout infancy while obstructive sleep apneas are more frequent in adolescents ( Cohen et al . , 2014 ) . PWS is caused by the loss of paternal expression of several genes of the 15q11-q13 region , including NECDIN . Necdin protein is a member of the Mage family , with proposed functions in differentiation ( Andrieu et al . , 2003; Takazaki et al . , 2002 ) , migration ( Kuwajima et al . , 2010; Miller et al . , 2009; Tennese et al . , 2008 ) , neurite growth ( Liu et al . , 2009; Tennese et al . , 2008 ) , axonal extension , arborization and fasciculation ( Pagliardini et al . , 2005 ) , and cell survival ( Aebischer et al . , 2011; Andrieu et al . , 2006; Kuwako et al . , 2005; Tennese et al . , 2008 ) . Among several mouse models of PWS , only those with Necdin deletion , Necdin ( Ndn ) -KO mouse models ( Ndntm1-Stw [Gérard et al . , 1999] and Ndntm1-Mus [Muscatelli et al . , 2000] ) , present breathing deficits . Newborns Ndn-KO showed severe arhythmia , apnea , and blunted responses to respiratory challenges that frequently result in early postnatal lethality ( Ren et al . , 2003; Zanella et al . , 2008 ) . This dyspnoeic phenotype is recapitulated in brainstem slices that contain the Inspiratory Rhythm Generator ( IRG ) , which display an irregular inspiratory rhythm and apneas ( Ren et al . , 2003; Zanella et al . , 2008 ) . Interestingly , 5-HT application , as well as other neuromodulators that are commonly co-released by medullary 5-HT neurons , such as substance P and thyrotropin-releasing hormone ( Hodges and Richerson , 2008; Holtman and Speck , 1994; Kachidian et al . , 1991; Ptak et al . , 2009 ) , stabilized the in vitro inspiratory rhythm ( Pagliardini et al . , 2005; Zanella et al . , 2008 ) . A role for serotonergic transmission in the genesis of respiratory dysfunction in the Necdin-KO model is supported by neuroanatomical studies: Pagliardini and colleagues report abnormal morphology and orientation of axonal fibers that contain large 5-HT/Substance P varicosities in the developing Ndntm1-Stw-KO medulla ( Pagliardini et al . , 2005; Pagliardini et al . , 2008 ) . Similarly , we have also previously found that 5-HT fibers contained ‘swollen 5-HT varicosities’ in the Ndntm1-Mus-KO model , and that Necdin is expressed in virtually all 5-HT neurons ( Zanella et al . , 2008 ) . These findings suggest a potential role for abnormalities in 5-HT metabolism and release as a potential mediator of respiratory dysfunction in the Necdin-KO model of PWS , but fall short of proving causality . Here , we demonstrate a causal link between the perturbed development of the 5-HT system in Ndntm1-Mus-KO mice ( referred to hereafter as Ndn-KO ) and their observed respiratory phenotype ( central apnea and hypercapnia ) . Our data implicate increased activity of serotonin transporter ( SERT ) as a key mediator of central apnea in this model , and that its inhibition restores normal breathing in Ndn-KO mice . Pet-EYFP mice expressing YFP under Pet1-promoter control , an early marker of developing 5-HT neurons ( Hawthorne et al . , 2010 ) , were used to show that Necdin is expressed from E10 . 5 in early post-mitotic 5-HT precursors and later on in all 5-HT neurons until adulthood ( Figure 1A , Figure 1—figure supplement 1A–I ) . We then assessed whether Necdin deficiency could induce alterations of 5-HT neuronal development . In wild-type mice rostral hindbrain 5-HT neurons project to the mesencephalon at E12 . 5 , and we observed a decrease in those ascending 5-HT projections in Ndn-KO embryos ( Figure 1—figure supplement 1J ) , confirming previous work ( Pagliardini et al . , 2008 ) . At E16 . 5 , when the 5-HT raphe nuclei reach their mature configuration , we observed misplaced 5-HT neurons in Ndn-KO embryos ( Figure 1B ) , with ~30% reduction in the total number of 5-HT neurons in the B1-B2 caudal raphe nuclei at birth ( Figure 1C ) . Our observations suggested a defect in 5-HT neuronal migration; which was tested using the Pet-EYFP model . In E10 . 5 WT embryos , Pet-EYFP neurons displayed typical bipolar morphology with oval-shaped somata aligned with two primitive processes attached to the ventricular and pial surfaces , required for somal translocation and involved in migration processes ( Hawthorne et al . , 2010 ) ( Figure 1D ) . In contrast cells were not correctly aligned and process orientation was significantly disturbed in Pet-EYFP/Ndn-KO embryos ( Figure 1D–E ) . Cell migration was also defective in organotypic slice cultures prepared from E12 . 5 embryos . Two-photon time-lapse imaging indicated that migratory behavior , based on somal translocation , was altered in Ndn-KO mice ( Figure 1F–H , Figure 1—video 1 and 2 ) with tracked cells exhibiting increased tortuosity ( Figure 1G ) and decreased velocity ( Figure 1H ) of their growth trajectories . Interestingly , a comparable migration defect has been described in primary cultures of Ndntm1-Stw-KO cortical neurons ( Bush and Wevrick , 2010 ) Here we revealed an alteration of cell migration of 5-HT precursors leading to misplaced 5-HT raphe nuclei in Ndn-KO mice . The acquisition of specific firing properties is considered a critical marker of 5-HT neuronal and circuit maturation ( Rood et al . , 2014 ) . Using visually guided patch-clamp recordings on brain slices ( P15 ) , we demonstrated a significant increase of spontaneous firing in Pet-EYFP/Ndn-KO cells ( Figure 1I–K ) suggesting a decreased availability of extracellular 5-HT ( Maejima et al . , 2013 ) . Overall , our results show that Necdin is responsible for the normal migration of 5-HT precursor neurons during development and exerts effects on their electrophysiological properties in post-natal life . We hypothesised that reduced availability of extracellular 5HT could have contributed to the excessive electrophysiological activity we observed in Pet-EYFP neurons in Ndn-KO animals and examined potential mechanisms through which extracellular 5-HT could be reduced . We compared the distributions of 5-HT- immunoreactive enLarged Punctiform Axonal stainings ( 5-HT LPAs , previously named ‘swollen large varicosities’ [Pagliardini et al . , 2005; Zanella et al . , 2008] ) in Ndn-KO and WT mice . In all regions analyzed we found significantly more 5-HT LPAs in Ndn-KO mice ( Figure 2A–B ) . These 5-HT LPAs could result from ( 1 ) an increase of 5-HT synthesis and/or ( 2 ) a decrease in 5-HT degradation and/or ( 3 ) an increase of 5-HT reuptake . HPLC analyses showed a similar level of L-Trp and 5-HT in Ndn-KO compared with WT mice , but a significant increase of 5HIAA product in mutants ( the ratio of 5HIAA/5-HT also being increased: Figure 2—figure supplement 1A–D ) . Noticeably , transcript levels of Tryptophan hydroxylase 2 , the enzyme that converts L-Trp to 5-HT , were similar in Ndn-KO and WT mice ( Figure 2—figure supplement 1E ) . These results suggest that the increase in 5-HT LPAs found in Ndn-KO brainstems probably result from an accumulation of intracellular 5-HT due to an increased 5-HT reuptake , since there is no increase of 5-HT synthesis but , on the contrary , an increase of 5-HT degradation . We hypothesised that overexpression of serotonin transporter ( SERT ) represents a plausible mechanism through which 5-HT could be accumulated in Ndn-KO mice , based on the observation that inactivation of Maged1 , another member of the Mage gene family , leads to overexpression of SERT ( encoded by the Slc6a4 gene ) ( Mouri et al . , 2012 ) . Indeed , we observed a 3 . 2 fold increase in SERT protein expression in the brainstems of Ndn-KO compared to WT pups ( Figure 2C–D ) , while Slc6a4 transcript levels were similar ( Figure 2—figure supplement 1F ) . This suggests post-transcriptional or post-translational dysregulation of Slc6a4/SERT in Ndn-KO . Subsequently , in 5-HT neurons of raphe primary cultures , we assessed SERT activity by live single cell uptake assay , using ASP+ ( 4 ( 4- ( dimethylamino ) styryl ) -N-methylpyridinium ) , a fluorescent substrate of SERT ( Lau et al . , 2015; Oz et al . , 2010 ) . Changes in the kinetics and saturation of ASP+ uptake were measured after 8 days in vitro culture in 5-HT neurons from neonatal ( P0 ) WT , Ndn-KO , and Slc6a4-KO mice ( Figure 2E–H , Figure 2—figure supplement 2A–B ) . As expected , cultures accumulated ASP+ over time in all conditions tested . However , kinetics experiments show that ASP+ accumulation was significantly faster ( greater mean velocity v ) in Ndn-KO compared to WT raphe neurons ( Figure 2E ) . Saturation experiments using increasing concentrations of ASP+ confirmed that ASP+ uptake is a saturable process ( Figure 2F ) and showed a Vmax ( Figure 2G ) and KM ( Figure 2H ) significantly higher in Ndn-KO than in WT or Slc6a4-KO neurons . ASP+ uptake was ~2 fold increased in Ndn-KO while it was null in Slc6a4-KO cell cultures . We conclude that there is an increase of ASP+ uptake in Ndn-KO neurons , specifically dependent on SERT activity , suggesting a mechanism for 5-HT LPAs accumulation in vivo . To determine whether in vivo deletion of Slc6a4 could suppress the 5-HT LPAs in Ndn-KO , we compared the number of 5-HT LPAs in Ndn-KO , Slc6a4-KO and Ndn/Slc6a4-double KO ( Ndn/Slc6a4-DKO ) neonates in various brain structures . The number of 5-HT LPAs was similar in brains of Ndn/Slc6a4-DKO and WT mice ( Figure 2A–B ) , indicating that the absence of Ndn is functionally compensated for by the lack of Slc6a4 . Together , our data show that increased SERT expression in Ndn-KO mice underlies an increase of 5-HT reuptake , which accumulates in 5-HT LPAs . In the absence of any increase in 5-HT synthesis ( and in fact increased 5-HT degradation ) , this sequence of events could be sufficient to cause a physiologically relevant decrease extracellular 5-HT . As exogenous 5-HT application stabilized respiratory rhythm of Ndn-KO mice in vitro , ( Zanella et al . , 2008 ) , we hypothesized that SERT dysregulation observed in Ndn-KO mice might underlie their respiratory phenotype . To further investigate this causal link , we compared breathing parameters in WT , Ndn-KO , Ndn/Slc6a4-DKO and in Ndn-KO pups treated with Fluoxetine , a selective 5-HT reuptake inhibitor ( SSRI ) used clinically to increase extracellular 5-HT ( Figure 3A–B ) . First , we confirmed that respiratory deficits , quantified as the percentage of mice exhibiting apnea ( Figure 3C ) , the number of apneas per hour ( Figure 3D ) , or the accumulated apnea duration ( Figure 3E ) , were significantly increased in Ndn-KO compared to WT mice . These deficits were suppressed by reducing SERT function either by constitutive genetic inactivation ( Ndn/Slc6a4-DKO pups ) or by 10 days of Fluoxetine treatment ( P5-P15; 10 mg/kg/day ) in Ndn-KO pups ( Figure 3C–E ) . Other basic respiratory parameters ( minute ventilation , frequency of breathing , tidal volume ) were unchanged between all genotypes ( Figure 3F–H ) . Therefore , our results show that increasing extracellular 5-HT is sufficient to suppress apneas in juvenile Ndn-KO mice . Since Fluoxetine treatment in early life has positive effects on apneas , we next questioned the long-term consequences of this treatment . Novel cohorts of WT , Ndn-KO and Ndn-KO pups were treated as above with Fluoxetine or vehicle and then submitted to plethysmography 0 , 15 and 45 days after treatment ( DAT ) ( Figure 3—figure supplement 1A–B ) . The positive effect of Fluoxetine on respiratory function in Ndn-KO pups at the end of treatment were confirmed in this cohort , but did not persist at 15 and 45 DAT ( Figure 3—figure supplement 1C–E ) . Other respiratory parameters ( minute ventilation , frequency of breathing , tidal volume ) measured at 45 DAT were unchanged between all genotypes ( Figure 3—figure supplement 1F–H ) . An altered ventilatory response to hypercapnia was previously observed in adult Ndn-KO mice ( Zanella et al . , 2008 ) , so we next investigated whether this deficit is apparent in P0-P1 pups . We examined the chemoreflex of Ndn-KO and WT neonates by initially subjecting them to a moderate hypercapnia ( 5 min; 4% CO2 ) ( Figure 4A–C ) . Under hypercapnic stress , WT but not Ndn-KO neonates progressively increased their respiratory frequency ( Rf ) ( Figure 4D ) , leading to an increase in minute ventilation ( volume breathed over 1 min , VE ) ( Figure 4F ) . In contrast , no significant effects of hypercapnia were detected on any respiratory variables in Ndn-KO pups and thus Ndn-KO pups appear relatively insensitive to hypercapnia . To determine whether altered central 5-HT transmission contributes to this effect , we performed electrophysiological recordings of rhythmic phrenic bursts using en bloc brainstem-spinal cord preparations from P0-P1 WT and Ndn-KO pups . During perfusion with physiological aCSF ( pH 7 . 4 ) , we found no significant difference in phrenic burst ( PB ) shape , amplitude or discharge frequency ( PBf ) between WT and Ndn-KO pups ( Figure 4G–H ) . As expected , PBf in WT preparations progressively increased upon acidosis ( pH = 7 . 1 , Figure 4I , L ) . However , this effect was not observed in Ndn-KO preparations ( Figure 4J , L ) . We then assessed whether increasing extracellular 5-HT could rescue chemoreflex sensitivity in this preparation . Bath application of Fluoxetine ( 20 µM ) prior to acidosis did not affect baseline PBf of Ndn-KO preparations ( Figure 4K , L ) , but instead significantly increased PBf responses to acidosis to levels indistinguishable from WT controls ( Figure 4K , L ) . Qualitatively similar responses were observed in experiments in which a 5-HT1A receptor agonist ( 8OHDPAT ) was substituted for Fluoxetine ( Figure 4—figure supplement 1A–D ) . We therefore conclude that the central chemoreceptor hyposensitivity characteristic of the Ndn-KO model can be restored by pharmacological manipulations that increase extracellular 5-HT and/or stimulate 5-HT1A-R activity . Although Fluoxetine had beneficial but transient effects on apnea incidence in Ndn-KO mice , we observed deleterious and long-lasting effects on respiratory function in WT controls . Early life Fluoxetine-treatment induced a significant increase in the number of apneic mice , the frequency of apneas , and the cumulative distribution of apneas at all timepoints measured ( 0 , 15 and 45 DAT , Figure 3—figure supplement 2A–E ) , such that measurements at 45 DAT in WT mice ( Figure 3—figure supplement 2 ) were similar to those obtained in Ndn-KO mice ( Figure 3—figure supplement 1 ) . The sensitivity of WT brainstem-spinal cord preparations , treated with Fluoxetine or with 8OHDPAT , to acute acidosis was similarly affected ( Figure 4—figure supplement 2A–D ) . In neutral aCSF , neither Fluoxetine ( Figure 4—figure supplement 2C ) or 8OHDPAT ( Figure 4—figure supplement 2D ) affected resting PBf of WT en bloc preparations but instead abolished the normal increases in PBf responses to acidosis . Thus , we confirm that Fluoxetine treatment abolishes the capacity of WT mice to respond to acidosis ( Voituron et al . , 2010 ) , and we propose a role for 5-HT1A-R activity in this response . We show here , for the first time , adverse effects of Fluoxetine on breathing outcomes . Previously , a pleiotropic function of Necdin has been reported in different neuronal populations and at different developmental stages . Concerning the 5-HT system , an expression of Necdin was observed in virtually all 5-HT neurons ( Zanella et al . , 2008 ) and an alteration of the 5-HT system in embryonic and postnatal development was partially described in both Ndn-KO ( Ndntm1-Stwand Ndntm1-Mus ) mouse models , with alterations in 5-HT axonal bundle projections ( Lee et al . , 2005; Pagliardini et al . , 2005 ) and 5-HT fibers containing swollen 5-HT ‘varicosities’ ( Pagliardini et al . , 2005; Zanella et al . , 2008 ) . Furthermore , an alteration of 5-HT metabolism ( Zanella et al . , 2008 ) was observed in mutant neonates suggesting that it might alter 5-HT modulation of the Respiratory Rhythm Generator . Finally , an in vitro exogenous application of 5-HT on brainstem-spinal cord preparations of Ndn mutant mice alleviates the incidence of apneas ( Pagliardini et al . , 2005; Zanella et al . , 2008 ) . Despite those observations , the pathological mechanism responsible for the serotonopathy in Ndn-KO mice and the causal link between this serotonopathy and the breathing alterations were not investigated . Here , we aimed to answer those questions . Noticeably , all previous studies have been performed on heterozygous Ndn-deficient mice , with a deletion of the Ndn paternal allele only ( Ndn+m/-p ) , the maternal allele being normally silent . However , we have shown that , due to a faint and variable expression of the Ndn maternal allele ( +m ) , Ndn+m/-p mice present a variability in the severity of respiratory phenotype compared with the Ndn-/- mice ( here named Ndn-KO ) ( Rieusset et al . , 2013 ) . For instance , reduction of 5-HT neurons was not previously found significant in the Ndn + m/-p mice ( Zanella et al . , 2008 ) but has been found significantly reduced in the Ndn-/- mice . In order to avoid such variability and to get consistent results , we chose here to study Ndn-/- mice . Here , we have shown that Necdin plays a pleiotropic role in the development of 5-HT neuronal precursors that guides the development of central serotonergic circuits and the physiological activity of mature 5-HT neurons . Our results suggest that Necdin controls the level of SERT expression in 5-HT neurons and that lack of Necdin increases the quantity and activity of SERT leading to an increased reuptake and intra-cellular accumulation of 5-HT , as visualized by 5-HT LPAs , leading to a reduction in available extracellular 5-HT . Importantly , in vivo inhibition of SERT activity , genetically or pharmacologically ( Fluoxetine treatment ) , is sufficient to prevent the formation of those 5-HT LPAs and suppresses the apnea observed in Ndn-KO mice . We also demonstrate , using an ex vivo approach , that the altered chemosensitivity to CO2/acidosis is caused by a central 5-HT deficit and is rescued by Fluoxetine-treatment . We conclude that an increase of 5-HT reuptake is the main cause of breathing deficits ( central apnea and hypercapnia response ) in Ndn-KO mice . Unexpectedly , we reveal an adverse and long-term effect of early life administration of Fluoxetine on the breathing ( apneas , chemosensitivity to CO2/acidosis ) of healthy mice . Previous adverse effects have been observed on anxiety and depression ( Glover and Clinton , 2016; Millard et al . , 2017 ) after an early postnatal administration of Fluoxetine but the respiratory deficits are reported here for the first time and should be further investigated in another study . Respiratory failure in patients with PWS constitute a challenging issue since it is the most common cause of death for 73% of infants and 49% of children , ( Butler et al . , 2017 ) . Death is often linked to respiratory infection or respiratory disorder and may be sudden , with some reported cases of sudden death occurring at night ( Gillett and Perez , 2016 ) . In PWS patients , any environmental acute respiratory challenge caused by , for instance , a respiratory tract infection , high altitude or intense physical activity further exacerbates their inherent disability ( blunted response to hypoxima/hypercapnia ) to adapt an respiratory response . Until now , the underlying pathology for respiratory failure remained elusive and did not appear to be impacted by recent advancements in treatment modalities ( Butler et al . , 2017 ) . Although oxygen treatment is efficient in preventing the hypoxemia induced by central apneas ( Urquhart et al . , 2013 ) , such treatment is physically constraining . Within the context of PWS , the current study points towards a critical link between Necdin , serotonopathy , and chemosensing , a function in which brainstem serotonergic circuits play a critical role . Since our study shows that Fluoxetine can suppress apnea and restore chemosensitivity , we propose that Fluoxetine might be an appropriate ‘acute’ treatment that could be considered for Prader-Willi infants/children when they present the first signs of any breathing difficulties . Mice were handled and cared for in accordance with the Guide for the Care and Use of Laboratory Animals ( N . R . C . , 1996 ) and the European Communities Council Directive of September 22th 2010 ( 2010/63/EU , 74 ) . Experimental protocols were approved by the institutional Ethical Committee guidelines for animal research with the accreditation no . B13-055-19 from the French Ministry of Agriculture . All efforts were made to minimize the number of animals used . Necdin is an imprinted gene , paternally expressed only ( Figure 2—figure supplement 3 and Figure 4—figure supplement 3 ) . In order to avoid a variability in our results due to a stochastic and faint expression of the maternal allele ( Rieusset et al . , 2013 ) , we worked with the Ndntm1-Mus strain and decided to study Ndn-/- mice ( named here Ndn-KO ) , instead of Ndn+m/-p mice as it has been done previously . Fluoxetine was obtained from Sigma ( Saint-Quentin Fallavier , France ) for cell culture and en bloc medullary experiments and from Mylan pharma for in vivo experiments . We bred ePet-EYFP-expressing ( Scott et al . , 2005a; Scott et al . , 2005b ) or Slc6a4-Cre Knock-in ( Zhuang et al . , 2005 ) mice with Ndn-KO ( Muscatelli et al . , 2000 ) mice , all on C57BL/6 background . Protocols of genotyping mice have been previously described for Pet-EYFP ( Hawthorne et al . , 2010 ) , Ndn-KO ( Rieusset et al . , 2013 ) and Sert-Cre Knock-in mice ( Zhuang et al . , 2005 ) , in which the Slc6a4 gene was replaced by Cre was referred to in the text as Slc6a4-KO . Breeding of Slc6a4-KO with Ndn-KO mice was referred to in the text as Ndn-Slc6a4-DKO . Tissue preparation and IHC were performed as previously described ( Rieusset et al . , 2013 ) . Antibodies used were: rabbit polyclonal anti-Necdin ( 07–565; Millipore , Bedford , MA , USA; 1:500 ) , mouse monoclonal anti-GFP ( Interchim , NB600-597; 1:500 ) , goat polyclonal anti-5HT ( Immunostar , 20079; 1:300 ) . Sections were examined on a Zeiss Axioplan two microscope with an Apotome module . Brainstem structures were sampled by selecting the raphe obscurus area and counting was performed on three sagittal sections/animal of 100 µm which represent the entire PET1-YFP positive cell population of the raphe obscurus ( ROb/B2 ) and pallidus ( RPa/B1 ) , both nuclei being difficult to separate . For each section , a Z-stack composed of 10 confocal images ( 8 µm focal spacing ) was acquired . For quantification , stereological method has been applied on each Z-stack image using the eCELLence software developed by Glance Vision Technologies ( Italy ) . The total cell number/per animal was obtained by summing the sub-total of cells counted for the 3 Z-stacks . Images of 5-HT LPAs were acquired using a confocal microscope ( Olympus ) . Between 4 and 8 fibers/brain region for each animal ( 3WT and 3 KO ) were analyzed for the presence of 5-HT LPAs ( >1 . 8 µm2 ) on 100 µm long fiber . The size of 5-HT LPAs was quantified using Image J . 5-HT LPA diameter has been defined ad arbitrium as the size of the largest 5-HT punctiform labelling found in the WT fibers . Slice cultures from E11 . 5 embryonic mouse brainstems were prepared from Pet-EYFP and Ndn KO/Pet-EYFP mice . Thick coronal sections ( 250 µm ) brainstem were cut using a tissue chopper and cultured in Neurobasal medium ( Thermofisher ) containing 2% B27 ( Thermofisher ) , 4% horse serum , 10 µg/ml insulin , 200 mM HEPES , 1% Antibiotic Antimycotic ( Thermofisher ) . For time lapse experiments , the dishes were mounted in a CO2 incubation chamber ( 5% CO2 at 37°C ) fitted onto an inverted confocal microscope ( LSM510 , Zeiss ) . Acquisitions of the region containing raphe Pet-EYFP +neurons were performed every 10 min for up to 15 hr . Cell coordinates , velocity , and tortuosity ( total length of the track/direct distance from the first to the last point ) were calculated using MtrackJ plugin of Image J . Sagittal slices that included the raphe ( 400 μm thick ) were cut from brainstems of 2 week old Pet-EYFP and Ndn-KO/Pet-EYFP mice . Whole-cell recordings were made from YFP+ cells in the region of the B4 raphe nucleus . During recordings , slices were continuously perfused with artificial cerebrospinal-fluid ( aCSF ) at 37°C . Patch pipettes ( 4–5 MΩ ) were filled with an internal solution with the following composition ( in mM ) : 120 KGlu , 10 KCl , 10 Na2-phosphocreatine , 10 HEPES , 1 MgCl2 , 1 EGTA , 2 ATP Na2 , 0 . 25 GTP Na; pH = 7 . 3 adjusted with KOH . Current clamp at i = 0 were recorded with a HEKA amplifier and acquired using PatchMaster software ( HEKA ) . Offline analysis was performed with Clamfit 10 . 3 . As previously reported ( Berner et al . , 2012 ) , the medulla and cervical cord of P0-P1 neonatal mice were dissected , placed in a 2 ml in vitro recording chamber , bubbled with carbogen , maintained at 27°C and superfused ( 3 . 5–4 . 5 ml per min ) with aCSFcomposed with ( mM ) : 129 . 0 NaCl , 3 . 35 KCl , 21 NaHCO3 , 1 . 26 CaCl2 , 1 . 15 MgCl2 , 0 . 58 NaH2PO4 , and 30 . 0 D-glucose ( ‘Normal aCSF’: pH 7 . 4 ) or using the same components except with 10 mM NaHCO3 ( ‘Acidified aCSF’: pH 7 . 1 ) . Inspiratory discharges of respiratory motoneurons were monitored by extracellular recording with glass suction electrodes applied to the proximal cut end of C4 and C3 spinal nerves roots . Axoscope software and Digidata 1320A interface ( Axon Instruments , Foster , CA , USA ) were used to collect electrophysiological data . Offline analysis was performed with Spike 2 ( Cambridge Electronic Design , UK ) and Origin 6 . 0 ( Microcal Software , Northampton , MA , USA ) software for PC . Burst frequency was analyzed and calculated as the number of C4 bursts per minute . The values of inspiratory burst frequency were calculated as the mean of the last 3 min of any condition: ACSF ( 7 . 4 ) and ACSF ( 7 . 1 ) . Standardized experiments in WT and Ndn-KO preparations were repeated on different preparations from different litters . For a given preparation , only one drug was applied and only one trial was performed . For RT-qPCR , mice were sacrificed at P1 , the brainstem dissected , and tissues were rapidly collected and frozen in liquid nitrogen prior to RNA isolation using standard conditions . RNA , reverse transcription and real time PCR were conducted as previously described ( Rieusset et al . , 2013 ) . Sequences of the various primer pairs used for qPCR , as well as the slope of the calibration curve established from 10 to 1 × 109 copies and qPCR efficiency E , were as follow: Tph2: F: 5’-GAGCTTGATGCCGACCAT-3’; R: 5’-TGGCCACATCCACAAAATAC-3’; Slc6a4: F:5’-CATATGCTACCAGAATGGTGG-3’; R:5’-AAGATGGCCATGATGGTGTAA-3’ . For each sample , the number of cDNA copies was normalized according to relative efficiency of RT determined by the standard cDNA quantification . Finally , gene expression was expressed as the cDNA copy number quantified in 5 µL aliquots of RT product . Newborn mice were sacrificed and brainstems were immediately dissected and snap-frozen in liquid nitrogen and stored at −80°C until protein extraction . Protein extraction was conducted as previously described ( Felix et al . , 2012 ) . Membranes were blocked with PBS containing 5% BSA for 1 hr , followed by an overnight incubation at 4°C with the following primary antibodies: guinea pig anti-SERT ( 1/2000 , Frontier Institute ) , mouse anti-B3 tubulin ( 1/2000 , ThermoFisher Scientific ) . Membranes were then washed and incubated 2 hr with either anti-guinea pig ( 1/1000 , ThermoFisher Scientific ) , or anti-mouse ( 1/2000; DAKO ) horseradish peroxidase-conjugated secondary antibodies . Visualisation was performed using the Super signal West-pico chemolumniscent substrate ( Pierce , Thermo Scientific , France ) . Quantification was performed using ImageJ . Pregnant mice were killed by cervical dislocation at gestational day E18 . 5 and fetuses were removed , decapitated , and the medulla dissected and stored at −80°C until measurements . Medullary 5-HT , its precursor L-tryptophan ( L-Trp ) , and its main metabolite , 5-hydroxy-indol acid acetic ( 5-HIAA ) , were measured with high-pressure liquid chromatography separation and electrochemical detection ( Waters System: pump P510 , electrochemical detector EC2465; Atlantis column DC18; mobile phase: citric acid , 50 mM; orthophosphoric acid , 50 mM; sodium octane sulfonic acid , 0 . 112 mM; EDTA , 0 . 06 mM; methanol , 5%; NaCl , 2 mM; pH 2 . 95 ) . Contents are expressed in nanograms per medulla . Cells were placed in a bath chamber on the stage of an inverted microscope ( Nikon eclipse TE300 ) and perfused ( 2 ml/min ) with Krebs medium ( mM ) : 150 NaCl; 2 . 5 KCl; 2 CaCl2; 2 MgCl2; 2 . 5 Hepes acide; 2 . 5 Hepes-Na; pH 7 , 4 . Time-lapse cell acquisition was started when ASP+ ( 1 , 2 , 5 , 10 , 15 or 20 µM ) was added to the perfusion . ASP+ was excited at 488 nm and fluorescence was captured at 607 nm every 10 s for 5 min using Metamorph software ( MolecularDevices ) . Each ASP+ concentration was tested on three different cultures for WT and Ndn-KO and one for Ndn/Slc6a4-DKO . Cells placed on the coverslip were replaced for each concentration tested . For each ASP+ cells , an ROI of the same surface was delineated on the soma in order to measure pixel intensity in arbitrary fluorescence units . 6 ROI were determined at each measurement . Data were background subtracted and ASP+ fluorescence intensity was expressed as a function of initial fluorescence intensity . Breathing of unrestrained , non-anesthetized mice was recorded using constant air flow whole-body plethysmography filled with air or 4% CO2 in air ( EMKA Technologies , Paris , France ) . Neonatal mice ( P0-P1 ) were recorded in 25 ml chambers ( calibrated by injecting 50 µl of air ) maintained at neonatal thermoneutral ambient temperature ( 32 ± 0 . 5°C ) . For adolescent and adult mice ( P15-P30-P60 ) , four plethysmography 200 ml chambers containing air or ( calibrated by injecting 1 ml of air ) maintained at 25 ± 0 . 5°C were used to allow simultaneous measurements . Analog signals were obtained using an usbAMP device equipped with four inputs and processed using EMKA technologies IOX software ( EMKA Technologies , Paris , France ) . For neonatal mice , we measured mean respiratory frequency ( Rf , expressed in cycles per minute ) during quiet periods when mice breathed air or 5 min after breathing hypercapnic air . For adolescent and adult mice respiratory parameters ( frequency , tidal volume , minute ventilation ) were recorded over 30 min after an initial 30 min period of stabilization in the apparatus . , Apnea was defined as a prolonged expiratory time ( four times eupneic expiratory time ) , which corresponds to a threshold of 1 s . Analyses were performed using two-tailed non-parametric statistical tools due to the size of the samples ( GraphPad , Prism software ) . Values are indicated as following: ( Q2 ( Q1 , Q3 ) , n; statistical test , p-value ) where Q2 is the median , Q1 is the first quartile and Q3 is the third quartile and scatter dot plots report Q2 ( Q1 , Q3 ) . Histograms report the mean ±SEM . The level of significance was set at a p-value less than 0 . 05 . Appropriate tests were conducted depending on the experiment and are indicated in the figure legends . Mann-Whitney ( MW ) test was performed to compare two unmatched groups: differences between WT and Ndn-KO ( Figure 1 and Figure 2—figure supplement 1 ) . Kolmogorov-Smirnov test was performed to compare the cumulative distribution of two unmatched groups: differences between WT and Ndn-KO in apnea accumulation over time ( Figure 3E; Figure 3—figure supplement 1E; Figure 3—figure supplement 2E ) . Chi-square test was performed to compare two groups of animal ( WT and Ndn-KO ) with categorical outcome variable ( apnea or no apnea ) ( Figure 3C; Figure 3—figure supplement 2C ) . Kruskal-Wallis ( KW ) followed by a post hoc test Dunn test was performed to compare three or more independent groups ( Figure 2G , H; Figure 3D , F–H ) ; Friedman test followed by a post hoc test Dunn test was performed to compare matched groups ( Figure 4—figure supplement 2C , D ) . Two-way ANOVA followed by Bonferroni post-hoc test was performed to compare two factors ( Figure 2B ) . Two-way repeated-measure ( RM ) ANOVA was performed to compare two factors ( genotype compared either to time , drug treatment or respiratory challenge ) with repeated measure matched by time or respiratory challenge ( Figure 3—figure supplement 1D; Figure 3—figure supplement 2D;Figure 4D–F , L and Figure 4—figure supplement 1D ) ; genotype and respiratory challenge . ANCOVA was performed to compare slopes of two regression lines ( WT versus Ndn-KO: Figure 2E ) . p<0 . 05; p<0 . 01; p<0 . 001; **p<0 . 0001 .	Prader-Willi syndrome results from the disruption of a cluster of neighboring genes , including one called Necdin . Symptoms begin in early infancy and worsen with age . Affected children tend to develop an insatiable appetite , which often leads to obesity . They also experience serious problems with their breathing . Chest infections , high altitude and intense physical activity can be dangerous for children with Prader-Willi syndrome . This is because a slight shortage of oxygen may trigger breathing difficulties that could prove fatal . The brain cells that produce a chemical messenger called serotonin help to control breathing . Several lines of evidence suggest that loss of Necdin may trigger breathing difficulties in Prader-Willi syndrome via effects on the serotonin system . First , serotonin neurons produce the Necdin protein . Second , laboratory mice that lack the gene for Necdin have abnormally shaped serotonin neurons . Third , these mice show breathing difficulties like those of individuals with Prader-Willi syndrome . But while this implies a connection between serotonin , Necdin and breathing difficulties , it falls short of establishing a causal link . Matarazzo et al . now reveal an increase in the quantity and activity of a protein called the serotonin transporter in mutant mice that lacked the gene for Necdin compared to normal mice . Serotonin transporter proteins mop up the serotonin that neurons release when they signal to one another . Neurons in the mutant mice take up more serotonin than their counterparts in normal mice; this means they have less serotonin available for signaling . This may make it harder for the mutant mice to regulate their breathing . Drugs called selective serotonin-reuptake inhibitors ( or SSRIs for short ) can block the serotonin transporter . These drugs , which include Fluoxetine ( also called Prozac ) , are antidepressants . Matarazzo et al . show that SSRIs temporarily restore normal breathing in young mice that lack the gene for Necdin . However , these drugs have harmful long-term effects on breathing in non-mutant mice . Further studies should test whether short-term use of SSRIs could offer immediate relief for breathing difficulties in infants and children with Prader-Willi syndrome .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Necdin shapes serotonergic development and SERT activity modulating breathing in a mouse model for Prader-Willi syndrome
Escaping aversive stimuli is essential for complex organisms , but prolonged exposure to stress leads to maladaptive learning . Stress alters neuronal activity and neuromodulatory signaling in distributed networks , modifying behavior . Here , we describe changes in dopaminergic neuron activity and signaling following aversive learning in a learned helplessness paradigm in mice . A single dose of ketamine suffices to restore escape behavior after aversive learning . Dopaminergic neuron activity in the ventral tegmental area ( VTA ) systematically varies across learning , correlating with future sensitivity to ketamine treatment . Ketamine’s effects are blocked by chemogenetic inhibition of dopamine signaling . Rather than directly altering the activity of dopaminergic neurons , ketamine appears to rescue dopamine dynamics through actions in the medial prefrontal cortex ( mPFC ) . Chemogenetic activation of Drd1 receptor positive mPFC neurons mimics ketamine’s effects on behavior . Together , our data link neuromodulatory dynamics in mPFC-VTA circuits , aversive learning , and the effects of ketamine . Major depressive disorder ( MDD ) is a prevalent mental illness linked to diminished quality of life and increased mortality . Persistent changes in mood and emotional reactivity represent fundamental features of MDD , extensively investigated in human subjects ( Rottenberg , 2017; Rottenberg et al . , 2005 ) . Reduced reactivity to both positive and negative stimuli has been consistently observed in clinically depressed patients ( Bylsma et al . , 2008; Rottenberg and Hindash , 2015 ) , suggesting that MDD may involve systematic changes in the processing of reward and aversion . These changes in reward-based and aversive responses can be modeled in animals ( Abler et al . , 2007; Heldt et al . , 2007; Nestler and Carlezon , 2006; Proulx et al . , 2014 ) . In animal models involving prolonged stress , the reactivity to positive valence ( e . g . social stimuli ) and negative valence ( e . g . tail suspension ) experiences is usually diminished ( Beyer and Freund , 2017 ) , suggesting that prolonged aversive experience induces maladaptive learning . One established model of aversive learning is learned helplessness ( LH ) ( Abramson et al . , 1978; Maier and Seligman , 1976; Seligman and Maier , 1967 ) . Following prolonged inescapable stress exposure , animals learn that outcomes are independent of their behavioral actions; this learning eventually diminishes attempts to escape from avoidable stressful stimuli ( Maier and Seligman , 2016 ) . This form of aversive learning has been reproduced in humans and other animals , including rodents ( Abramson et al . , 1978; Chourbaji et al . , 2005; Maier , 1984; Maier and Seligman , 2016 ) . Reduced reactivity to aversive stimuli after LH is reversed by antidepressant treatments in animal models ( Belujon and Grace , 2014; Chourbaji et al . , 2005; Krishnan and Nestler , 2011 ) . Several studies have implicated the involvement of neuromodulatory systems , including dopamine , norepinephrine , and serotonin in the acquisition of LH and its expression over time ( Belujon and Grace , 2017; Eley et al . , 2004; Nestler and Carlezon , 2006; Nutt et al . , 2006 ) . Dopaminergic ( DA ) neurons in the ventral tegmental area ( VTA ) primarily encode reward and aversion , responding to both types of stimuli ( Lammel et al . , 2014; Morales and Margolis , 2017; Tan et al . , 2012; Tsai et al . , 2009; Watabe-Uchida et al . , 2017 ) . The activation of DA neurons is important in the development of aversive conditioning ( de Jong et al . , 2019 ) , and the activity of DA neurons is differentially modulated by acute and chronic stress ( Grace , 2016; Hollon et al . , 2015; Russo and Nestler , 2013 ) . Establishing a causal connection between DA neuron activity and depressive-like behavior , optogenetic activation of VTA DA neurons increases behavioral resilience to social defeat stress ( Chaudhury et al . , 2013 ) and promotes active coping actions ( Tye et al . , 2012 ) . How VTA DA neurons adjust their activity during aversive learning and how these activity changes relate to reduced reactivity to aversive stimuli after learning remains unclear . Widely used antidepressants , mostly targeting monoamine reuptake systems , are limited by delayed onset of efficacy , incomplete remission , and low remission rates ( Walker et al . , 2015 ) . Ketamine acts primarily as an antagonist at the glutamatergic N-methyl-D-aspartate ( NMDA ) receptors , showing rapid onset anti-depressant effects in depressed patients ( Berman et al . , 2000; Daly et al . , 2018 ) . In addition , mechanisms beyond direct NMDAR antagonism likely participate in the rapid behavioral effects of ketamine , implicating other classes of glutamate receptors , neuromodulators , and emergent circuit-level dynamics ( Ali et al . , 2020; Chatterjee et al . , 2012; Duman , 2018; Hare et al . , 2020; Lorrain et al . , 2003; Zanos et al . , 2016 ) . Ketamine ameliorates depressive-like behaviors in animal models of acute and prolonged stress ( Duman et al . , 2016; Duman and Aghajanian , 2012; Fuchikami et al . , 2015; Harmer et al . , 2017; Krishnan and Nestler , 2011 ) , and rescues escape actions in response to aversive stimuli after LH induction ( Belujon and Grace , 2014 ) . A recently published meta-analysis suggests that acute sub-anesthetic doses of ketamine may increase DA levels in the cortex , dorsal striatum , and nucleus accumbens ( Kokkinou et al . , 2018 ) . In vivo recordings from electrophysiologically identified VTA DA neurons in rats highlight ketamine’s modulation of neuronal firing ( Belujon and Grace , 2014 ) . The medial prefrontal cortex ( mPFC ) , essential for higher order cognitive functions including the control of emotional processing , is one key site for ketamine effects in the brain . Ketamine has been shown to modulate mPFC activity and plasticity to rescue depressive-like behaviors ( Li et al . , 2010; Lorrain et al . , 2003; Moda-Sava et al . , 2019; Ng et al . , 2018; Shirayama and Hashimoto , 2017; Wu et al . , 2021a ) . Given the bidirectional connectivity between VTA and mPFC ( Beier et al . , 2015 ) , ketamine may regulate VTA DA activity through actions in mPFC , proposed to potentiate the activity of pyramidal neurons rapidly through disinhibition by suppressing inhibitory interneurons ( Ali et al . , 2020; Hare et al . , 2020; Homayoun and Moghaddam , 2007 ) . Despite some reported and suggested links between ketamine and DA systems , the causal relationship between the DA system and ketamine’s effects on behavior remains to be elucidated . Here , we use fiber photometry to record the responses of VTA DA neurons to aversive stimuli across different phases of learning in LH . By leveraging the tunability of LH induction parameters , we also design a modified , weaker LH paradigm to reveal the activity patterns of VTA DA neurons for distinct behavioral outcomes , correlating the activity signatures with future sensitivity to ketamine treatment . By using chemogenetic inhibition , we demonstrate that VTA DA activity and downstream signaling is necessary for the behavioral effects of ketamine . Finally , combining fiber photometry , anatomical tracing , and chemogenetics , we find that mPFC serves as an action site of ketamine to restore DA dynamics and escape actions . To define the function of midbrain DA neurons during aversive learning , we used a variant of learned helplessness ( LH ) ( Chourbaji et al . , 2005 ) . A shuttle box with two compartments connected by a door allows animals to escape from one side to the other when an electric foot shock is delivered to either compartment . Prior to LH induction , mice were exposed to 30 escapable shocks to test baseline escape behaviors . Initially , mice escape from electric foot shocks . However , following repeated exposure to inescapable foot shocks , mice reduce escapes from avoidable 10 s-long foot shocks ( Figure 1a and b ) . Our data and prior publications ( Belujon and Grace , 2014; Beurel et al . , 2011; Maeng et al . , 2008 ) show that a single low dose of racemic ketamine 4 hr prior to the test ( 10 mg/kg , b . w . , i . p . ) is sufficient to rescue escape behavior in this LH paradigm ( Figure 1b ) . A separate group of mice that received saline instead of ketamine following the same experimental design did not decrease escape failures after LH ( Figure 1—figure supplement 1a ) . Thus , reduced failures to escape after ketamine treatment are not simply a function of time elapsed since LH induction ( i . e . spontaneous fear extinction ) . Reduced escape behavior after LH induction , as well as its reversal by ketamine , are not strictly context-dependent , since the proportion of failures to escape was similar regardless of whether behavioral evaluation was carried out in the induction context or in a novel environment ( Figure 1—figure supplement 1b ) . To understand the relationship between VTA DA neuron activity and ketamine’s rescue of escape behavior , we used fiber photometry to monitor the activity of the genetically encoded calcium indicator GCaMP6f in VTA DA neurons . DATicre neonates were virally transduced with Cre-dependent AAV1 . CAG . FLEX . GCaMP6f . Four to six weeks after transduction , optical fibers were implanted in the VTA ( Figure 1c ) , guided by real-time photometry ( Figure 1—figure supplement 2a ) . During fiber implant procedure , we observed that surgical anesthesia was associated with ~1 Hz oscillations in VTA calcium transients ( Figure 1—figure supplement 2b ) . After mice recovered from surgery , we evaluated the activity of VTA DA neurons during aversive learning in young adult mice ( P40-60 ) of both sexes . We first recorded Ca2+ transients of VTA DA neurons in response to brief , inescapable foot shocks ( 3 s , 0 . 3 mA ) during the learning period and after ketamine treatment . At the start of learning ( Induction start ) , the activity of VTA DA neurons first decreased during the aversive foot shocks and then rose after the termination of the shock ( Figure 1d and e ) . This biphasic response is consistent with recently published observations of VTA DA activity during other forms of aversive conditioning ( de Jong et al . , 2019 ) . Notably , the responses of VTA DA neurons to inescapable foot shocks were blunted at the end of the second-day induction ( Induction end ) . A single low dose of racemic ketamine ( 10 mg/kg , b . w . , i . p . , LH + KET ) largely restored the characteristic Ca2+ transient features , in parallel to the behavioral rescue ( Figure 1d and e; Figure 1—figure supplement 2d , e ) . Visualizing sequential Ca2+ traces from individual trials illustrates ( 1 ) the prevalence of after-shock peaks at the start of learning , ( 2 ) their decreased latency relative to shock onset during the second day of training , and ( 3 ) the recovery of Ca2+ transient latency to peak following ketamine administration ( Figure 1—figure supplement 2f ) . To quantify this temporal structure , we computed the latency to peak on sequential averaged trials ( n = 10 trials/avg ) and plotted the data as a scatter plot and cumulative distribution ( Figure 1—figure supplement 2g ) . No significant transients were observed in GFP-expressing controls in this behavioral assay ( Figure 1—figure supplement 2c ) . Since the activity of DA neurons has been linked to movement ( da Silva et al . , 2018; Howe and Dombeck , 2016 ) , one potential explanation for how aversive learning could modulate Ca2+ activity of VTA DA neurons involves altered locomotor behavior . While ketamine acutely changes locomotion , this effect normally occurs within tens of minutes following administration ( Ali et al . , 2020 ) , and resolves by the time clinically relevant changes in affective behavior are observed ( Berman et al . , 2000; Peltoniemi et al . , 2016 ) . We found no differences in open-field locomotion across phases of learning ( Figure 1—figure supplement 2h ) . Additionally , changes in VTA DA neuron activity were not associated with motion transitions , including onset and offset of locomotion ( Figure 1—figure supplement 2i ) . This result is not surprising , since movement transitions are typically associated with the activity of DA neurons in the Substantia Nigra pars Compacta rather than the VTA ( da Silva et al . , 2018; Howe and Dombeck , 2016 ) . Together , our data demonstrate that VTA DA neuron activity is restructured by LH , which is rescued following ketamine treatment . To better understand the relationship between VTA DA neuron activity and specific behavioral responses across phases of learning , we designed a weaker learning paradigm ( wLH ) , which allowed us to compare Ca2+ transients across distinct behavioral outcomes , escapes versus failures . wLH includes a larger number of brief escapable foot shocks as the test stimuli ( 3 s long , 100 trials ) , with only a single day of LH induction with inescapable shocks ( Figure 2a ) . As anticipated , a weaker form of LH , characterized by a lower average failure rate , was observed in this paradigm compared to stronger LH ( sLH , as in Figure 1 ) ( two-way ANOVA for sLH and wLH across behavioral states , Sidak’s multiple comparison , sLH 78 % vs wLH 48% , p = 0 . 0022 ) . Again , escape behavior recovered after ketamine treatment , compared to saline-treated animals ( Figure 2b , right panel ) . Another group of control animals underwent test sessions but no induction; they showed no changes in escape behavior over the time of the experiment ( one-way ANOVA , p = 0 . 9882 ) . A three-state transition model depicts the probability of transitioning between behavioral responses , allowing us to compare patterns of response sequences across learning ( Figure 2—figure supplement 1a ) . In addition to escapes and failures , responses were labeled ‘spontaneous’ when the animal spontaneously ran to the other side of the chamber during the random length pretrial time . Baseline and post-ketamine sequences of behavioral responses were more similar to each other than either was to wLH . In addition to changing the average probabilities of pairs of responses , wLH also altered longer sequences of outcomes within animals , reflected in decreased probability and length of successive escapes , as well as increased successive failures . Ketamine treatment partially recovered the probability and length of successive escapes , while a more prominent effect was observed on decreasing successive failures back to baseline levels ( Figure 2—figure supplement 1b ) . Altogether , these analyses suggest that ketamine specifically restructures behavioral sequences that are altered by wLH learning . The weaker LH paradigm , along with a large number of escapable foot shocks as the test stimuli , enabled broad sampling of Ca2+ transients during both escape and failure trials across conditions . Similar reductions in VTA DA responses to inescapable foot shocks were observed during wLH induction , as for strong LH ( Figure 2—figure supplement 1c ) . For escapable shocks , separately plotting VTA DA neuron activity for trials where the animal escaped or failed to escape revealed that each behavioral response is associated with distinct Ca2+ transient shapes ( Figure 2c ) . Failure trial transients were biphasic , where fluorescence decreased during the shock and increased afterwards . In contrast , monophasic transients accompanied escape trials , with an increase in fluorescence following successful transitions to the other side of the box . After learning , an increase in the similarity of the activity patterns between escape and failure trials was observed ( Figure 2c and d ) . For successful escape responses , VTA DA Ca2+ transients did not change in wLH and wLH+ ketamine . The increased similarity of Ca2+ transients between outcomes after learning was explained by the reduction in negative transients associated with failures . These failure-linked negative transients recovered following ketamine treatment ( Figure 2d ) . The timing of peaks in failure outcomes varied across conditions ( Figure 2—figure supplement 2a , b ) . Outcome-specific GCaMP6f transients , for either successful escapes or failures , did not vary across days in control mice without wLH induction , suggesting that the exposure to inescapable shocks is necessary to shape the responses of VTA DA neurons ( Figure 2—figure supplement 3a , b ) . In a separate control group , saline treatment did not change DA transients in either escape or failure trials after wLH ( Figure 2—figure supplement 4a , b ) . In humans and other animals , individual variability in stress susceptibility and in responsiveness to neuroactive compounds is broadly acknowledged ( Honey et al . , 2008 ) . To visualize behavioral trajectories , learning curves were constructed by starting at 0 and incrementing by one for every escape trial , decrementing by one for every failure , and keeping constant for spontaneous transitions ( Figure 2e ) . These learning curves highlight individual variability in learning and ketamine responsiveness . To depict the relationship between an animal’s ketamine responsiveness and VTA DA activity , we plotted the Euclidean norm of the difference between their Ca2+ transients during escape and failure trials ( \|\|escape-failure\|\|2 ) across learning ( Figure 2f , Figure 2—figure supplement 5a , b ) . We selected this trace distance metric because it is agnostic to amplitude and kinetics of the response , specifically assaying the degree to which the activity of VTA DA neurons differentiates specific trial outcomes . Figure 2g shows a correlation between behavioral response ( % failures ) and trace distance following ketamine . Stronger ketamine behavioral effects were linked to more distinct Ca2+ transients when comparing escape and failure trials ( R2 = 0 . 399 ) . Animals characterized by a larger trace distance in the baseline were more resilient to inescapable stress during LH illustrated by fewer failures to escape after wLH ( Figure 2h , R2 = 0 . 629 ) . Intriguingly , individuals with larger differences between DA GCaMP signals on escape and failure trials in the baseline showed stronger responses to future ketamine treatment ( Figure 2i , R2 = 0 . 715 ) . The association between DA transients , behavioral states , and trial outcomes raises the possibility that particular signatures of DA signals in response to aversive stimuli are required for escape actions . To test whether VTA DA activity is required for escape actions in general , or alternatively , whether it becomes important for restoring escapes after LH , we conditionally expressed an inhibitory DREADD ( designer receptor exclusively activated by designer drug ) hM4Di in VTA DA neurons . This engineered muscarinic receptor is activated by clozapine-N-oxide ( CNO ) to drive Gαi-coupled pathways ( Armbruster et al . , 2007; Kozorovitskiy et al . , 2015; Kozorovitskiy et al . , 2012; Roth , 2016 ) . We used a CBA promoter-driven Cre-conditional hM4Di AAV , with expression in VTA DA neurons confirmed by immunofluorescence ( Figure 3a ) . Efficacy was evaluated using cell-attached recordings of hM4Di-mCherry+ VTA DA neurons , with a flow in and a washout of 1 µM CNO ( Figure 3b ) . Mice conditionally expressing hM4Di in VTA DA neurons were compared to their Cre- littermate controls in the strong LH paradigm . Here , two ketamine treatments were administered on sequential days—the first one along with CNO , and the second one without ( Figure 3c ) . The expression of hM4Di in DA neurons did not change baseline behavior or learned helplessness induction ( Figure 3d ) . However , inhibiting DA neurons by CNO application ( 3 mg/kg , b . w . , i . p . ) co-administered with ketamine ( 10 mg/kg , b . w . , i . p . ) blocked ketamine’s rescue of escape behaviors . This effect was evident 4 hr after CNO/Ketamine treatment and persisted at 24 hr . Yet , when ketamine was administered alone on the following day , the behavioral rescue in hM4Di+ mice was successful . Persistent LH was evident in multiple animals of both groups 72 hr following the ketamine only treatment ( last two bars , Figure 3d ) . Comparing behavioral responses following ketamine alone versus ketamine plus CNO , within animals at two separate time-points , showed that the responses of hM4Di- animals lie around the unity line ( Figure 3d ) . In contrast , hM4Di+ responses were above the unity line , reflecting selective efficacy of ketamine treatment in the absence of CNO . Locomotor behavior in hM4Di-expressing animals was not grossly altered by a single CNO administration , suggesting that the observed changes in escape actions are not due to reduced movement ( Figure 3e ) . Although chemogenetic inhibition of DA activity blocked the behavioral effect of ketamine after aversive learning , the administration of CNO in the absence of ketamine in separate groups of hM4Di+ mice did not alter the proportion of failures to escape in the baseline condition or after sLH ( Figure 3f ) . These data support the alternative hypothesis that the effect of ketamine after LH specifically requires the associated restoration of DA signals . Although VTA DA activity varies with behavioral outcomes before LH , innate escape actions may not fundamentally require DA activity . In addition , CNO co-administration blocked the behavioral effects of ketamine in the relatively weaker learning paradigm , as expected ( wLH , Figure 3g ) . These data support the necessity of VTA DA signaling for the behavioral reversal of aversive learning by ketamine , rather than for native escape responses in naïve animals . In the next series of experiments , we sought to address the mechanisms underlying ketamine modulation of VTA DA activity and escape actions . In vivo ketamine treatment may modulate DA neuronal activity in a cell-autonomous manner locally in the VTA , or alternatively , indirectly , through effects on brain regions interconnected with the VTA . To determine whether ketamine application changes VTA DA GCaMP activity locally , we performed two-photon calcium imaging of VTA DA neurons with and without ex vivo ketamine application . DATicre neonates were transduced with AAV1 . CAG . FLEX . GCaMP6f , and acute brain slices of VTA were prepared and imaged 4–6 weeks after viral transduction ( Figure 4a and b ) . We observed spontaneous Ca2+ oscillations of DA neurons ex vivo ( Figure 4c ) , which have been shown to match action potentials in previous reports ( Engelhard et al . , 2019 ) . Further , bath application of ketamine ( 50 μM ) in acute VTA brain slices did not alter the power or frequency tuning of spontaneous Ca2+ dynamics in DA neurons ( Figure 4C ) , suggesting that the effect of in vivo ketamine treatment on VTA DA Ca2+ dynamics is not likely to be caused by direct cell-autonomous regulation . To reveal whether ex vivo ketamine application regulates VTA DA firing or spontaneous synaptic inputs , we also carried out cell-attached and whole-cell voltage clamp recordings from genetically targeted VTA DA neurons . DATicre neonates were transduced with AAV1 . FLEX . EGFP , and EGFP-expressing cells were recorded in acute brain slices from the VTA ( Figure 4g ) . Consistent with imaging results , ex vivo application of ketamine did not change spontaneous firing rate of VTA DA cells ( Figure 4h and i ) , or the amplitudes and inter-event intervals of spontaneous EPSCs and IPSCs ( Figure 4J-o ) . Altogether , our data suggest the in vivo effects of ketamine on DA activity and behavior are not likely caused by local regulation in the VTA . The absence of local modulation of DA neuron activity by ketamine supports the possibility of circuit-level mechanisms . In addition to spontaneous tonic activity , VTA DA neurons are driven by multiple sources of glutamatergic excitation , including inputs from mPFC , PPTg , and PAG , among others ( Morales and Margolis , 2017 ) . To explore whether ketamine acts through one or more among core glutamatergic VTA inputs to rescue escape actions , we locally infused ketamine into three different brain regions ( mPFC , PAG , and PPTg ) , along with the VTA itself , in separate experiments ( Figure 5a ) . Local infusion of ketamine into mPFC alone sufficed to rescue escape behavior after LH ( Figure 5b , Figure 5—figure supplement 1a ) . In contrast , ketamine infusion into the periaqueductal gray ( PAG ) and the pedunculopontine tegmental nucleus ( PPTg ) did not rescue escape actions . Local VTA infusion of ketamine also failed to rescue escape behavior after LH , further supporting the idea that the in vivo effects of ketamine on behavior are not implemented through local regulatory mechanisms in the VTA ( Figure 5b ) . To test whether local ketamine effects in mPFC rescue VTA DA dynamics after aversive learning in parallel to the behavioral changes , we infused ketamine locally into mPFC ( 12 . 5 µg , 500 nl ) , with fiber photometry as a readout of VTA activity ( Figure 5c ) . In vivo local infusion of ketamine into mPFC sufficed to recover Ca2+ activity signatures in VTA DA neurons after aversive learning , along with behavioral rescue ( Figure 5d ) , while infusion of ACSF did not recover VTA DA activity ( Figure 5—figure supplement 1b , c ) . These data support the idea that ketamine rescues VTA DA dynamics through circuit-level effects involving mPFC . Next , in order to define the mPFC cell populations that project to the VTA , we combined retrograde tracing with fluorescence in situ hybridization ( FISH ) . We injected red retrograde tracer fluorescent beads ( retrobeads , RTB ) into the VTA ( Figure 6a ) and observed red RTB fluorescence in mPFC 7–9 days after injection ( Figure 6b ) . Results from FISH experiments demonstrate that the majority of RTB+ neurons in deep layers ( layers 5/6 ) of mPFC express Drd1a mRNA . In contrast , only a small fraction of the retrobead+ neurons in layers 2/3 of mPFC have Drd1a mRNA ( Figure 6b and c ) . To determine whether the deep layer Drd1+ population in mPFC projects to the VTA , we expressed AAV8 . FLEX . EGFP in Drd1Cre ( FK150 ) animals and found that VTA neurons reside within fields of dense projections from mPFC Drd1+ neurons ( Figure 6d ) . These anatomical tracing results are consistent with previous reports showing that VTA-targeting mPFC neurons may also receive VTA DA inputs ( Beier et al . , 2015; Morales and Margolis , 2017 ) . To determine whether ketamine modulates the activity of DA-sensing mPFC Drd1+ neurons in vivo , we used fiber photometry to monitor the activity of Drd1+ neurons in mPFC before and after ketamine treatment . To restrict the expression of GCaMP6f , we transduced AAV1 . CAG . FLEX . GCaMP6f in Drd1Cre ( FK150 ) mice ( Figure 6—figure supplement 1a , b ) . We observed a rapid enhancement in population activity of cortical Drd1+ neurons after i . p . ketamine treatment ( Figure 6—figure supplement 1c , d ) . This ketamine-induced enhancement of population activity is likely mediated by suppressing the activity of local inhibitory interneurons ( Ali et al . , 2020; Hare et al . , 2020; Homayoun and Moghaddam , 2007 ) . If ketamine effect on mPFC Drd1+ neurons is crucial for behavioral changes , then the activation of mPFC Drd1+ neurons should rescue escape actions . We used the Gαq-coupled hM3D to directly enhance the excitability of Drd1 expressing neurons in mPFC following aversive learning ( Figure 6e ) . The expression and functional activation of hM3Dq were validated by immunohistochemistry and electrophysiology ( Figure 6f ) . The expression of hM3Dq also did not change baseline escape/failure rates , or the magnitude of aversive learning . After LH induction , a single i . p . dose of CNO was sufficient to rescue escape behavior rapidly within 2 hr , with a larger effect observed 4 hr after CNO treatment ( Figure 6g ) . At the end of the experiment , the successful enhancement of neuronal activity was further validated by evaluating immediate early gene product c-fos expression in mCherry+ neurons ( Figure 6h ) . The activation of hM3Dq in Drd1-positive mPFC neurons did not alter locomotion , suggesting the rescue of escape action is not due to hyperlocomotor activity ( Figure 6i ) . In addition to our results , a recently published study showed that optogenetic activation of Drd1+ mPFC neurons decreases immobility time in the forced swim test , suggesting the possibility that these Drd1-expressing neurons may broadly regulate aversive or active coping responses ( Hare et al . , 2019 ) . Furthermore , consistent with prior findings ( Hare et al . , 2019 ) , chemogenetic inhibition of mPFC Drd1+ neuronal excitability through Gαi-coupled hM4D blocked the behavioral effects of ketamine after LH ( Figure 6—figure supplement 2a , b ) . Altogether , our data demonstrate that mPFC serves as an action site of ketamine to rescue VTA DA dynamics , which is necessary to drive escape actions after LH . Multivalenced encoding of information by VTA DA neurons and their projections is now well established ( Berridge and Kringelbach , 2015; Lammel et al . , 2014; Morales and Margolis , 2017; Nestler and Carlezon , 2006; Schultz , 2016 ) . Prior reports demonstrate that stress and aversive learning change DA neuron activity in both acute and chronic paradigms ( Chaudhury et al . , 2013; de Jong et al . , 2019; Lammel et al . , 2014; Lammel et al . , 2012; Lammel et al . , 2011 ) . Our findings here show that VTA GCaMP transients are shaped both by trial outcomes—escape or failure to escape from an avoidable stimulus—as well as by the general behavioral state induced by LH learning and ketamine treatment . This malleability and multifaceted activity of DA signaling opens the possibility that particular patterns and levels of VTA DA activity map to specific behavioral outcome distributions . In this framework , differences in VTA DA activity between success and failure outcomes map to the tendency to escape aversive stimuli . Accordingly , decreases in the differences between VTA DA activity on escape and failure trials after aversive learning are associated with increased failures to escape avoidable stressful stimuli . Future behavioral responses to ketamine are correlated with trace distances between escape and failure-associated GCaMP transients before learning occurs . Building on these correlative observations , chemogenetic inhibition of VTA DA neuron activity supports a causal relationship between learning , DA activity , and ketamine effects . These data demonstrate that DA activity changes in LH paradigms and perturbations of DA dynamics shape behavioral outcomes in escape behavior . The question of what key biological parameters distinguish resilient and susceptible animals remains enigmatic . In this study , we observe substantial individual variability of behavioral expression after LH . Other studies have mapped this type of variability to a binary structure of resilience and susceptibility ( Bagot et al . , 2017; Chaudhury et al . , 2013; Friedman et al . , 2014; Peña et al . , 2019; Wang et al . , 2017 ) . While there is some utility in forcing a binary behavioral outcome , in this relatively large dataset ( n > 130 mice ) , we noted a broad ( and not bimodal ) variance in LH outcomes for both strong and weaker LH , reflecting the diversity of individual learning experiences ( Figure 6—figure supplement 3a , b ) . Notably , we also observe biologically significant variability in VTA DA responses , correlated with future behavioral outcomes after learning and ketamine treatment . The biological nature of this variability may relate to resilience or susceptibility to stress . One interesting possibility is that projections and activity profiles of DA neuron subgroups could be related to early-life and adolescent experience ( O’Donnell , 2010; Peña et al . , 2019; Peña et al . , 2017; Wahlstrom et al . , 2010 ) , resulting in variability of the overall low-dimensional activity pattern of the VTA DA population recorded using photometry . Based on this observation , it is expected that task-specific VTA DA activity may serve as a potential biomarker for the degree of susceptibility to stressful events . In the context of human populations , PET studies have been used to measure aspects of DA system function ( Jönsson et al . , 1999; MacDonald et al . , 2012; Okubo et al . , 1997; Volkow et al . , 1996 ) . Importantly , genetic variability which can drive differences in DA signaling ( Jönsson et al . , 1999 ) , could provide additional biomarkers to predict clinical ketamine efficacy . Animals used in this study derive from a genetically homogenous background , and one can expect larger variances in DA-associated genes in the human population . DA signaling has been associated with positive and negative emotionality and related to variability in psychopathology ( Felten et al . , 2011 ) . Polymorphisms in DA genes regulate DA transmission throughout the brain , influencing depression-related phenotypes ( Frisch et al . , 1999; Haeffel et al . , 2008; Pearson-Fuhrhop et al . , 2014 ) . Variance in DA genes and their interactors , along with early life experience , may shape activity patterns of DA neurons and DA release in response to stressful events , defining depression susceptibility in humans . Multiple studies demonstrate that mPFC is broadly involved in the rapid and lasting behavioral effects of ketamine . NMDA receptor antagonists , including ketamine , have been shown to rapidly enhance neuronal and dendritic activity of cortical pyramidal neurons through disinhibition ( Ali et al . , 2020; Hare et al . , 2020; Homayoun and Moghaddam , 2007; Picard et al . , 2019; Zanos and Gould , 2018 ) . Further , on a relative longer timescale , increases in cortical dendritic spine density , induced by ketamine , maintain behavioral effects after ketamine clearance in vivo ( Li et al . , 2010; Moda-Sava et al . , 2019 ) . The enhancement in cortical spinogenesis requires mTORC1 , BDNF , subtypes of glutamate receptors , as well as dopaminergic signaling ( Ali et al . , 2020; Duman et al . , 2012; Li et al . , 2010; Liu et al . , 2013; Sarkar and Kabbaj , 2016; Wu et al . , 2021a ) . Optogenetic stimulation of mPFC neurons reverses depressive-like behaviors ( Covington et al . , 2010; Fuchikami et al . , 2015; Kumar et al . , 2013 ) . Evidence from this study and others has consistently demonstrated that the activation of mPFC Drd1 neurons is important for ketamine’s behavioral effects . Consistent with diverse outputs of mPFC , local ketamine actions there may subsequently change the activity of multiple downstream brain regions , including the VTA , but also basolateral amygdala , nucleus accumbens ( NAc ) , and lateral habenula , among others . These brain regions likely contribute to specific behavioral effects of ketamine observed in different behavioral paradigms ( Belujon and Grace , 2014; Hare et al . , 2019; Kumar et al . , 2013; Yang et al . , 2018a ) . Besides modulating mPFC Drd1+ neurons , DA signaling in many brain regions regulates escape actions and avoidance behaviors . For example , the release and postsynaptic function of DA in the NAc and amygdala have been extensively studied in conditioned avoidance learning ( Antunes et al . , 2020; Darvas et al . , 2011; McCullough et al . , 1993; Oleson et al . , 2012; Oleson and Cheer , 2013; Stelly et al . , 2019; Wenzel et al . , 2018 ) . In this study , we found that VTA DA activity is not necessarily critical for innate escape actions , since chemogenetic suppression of VTA DA activity does not increase failures to escape from shock stimuli in the absence of prior learning . Importantly , after LH , the responses of VTA DA neurons to both inescapable and escapable shocks are significantly changed . Since DA release has been shown to be highly correlated with somatic DA neuronal activity ( de Jong et al . , 2019; Lee et al . , 2021; Patriarchi et al . , 2018 ) , we predict that downstream DA release in the NAc and amygdala is also modulated after LH and ketamine treatment . Which specific projections carry information about different aversive stimuli ( e . g . , conditioned cues or unpredicted aversive events ) and whether they derive from the same or different sub-populations of VTA DA neurons , remains an active area of research ( Beier et al . , 2015; de Jong et al . , 2019; Lammel et al . , 2011; Lammel et al . , 2012; Yang et al . , 2018b ) . VTA DA neurons are highly heterogenous based on their input/output anatomy and transcriptional profiles ( Beier et al . , 2015; Morales and Margolis , 2017 ) . Prior studies have demonstrated that VTA DA neurons respond to foot shocks differently , based on the brain regions they project to ( de Jong et al . , 2019 ) . In this study , we do not distinguish responses of DA neurons based on their projections or transcriptional profiles . Thus , the observed effects are sufficiently powerful to be seen on the background of a mixed DA population that maintains projection target diversity . Given the absence of local ketamine effects on VTA DA neurons , as shown in ex vivo calcium imaging and electrophysiological recordings , along with in vivo infusion experiments , the modulation of VTA DA activity by ketamine in vivo must depend on changes in the activity of projections from other brain regions . Based on evidence from electron microscopy reconstructions and anatomical tracing , VTA DA neurons that project to mPFC receive inputs from mPFC pyramidal neurons , forming excitatory connections ( Beier et al . , 2015; Carr and Sesack , 2000 ) . Trans-synaptic tracing analyses and optogenetic assays provide evidence that mPFC pyramidal neurons send excitatory glutamatergic projections to VTA ( Beier et al . , 2015; Lodge , 2011; Xiao et al . , 2018; Yang et al . , 2018b ) . Prior data from our lab ( Xiao et al . , 2018 ) and others ( Beier et al . , 2015; Carr and Sesack , 2000; Geisler et al . , 2007 ) confirm that VTA DA neurons receive excitatory mPFC inputs . Glutamate release from mPFC terminals elicits excitatory responses in VTA DA neurons ( Gariano and Groves , 1988; Kumar et al . , 2013; Tong et al . , 1996; You et al . , 2007 ) , driving synchronized activity across VTA and mPFC , as well as other limbic structures ( Kumar et al . , 2013 ) . Although the mPFC and VTA are the not only brain regions that control escape behaviors , the enhanced activity within these two brain regions is likely one important mechanism underlying behavioral effects of ketamine . Our data show that local mPFC ketamine infusion rescues both VTA DA activity and behavior ( Figure 5 ) . In vivo ketamine treatment rapidly enhances Drd1 neuronal activity in mPFC ( Figure 6—figure supplement 1 ) and chemogenetic activation of mPFC Drd1+ neurons rescues escape behaviors ( Figure 6g ) . Moreover , in recently published work , we show that DA signaling through Drd1 activation is necessary for ketamine’s action on restoring glutamate-evoked dendritic spinogenesis in mPFC ( Wu et al . , 2021a ) , which has been shown to maintain the lasting behavioral effects of ketamine after corticosterone treatment ( Moda-Sava et al . , 2019 ) . Chemogenetic activation of Gαs-coupled signaling cascades downstream of Drd1 activation in mPFC also rescues escape behavior after LH ( Wu et al . , 2021a ) . Based on the sum of these data across multiple laboratories and experimental modalities , we hypothesize that the initial enhancement of activity induced by ketamine in mPFC may be amplified within recurrent mPFC-VTA circuit , which may help extend ketamine behavioral effects beyond its in vivo bioavailability . Future studies are necessary to explicitly test this model . Other potential circuit mechanisms for ketamine’s enhancement of VTA DA activity may involve ventral hippocampal inputs to nucleus accumbens and lateral habenula projections to the VTA , both of which are modulated by stressful experience and ketamine ( Bagot et al . , 2015; Belujon and Grace , 2014; Li et al . , 2011; Pignatelli et al . , 2020; Yang et al . , 2018a ) . One such potential mechanism could involve actions in the lateral habenula , which sends glutamatergic inputs to GABAergic neurons that suppress VTA DA activity ( Gonçalves et al . , 2012 ) . Excitatory synapses onto VTA-projecting lateral habenula neurons are potentiated in rats in behavioral and genetic models of LH ( Li et al . , 2011 ) . Ketamine could disinhibit VTA DA neurons by suppressing the activity of lateral habenula glutamatergic neurons ( Yang et al . , 2018a ) . Rapid and lasting behavioral effects of ketamine likely result from synergistic actions across many brain regions , involving multiple molecular pathways . Drugs that modulate the dopaminergic system represent first line therapies for a large number of diverse neurological and mental health conditions that are comorbid with major depressive disorder , including Parkinson’s disease , schizophrenia , OCD , ADHD , and eating disorders ( Biederman et al . , 1998; Corcos et al . , 2000; Masellis et al . , 2003; Mayeux et al . , 1981; Siris , 2000 ) . The recent FDA approval of esketamine for the treatment of major depressive disorder is a major expansion of its clinical use . Our results on dopaminergic mediation of ketamine’s effects on behavior and plasticity suggest the possibility of ketamine’s differential clinical effects in patients receiving exogenous DA precursors ( e . g . L-DOPA ) , re-uptake inhibitors , and dopaminergic receptor agonists and antagonists . Animals were handled according to protocols approved by the Northwestern University Animal Care and Use Committee . Weanling and young adult male and female mice ( postnatal days 40–80 ) were used in this study . Approximately equal numbers of males and females were used for every experiment . All mice were group-housed , with standard feeding , 12 h light and 12 hr dark cycle ( 6:00 or 7:00 lights on ) , and enrichment procedures . Littermates were randomly assigned to conditions . C57BL/6 mice used for breeding and backcrossing were acquired from Charles River ( Wilmington , MA ) , and all other mouse lines were acquired from the Jackson Laboratory ( Bell Harbor , ME ) and bred in house . B6 . SJL-Slc6a3tm1 . 1 ( cre ) Bkmn/J mice , which express Cre recombinase under control of the dopamine transporter promoter , are referred to as DATiCre ( Bäckman et al . , 2006 ) ; B6 . FVB ( Cg ) -Tg ( Drd1-cre ) FK150Gsat/Mmucd mice , which express Cre recombinase under control of the dopamine Drd1a receptor promoter , are referred to as Drd1Cre ( FK150 ) ( Gong et al . , 2007 ) . All transgenic animals were backcrossed to C57BL/6 for several generations . Heterozygous Cre+ mice were used in experiments . Standard genotyping primers are available on the Jackson Lab website . Conditional expression of target genes in Cre-containing neurons was achieved using recombinant adeno-associated viruses ( AAVs ) using the FLEX cassette or encoding a double-floxed inverted open-reading frame ( DIO ) of target genes , as described previously ( Kozorovitskiy et al . , 2015 ) . For fiber photometry and ex vivo two-photon calcium imaging experiments in the VTA , DATiCre mice were transduced with AAV1 . CAG . FLEX . GCaMP6f . WPRE-SV40 ( 1 . 33 × 1013 GC/ml ) from the UPenn viral core ( Philadelphia , PA , a gift from the Genetically Encoded Neuronal Indicator and Effector Project ( GENIE ) and Douglas Kim; Addgene viral prep #100835-AAV1 ) ( Chen et al . , 2013 ) or AAV8 . CAG . FLEX . EGFP ( 3 . 1 × 1012 GC/ml , UNC vector core , Dr . Ed Boyden ) . For fiber photometry experiments in the mPFC , Drd1Cre ( FK150 ) mice were transduced with AAV1 . CAG . FLEX . GCaMP6f . WPRE . SV40 ( 1 . 33 × 1013 GC/ml ) . For chemogenetic experiments in mPFC , Drd1Cre ( FK150 ) mice were transduced with AAV2 . hSyn . DIO . hM3Dq . mCherry ( 8 . 6 × 1012 GC/ml , Addgene viral prep #44361-AAV2 , Dr . Bryan Roth ) ( Krashes et al . , 2011 ) . For chemogenetic experiments in VTA , DATiCre mice were transduced with a custom built AAV1 . CBA . DIO . hM4Di . mCherry ( 1 . 28 × 1013 GC/ml , Vigene Biosciences , Rockville , MD , plasmid a gift from Dr . Bernardo Sabatini ) ( Hou et al . , 2016 ) . For retrograde labeling , mice were intracranially injected with Red Retrobeads ( Lumafluor Inc ) in VTA . Neonatal viral transduction was carried out to minimize invasiveness and increase surgical efficiency ( Bariselli et al . , 2016; He et al . , 2018; Kozorovitskiy et al . , 2015; Kozorovitskiy et al . , 2012; Peixoto et al . , 2016 ) . P3-6 mice were cryoanesthetized , received ketoprofen for analgesia and were placed on a cooling pad . Virus was delivered at a rate of 100 nl/min for up to 150–200 nl using an UltraMicroPump ( World Precision Instruments , Sarasota , FL ) . Medial prefrontal cortex ( mPFC ) was targeted in the neonates by directing the needle immediately posterior to the eyes , 0 . 3 mm from midline , and 1 . 8 mm ventral to skin surface . Ventral tegmental area ( VTA ) was targeted in the neonates by directing the needle approximately ±0 . 2 mm lateral from Lambda and 3 . 8 mm ventral to skin surface . Coordinates were slightly adjusted based on pup age and size . Following the procedure , pups were warmed on a heating pad and returned to home cages . For intracranial injections , mice were anesthetized with isoflurane ( 3 % for induction value , 1 . 5%–2% for maintenance ) or ketamine:xylazine ( 100:12 . 5 mg/kg b . w . ) , received ketoprofen for analgesia , and were placed on a small animal stereotax frame ( David Kopf Instruments , Tujunga , CA ) . AAVs or retrobeads were delivered through a pulled glass pipette at a rate of 100–150 nl/min using an UltraMicroPump ( World Precision Instruments , Sarasota , FL ) . Ketamine , ACSF , or Chicago Sky Blue 6B were delivered intracranially at a rate of 150 nl/min using the UltraMicroPump . Injection coordinates for VTA , 2 . 8 mm posterior to bregma , 0 . 4 mm lateral , and 4 . 3–4 . 5 mm below the pia; for mPFC , 2 . 3 mm anterior to bregma , 0 . 4 mm lateral , and 1 . 3–1 . 6 mm below the pia; for PAG , 4 . 2 mm posterior to bregma , 0 . 1 mm lateral , and 2 . 2 below the pia; for PPTg , 4 . 5 mm posterior to bregma , 1 . 2 mm lateral , and 2 . 5 below the pia . Pipettes were held at the injection location for 15 min following AAV or retrobead release . Coordinates were slightly adjusted based on mouse age and size . For photometry fiber placement in the VTA or mPFC , mice were implanted with a 400 µm diameter 0 . 48 NA single mode optical fiber ( Doric lenses , Quebec City , QC , Canada ) directly above the VTA at –2 . 8 mm ( AP ) ; + 0 . 4 mm ( ML ) ; 4 . 3–4 . 5 mm ( DV ) , mPFC +2 . 0 mm ( AP ) ; + 0 . 4 mm ( ML ) ; 1 . 3–1 . 6 mm ( DV ) , 4–6 weeks after viral transduction . Behavioral experiments were conducted 7–12 days after implantation . Real-time photometry recording was performed during optical fiber implant above the VTA , for optimal targeting . When the fiber tip approached the VTA region , a continuous increase of fluorescence intensity was observed . The final position of implantation was determined by the cessation of further increases in fluorescence intensity . Mice recovered for at least 7 days after implantation . Hardware was created based on open-source resources made available by Dr . Thomas Davidson ( https://drive . google . com/drive/folders/0B7FioEJAlB1aNmdPOEsxTjhxajg ) ( Lerner et al . , 2015 ) . Briefly , a custom-built setup was created combining Doric fluorescence mini-cube ( Doric , Westport , CT ) and a 2151 Femtowatt photoreceiver with a lensed adapter ( Newport , Irvine , CA ) . All downstream hardware including fiberoptic cannuale and patch cords , except for LEDs and drivers ( Thorlabs , Newton , NJ ) , is readily available from Doric . A conventional single-cell electrophysiology recording system ( DAQ+ software ) was used to acquire signal and drive the LED , with a modified version of MATLAB based Scanimage ( Pologruto et al . , 2003 ) adapted for electrophysiology recordings by Dr . Bernardo Sabatini ( https://github . com/bernardosabatinilab ) . Signals were sampled at 1 kHz and downsampled to 250 Hz for time-locked and to 10 Hz for non-time-locked analyses . Fluorescence signal was baseline adjusted in non-overlapping 100 sec windows as ( signal-median ( signal ) ) / median ( signal ) , denoted as dF/F . Recordings were made during a subset of the sessions of LH and open-field locomotion , as noted in the text . Additional analyses details and link to analysis code are below . Coronal brain slice preparation was modified from previously published procedures ( Kozorovitskiy et al . , 2015; Kozorovitskiy et al . , 2012; Xiao et al . , 2017 ) . Animals were deeply anesthetized by inhalation of isoflurane , followed by a transcardial perfusion with ice-cold , oxygenated artificial cerebrospinal fluid ( ACSF ) containing ( in mM ) 127 NaCl , 2 . 5 KCl , 25 NaHCO3 , 1 . 25 NaH2PO4 , 2 . 0 CaCl2 , 1 . 0 MgCl2 , and 25 glucose ( osmolarity 310 mOsm/L ) . After perfusion , the brain was rapidly removed , and immersed in ice-cold ACSF equilibrated with 95%O2/5%CO2 . Tissue was blocked and transferred to a slicing chamber containing ice-cold ACSF , supported by a small block of 4 % agar ( Sigma-Aldrich ) . Bilateral 250 or 300 µm-thick slices were cut on a Leica VT1000s ( Leica Biosystems , Buffalo Grove , IL ) in a rostro-caudal direction and transferred into a holding chamber with ACSF , equilibrated with 95%O2/5%CO2 . Slices were incubated at 34°C for 30 min prior to electrophysiological recording and two-photon calcium imaging . Slices were transferred to a recording chamber perfused with oxygenated ACSF at a flow rate of 2–4 ml/min at room temperature . To assess spontaneous firing rate of dopaminergic neurons in the VTA , cell-attached recordings were performed . Cell-attached recording electrode pipettes were filled with the internal solution for voltage clamp recordings to monitor spontaneous break in , with pipette resistance varying between 3 and 7 MΩ . To assess spontaneous synaptic inputs ( sEPSC/sIPSC ) , voltage clamp recordings were performed . Dopaminergic neurons were identified by the expression of mCherry or GFP in DATiCre mice . Recording electrodes contained the following ( in mM ) : Cell-attached recordings and voltage clamp for sEPSCs: 120 CsMeSO4 , 15 CsCl , 10 HEPES , 10 Na- phosphocreatine , 2 MgATP , 0 . 3 NaGTP , 10 QX314 , and 1 EGTA ( pH 7 . 2–7 . 3 , ~ 295 mOsm/L ) ; Voltage clamp for sIPSCs ( high chloride internal solution ) : 100 CsCl , 35 CsF , 4 MgCl2 , 10 HEPES , 10 Na-phosphocreatine , 4 MgATP , 0 . 4 Na2GTP , and 1 EGTA ( pH 7 . 2 , 295 mOsm/L ) . In voltage clamp recordings , cells were held at –70 mV . Recordings were made using 700B amplifiers ( Axon Instruments , Union City , CA ) ; data were sampled at 10 kHz and filtered at 4 kHz with a MATLAB-based acquisition script ( MathWorks , Natick , MA ) . Offline analysis of electrophysiology data was performed using MATLAB ( Mathworks , Natick , MA ) , and Clampfit 11 . 2 ( Molecular Devices , San Jose , CA ) . Calcium sensor imaging was accomplished on a custom-built microscope combining two-photon laser-scanning microscopy ( 2PLSM ) , as previously described ( Banala et al . , 2018; Kozorovitskiy et al . , 2015; Kozorovitskiy et al . , 2012; Xiao et al . , 2018; Xiao et al . , 2017 ) . A mode-locked Ti:Sapphire laser ( Mai Tai eHP DeepSee , Spectra-Physics , Santa Clara , CA ) was tuned to 910 nm for GCamp6f . The intensity of the laser was controlled by Pockels cells ( Conoptics , Danbury , CT ) . A modified version of Scanimage software was used for data acquisition ( Pologruto et al . , 2003 ) . Calcium imaging of GCaMP6f expressing DA neurons in acute brain slices of the VTA was done at 910 nm and sampled at 12 Hz . Spontaneous activity was imaged for 5 min in the baseline at 34°C , followed by a 5 min recording after ex vivo ketamine application ( 50 μM , with 15–40 min delay ) . Pharmacological agents were acquired from Vedco ( St . Joseph , MO ) or Sigma-Aldrich ( St . Louis , MO ) . In vivo injections included intraperitoneal and subcutaneous injections of ketamine ( 10 mg/kg , Vedco , St . Joseph , MO ) and Clozapine N-oxide ( 3 mg/kg in vivo , 1 μM in vitro , Sigma-Aldrich ) ; intracranial injections of ketamine ( 12 . 5 µg in 500 nl ACSF ) and Chicago Sky Blue 6B ( 50 µg in 500 nl ACSF , Tocris , Bristol , United Kingdom ) . Ex vivo applications included SR95531 ( Gabazine , 10 μM , Tocris ) , CNQX ( 10 μM , Tocris ) , and ketamine ( 50 μM ) . Mice were deeply anesthetized with isoflurane and transcardially perfused with 4% paraformaldehyde ( PFA ) in 0 . 1 M phosphate buffered saline ( PBS ) . Brains were post-fixed for 1–5 days and washed in PBS , prior to sectioning at 50–100 µm on a vibratome ( Leica Biosystems ) . Sections were pretreated in 0 . 2% Triton X-100 for an hour at RT , then blocked in 10% bovine serum albumin ( BSA , Sigma-Aldrich , ST Louis , MO ) :PBS with 0 . 05 % Triton X-100 for 2 hr at RT , and incubated for 24–48 hr at 4°C with primary antibody solution in PBS with 0 . 2 % Triton X-100 . On the following day , tissue was rinsed in PBS , reacted with secondary antibody for 2 hr at RT , rinsed again , then mounted onto Superfrost Plus slides ( ThermoFisher Scientific , Waltham , MA ) . Sections were dried and coverslipped under ProLong Gold antifade reagent with DAPI ( Molecular Probes , Life Technologies , Carlsbad , CA ) or under glycerol:TBS ( 9:1 ) with Hoechst 33342 ( 2 . 5 µg/ml , ThermoFisher Scientific ) . Primary antibodies used in the study were rabbit anti-tyrosine hydroxylase ( 1:1000; AB152 , Millipore , Burlington , MA ) , mouse anti-tyrosine hydroxylase ( 1:1000; AB129991 , Abcam , Cambridge , UK ) , rabbit anti-RFP ( 1:500 , 600-401-379 , Rockland , Limerick , PA ) , and rabbit anti-c-Fos ( 1:5000; Synaptic Systems , Goettingen , Germany ) . Alexa Fluor 488- , Fluor 594- , or Fluor 647-conjugated secondary antibodies against rabbit or mouse ( Life Technologies , Carlsbad , CA ) were diluted 1:500 . Whole sections were imaged with an Olympus VS120 slide scanning microscope ( Olympus Scientific Solutions Americas , Waltham , MA ) . Confocal images were acquired with a Leica SP5 confocal microscope ( Leica Microsystems ) . Depth-matched z-stacks of 2 µm-thick optical sections were analyzed in ImageJ ( FIJI ) ( Schindelin et al . , 2012; Schneider et al . , 2012 ) . For c-fos quantification , every four adjacent z stack slices were combined , for a total of 6 µm thickness . mCherry signal was used to localize cell bodies of hM3Dq-expressing neurons . Laser intensity and all imaging parameters were held constant across samples , and the same threshold was applied for subtracting background immunofluorescence . C-fos+ neurons were identified by an experimenter blind to the conditions . Quantitative fluorescence in situ hybridization ( FISH ) was conducted following previously published procedures ( Xiao et al . , 2018; Xiao et al . , 2017 ) . Mice were deeply anesthetized by inhalation of isoflurane and decapitated . Brains were quickly removed and frozen in tissue-freezing medium on a mixture of dry ice and ethanol for 5–15 min prior to storage at 80°C . Brains were subsequently cut on a cryostat ( Leica CM1850 , Leica Biosystems ) into 20-µm-thick sections , adhered to Superfrost Plus slides , and frozen at 80°C . Samples were fixed with 4% PFA in 0 . 1 M PBS at 4°C for 15 min , processed according to the manufacturer’s instructions in the RNAscope Fluorescent Multiplex Assay manual for fresh frozen tissue ( Advanced Cell Diagnostics , Newark , CA ) , and coverslipped with ProLong Gold antifade reagent with DAPI ( Molecular Probes ) . Drd1a receptor channel 2 ( Drd1a ) or Slc6a3 ( Dat ) probes were added to slides in combination , and Amp4-b fluorescent amplification reagent was used for all experiments . Sections were subsequently imaged on a Leica SP5 confocal microscope in four channels with a ×40 objective lens at a zoom of 1 . 4 and resolution of 512 × 512 pixels with 1 . 0 µm between adjacent z sections . FISH images were analyzed using FIJI ( Schindelin et al . , 2012 ) . Briefly , every four adjacent z stack slices were combined , for a total of 3 µm thickness , in order to minimize missed colocalization , while decreasing false positive colocalization driven by signal from cells at a different depth in a similar x-y position . All channels were thresholded . Cellular ROIs were defined using the retrobead+ channel information to localize cell bodies . FISH molecule puncta were counted within established cell boundaries . Whether a cell was considered positive for a given marker was determined by setting a transcript-dependent threshold of the number of puncta ( e . g . over five puncta/soma for Drd1a+ ) . These stringent parameters for co-localization and the challenges of quantifying low abundance receptor transcripts likely lead to underestimation of receptor-positive populations . For evaluating locomotor behavior , Toxtrac ( Rodriguez et al . , 2018 ) was used to track the animal’s position , defined by its body center position , and quantify the distance travelled in each session . To detect motion onset/offset , movement was defined by the body center moving >30 mm/s for at least 0 . 5 s . The associated Ca2+ transients were then aligned to transitions between motion start and stop times and averaged across all animals . Video was recorded and the mouse was tracked with Toxtrac . Three state transition models were constructed for a given condition and behavioral response , by counting the occurrence of each of the three possible 2-response sequences in all animals and then dividing by the total number of the given response . For example , an escape can be followed by another escape , a failure , or a premature trial . The probability of a failure following an escape was calculated by counting the number of failures that followed escapes and dividing it by the total number of escape trials . The graphs for the transition models were constructed using GraphViz ( Gansner and North , 2000 ) . The similarity between transition models was calculated as one minus the Frobenius norm ( the Euclidean norm of a matrix ) of the difference between models ( 1-norm ( modela- modelb ) ) . To depict individual learning trajectories , starting at zero , trajectory value incremented by one for each escape trial , decremented by one for each failure trial , and kept constant for spontaneous transitions in the shuttle box . For the analysis of distributions of % failure to escape , the density histograms of percentage of failures were generated with a bin width = 4 . Density histograms were fitted with the Gaussian distribution using fitdistrplus R package using a maximum likelihood estimation . The heatmap of Ca2+ transients was constructed by plotting single trial transients with signal amplitude depicted by color . The latency to peak was defined as the time from shock onset to the maximum Ca2+ transient value within 8 s of shock onset . The negative and positive area under the curve ( AUC ) was calculated based on the direction of the peaks . Peaks were omitted from analyses if they were less than 5% of the distance from minimum to maximum dF/F or less than three times the standard deviation of the baseline period . The baseline was chosen by using the mean dF/F of the 5 s segment before the shock onset . For behavioral response-specific Ca2+ transients , the transients for each behavioral response were averaged within a single subject , and then these data were averaged across all animals . The distance between escape and failure traces for each animal was computed as the Euclidean norm of the difference between the average of all their Ca2+ transients on escape trials and the average of all their Ca2+ transients on failure trials ( \|\|mean ( escape trials ) -mean ( failure trials ) \|\|2 ) . ROIs of dopaminergic neuron somata were defined manually . Raw fluorescence intensity for all frames during 5-min recording sessions was extracted using FIJI ( Schindelin et al . , 2012 ) . For each neuron , the power spectral density of its baseline adjusted fluorescence was computed using Welch’s method ( Welch , 1967 ) with a Hann window and 50% overlap . Required sample sizes were estimated based on previous publications and past experience . The number of replicates was reported , and several internal replications are present in the study . No data were excluded after analyses . Animals were randomly assigned to treatment groups . Group statistical analyses were done using GraphPad Prism 7 and 8 software ( GraphPad , LaJolla , CA ) . For N sizes , the number of trials or cells recorded , as well as the number of animals are provided . All data are expressed as mean ± SEM , or individual plots . Probabilities are expressed as aggregate probabilities within individuals . For two-group comparisons , statistical significance was determined by two-tailed Student’s t-tests . For multiple group comparisons , one-way or two-way analysis of variance ( ANOVA ) tests were used for normally distributed data , followed by post hoc analyses . For non-normally distributed data , non-parametric tests for the appropriate group numbers were used . Pearson regression was used to detect the correlation between two groups of data . p < 0 . 05 was considered statistically significant . All computer code generated in the current study is available at https://github . com/KozorovitskiyLaboratory/Wu_et_al_2021 ( Wu et al . , 2021b; copy archived at swh:1:rev:459ff15a97a2af4e955d20531235b399b4be5b22 ) .	Over 264 million people around the world suffer from depression , according to the World Health Organization ( WHO ) . Depression can be debilitating , and while anti-depressant drugs are available , they do not always work . A small molecule drug mainly used for anesthesia called ketamine has recently been shown to ameliorate depressive symptoms within hours , much faster than most anti-depressants . However , the molecular mechanisms behind this effect are still largely unknown . Most anti-depressant drugs work by restoring the normal balance of dopamine and other chemical messengers in the brain . Dopamine is released by a specialized group of cells called dopaminergic neurons , and helps us make decisions by influencing a wide range of other cells in the brain . In a healthy brain , dopamine directs us to rewarding choices , while avoiding actions with negative outcomes . During depression , these dopamine signals are perturbed , resulting in reduced motivation and pleasure . But it remained unclear whether ketamine’s anti-depressant activity also relied on dopamine . To investigate this , Wu et al . used a behavioral study called “learned helplessness” which simulates depression by putting mice in unavoidable stressful situations . Over time the mice learn that their actions do not change the outcome and eventually stop trying to escape from unpleasant situations , even if they are avoidable . The experiment showed that dopaminergic neurons in an area of the brain that is an important part of the “reward and aversion” system became less sensitive to unpleasant stimuli following learned helplessness . When the mice received ketamine , these neurons recovered after a few hours . Individual mice also responded differently to ketamine . The most ‘resilient’ , stress-resistant mice , which had distinct patterns of dopamine signaling , also responded most strongly to the drug . Genetic and chemical manipulation of dopaminergic neurons confirmed that ketamine needed intact dopamine signals to work , and revealed that it acted indirectly on dopamine dynamics via another brain region called the medial prefrontal cortex . These results shed new light on how a promising new anti-depressant works . In the future , they may also explain why drugs like ketamine work better for some people than others , ultimately helping clinicians select the most effective treatment for individual patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2021	Attenuated dopamine signaling after aversive learning is restored by ketamine to rescue escape actions
Almost all critical functions in cells rely on specific protein–protein interactions . Understanding these is therefore crucial in the investigation of biological systems . Despite all past efforts , we still lack a thorough understanding of the energetics of association of proteins . Here , we introduce a new and simple approach to predict binding affinity based on functional and structural features of the biological system , namely the network of interfacial contacts . We assess its performance against a protein–protein binding affinity benchmark and show that both experimental methods used for affinity measurements and conformational changes have a strong impact on prediction accuracy . Using a subset of complexes with reliable experimental binding affinities and combining our contacts and contact-types-based model with recent observations on the role of the non-interacting surface in protein–protein interactions , we reach a high prediction accuracy for such a diverse dataset outperforming all other tested methods . Interactions between proteins play a central role in the processes happening in the cells , from DNA replication to protein degradation ( Jones and Thornton , 1996; Alberts , 1998; Nooren and Thornton , 2003; Perkins et al . , 2010 ) . Perturbations in those interaction networks can lead to disease ( Stites , 1997; Sugiki et al . , 2014 ) . Characterizing these protein–protein interactions ( PPIs ) is therefore crucial for a proper understanding of mechanisms in biological processes for disease research and for drug development , as most common targets of drugs are proteins ( such as enzymes , ion channels , and receptors ) . Exploring recognition processes at atomic level requires knowledge of the three-dimensional ( 3D ) structure of the associated molecular complexes . It is however the binding affinity ( BA ) ( i . e . , the natural inclination of molecules to associate ) that defines whether or not complex formation will occur . The BA is therefore the key for understanding and predicting recognition , association and dysfunction phenomena related to protein complexes . It has been shown that changes in BA caused by mutations or post-translational modification errors lead to various diseases ( Vidal et al . , 2011 ) . Commonly , the affinity of an interaction is described through the equilibrium dissociation constant Kd , or , in thermodynamic terms , the Gibbs free energy ΔG ( ΔG = RT ln Kd ) . Measuring Kd values experimentally is a time-consuming and expensive process . Many computational methods aimed at predicting BA have been developed . Gaining the ability to predict BA is indeed an urgent need as it offers great opportunities not only to control interactions and develop innovative therapeutics ( Keskin et al . , 2005; Aloy and Russell , 2006; Beltrao et al . , 2007; Kiel et al . , 2008; Dell'Orco , 2009 ) , but also for other applications such as protein engineering ( Kortemme et al . , 2004; Sharabi et al . , 2011 ) , computational mutagenesis ( Ben-Shimon and Eisenstein , 2010 ) , and docking ( Halperin et al . , 2002 ) . Different methods aimed at predicting BA have been proposed throughout the years , varying greatly in terms of accuracy and computational cost . Exact methods such as free energy perturbation and thermodynamics integration can be very accurate , but due to their computational costs their application is extremely limited ( mostly to low throughput studies and mainly for small drug binding or mutations ) . Methods based on empirical functions ( empirical , force-field-based potentials , statistical potentials , and scoring functions used in docking ) are much faster ( Jiang et al . , 2002; Ma et al . , 2002; Zhang et al . , 2005; Audie and Scarlata , 2007; Su et al . , 2009; Bai et al . , 2011; Moal et al . , 2011; Qin et al . , 2011; Moal and Bates , 2012; Tian et al . , 2012; Kastritis et al . , 2014; Luo et al . , 2014 ) . However , even if some have been very successful on small training sets ( Horton and Lewis , 1992; Audie and Scarlata , 2007 ) , most published models still fail to systematically predict BA ( Kastritis and Bonvin , 2010 ) for large datasets or discriminate between binders and non-binders ( Sacquin-Mora et al . , 2008; Fleishman et al . , 2011 ) . The main weaknesses of these methods are that they usually neglect factors such as conformational changes upon binding , allosteric regulation , and solvent and co-factor effects , which may all contribute to the binding strength . Binding between two proteins is mainly defined by their contact region , the interface , and it is indeed the network of contacts between surface residues that holds complexes together , defines their specificity and contributes to their interaction strength . The importance of such inter-residue contacts has already been established in docking . In the Critical Assessment of Prediction of Interactions ( CAPRI ) experiment ( Janin et al . , 2003 ) , for instance , assessment of the accuracy of the docked models is based on a combination of positional root mean square deviation ( RMSD ) criteria and conservation of intermolecular contacts with respect to the native structure ( Lensink et al . , 2007 ) . In the context of scoring , considering the conservation of contacts at the interface across docking models has been shown to improve the ranking of docked models ( Oliva et al . , 2013; Chermak et al . , 2014 ) . The atom contact frequency in a set of predictions , a similar concept , has also been included in the ZRANK docking pipeline ( Hwang et al . , 2010 ) . Next to their use in scoring , contacts have been introduced as a way to cluster docking solutions based on the fraction of common inter-residue contacts among a set of decoys ( Rodrigues et al . , 2012 ) . However , in addition to properties of the interface , a recent work has also demonstrated an effect of the non-interacting surface ( NIS ) on BA ( Kastritis et al . , 2014 ) , a finding that has been corroborated in a recent study by Marillet et al . ( Marillet et al . , 2015 ) . Here we show how the network of contacts at the interface of a protein–protein complex can help in describing the BA of the interaction . Based on the number contacts at the interface , we propose an innovative and very simple method to predict BA . To this end we used the protein–protein BA benchmark ( Kastritis et al . , 2011 ) consisting of 144 non-redundant protein–protein complexes with experimentally determined Kd ( ΔG ) and available 3D structures . Our results show that interfacial contacts , which have so far been neglected in the rationalization and prediction of BA , can be considered the best structural property to describe binding strength . Based on this observation , we describe an extremely simple BA predictor that accounts for different types of contacts and shows the best performance reported so far ( to our knowledge ) on such a large and diversified dataset of complexes . Its performance is compared to other previously published predictor methods ( Moal et al . , 2011 ) . Further , we analyze the impact of the experimental method used to characterize BA on the prediction performance . In a protein complex the interactions are usually of relatively short range . Recently , however , Kastritis et al . ( 2014 ) revealed the unexpected contribution of NIS residues to BA . We therefore systematically evaluated the effect of the distance on defining the inter-residues network by varying the cut-off distance between 3 Å and 20 Å ( see ‘Materials and methods’ ) . For each distance cut-off the number of ICs was correlated with the experimental ΔGs . The results , reported in Figure 1 , show that the highest correlation is achieved at a cut-off of 4 . 0 Å , ( RICs = −0 . 50 , ρ < 0 . 0001 ) . This correlation decreases slowly until 8 . 0 Å ( RICs = −0 . 41 , ρ < 0 . 0001 ) and drops at higher distances . We also evaluated the ranking power of the ICs , expressed through the Spearman's correlation coefficient S , which follows the same trend as R with slightly higher absolute values ( Figure 1 ) . 10 . 7554/eLife . 07454 . 003Figure 1 . Correlation between number of inter-residue contacts and binding affinity ( ΔGs ) as a function of the distance cut-off used to calculate the contacts . Both the Pearson's R ( dark grey bars ) and the Spearman's S ( light grey-patterned bars ) correlation coefficient are reported . DOI: http://dx . doi . org/10 . 7554/eLife . 07454 . 003 Many different experimental methods can be used to determine the ΔG of a protein–protein complex . Each presents different characteristics so that the measured ΔG values depend on the method used , its sensitivity and the strength of the interactions that are being measured . For the set of 122 complexes used in this work , ten different experimental methods have been used to detect the ΔGs: stopped-flow fluorimetry ( 8 cases ) , surface plasmon resonance ( SPR ) ( 40 cases ) , high performance liquid chromatography ( HPLC ) /UV absorption spectroscopy ( 14 cases ) , sedimentation ( 1 case ) , radioligand binding ( competitive binding experiments ) ( 2 cases ) , potentiometry ( 1 case ) , reduction assay ( 1 case ) , isothermal titration calorimetry ( ITC ) ( 19 cases ) , inhibition assay ( 17 cases ) , fluorescence spectrophotometric assays ( 19 cases ) . We analyzed separately the BAs from the various experimental methods with enough data points ( ≥8 ) and compared the prediction performance with the full data set . As reported in Table 1 , the correlations between ICs and experimental ΔGs increased to RICs = −0 . 70 ( ρ = 0 . 03 ) , RICs = −0 . 53 ( ρ = 0 . 0003 ) , RICs = −0 . 65 ( ρ = 0 . 006 ) and RICs = −0 . 55 ( ρ = 0 . 006 ) in the case of ΔGs determined by stopped-flow fluorimetry , SPR , spectroscopic methods and ITC , respectively . For the 17 cases measured by inhibition assays and the 19 by fluorescence spectrophotometry techniques the correlations were meaningless ( RICs = 0 . 04 with ρ = 0 . 9 and RICs = 0 . 05 with ρ = 0 . 8 , respectively ) . These are indirect methods useful in calculating relative binding strengths ( known as IC50s ) , but these have limitations when used to calculate absolute BA values ( Lazareno and Birdsall , 1993; Wilkinson , 2004; Masi et al . , 2010 ) . 10 . 7554/eLife . 07454 . 004Table 1 . Pearson's correlations and p-values ( ρ ) between inter-residue contacts ( ICs ) and buried surface area ( BSA ) and experimental binding affinities ( ΔGs ) for the entire dataset and subsets corresponding to various experimental methodDOI: http://dx . doi . org/10 . 7554/eLife . 07454 . 004Class#ComplexesRICsRBSAAll122−0 . 50 ( ρ < 0 . 0001 ) −0 . 32 ( ρ = 0 . 002 ) Stopped-flow8−0 . 70 ( ρ = 0 . 03 ) −0 . 55 ( ρ = 0 . 08 ) SPR39−0 . 53 ( ρ = 0 . 0003 ) −0 . 44 ( ρ = 0 . 002 ) Spectroscopy14−0 . 65 ( ρ = 0 . 006 ) −0 . 27 ( ρ = 0 . 2 ) ITC20−0 . 55 ( ρ = 0 . 006 ) −0 . 64 ( ρ = 0 . 001 ) Inhibition assay170 . 05 ( ρ = 0 . 4 ) −0 . 08 ( ρ = 0 . 4 ) Fluorescence190 . 04 ( ρ = 0 . 4 ) 0 . 34 ( ρ = 0 . 1 ) The ICs were calculated for a 4 . 0 Å cut-off . Removing the cases from inhibition assays and fluorescence spectrophotometry methods , and all others for which only a few data points were reported ( potentiometry , radioligand , reduction assay , and sedimentation ) , we end up with a reliable dataset of 81 structures ( a ‘cleaned’ dataset ) showing an increased correlation of RICs = −0 . 59 ( ρ < 0 . 0001 ) at the re-optimized distance threshold of 5 . 5 Å to define a contact ( Figure 2 ) . All further results will therefore refer to the 5 . 5 Å cut-off to define ICs . 10 . 7554/eLife . 07454 . 005Figure 2 . Plots of inter-residue contacts ( ICs ) vs experimentally determined binding affinities ( ΔGs ) of protein–protein complexes . ( A ) Full dataset ( 122 complexes ) , ( B ) reliable experimental methods only ( stopped-flow , surface plasmon resonance , spectroscopy , isothermal titration calorimetry ) ( 81 complexes ) , and ( C ) non-reliable experimental methods ( inhibition assay and fluorescence ) ( 36 complexes ) . The trend line and corresponding Pearson correlation coefficients and p-values ( ρ ) are reported in each plot; binding affinities are reported as absolute values . DOI: http://dx . doi . org/10 . 7554/eLife . 07454 . 005 In order to assess which structural property might be the best descriptor for the binding strength , we calculated on the ‘cleaned’ dataset values for the widely used buried surface area ( BSA ) , the NIS characteristics ( recently shown to contribute in the BA ) ( Kastritis et al . , 2014 ) , and our ICs . We further classified these properties based on the amino acid type—polar/apolar—for BSA and NIS , and contact types—polar/polar , polar/charged , polar/apolar , charged/charged , charged/apolar , apolar/apolar—for the ICs . For the latter we also considered the hydrophobic/hydrophilic classification . For all these , we evaluated whether this finer classification ( resulting , of course , in more parameters in our model ) improves the correlations . It is clear from the results summarized in Table 2 that the number of ICs is a better descriptor than the BSA , with RICs_total = −0 . 59 ( ρ < 0 . 0001 ) vs RBSA_total = −0 . 46 ( ρ < 0 . 0001 ) . When distinguishing between the amino acid properties , the factors that contribute the most to BA are the number of ICs between polar and apolar residues ( R = −0 . 56 , ρ < 0 . 0001 ) and between hydrophilic residues ( R = −0 . 53 , ρ < 0 . 0001 ) . However , none of these individual classes shows better correlation than the total ICs and BSA . All the calculated data are provided in Supplementary file 2 . 10 . 7554/eLife . 07454 . 006Table 2 . Pearson's correlations and p-values between experimental binding affinities and the inter-residue contacts ( ICs ) , buried surface area ( BSA ) and non-interacting surface ( NIS ) ( Kastritis et al . , 2014 ) properties calculated on the ‘cleaned’ datasetDOI: http://dx . doi . org/10 . 7554/eLife . 07454 . 006PropertyRp-valueICs_total−0 . 59<0 . 0001ICs_charged/charged−0 . 17=0 . 06ICs_charged/polar−0 . 26=0 . 009ICs_charged/apolar−0 . 45<0 . 0001ICs_polar/polar−0 . 13=0 . 1ICs_polar/apolar−0 . 56<0 . 0001ICs_apolar/apolar−0 . 34=0 . 001ICs_hydrophilic/hydrophilic−0 . 53<0 . 0001ICs_hydrophilic/hydrophobic−0 . 34=0 . 001ICs_hydrophobic/hydrophilic−0 . 05=0 . 3BSA_total−0 . 46<0 . 0001BSA_polar−0 . 36=0 . 0005BSA_apolar−0 . 47<0 . 0001%NIS_polar0 . 07=0 . 06%NIS_apolar−0 . 33=0 . 001%NIS_charged0 . 28=0 . 006A fine classification of those properties based on the polar/apolar/charged and hydrophobic/hydrophilic nature of the amino acids is also reported . The property with the highest R value is highlighted in bold . The corresponding data are provided in Supplementary file 2 . In order to assess the predictor power of the above-described structural properties , we built different predictor models ( contacts-based , BSA-based , NIS-based and combinations of these ) , optimizing the following model: ( 1 ) Model N: ΔGcalc=w1P1+w2P2+ . . . . +Q , where PN values are the properties used to train Model N , wN values are the weight and Q is a shift value . To avoid the problem of over-fitting when many variables are used ( >3 ) , we applied the Akaike's Information Criterion ( AIC ) stepwise selection method implemented in R ( R Development Core Team , 2014 ) in order to identify ( and consider only ) the significant variables among the training ones . Each derived model , with associated weights wN and performance , is reported in Table 3 . 10 . 7554/eLife . 07454 . 007Table 3 . Optimization of binding affinity predictor models based on the regression model ΔGcalc = w1P1 + w2P2 + … . + QDOI: http://dx . doi . org/10 . 7554/eLife . 07454 . 007Properties ( PN ) Model 1Model 2Model 3Model 4Model 5Model 6ICs_total0 . 07782-----ICs_charged/charged--/--0 . 09420ICs_charged/polar--/--/ICs_charged/apolar--0 . 11627--0 . 10038ICs_polar/polar--−0 . 12655--−0 . 19522ICs_polar/apolar--0 . 23595--0 . 22609ICs_apolar/apolar--/--/ICs_hydrophil/hydrophil---0 . 09055--ICs_hydrophil/hydrophob---0 . 05726--ICs_hydrophob/hydrophil---0 . 06037--BSA_total-0 . 00278----BSA_polar----0 . 00131-BSA_apolar----0 . 00400-%NIS_polar-----/%NIS_apolar-----−0 . 18786%NIS_charged-----−0 . 13872Intercept ( Q ) 4 . 788395 . 660325 . 137664 . 904525 . 4480915 . 9433R−0 . 59−0 . 46−0 . 67−0 . 60−0 . 48−0 . 73p-value<0 . 0001<0 . 0001<0 . 0001<0 . 0001<0 . 0001<0 . 0001RMSE ( kcal mol−1 ) 2 . 252 . 462 . 082 . 222 . 451 . 89The weights wN are reported for each properties PN used to train Model N . Properties that have not been used for training the Model or have been evaluated as not relevant from the Akaike's An Information Criterion ( AIC ) evaluation are reported as ‘-’ and ‘/’ , respectively . At the bottom of the table , the correlation coefficient and prediction error ( expressed as R and RMSE ) are reported . Models 1 and 2 were trained on ICs_total and BSA_total , respectively , with a better performance of the ICs-based model ( as already reported in Table 2 ) ( root mean square errors [RMSEs] of 2 . 25 and 2 . 46 kcal/mol , for ICs and BSA , respectively ) . Models 3 and 4 have been trained using the ICs classified by residue type ( polarity in Model 3 , hydrophobicity in Model 4 ) . While single amino-acid-type ICs properties do not improve the correlations , their linear combination results in a significant improvement from R = −0 . 59 ( ρ < 0 . 0001 ) for Model 1 to R = −0 . 67 ( ρ < 0 . 0001 ) for Model 3 and R = −-0 . 60 for Model 4 ( ρ < 0 . 0001 ) . Model 5 has been trained on the polar/apolar classification of the Horton and Lewis BSA model , ( Horton and Lewis , 1992 ) , with a slightly improved performance compared with Model 2 based on the total BSA , but it is still worse than any of the contact-based models ( i . e . , Model 1 , Model 3 , and Model 4 ) . Among the models that are based on properties of the interface of the binding site , the one based on polarity-classification of ICs ( i . e . , Model 3 ) shows the best performance . We therefore added to it the NIS properties in order to obtain a full description of the structural properties of the complex , resulting in Model 6 . After AIC evaluation , we obtained the following linear equation: ( 2 ) ΔGcalc=0 . 09459 ICscharged/charged+0 . 10007 ICscharged_apolar−0 . 19577 ICspolar/polar+0 . 22671 ICspolar/apolar−0 . 18681%NISapolar−0 . 13810%NIScharged+15 . 9433 . Fourfold cross-validation results for this model ( repeated 10 times ) are reported in Supplementary file 3 , showing consistency in terms of coefficient and prediction accuracy . A scatter plot of predicted vs experimental affinities is reported in Figure 3 . The most relevant contributions to BA are the number of ICs made by charged and polar residues ( ICs_charged/charged and ICs_polar/polar in Equation 2 ) , while the apolar residues are only counted when interacting with charged and polar ones ( ICs_charged/apolar and ICs_polar/apolar in Equation 2 ) . This ICs/NIS-based model show the best performance of any model developed so far , with R = −0 . 73 and RMSE = 1 . 89 kcal mol−1 . 10 . 7554/eLife . 07454 . 008Figure 3 . Scatter plot of predicted vs experimental binding affinities . The predictions were made according to the inter-residue contacts ( ICs ) /non-interacting surface ( NIS ) -based model ( Model 6 , Equation 2 ) for the cleaned dataset of 81 protein–protein complexes . The correlation for all 81 complexes yields an R of −0 . 73 ( ρ < 0 . 0001 ) with a RMSE of 1 . 89 kcal mol−1 . When only rigid cases ( interface RMSD between superimposed free and bound components ≤1 . 0 Å , red triangles ) are considered , the correlation increases to R = −0 . 75 ( ρ < 0 . 0001 ) with a RMSE of 1 . 88 kcal mol−1 , while for flexible cases ( interface RMSD >1 . 0 Å; yellow rhombus ) R = −0 . 73 ( ρ < 0 . 0001 ) with a RMSE of 1 . 88 kcal mol−1 . The x = y line is shown as reference; binding affinities are reported as absolute values . DOI: http://dx . doi . org/10 . 7554/eLife . 07454 . 008 In many assemblies , the structure of free monomers differs from their structure in the oligomeric state ( the ‘bound’ form ) due to the association process . The affinity benchmark also reports the interface RMSD ( i_rmsd ) between the unbound and bound structures . This is a measure of how much conformational change takes place upon association . We investigated if our model would have a higher predictive power when classifying the complexes according to their amplitude of conformational change upon binding . Predictions made with our combined contacts and NIS model ( Model 6 , Equation 2 ) are much less sensitive to conformational changes than all previous models ( Figure 3 and Figure 4 ) , with only minor differences in performance between rigid ( i_rmsd ≤ 1 . 0 Å , R = −0 . 75 and RMSE = 1 . 88 kcal mol−1 ) and flexible cases ( i_rmsd > 1 . 0 Å , R = −0 . 73 and RMSE = 1 . 89 kcal mol−1 ) . This indicates that , in contrast to previous predictors , the number of interface residue contacts is a rather robust predictor that is less sensitive to conformational changes . 10 . 7554/eLife . 07454 . 009Figure 4 . Comparison of the performance of our ICs/NIS-based model ( Model 6 , Equation 2 ) with other predictor models reported by Moal et al . ( 2011 ) and the CCHarPPI ( Moal et al . , 2015a , b ) webserver . The performance is expressed as Pearson's Correlation coefficient between experimental and predicted binding affinities . Predictions were made on the common set of 79 complexes between our cleaned dataset , the data tested by Moal et al . ( 2011 ) and the CCHarPPI ( Moal et al . , 2015a , b ) pre-calculated data . Correlations for the entire set and the rigid ( 43 ) and flexible ( 36 ) complexes are reported as absolute values for easier comparison ( methods marked with asterisk showed original negative correlations ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07454 . 009 In order to perform a fair comparison with other previously published methods , we calculated their performance on the complexes that are in common between our clean dataset , the one reported by Moal et al . ( 2011 ) , and the pre-calculated data on the Computational Characterisation of Protein–Protein Interactions ( CCHarPPI ) web server , ending up in 79 protein–protein complexes ( Figure 4 ) . The considered models include the ‘global surface model’ of Kastritis et al . ( 2014 ) , the BSA-based model of Horton and Lewis ( 1992 ) , the top three best performing methods reported by Moal et al . ( 2011 ) ( their consensus model , DFIRE [Liu et al . , 2004] and PMF[Su et al . , 2009] ) and the composite scoring functions reported by the CCHarPPI webserver [Moal et al . , 2015a , b] , such as ZRANK [Pierce and Weng , 2007] , ZRANK2 [Pierce and Weng , 2008] , RosettaDock [Chaudhury et al . , 2010] , PyDock [Cheng et al . , 2007] , FireDock [Andrusier et al . , 2007] , PISA [Viswanath et al . , 2013] , PIE [Ravikant and Elber , 2010] , and SIPPER [Pons et al . , 2011] . As shown in Figure 4 , our ICs/NIS-based model ( Model 6 ) outperforms all other methods tested . It is also less sensitive to conformational changes . All associated data are provided in Supplementary file 4 . In addition to the composite scoring function of CCharPPI , none of the other 99 intermolecular parameters reported by CCharPPI outperformed our model , even if some show correlations above −0 . 50 . The ICs introduced in this work to describe BA seem to be a ‘higher level definition’ structural parameter than the BSA since they express not only the contact surface but also the pairwise non-bonded interactions that the two proteins make , which is related to the packing of the interface . Indeed , a weak complex is expected to be less tightly packed than a strong one , a difference that should be better reflected in the ICs than in the BSA . In particular , the origin of this difference might reside in the fact that the contribution of each residue to the BSA will greatly depend on the solvent-accessible surface area of the residue in the free form , whereas this does not affect ICs . To illustrate this point we checked the main differences between ICs and BSA for the complex between Fab D3H44 and Tissue factor ( PDB code: 1JPS [Faelber et al . , 2001]; ΔGexp = −13 . 6 kcal mol−1 ) . This complex has a BSA of 1852 Å2 and 83 ICs at 5 . 5 Å , resulting in a ICs-based affinity prediction of −12 . 8 kcal mol−1 and a BSA-based one of −10 . 3 kcal mol−1 . The relative contribution of each interfacial Fab residue to the total BSA and number of ICs is shown in Figure 5A—figure supplement 1A: The contributions of the various residues to the ICs are more equally distributed than for the BSA . The latter shows high contributions for some residues , which is closely related to their solvent-accessible surface area ( ASA ) in the free form ( defined here as the conformation extracted from the complex , that is , without any conformational changes—see ‘Materials and methods’ ) ( see Figure 5B—figure supplement 1 ) and the surface representation in Figure 5 . Indeed , because BSAFab = ASAFab_free −ASAFab_complex , when a residue is at the core of the binding interface ( in other words totally shielded by Tissue factor ) the ASAFab_complex will be close to 0 , resulting in BSAFab ∼ASAFab_free . In contrast , residues already almost fully buried in the free form will not contribute to the BSA , whereas they might be able to form contacts in our ICs model . 10 . 7554/eLife . 07454 . 010Figure 5 . Surface representation of Fab D3H44; residues at the interface are colored according to their contribution ( in percentage ) to ( A ) the buried surface area ( BSA ) of Fab upon complex formation and ( B ) the total number of inter-residue contacts ( ICs ) made . Increasing graduation of pink is used for the ranges 0–2% , 2–4% , 4–6% , and above 6% of BSA/ICs contribution . ( C ) Surface representation of Fab D3H44 ( gray ) in complex with Tissue factor ( light blue ) , PDB code: 1JPS ( Faelber et al . , 2001 ) . Fab D3H44 is represented in all panels with the same orientation . Values of residues BSA/ICs contribution are reported in Supplementary file 5 . The following figure supplement is available for Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 07454 . 01010 . 7554/eLife . 07454 . 011Figure 5—figure supplement 1Comparison between BSA and ICs relative contribution . ( A ) Relative contribution ( percentage ) of each Fab D3H44 interfacial residues to the total BSA ( hot pink ) and ICs ( green ) . ( B ) Corresponding solvent-accessible surface area in Å2 of the Fab D3H44 residues in the free form ( separated proteins taken from the complex ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07454 . 011 To report an example of this , the three residues GLU_H31 , TYR_H32 , and TYR_H33 located at the core of binding site ( see Figure 5 ) contribute equally to the ICs calculation ( 6% ) , while they account for 10 . 5% , 5 . 6% , and 8 . 1% of the BSAFab , respectively . On the other hand , residues such as ILE_H29 , TYR_H53 and ASP_H99 , which are already highly buried in the free form and therefore contribute less than 0 . 1% to the BSAFab , are still making contacts , contributing between 1 . 2% and 2 . 4% in the calculation of ICs . In general , it seems that our ICs-based model is accounting more evenly than the BSA model for the contributions of highly solvent-accessible and buried residues , which leads to a higher prediction accuracy . Our new ICs/NIS-based BA model predicts BAs with an unprecedented accuracy ( R = −0 . 73 , ρ < 0 . 0001; RMSE = 1 . 89 kcal mol−1 ) , on a large , various and reliable dataset of 81 complexes . It achieves this by making use of only two structural features: the interfacial residue–residue contacts and the contribution of the NIS . Unsurprisingly , the higher the number of interfacial contacts , the stronger the binding strength , which is consistent both with the previously reported evidence that interfaces tend to be larger and more tightly packed with increasing interaction strength ( Nooren and Thornton , 2003 ) and with the simple BSA models introduced by Chothia and Janin ( 1975 ) and Horton and Lewis ( 1992 ) . BSA and the number of contacts at the interface are of course somewhat related , but the number of interface contacts shows much better correlations with binding strength than the BSA . In summary , our study demonstrated that interface contacts , decomposed into their polar/apolar/charged characteristics , and combined with contributions of the NIS based on the recent work of Kastritis et al . ( 2014 ) ( in particular the percentage of apolar and charged surface ) , lead to the best BA predictor for protein–protein complexes reported to date . Importantly , these are less sensitive to conformational changes occurring upon binding , which are one of the challenging aspects to deal with for both structure and affinity prediction . In order to evaluate the relationship between the contacts at the interface and the experimental BA in protein–protein complexes , we used the bound structures from the structure-based protein–protein BA benchmark of Kastritis et al . ( 2011 ) . It contains 144 non-redundant protein–protein complexes with known 3D structures ( of both the unbound and bound components ) and associated experimental ΔG values . From this dataset we removed: ( i ) three cases ( PDB codes: 1NSN , 1UUG , and 1IQD ) for which the ΔG has not a unique value ( reported as > −14 kcal/mol , > −18 kcal/mol , and > −15 kcal/mol , respectively ) , and ( ii ) all the complexes that show gaps or unresolved fragments at the binding interface ( considering a gap to be a missing segment longer than two residues ) . 19 cases were discarded in total ( for details see Supplementary file 1 ) . This resulted in a dataset of 122 complexes , covering diverse types of biological functions including antibody/antigen ( A or AB with bound antibody , 10 cases ) , enzyme/inhibitors ( EI , 34 cases ) , enzyme/substrate ( ES , 9 cases ) , enzyme/regulatory subunit ( ER , 7 cases ) , G-protein containing ( OG , 15 cases ) , membrane receptors ( OR , 7 cases ) , miscellaneous ( OX , 26 cases ) , and non-cognate complexes ( NC , 9 cases ) . The dataset includes both weak and strong complexes in terms of interaction strength , with ΔG values ranging between −4 . 3 and −18 . 6 kcal/mol . The published benchmark also reports for each entry the interface C-alpha RMSD ( i_rmsd ) between unbound and bound form , which provides an estimate of the amplitude of the conformational changes that take place upon binding . Our clean dataset has i_rmsd ranging between 0 . 17 Å and 4 . 90 Å . The interacting area , expressed in terms of BSA upon complex formation , ranges from 808 Å2 to 3370 Å2 . We calculated the number of interface residue pair-wise contacts ( ICs ) for each complex using the COCOMAPS web tool ( Vangone et al . , 2011 ) . Two residues are considered in contact if a pair of ( any ) atoms belonging to two residues is closer than a defined cut-off distance . To systematically evaluate the impact of the cut-off distance on the correlation between ICs and BA , we varied the cut-off between 3 Å and 20 Å in steps of 0 . 5 Å in the range 5–8 Å and 1 . 0 Å from 8 Å and above . The BSA upon complex formation was calculated using NACCESS ( Hubbard and Thornton , 1993 ) as: ( 3 ) BSA= ( ASAprotein1+ASAprotein2 ) −ASAcomplex , where ASAprotein1 and ASAprotein2 are the solvent-accessible surface areas calculated from the free components ( i . e . , the separated bound conformation of the proteins—note that this is different from the unbound form of the protein ) using a 1 . 4 Å radius sphere . The NIS properties ( i . e . , percentage of polar , apolar and charged residues on the NIS ) were calculated as described in Kastritis et al . ( 2014 ) . Residues were classified based on their physico-chemical properties as follow:polar: C , H , N , Q , S , T , Y , Wapolar: A , F , G , I , V , M , Pcharged: E , D , K , R . The hydrophobic nature of the residues was defined according to the Kyte–Doolittle hydrophobicity index ( Kyte and Doolittle , 1982 ) . The scripts for calculation of ICs ( polar/apolar/charged divided ) and NIS are available at: http://bonvinlab . org/software . A description of how to predict binding affinity with our approach is described in details in Bio-Protocol ( Vangone and Bonvin , 2017 ) . To assess the linear dependence between two variables ( i . e . , the experimental BA and the structural properties tested , such as ICs ) , the Pearson product-momentum correlation coefficients ( R ) were calculated . The ranking power of ICs with BA was also calculated as reported by the Spearman's correlation parameter S . We trained different models ( Equation 1 ) using linear regression in R ( R Development Core Team , 2014 ) ; to avoid problem of over-fitting when many variables were used ( >3 ) we applied the AIC stepwise selection approach ( backward and forward ) in order to identify the significant terms and calculate weights only for them . Cross-validation on the final model was performed by partitioning the dataset into four complementary subsets , training on the 75% of the data ( training set ) and validating on the other 25% ( prediction set ) . Such fourfold cross-validation was repeated 10 times . We compared our method with other BA predictors , potentials and composite scoring functions . Their performance is reported as correlation ( R ) between the predicted BA ( or potential ) and the experimental BAs . The comparison was made for 79 protein–protein complexes that are in common between our cleaned dataset of 81 structures and the 137 complexes reported by Moal et al . ( 2011 ) . Predicted BAs for the ‘global surface model’ developed by Kastritis et al . ( 2014 ) have been calculated through the program provided by the authors; the Horton & Lewis BSA-based model ( 1992 ) ( Horton and Lewis , 1992 ) was re-trained on our clean dataset ( reported as Model 5 ) ; data for the consensus model reported by Moal et al . ( 2011 ) , DFIRE ( Liu et al . , 2004 ) , and PMF ( Su et al . , 2009 ) are provided in ( Moal et al . , 2011 ) ; data of the composite scoring ZRANK ( Pierce and Weng , 2007 ) , ZRANK2 ( Pierce and Weng , 2008 ) , RosettaDock ( Chaudhury et al . , 2010 ) , PyDock ( Cheng et al . , 2007 ) , FireDock ( the total energy , the antibody–antigen energy and the enzyme-inhibitor energy ) ( Andrusier et al . , 2007 ) , PISA ( Viswanath et al . , 2013 ) , PIE ( Ravikant and Elber , 2010 ) , and SIPPER ( Pons et al . , 2011 ) were obtained as pre-calculated data from the CCHarPPI webserver ( Moal et al . , 2015a , b ) . Apart from the composite scoring function , CCHarPPI reports 99 additional intermolecular parameters , such as potential functions , energy functions , and various descriptors . The performance of each of them has been compared with our model .	Proteins help to copy DNA , transport nutrients and perform many other important roles in cells . To perform these tasks , proteins often interact with other proteins and work together . These interactions can be very complex because each protein has a three-dimensional shape that may change when it binds to other proteins . Also , two proteins may form several connections with each other . It is possible to carry out experiments to calculate how likely it is that two proteins will physically interact with each other and how strong their connections will be . However , these measurements are time consuming and costly to do . Some researchers have developed computer models to help predict the interactions between proteins , but these models are often incorrect because they leave out some of the chemical or physical properties that influence the ability of proteins to interact . With the aim of making a better model , Vangone and Bonvin examined 122 different combinations of proteins whose abilities to interact had previously been experimentally measured . Vangone and Bonvin found that the number of connections between each pair of proteins was a strong predictor of how tightly the proteins connect to each other . Particular features of the surface of the proteins—specifically , the region defined as the non-interacting surface—can also influence how strong the interaction is . Vangone and Bonvin used this information to develop a new model that predicts how tightly proteins interact with each other based on the number of connections between the two proteins and the characteristics of the non-interacting surface . The model is simple , and Vangone and Bonvin show that it is more accurate than previous models . Defects in the interactions between proteins can lead to many diseases in humans , so this model may be useful for the development of new drugs to treat these conditions .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2015	Contacts-based prediction of binding affinity in protein–protein complexes
Activation triggers the exchange of subunits in Ca2+/calmodulin-dependent protein kinase II ( CaMKII ) , an oligomeric enzyme that is critical for learning , memory , and cardiac function . The mechanism by which subunit exchange occurs remains elusive . We show that the human CaMKII holoenzyme exists in dodecameric and tetradecameric forms , and that the calmodulin ( CaM ) -binding element of CaMKII can bind to the hub of the holoenzyme and destabilize it to release dimers . The structures of CaMKII from two distantly diverged organisms suggest that the CaM-binding element of activated CaMKII acts as a wedge by docking at intersubunit interfaces in the hub . This converts the hub into a spiral form that can release or gain CaMKII dimers . Our data reveal a three-way competition for the CaM-binding element , whereby phosphorylation biases it towards the hub interface , away from the kinase domain and calmodulin , thus unlocking the ability of activated CaMKII holoenzymes to exchange dimers with unactivated ones . Ca2+/calmodulin-dependent protein kinase II ( CaMKII ) is a Ser/Thr kinase that is particularly important in neuronal signaling and cardiac function ( Colbran et al . , 1989a; Giese and Mizuno , 2013; Hook and Means , 2001; Kandel et al . , 2014; Lisman and Raghavachari , 2015; Lisman et al . , 2002; Luo and Anderson , 2013 ) . CaMKII has a unique architecture , in which the catalytic domains are linked flexibly to a hub domain , also called the association domain , which forms a dodecameric or tetradecameric assembly ( see Figure 1A for a description of the domains of CaMKII ) ( Chao et al . , 2011; Kanaseki et al . , 1991; Kolb et al . , 1998; Kolodziej et al . , 2000; Morris and Török , 2001; Rosenberg et al . , 2006; Stratton et al . , 2013; Woodgett et al . , 1983 ) . This organization , in which twelve or more kinase domains are maintained in close proximity , enables the highly cooperative activation of CaMKII by Ca2+/calmodulin ( Ca2+/CaM ) and the integration of calcium inputs ( Bradshaw et al . , 2003; Chao et al . , 2010 , 2011; De Koninck and Schulman , 1998 ) . 10 . 7554/eLife . 13405 . 003Figure 1 . The domains of human CaMKII-α . ( A ) The architecture of the CaMKII holoenzyme is shown on the right . The hub domain forms a dodecamer or tetradecamer , and the kinase domains are flexibly linked to it by the autoinhibitory segment . ( B ) The autoinhibitory segment contains three critical sites of phosphorylation: Thr 286 , Thr 305 and Thr 306 . Activation by Ca2+/CaM results in phosphorylation at Thr 286 , which renders the enzyme independent of Ca2+/CaM . Further phosphorylation of Thr 305 and Thr 306 prevents re-binding of Ca2+/CaM . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 003 The activity of CaMKII is regulated by autophosphorylation ( Lai et al . , 1986; Lou et al . , 1986; Miller and Kennedy , 1986; Schworer et al . , 1986 ) . In most kinases , phosphorylation of an “activation loop” that is located at the active site stabilizes the active conformation of the catalytic domain ( Huse and Kuriyan , 2002; Johnson et al . , 1996; Taylor et al . , 1992 ) . The activation loop of the CaMKII kinase domain has no phosphorylation site , and is in an active conformation ( Bulleit et al . , 1988; Hanley et al . , 1987; Lin et al . , 1987; Rosenberg et al . , 2005 ) . The three principal sites of autophosphorylation in CaMKII are located within the autoinhibitory segment , which follows the N-terminal kinase domain and blocks the catalytic site in the unactivated state of the enzyme ( Figure 1 ) . One of the autophosphorylation sites , Thr 286 in the α isoform of human CaMKII ( CaMKII-α ) , prevents rebinding of the autoinhibitory segment to the kinase domain upon phosphorylation . Phosphorylation of Thr 286 is the key step by which CaMKII gains calcium-independence ( autonomy ) after stimulation by Ca2+/CaM ( Colbran et al . , 1989b; Lou and Schulman , 1989; Miller et al . , 1988; Schworer et al . , 1988; Thiel et al . , 1988; Waldmann et al . , 1990 ) ( Figure 1B ) . The other two sites , Thr 305 and Thr 306 , are located within an 18 residue CaM-binding element that is part of the autoinhibitory segment ( Figure 1B ) ( Meador et al . , 1993; Rellos et al . , 2010 ) . Phosphorylation of these two sites prevents the rebinding of Ca2+/CaM to CaMKII ( Elgersma et al . , 2002; Hanson et al . , 1994; Hashimoto et al . , 1987; Lickteig et al . , 1988; Lou and Schulman , 1989; Mukherji and Soderling , 1994 ) . The importance of phosphorylation at these three sites is underscored by the fact that mice in which these sites are mutated exhibit distinct defects in learning and memory ( Elgersma et al . , 2002; Giese et al . , 1998; Silva et al . , 1992; Wayman et al . , 2008 ) . We reported recently that the activation of CaMKII-α by ATP and Ca2+/CaM triggers the exchange of activated subunits between different holoenzyme assemblies ( Stratton et al . , 2014 ) . The replacement of Thr 286 by aspartate , which confers constitutive activity on the kinase ( Waldmann et al . , 1990 ) , results in robust subunit exchange without Ca2+/CaM , and the spread of activation to unactivated subunits . Our observation of subunit exchange connects with previous speculation that such a mechanism might allow maintenance of some level of activated CaMKII long after withdrawal of the activating stimulus ( Crick , 1984; Lisman , 1985; Lisman and Goldring , 1988; Lisman and Raghavachari , 2015; Miller and Kennedy , 1986 ) . It might also provide a mechanism to potentiate the effects of calcium stimuli under conditions where Ca2+/CaM is limiting with respect to CaMKII , as in the dendritic spine ( Pepke et al . , 2010; Persechini and Stemmer , 2002 ) . We now present the results of experiments that clarify the molecular mechanism of activation-triggered subunit exchange in CaMKII . The two principal isoforms of CaMKII in the brain are the α and β isoforms , and we show that the β isoform also undergoes activation-dependent subunit exchange . We show that the intact human CaMKII-α holoenzyme exists as a mixture of dodecameric and tetradecameric forms , an observation that provides a link to a possible mechanism for subunit exchange . We demonstrate that the CaM-binding element of CaMKII is subject to a three-way competition for binding , between the kinase domain , Ca2+/CaM and the hub . A structural understanding of how the CaM-binding element interacts with the hub and weakens its integrity emerged , unexpectedly , from studies on CaMKII homologs found in two distantly diverged species , the sea anemone Nematostella vectensis and the choanoflagellate Salpingoeca rosetta . These structures suggest that the docking of peptide segments at the interfaces between hubs can break the integrity of the ring . CaMKII is widely described as a dodecameric assembly , with six-fold cross-sectional symmetry . This view is based on negative-stain electron microscopic analyses ( Gaertner et al . , 2004; Kolodziej et al . , 2000; Morris and Török , 2001; Woodgett et al . , 1983 ) and was reinforced by a crystal structure of an intact dodecameric holoenzyme ( Chao et al . , 2011 ) . It is therefore puzzling that crystal structures of the isolated hub domains of CaMKII from C . elegans and mammalian species show them to be assembled into both dodecamers ( Rellos et al . , 2010 ) and tetradecamers ( Hoelz et al . , 2003; Rosenberg et al . , 2006 ) . To determine the stoichiometry of hub assemblies in solution , we analyzed the human CaMKII-α hub domain by native electrospray ionization mass spectrometry ( ESI-MS ) ( Chowdhury et al . , 1990; Heck , 2008; Sharon and Robinson , 2007 ) . The mass spectra demonstrate that the isolated hub assembly exists as a ~1:1 mixture of dodecamers and tetradecamers in solution . Collision-induced dissociation ( CID ) MS/MS of the mixture of hub parent ions ( at 30 V collision energy ) shows the presence of fragment ions corresponding to a hub monomer and a mixture of 11-subunit and 13-subunit species ( Figure 2A ) . Collisional activation of intact gaseous protein complexes typically results in asymmetric dissociation , in which loss of a highly charged monomer subunit occurs as a result of structural changes and charge partitioning in the activated complex ( Jurchen and Williams , 2003 ) . This validates the mixed stoichiometry of the parent ion . Thus , the crystal structures of dodecameric and tetradecameric hubs are not artifacts of crystallization , but reflect instead a natural variation in the stoichiometry of assembly of the hub . 10 . 7554/eLife . 13405 . 004Figure 2 . Human CaMKII-α forms both dodecamers and tetradecamers . ( A ) Native ESI-MS spectra of the human CaMKII-α hub reveals a ~1:1 mixture of dodecamers and tetradecamers ( top panel ) . Collision-induced dissociation ( CID ) MS/MS of hub parent ions shows the presence of fragment ions corresponding to a highly charged hub monomer and a mixture of 11-subunit and 13-subunit species , clearly showing the mixed stoichiometry of the parent ion ( bottom panel ) . ( B ) Two-dimensional class averages of images from negative-stain electron microscopy for human CaMKII-α show the existence of holoenzyme particles with both six-fold and seven fold symmetry . A subset of 32 of the 50 total class averages is shown , and an expanded view of two class averages is shown , one with seven-fold and one with six-fold symmetry . ( C ) A possible pathway for the transition between the dodecameric and tetradecameric species through the addition or loss of a dimer unit . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 004 Due to poor signal in ESI-MS of full-length CaMKII holoenzyme , we examined unactivated human CaMKII-α holoenzyme by negative-stain electron microscopy ( EM ) in order to determine its stoichiometry , as described in Methods . An important aspect of our analysis is that no symmetry was imposed on the particles at any stage of the generation of class averages . The hub assemblies are clearly resolved in the EM micrographs , but the kinase domains are not , as is common for CaMKII . Visual inspection of the two-dimensional class averages clearly reveals a population of holoenzyme particles with seven-fold symmetry , in addition to those with the expected six-fold symmetry . It is unclear why particles with seven-fold symmetry were not reported in the previous EM analyses of CaMKII-α , which focused on dodecameric species ( Kolodziej et al . , 2000; Morris and Török , 2001 ) . In our analysis , we could discern clear evidence , by visual inspection , for either six-fold or seven-fold symmetry in seven out of 50 classes each ( the symmetry of the other class averages was not obvious ) . Based on the number of particles contributing to each of these 14 classes , we estimate the ratio of dodecameric to tetradecameric species to be roughly 55:45 ( Figure 2B ) . The observation that full-length human CaMKII-α exists in both dodecameric and tetradecameric forms has guided our thinking about how the exchange process might occur . If CaMKII dimers are the unit of exchange , as we hypothesized previously , then the dodecameric and tetradecameric species can interconvert by releasing and capturing dimers ( Figure 2C ) . The release of dimers , rather than monomers , is also potentially significant for the maintenance of autonomous ( Ca2+/CaM independent ) activity . Phosphorylation of Thr 286 can only occur in trans ( Hanson et al . , 1994; Rich and Schulman , 1998 ) , and a dimeric unit may be able to maintain this phosphorylation whereas a monomer could not . To analyze the mechanism of subunit exchange , we focused initially on potential interactions between the CaM-binding element and the hub . We had shown previously that mutation of the CaM-binding element blocks subunit exchange , and that phosphorylation of Thr 305 and Thr 306 in the CaM-binding element potentiates exchange ( Stratton et al . , 2014 ) . We prepared both phosphorylated and unphosphorylated forms of peptides spanning the CaM-binding element of CaMKII-α ( see Materials and methods ) . The peptides were labeled with a fluorophore ( Bodipy-FL maleimide ) at the C-terminal end , and binding was monitored by changes in fluorescence polarization ( Figure 3A ) . To a fixed volume of labeled peptide ( 2 nM ) , increasing concentrations of the hub was added , and the change in fluorescence polarization was monitored ( Figure 3B ) . This titration showed evidence for some degree of non-saturable binding at high hub concentrations . We therefore used a competition assay to determine the peptide affinities for the hub . 10 . 7554/eLife . 13405 . 005Figure 3 . Binding of the CaM-binding element to the human CaMKII-α hub . ( A ) Schematic diagram of the binding experiments . ( B ) Direct titration of the hub to the labeled peptide A . Fluorescence polarization of the labeled peptide increases as a function of total hub concentration . The value of KD derived from these data is ~100 μm , but due to evidence for non-saturable binding , we relied instead on competition experiments to obtain KD values . ( C–E ) Competition experiments with unlabeled peptides . Addition of labeled peptide is followed by competition with excess unlabeled peptide of the same sequence ( except for the competition with peptide B , where labeled Peptide A was used for initial binding ) . ( F ) Competition with mutant peptides . The hub is bound to labeled peptide A and fluorescence polarization is measured as a function of the total concentration of unlabeled mutant peptides . The sites of mutation in these peptides are highlighted in red and underlined . Fluorescence polarization is a ratio involving light intensities that are parallel and perpendicular to the plane of linearly polarized excitation light and is therefore reported as a dimensionless number . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 00510 . 7554/eLife . 13405 . 006Figure 3—figure supplement 1 . Calculation of effective concentration of the CaM-binding element by assuming that the linker restricts it to within 50 Å of the hub . This corresponds to an effective concentration of ~3 mM . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 006 For the competition experiments , 2 nM of various labeled peptides were bound to 30 μM of the hub , resulting in high fluorescence polarization . Upon titration with the corresponding unlabeled versions of the peptides , the labeled peptides were competed off , and the fluorescence polarization decreased to the baseline value ( Figure 3C–E ) . The value of the inhibition constant , IC50 , is ~100 μM for the unlabeled peptides , whether phosphorylated or not . The values of the corresponding dissociation constants ( KD ) , derived from non-linear least squares fit of the data by assuming that each subunit has a single peptide-binding site , are ~90 μM , ~60 μM and ~50 μM , for the unphosphorylated peptide , and peptides phosphorylated on Thr 305 and Thr 306 , respectively ( see Materials and methods ) . These data show that phosphorylation of the peptide at either Thr 305 or Thr 306 , separately , does not alter the affinity for the hub substantially ( see Figure 3D–E ) . We could not obtain the doubly phosphorylated peptide , due to difficulty in its synthesis . Peptides spanning the CaM-binding element are inhibitors of the CaMKII kinase domain , which prevented generation of the doubly phosphorylated species through enzymatic means . Prominent grooves are formed at the vertical interfaces between adjacent hub domains , and these can dock peptides , as discussed below . The edges of these grooves are lined by four acidic residues , two on each side of the interface: Glu 355 , Glu 359 , Glu 390 and Asp 393 in human CaMKII-α , using the numbering in Protein Databank ( PDB ) entry 1HKX . The CaM-binding element contains the sequence 291KKFNARRKLK300 . If this peptide is docked as an additional strand on the open β sheet presented by one of the subunits , then the positively charged residues of the peptide could interact with the negatively charged residues in the hub . We carried out competition experiments using unlabeled peptides in which positively charged residues in the KKFNARRKLK motif in the CaM-binding element were replaced by negatively charged residues ( i . e . , the variant peptide sequences are KKFNAERKLK , KKFNARRKLE , and KEFNAERKLE; the mutated residues are underlined ) . The triple-mutant peptide fails to compete the binding of the labeled peptide ( Figure 3F ) . The two peptides bearing mutations at single sites have diminished ability to compete , compared to the unlabeled peptides . The IC50 values of these peptides are increased by factors of about 2 and 4 , relative to the wild type peptide . A more rigorous identification of the hub residues that constitute the peptide binding groove awaits further analysis of hub mutants . Although the CaM-binding element has low affinity for the hub when added as a separate peptide , its effective concentration with respect to the hub is very high in the intact holoenzyme . We estimated the local concentration of the CaM-binding element by assuming that the linker restricts it to within 50 Å of the hub . This corresponds to an effective concentration of ~3 mM ( see Figure 3—figure supplement 1 ) . Thus , the observed KD of ~100 μM would allow the CaM-binding element to successfully engage the hub in the holoenzyme . Our finding that the CaM-binding element binds to the hub raises two questions: what prevents this element from interacting with the hub in unactivated CaMKII holoenzymes , and what is the role of phosphorylation in gating this interaction ? Although phosphorylation of the CaM-binding element is not required for binding to the hub , the CaM-binding element will be sequestered by either the kinase domain or by Ca2+/CaM in the absence of phosphorylation , and thereby prevented from interacting with the hub . The unphosphorylated autoinhibitory segment , which includes the CaM-binding element , binds to Ca2+/CaM with sub-picomolar affinity ( Tse et al . , 2007 ) . Thr 305 and Thr 306 are buried within this high affinity interface ( Crivici and Ikura , 1995; Meador et al . , 1992 ) , and phosphorylation would therefore disrupt the interaction with Ca2+/CaM . The autoinhibitory segment also binds to the kinase domain , which it inhibits with an IC50 value of ~1 μM , when added in trans as a peptide ( Colbran et al . , 1988 ) . Phosphorylation on Thr 286 increases the affinity of CaMKII for Ca2+/CaM by more than 1000-fold , by fully or partially releasing the autoinhibitory segment from the kinase domain ( Meyer et al . , 1992 ) . Thus , the kinase domain sequesters the autoinhibitory segment , including the CaM-binding element , in the absence of phosphorylation , reducing its effective concentration with respect to the hub . We tested the ability of the CaM-binding element to mediate subunit exchange when it is released from the kinase domain and Ca2+/CaM . To do this , we made a construct comprised of the hub , the linker and the autoinhibitory segment of CaMKII-α , without the kinase domain ( denoted hub-△kinase ) . The construct includes an N-terminal SUMO tag , which reduced proteolysis during purification . We monitored the gain of fluorescence resonance energy transfer ( FRET ) in order to test subunit exchange kinetics for this construct , by mixing samples labeled separately with donor and acceptor fluorophores , at a subunit concentration of ~5 μM , as described ( Stratton et al . , 2014 ) . We observed that the hub-△kinase construct undergoes robust subunit exchange ( Figure 4A ) . In contrast , the isolated hub of CaMKII-α , lacking the autoinhibitory segment and the linker , does not exchange subunits ( Stratton et al . , 2014 ) . We also made a construct that spans the linker segment and the hub ( residues 314 to 475 , see Figure 1A; referred to as hub-linker ) . We observe robust subunit exchange only when the CaM-binding element is present in both samples ( Figure 4—figure supplement 1 ) . That is , a hub construct bearing the CaM-binding element cannot efficiently break a hub assembly lacking this element . We have shown previously that activated CaMKII holoenzyme , including the kinase domains , undergoes subunit exchange with unactivated holoenzymes ( Stratton et al . , 2014 ) . The role of the kinase domains in destabilization of unactivated assemblies requires further study . 10 . 7554/eLife . 13405 . 007Figure 4 . Analysis of kinase domain deletion and hub mutation . ( A ) Spontaneous subunit exchange in CaMKII lacking only the kinase domain ( hub-△kinase ) . Increase in the FRET-ratio ( see Materials and methods ) is graphed as a function of time for the hub-△kinase construct ( also see Figure 4—figure supplement 1 ) . The increase in FRET upon mixing donor and acceptor-labeled proteins is indicative of subunit exchange ( Stratton et al . , 2014 ) . ( B ) Analytical gel filtration of isolated hub domains and hub-△kinase from human CaMKII-α . The isolated hub shows one peak , with a retention volume corresponding to dodecamer/tetradecamer , as estimated from calibration data ( see Figure 4—figure supplement 4 ) . The hub-△kinase construct shows two peaks , corresponding to dodecamer/tetradecamer and dimer , respectively . Note that the hub-△kinase construct includes a SUMO tag , which accounts for its smaller retention volume compared to that for the hub alone . The gel filtration traces are normalized so that they have the same area under the curve . ( C ) A view of the lateral interface between subunits in adjacent vertical dimers of CaMKII-α ( PDB code: 1HKX ) . ( D ) Replacement of Phe 397 at the hub interface by alanine leads to release of dimers from the isolated hub ( left ) and the holoenzyme ( right ) . Pre-incubation of F397A holoenzyme with Mg2+-ATP enhances the amount of dimers released , which is correlated with phosphorylation of Thr 305 and Thr 306 ( data not shown ) . See also mass spectrometric analysis of the F397A mutant hub ( Figure 4—figure supplement 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 00710 . 7554/eLife . 13405 . 008Figure 4—figure supplement 1 . Subunit exchange between various truncation constructs of CaMKII lacking the kinase domain ( hub-△kinase ) , the kinase domain and the autoinhibitory segment ( hub-linker ) and the hub alone . Increase in the FRET-ratio ( see Materials and methods ) is graphed as a function of time . The increase in FRET upon mixing donor and acceptor-labeled proteins is indicative of subunit exchange ( Stratton et al . , 2014 ) . The variability in the FRET ratio between experiments is due to differences in the percentage of labeled cysteine residues and fluorophore environments in different protein preparations . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 00810 . 7554/eLife . 13405 . 009Figure 4—figure supplement 2 . Collection and re-injection of the dodecameric/tetradecameric peak for the human CaMKII-α F397A mutant re-releases dimers . The fluorescence intensity for the first injection is shown on the left y-axis and that for the re-injection of the dodecamer/tetradecamer peak is shown on the right y-axis . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 00910 . 7554/eLife . 13405 . 010Figure 4—figure supplement 3 . Mass spectra of 200 μM mutant hub in 1 M ammonium acetate ( top ) , and 2 μM mutant hub in 25 mM Tris , 1 mM TCEP , pH 8 ( middle ) and 250 mM ammonium acetate , 25 mM Tris , 1 mM TCEP , pH 8 ( bottom ) . Heterogeneity is observed in the dimer and tetradecamer species due to incomplete cleavage of the His tag from the monomer . The measured mass of each species was: dimer ( red ) 34 , 839 ± 1 Da , dimer-1 His tag ( purple ) 32 , 825 ± 1 Da , tetradecamer-1 His tag ( orange ) 242120 ± 78 Da , tetradecamer-2 His tags ( green ) 240104 ± 16 Da , and tetradecamer-3 His tags ( blue ) 238048 ± 23 Da . Oligomer identity was confirmed by CID ( right panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 01010 . 7554/eLife . 13405 . 011Figure 4—figure supplement 4 . Calibration curve for the analytical gel filtration column . Beta amylase ( 200 kDa ) , gamma globulin ( 158 kDa ) , alcohol dehydrogenase ( 150 kDa ) , bovine serum albumin ( 66 kDa ) , chicken ovalbumin ( 44 kDa ) , carbonic anhydrase ( 29 kDa ) , gamma phosphatase ( 25 kDa ) , myoglobin ( 17 kDa ) , and cytochrome C ( 12 . 4 kDa ) were used as molecular weight standards to calibrate the column and their retention volumes are indicated by yellow squares . The fitting equation is also shown on the plot . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 011 In order to check whether subunit exchange is accompanied by the release of subunits from the hub , we analyzed the hub assembly by analytical gel filtration ( see Materials and methods ) . These experiments were typically done with an injection concentration corresponding to 700 nM CaMKII subunits; the concentration within the column itself is expected to be about ten-fold lower , based on the concentration of fractions recovered from the column . We monitored tryptophan fluorescence in order to obtain reliable protein signals at these low concentrations . The isolated hub runs at a retention volume corresponding to a dodecamer or tetradecamer , showing no evidence for lower molecular weight species ( Figure 4B ) . Strikingly , for the hub-△kinase construct , gel filtration reveals two peaks ( Figure 4B ) . The major peak ( ~65% of the sample ) corresponds to the dodecamer/tetradecamer , and the minor peak ( ~35% of the sample ) corresponds to a dimeric species . This suggests that the integrity of the hub is compromised by the presence of the linker and the autoinhibitory segment . Reinjection of the eluate corresponding to the major peak results in two peaks at the same retention times as in the original experiment , suggesting that there is an equilibrium between dodecamers/tetradecamers and dimers ( data not shown ) . The CaMKII hub consists of two rings that are joined together at the equatorial plane , each consisting of either six or seven subunits in dodecamers and tetradecamers , respectively . Given that the upper and lower rings do not separate readily ( Stratton et al . , 2014 ) , we sought to check whether the hub would release monomers or “vertical dimers” , comprising one subunit each from the upper and lower rings ( see Figure 1A ) . Each subunit has three aromatic residues that form the core of the interface between adjacent subunits in the same ring ( Figure 4C ) . We mutated one of these ( Phe 397 , the numbering is according to PDB code: 1HKX ) to alanine in the isolated hub as well as in the holoenzyme , with the expectation that it would destabilize interactions between adjacent vertical dimers . We then subjected the mutant proteins to analytical gel filtration . Both the mutant hub and holoenzyme exhibited a substantial release of dimers ( ~40% for the hub and ~60% for the holoenzyme ) , even at an elevated injection concentration of 3–5 μM ( Figure 4D ) . Isolation and reinjection of the peak corresponding to the dodecamer/tetradecamer in the holoenzyme resulted in further release of dimers , suggesting that there was an equilibrium ( Figure 4—figure supplement 2 ) . Mass spectrometry of the F397A mutant hub confirmed the presence of dimers ( Figure 4—figure supplement 3 ) . This is consistent with vertical dimers being the unit of assembly of the CaMKII hub , with each vertical dimer contributing one subunit each to the upper and lower rings . CaMKII-α and CaMKII-β are the predominant species in the brain ( Tombes et al . , 2003 ) . The principal difference between these two isoforms is the length of the linker , which is 218 residues long in CaMKII-β , compared to 30 residues in CaMKII-α . We purified CaMKII-β using a bacterial expression system , and tested its ability to undergo subunit exchange . We mixed proteins , labeled separately with donor and acceptor fluorophores and at a subunit concentration of ~5 μM , and monitored the development of FRET ( Figure 5A ) . This experiment shows that CaMKII-β exchanges subunits , but only after stimulation with Ca2+/CaM and ATP ( Figure 5B ) . Thus , the phenomenon of activation-triggered subunit exchange is not limited to just the α isoform of CaMKII ( Figure 5B–C ) . CaMKII-β can also exchange subunits with CaMKII-α in an activation-dependent manner , leading to the formation of CaMKII heterooligomers ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 13405 . 012Figure 5 . Subunit exchange in CaMKII-β . ( A ) Schematic diagram illustrating the design of solution FRET experiments for monitoring subunit exchange in CaMKII ( Stratton et al . , 2014 ) . ( B , C ) FRET ratio as a function of time for human CaMKII-β and human CaMKII-α . The difference in the extent of FRET between the experiments shown in Panels B and C is due to differences in the percentage of labeled cysteine residues and the fluorophore environments in different proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 01210 . 7554/eLife . 13405 . 013Figure 5—figure supplement 1 . Subunit exchange between human CaMKII-β and CaMKII-α . FRET ratio as a function of time is graphed . The replicate data are shown on two different y-axes ( set1 on left and set2 on right y-axis ) due to the large differences in the absolute values of FRET ratio from the two different protein preparations used for this analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 013 We now describe three new structures of hub assemblies that , together , provide support for the hypothesis that the CaM-binding element docks at the interfaces between vertical dimers , and suggest how such docking might destabilize and open the closed-ring assembly of the hub . The first insight came from a crystal structure of a dodecameric human CaMKII-α hub , for which the crystallization construct contained an additional 22 residues from an uncleaved expression tag . The tag contains an N-terminal hexahistidine sequence , followed by a PreScission protease cleavage site ( GSSHHHHHHSSGLEVLFQGPHM ) ( Waugh , 2011 ) . To our surprise , we found well-resolved electron density corresponding to the expression tag at two out of the twelve intersubunit interfaces . There was partial density for the tag at four more intersubunit interfaces . The tag binds the hub almost precisely as we had predicted the CaM-binding element would , and it extends the central β-sheet of the hub domain by forming an additional parallel strand ( Figure 6A ) . This docking allows the C-terminal end of the tag to connect to the N-terminal end of the first α-helix in the hub domain proper . The mouse CaMKII-α hub , which is 100% identical in sequence over the region spanning the human hub , was crystallized previously as a tetradecamer , rather than the dodecamer seen here ( PDB code: 1HKX ) ( Hoelz et al . , 2003 ) . The groove between adjacent subunits is necessarily narrower in the tetradecamer than in the dodecamer , because of the hinging apart of the subunits , and the presence of the expression tag could favor the dodecamer for this reason . 10 . 7554/eLife . 13405 . 014Figure 6 . Docking of peptide segments onto the central β-sheet of the hub domain . ( A ) The N-terminal expression tag in human CaMKII-α docked on the open β-strand of the central β-sheet within the same hub subunit . ( B ) Linker residues in N . vectensis CaMKII-B hub ( at pH 4 . 2 ) docked on the β-sheet of the hub subunit . ( C ) One subunit of the N . vectensis CaMKII-B hub , including the docked linker segment , is superimposed on the tetradecameric mouse CaMKII-α hub ( PDB code: 1HKX ) . The N . vectensis hub domain is in magenta with the docked linker shown in cyan , and the CaMKII-α hub is in blue . The expanded view shows that the sidechain of Ile 335 in the linker of N . vectensis CaMKII-B hub collides with Asp 393 from helix D in the adjacent subunit of the mouse tetradecameric hub . ( D ) The distortion in the structure of N . vectensis CaMKII-B hub at pH 4 . 2 . The spiral arrangement of the subunits is shown in the schematic diagram on the left , with each subunit labeled . The deviation between the closed-ring structure of N . vectensis CaMKII-A ( pH 7 . 0 ) and the spiral form of the N . vectensis CaMKII-B structure is illustrated on the right , with the two structures aligned using subunits E , F , G and H . Arrows indicate displacement of the Cα atoms in going from the closed-ring CaMKII-A hub structure to the opened-ring CaMKII-B hub . Each arrow is scaled by a factor of 2 and arrows are only shown for Cα displacements greater than or equal to 2 Å . The structure shown is that of N . vectensis CaMKII-B hub . The M and N subunits are not shown for clarity . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 01410 . 7554/eLife . 13405 . 015Figure 6—figure supplement 1 . Crystallographic axis of two-fold symmetry runs through the middle of the tetramer formed by the E-F and G-H dimers in N . vectensis pH 4 . 2 structure ( CaMKII-B hub ) . The left-handed “lock-washer” architecture allows 12 of the 14 subunits in the pH 4 . 2 structure to obey this axis of symmetry . The M-N dimer cannot obey the symmetry , because of the spiral geometry . The symmetry axis generates two copies of the M-N dimer . One copy is “joined” to the A-B dimer , and dislocated from the K-L dimer . The other copy is “joined” to the K-L dimer , but dislocated from the A-B dimer ( see Supplementary files 2 and 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 015 The role of peptide docking in destabilizing the hub emerged from structures of the hubs of two CaMKII isoforms , denoted CaMKII-A and CaMKII-B , present in the sea anemone Nematostella vectensis ( Miller and Ball , 2008; Putnam et al . , 2007 ) . N . vectensis , a member of the phylum Cnidaria , is an early branching metazoan ( Nakanishi et al . , 2012 ) . The sequence identity between the hub domains in the two N . vectensis CaMKII isoforms is 62% , and these are 55% and 53% identical to the hub domain of human CaMKII-α respectively . The hub domain of CaMKII-A crystallized at pH 7 , while that of CaMKII-B crystallized at pH 4 . 2 . In each case , the construct used is very similar to that used to crystallize the mouse CaMKII-α hub ( PDB code: 1HKX ) , and includes the last 12 residues of the linker . The N-terminal α helix of the CaMKII-B hub is preceded by a 12-residue segment of the variable linker , including several acidic residues . This portion of the linker is highly conserved in metazoan CaMKII sequences . This segment is either absent in other crystal structures , or forms an extension of the N-terminal helix . In the structure of the N . vectensis CaMKII-B hub at pH 4 . 2 , this segment is ordered in every subunit , and it folds back to dock onto the central β-sheet of the same hub subunit , extending the β-sheet by one strand , as seen in the structure described earlier for human CaMKII-α ( Figure 6B ) . The acidic sidechains in this segment are located close to the acidic residues in the hub . This particular configuration is unexpected , and is presumably possible only because some or all of the acidic residues are protonated at the low pH of crystallization . Comparison of the N . vectensis CaMKII-B hub at pH 4 . 2 with that of tetradecameric or dodecameric hubs shows that the linker peptide cannot be accommodated within the grooves of the closed-ring forms without steric clashes . In particular , the sidechain of Ile 335 in the N . vectensis docked linker would collide with the Asp 393 in helix D in the adjacent subunit if the geometries of either the tetradecameric or dodecameric hubs were preserved ( Figure 6C ) . These clashes are relieved in the pH 4 . 2 N . vectensis structure by a small distortion of the geometry of the assembly , away from a planar-ring configuration and into left-handed spiral geometry . The assembly is distorted at each interface in a systematic way throughout . This leads to the formation of a spiral , similar to a “lock-washer” , with a slight opening between the A-B and M-N dimers ( subunit notation is indicated in Figure 6D , Figure 6—figure supplement 1 ) . To illustrate this distortion from planarity , we aligned the pH 4 . 2 N . vectensis CaMKII-B hub structure onto the pH 7 . 0 CaMKII-A hub structure by superimposing the four subunits that are furthest from the break ( E , F , G , and H ) . We then illustrate the deviation between the two structures with arrows that indicate the displacement of Cα atoms from the flat closed-ring structure ( pH 7 . 0 ) to the slightly spiral , ring-opened , structure ( pH 4 . 2; the lengths of the arrows are scaled by a factor of 2 for clarity; Figure 6D ) . This analysis reveals that the A-B dimer is displaced in a direction opposite to that of the M-N dimer . This dislocation results in rupture of the interface between these two dimers . Apart from the rupture at the A-B/M-N interface , the rest of the CaMKII-B hub assembly is symmetrical . At each interface , helix D from one subunit packs against the β-sheet of the other subunit ( Figure 7A ) . A very similar interfacial interaction is seen in both the tetradecameric mouse CaMKII-α hub ( PDB code: 1HKX ) and the dodecameric human CaMKII-γ hub ( Rellos et al . , 2010 ) ( PDB code: 2UX0 ) . This interaction is ruptured at the A-B/M-N interface in N . vectensis CaMKII-B at pH 4 . 2 ( Figure 7B ) . For example , at the interface between subunits B and N , the sidechains of Gln 436 , Thr 446 and Asn 411 are pulled away from their hydrogen bonding partners in the other subunit . 10 . 7554/eLife . 13405 . 016Figure 7 . Analysis of the ring-opened N . vectensis CaMKII-B hub ( pH 4 . 2 ) and its implications for docking of the CaM-binding element . ( A ) Interfacial hydrogen-bonding interactions at an intact interface in the N . vectensis CaMKII-B hub . ( B ) As in ( A ) , but for one of the cracked interfaces . Hydrogen-bonding interactions are disrupted at this interface . ( C ) Modeling of the RRKLK motif of the CaM-binding element docked onto a hub interface in the tetradecameric mouse CaMKII-α hub ( PDB code: 1HKX ) , based on the N . vectensis CaMKII-B hub structure . The surface electrostatic potential of the CaMKII-α hub , calculated using the Adaptive Poisson-Boltzmann Solver ( APBS ) in PyMOL , is shown , with red and blue representing negative and positive electrostatic potential , respectively . The expanded view shows interactions between the modeled CaM-binding element and residues in the hub . ( D ) Left , structural superposition of the closed-ring tetradecameric mouse CaMKII-α hub ( PDB code: 1HKX ) on the ring-opened N . vectensis CaMKII-B hub . The structures were aligned using subunit B . Note that the disposition of helix D is different at the two interfaces ( circled ) . The expanded views show instantaneous structures ( at 1 ns and 29 ns ) from a molecular dynamics trajectory of the mouse hub , with the modeled CaM-binding element . Note that helix D in the simulation moves from its initial position to one that is closer to the position of this helix in the ring-opened N . vectensis structure . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 016 The segment that is bound at the interfacial grooves in the ruptured N . vectensis structure is not the CaM-binding element , which was not part of the crystallization construct . Instead it is the C-terminal portion of the linker , which is presumably accommodated due to the low pH of crystallization . We modeled how the CaM-binding element might dock on the hub by aligning a hub subunit from the N . vectensis CaMKII-B structure on to subunits of dodecameric human CaMKII-γ ( PDB code: 2UX0 ) and tetradecameric mouse CaMKII-α ( PDB code: 1HKX ) . This allowed us to transfer the docked peptides on to the models for the closed-ring hubs , as shown in Figure 7C for the tetradecameric ring . We then altered the sequence of the peptide to an RRKLK motif that is present in the CaM-binding element . This results in good electrostatic complementarity between the peptide sidechains and acidic sidechains in the hub ( Figure 7C ) . We initiated molecular dynamics trajectories from these structures . For the tetradecameric simulations ( two independent trajectories , 100 ns each ) , a conformational change occurs at the interface in both trajectories ( Figure 7D ) ; no corresponding change was seen in the dodecamer in these relatively short trajectories . Remarkably , the structural change closely resembles the difference seen between the ring-opened N . vectensis CaMKII-B hub and the closed tetradecameric rings in N . vectensis CaMKII-A and mouse CaMKII-α . Thus , we propose that electrostatic complementarity between the CaM-binding element and the hub interface allows the docking of this element onto the hub . Engagement of the interface by the CaM-binding element then results in a conformational change that favors a ring-opened “lock-washer” configuration of the assembly , rather than a closed ring . Choanoflagellates are the closest living relatives of metazoans ( Carr et al . , 2008; King et al . , 2008; Richter and King , 2013 ) . The choanoflagellate S . rosetta has one CaMKII gene ( Burkhardt et al . , 2014; Fairclough et al . , 2013 ) . The S . rosetta kinase domain , including the autoinhibitory segment , and the hub are 52% and 41% identical in sequence to the corresponding domains of human CaMKII-α . Although S . rosetta CaMKII contains a residue that is equivalent to Thr 286 , it lacks the two inhibitory phosphorylation sites within the CaM-binding element of the autoinhibitory segment ( Thr 305 and Thr 306 ) , as well as sites of regulatory glycosylation , nitrosylation and oxidation ( Coultrap et al . , 2014; Erickson et al . , 2008 , 2013 , 2015 ) . We expressed , purified , and crystallized the kinase and hub domains of S . rosetta CaMKII separately . The structure of the autoinhibited kinase domain was determined at 2 . 9 Å resolution , and is very similar to that of autoinhibited human CaMKII-δ kinase domain ( Rellos et al . , 2010 ) ( PDB code: 2VN9; Figure 8—figure supplement 1 ) . The S . rosetta CaMKII hub forms a vertical dimer that closely resembles those formed by the hubs of mammalian ( PDB codes: 1HKX and 2UX0 ) , the nematode C . elegans ( Rosenberg et al . , 2006 ) ( PDB code: 2F86 ) and N . vectensis CaMKII proteins . The mouse CaMKII-α hub dimer can be superimposed on the S . rosetta dimer with a rms deviation in Cα positions of 1 . 4 Å over 197 residues . Note , however , that the close overlap of the overall structure of the dimer masks an important conformational difference within each domain , which we discuss below . The quaternary assembly of the S . rosetta hub is strikingly different from that of all other hubs in that it forms a right-handed spiral , corresponding to a hexagonal P61 screw axis in the crystal lattice , instead of a closed ring ( Figure 8A ) . On looking at a projection down the hexagonal screw axis , the dimers are arrayed in a circle with almost the same diameter as that formed by the dodecameric forms of human CaMKII . An orthogonal view shows that the helical pitch of the assembly is such that the sixth dimer is displaced vertically by the length of a dimer , with respect to the first one ( Figure 8A ) . This positions the seventh dimer directly above the first one . The spiral formed by hub dimers continues in the crystal lattice . 10 . 7554/eLife . 13405 . 017Figure 8 . Structure and stoichiometry of the S . rosetta CaMKII . ( A ) Structural comparison of the hub assemblies of human CaMKII-α and S . rosetta CaMKII . Two views are shown for comparison and the six dimeric units are labeled 1–6 . ( B ) Analytical gel filtration of S . rosetta CaMKII holoenzyme ( blue ) shows one predominant peak , corresponding to a dimeric species , as estimated from calibration data ( see Figure 4—figure supplement 4 ) . ( C ) Native electrospray-ionization mass spectrometry ( ESI-MS ) for the S . rosetta CaMKII holoenzyme at different concentrations . The proportion of higher order oligomeric species increases with increasing concentration , with an even number of subunits in each major species . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 01710 . 7554/eLife . 13405 . 018Figure 8—figure supplement 1 . ( A ) Structural superposition of the autoinhibited kinase domain from S . rosetta CaMKII on the autoinhibited human CaMKII-δ kinase domain ( PDB code: 2VN9 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 01810 . 7554/eLife . 13405 . 019Figure 8—figure supplement 2 . Multi-angle light scattering ( MALS ) analysis at a concentration of ~200 µM reveals the existence of tetramers in S . rosetta CaMKII hub . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 019 The open-ended spiral assembly of S . rosetta hub dimers suggests that assemblies formed by the S . rosetta CaMKII holoenzyme would not possess defined stoichiometry . At an injection concentration of 700 nM , full-length S . rosetta CaMKII forms a dimer , as revealed by analytical gel filtration ( Figure 8B ) . Multi-angle light scattering ( MALS ) analysis at much higher concentration ( ~200 μM ) reveals the existence of tetramers ( Figure 8—figure supplement 2 ) . We also examined the concentration dependence of the stoichiometry of S . rosetta CaMKII by native ESI-MS . In contrast to the human CaMKII-α holoenzyme , for which good mass spectral signal and resolution could not be obtained , full-length S . rosetta CaMKII yielded high quality data . At 3 μM , the mass spectrum reveals mainly monomers and dimers , and a low abundance of tetramers ( Figure 8C ) . At 12 μM , the tetramer population increases , and a small population of hexamers also emerges . At 48 μM , the tetramer and hexamer populations increase further , and a population of octamers now appears . These results are consistent with the open-ended spiral assembly of the S . rosetta hub , which can gain or lose dimer units without steric constraints . The variability in the stoichiometry of S . rosetta CaMKII suggests that it should undergo subunit exchange readily , without activation . We verified that this is the case by using the FRET-based exchange assay . Separate samples of full length S . rosetta CaMKII were labeled with donor ( Alexa-488 ) and acceptor ( Alexa-594 ) fluorophores , respectively , followed by mixing at ~5 μM final subunit concentration for each sample , without ATP or Ca2+/CaM . This results in a steady increase in FRET , consistent with subunit exchange ( Figure 9—figure supplement 1 ) . We also monitored subunit exchange using native ESI-MS , which revealed that exchange proceeds through dimer units . We engineered a variant of S . rosetta CaMKII in which the linker connecting the kinase domain to the hub is deleted . This shorter variant can now be distinguished from the wild type in mass spectra . For the exchange experiment , short ( no linker ) and long ( wild type ) forms of S . rosetta CaMKII are mixed at subunit concentrations of ~280 μM each and incubated for 15 min at 37°C . This was followed by buffer exchange into 1 M ammonium acetate ( neutral pH ) and the spectrum was acquired at a final concentration of ~10 μM ( Figure 9A ) . This results in the generation of oligomers of variable stoichiometry , containing both short and long forms , in even stoichiometry ( Figure 9B ) . 10 . 7554/eLife . 13405 . 020Figure 9 . Subunit exchange in S . rosetta CaMKII monitored by ESI-MS . ( A ) Setup of the subunit exchange experiment for S . rosetta CaMKII . Wild-type CaMKII is mixed with a shorter version in which the linker is deleted , followed by mass spectrometric analysis . ( B ) Mass spectrum showing a mixture of multiple oligomeric species , ranging from monomer to hexamer . ( C ) Expansion of the mass spectrum in the region of the intact tetramer shows the existence of even numbers of long and short species . This indicates that subunit exchange in S . rosetta CaMKII occurs predominantly through the exchange of dimer units . 1L , 2L , 3L and 4L refer to CaMKII assemblies with one , two , three or four long ( wild-type ) subunits , respectively . Likewise , 1S , 2S , 3S and 4S refer to the corresponding numbers of short ( no linker ) subunits . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 02010 . 7554/eLife . 13405 . 021Figure 9—figure supplement 1 . Spontaneous subunit exchange in S . rosetta CaMKII without activation . Increase in the FRET ratio is graphed as a function of time , exhibiting spontaneous subunit exchange upon mixing donor and acceptor-labeled proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 021 The mass spectra reveal that subunit exchange occurs exclusively through exchange of dimers . For example , examination of the peaks corresponding to dimeric and tetrameric assemblies shows that the mixed species are composed predominantly of even numbers of short and long subunits ( data for tetrameric species are shown in Figure 9C ) . This is generally true for the hexameric species as well . Although a small amount of trimeric species is detected , they are composed predominantly of all short or all long subunits ( data not shown ) . This suggests that the trimer represents a species that is not in equilibrium with other species . Despite the striking dissimilarity in quaternary structure , the interface that holds dimers together in all metazoan CaMKII hubs is very closely preserved in the S . rosetta hub ( Figure 10A–C ) , with helix D from one hub domain packed against the β-sheet of the adjacent domain . When an interfacial region in the mouse tetradecameric hub assembly is superimposed on a corresponding region in the S . rosetta hub , the rms deviation in Cα positions is 0 . 35 Å over 22 residues . The interfacial hydrogen bonding network is somewhat weakened in the S . rosetta hub assembly due to replacement of a tyrosine residue by a phenylalanine ( Figure 10C ) . 10 . 7554/eLife . 13405 . 022Figure 10 . How the CaMKII hub accommodates assemblies with variable stoichiometry . ( A–C ) Comparison of interfacial interactions at the interface between vertical dimers in the mouse tetradecameric hub ( PDB code: 1HKX ) , human dodecameric hub ( PDB Code: 2UX0 ) and the S . rosetta hub respectively . ( D ) Structural comparison between a subunit of the S . rosetta spiral hub and a human dodecameric hub . The schematic diagram represents a hub domain in terms of two layers , one formed by the β-sheet and one by the α-helices . The conformational change in the hub in different assemblies corresponds to a change in the twist of the β-sheet . ( E ) Structural comparison of tetradecameric and dodecameric closed-ring hubs . Changes in the curvature of the β-sheet allow interfacial packing between adjacent subunits to be maintained in each case . The schematic illustration shows how a change in curvature of the β-sheet explains the transition from the dodecameric to the tetradecameric closed-ring forms , preserving the interfacial interactions . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 022 We wondered how the CaMKII hub can form different assemblies while maintaining the interactions that hold vertical dimers together . To answer this , we compared the S . rosetta hub structure with that of other hubs ( Figure 10D ) . Each hub domain consists of two layers: an inner layer that consists of a four-stranded antiparallel β-sheet and an outer layer formed by three α-helices , which precede the inner layer in sequence ( two shorter additional strands of the β-sheet form a segment that intervenes between two of the α-helices ) . The principal structural difference between a subunit in a closed-ring hub and one in the spiral S . rosetta hub is a rotation in the outer edge of the β-sheet with respect to the inner edge of the sheet and the α-helices ( see Figure 10D ) . The interfacial interactions between the vertical dimers of the hub are between residues on the outer edge of the β-sheet in one subunit , and residues presented by helix D in the other . These interfacial interactions are preserved in the S . rosetta hub , but a change in the curvature of the β-sheet results in a change in the orientation of helix D in each subunit . This results in a change in the quaternary assembly as the β-sheet/helix D interaction is propagated from subunit to subunit . A variation in the curvature of the β-sheet explains the variation in the assembly of all the hubs we have examined: the dodecameric and tetradecameric closed-ring hubs , the slightly ruptured left-handed spiral of the N . vectensis CaMKII-B hub and the open right-handed spiral of the S . rosetta hub ( see Figure 10E for a schematic illustration of how a change in curvature of the β-sheet explains the transition from the dodecameric to the tetradecameric closed-ring forms ) . In each case , as the extent of curvature of the β-sheet changes , the interfaces between the subunits are maintained while changing the geometry of the oligomeric assembly ( as illustrated in Figure 10E ) . Importantly , the docking of a peptide between two subunits can force a change in the curvature of the β-sheet , thereby forcing a change in the oligomeric assembly . Because the spiral S . rosetta structure so closely preserves the interactions at the interfaces , we wondered whether the human hub might be able to adopt a spiral form . We considered whether this is possible by carrying out normal mode analyses and molecular dynamics . We calculated the normal modes of a mouse CaMKII-α hub vertical dimer by using an elastic network model , as implemented in elNémo ( Bahar and Rader , 2005; Suhre and Sanejouand , 2004 ) ( see Materials and methods ) . The lowest-frequency internal normal mode corresponds to a change in the curvature of the β-sheet , such that one end of the β-sheet alters its distance from the helical layer , which stays relatively rigid . If the structure of the mouse hub domain is displaced along this normal mode , then the changes in structure recapitulate the essential differences between the structures of the subunits in different hub assemblies , including the S . rosetta spiral , to a remarkable extent ( Figure 11A ) . Low-frequency normal modes are indicative of low-energy deformations of the protein fold ( Bahar and Rader , 2005 ) , and so this suggests that the mammalian hub domains are intrinsically capable of the structural deformation that is required for the formation of an open spiral . 10 . 7554/eLife . 13405 . 023Figure 11 . Normal mode calculation and molecular dynamics simulations for the mouse and human hub respectively . ( A ) Movement of the mouse CaMKII-α hub ( PDB code: 1HKX ) along the lowest-frequency internal normal mode recapitulates the structural differences in the S . rosetta hub . Normal modes were calculated using a vertical dimer from the mouse CaMKII-α hub , using an elastic network model ( Suhre and Sanejouand , 2004 ) . In the diagram on the left , displacement vectors corresponding to the lowest-frequency internal normal mode are shown by arrows . The structures shown in magenta and blue correspond to excursions of the mouse hub along this mode . In the middle , the structures of the mouse hub and the S . rosetta hub are compared , and the arrows show the displacement vectors between the two structures . The diagram on the right shows the structures of the mouse hub and the S . rosetta hub , and excursions of the mouse hub along the lowest-frequency internal normal mode ( green ) . Note that the normal mode , calculated without reference to the S . rosetta structure , captures the essential features of the structural difference between the mouse and S . rosetta hub . ( B ) Comparison of the S . rosetta hub structure to instantaneous structures from a molecular dynamics trajectory for a decameric human hub assembly ( the open decameric assembly is generated by removing a vertical dimer from the dodecameric hub ring; PDB code: 2UX0 ) . Helices A to D in the S . rosetta hub are aligned to the corresponding helices in the instantaneous simulated structures . Three instantaneous structures from the trajectory , at 1 , 7 and 41 ns , are shown . Note that the strands of the β-sheet in structures at 7 and 41 ns are closely aligned with the S . rosetta structure . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 023 We had previously described molecular dynamics simulations in which the constraint of ring closure had been released by removing one dimer from a human dodecameric hub ( Stratton et al . , 2014 ) . We had then carried out two simulations of this system , extending for 100 ns and 50 ns , respectively . Inspection of instantaneous structures sampled from these trajectories shows that the structures of individual hub subunits undergo transient excursions in which the curvature of the β sheet is close to that seen in the S . rosetta structure ( Figure 11B ) . The hub does not go out of plane during these relatively short simulations . Presumably , adoption of the spiral form would require all subunits to change the curvature of the β-sheet in a correlated manner , which would require much longer simulation times . Finally , we generated a model for a human hub in the spiral configuration by taking the S . rosetta spiral structure and replacing the sequence of each domain by the human one . We then initiated molecular dynamics trajectories from these modeled spiral structures ( two replicates; 150 ns and 100 ns , respectively ) . The spiral geometry is stable in these trajectories , and the packing interactions at the hub interfaces remain intact ( data not shown ) . This indicates that the human sequence is consistent with the spiral architecture seen in the S . rosetta structure . The S . rosetta spiral may represent a particularly extreme distortion of the hub assembly – a smaller spiral distortion in human CaMKII than seen in S . rosetta may suffice to allow the release and capture of dimer units without the steric hindrance that is a feature of the nearly-closed N . vectensis spiral . Our results provide an understanding of how the structural integrity of CaMKII holoenzymes is altered by activation , thereby triggering subunit exchange . Subunit exchange in CaMKII requires a remarkable degree of architectural flexibility in the hub domain , which has long been an under-appreciated component of CaMKII . Although it gives the holoenzyme its distinctive shape and symmetry , the hub domain has neither the catalytic activity nor the sites of regulatory post-translational modification that has focused attention on the kinase domain and the autoinhibitory segment . We now show that the hub domain enables the CaMKII holoenzyme to switch between closed-ring dodecameric and tetradecameric states , and to potentially adopt ring-opened spiral configurations that can reversibly add and release dimer units to enable subunit exchange and interconversion between these states ( Figure 12 ) . 10 . 7554/eLife . 13405 . 024Figure 12 . Model for the interconversion of dodecameric and tetradecameric assemblies by the CaMKII hub domain . This could potentially support an efficient subunit exchange mechanism that does not require complete disassembly and reassembly . DOI: http://dx . doi . org/10 . 7554/eLife . 13405 . 024 The plasticity in the hub domain presumably emerged as it diverged in evolution from a class of enzymes typified by the keto-steroid isomerases ( Hoelz et al . , 2003; Wu et al . , 1997 ) . The fold of these enzymes , which is also shared by binding proteins such as nuclear transport factor 2 ( Stewart et al . , 1998 ) , creates an active site between the helical layer and the β sheet . As noted earlier , this cavity is deeper in CaMKII than in the enzymes to which it is related ( Hoelz et al . , 2003 ) . The cavity in the CaMKII hub contains several highly conserved charged residues , including three arginines . The positioning of these charged residues in an invagination between the two layers of the domain is what makes the hub domain so flexible – the cavity between the layers is solvated , and therefore able to readily accommodate changes in shape . The regulation of CaMKII activity by Ca2+/CaM occurs through a mechanism that is common to other Ca2+-regulated kinases – the autoinhibitory segment serves as a pseudosubstrate that occupies the substrate-binding channel , and it is released from the kinase domain by the binding of Ca2+/CaM or a related protein ( Goldberg et al . , 1996; Hu et al . , 1994; Pearson et al . , 1988 ) . Our data suggest that the CaM-binding element has an additional function , which is to dock at the interfaces between hub dimers , and thereby distort the hub assembly . Our data are consistent with the idea that the CaM-binding element induces the hub to transition from a closed-ring form to a ring-opened lock-washer configuration , which can release or capture dimer units . One attractive feature of this model , which is not directly tested here , is that subunit exchange need not involve complete disassembly and reassembly of the hub , which is likely to be a much less efficient process . The structures of CaMKII homologs from two distantly diverged organisms , the sea anemone N . vectensis and the choanoflagellate S . rosetta , provided crucial insight into the likely mechanism of subunit exchange . CaMKII , like many signaling proteins , is so highly conserved in higher metazoans that alteration in structure due to sequence variation is rather limited . Recent advances in the sequencing of genomes of early-branching metazoans and their closest relatives gives us an unprecedented opportunity to understand the core mechanisms that underlie the functions of critical signaling proteins , such as CaMKII , that are common to all metazoans ( Richter and King , 2013 ) . Activation-triggered subunit exchange in macromolecular assemblies has been reported only very rarely . One system in which such a process is thought to occur is the molecular clock formed by the KaiABC complex , which maintains diurnal rhythm in cyanobacteria ( Kageyama et al . , 2006; Markson and O'Shea , 2009 ) . The exchange of histone subunits between existing nucleosomes and newly synthesized ones may be part of the mechanism by which epigenetic modifications are maintained during replication and transcription ( Campos et al . , 2014; Xu et al . , 2010 ) . A form of activation-dependent subunit exchange underlies the function of the Mad2 protein in controlling chromosome separation ( Musacchio , 2015 ) . Certain oligomeric enzymes undergo alterations in the stoichiometry and architecture of their quaternary structures in response to the binding of allosteric effectors , a process that involves disassembly and reassembly of the subunits ( Selwood and Jaffe , 2012 ) . Activation-triggered subunit exchange provides a general mechanism to synchronize the functional states of a population of molecules , so that changes initiated by an input signal can spread from a small number of activated assemblies to many . CaMKII functions at synapses and in neuromuscular junctions , and our findings explain how its activation triggers the release of dimers that can spread and maintain the effects of calcium spikes . We anticipate that the structural mechanism we have delineated will enable future experiments in which mutations designed to alter subunit exchange in CaMKII are correlated with changes in its many crucial signaling functions . Full length CaMKII from S . rosetta and the CaMKII hub domains from N . vectensis were cloned from their respective cDNA libraries ( Fairclough et al . , 2013; Putnam et al . , 2007 ) ( S . rosetta cDNA provided by Nicole King , UC Berkeley and N . vectensis cDNA provided by Nathaniel Clarke & Christopher Lowe , Stanford University ) using standard PCR techniques . These genes , as well as DNA corresponding to the human hub , were cloned into a pET-28 vector ( Novagen ) , modified to contain a PreScission Protease ( Pharmacia ) site between the N-terminal 6-histidine tag and the coding sequence . Human full-length CaMKII-α and its variants as well as CaMKII-β were cloned into a pSMT-3 vector containing an N-terminal SUMO expression tag ( LifeSensors , Malvern , PA ) . Mutants were generated using Quikchange protocol ( Agilent Technologies , Santa Clara , CA ) and the constructs with domain truncations were made using standard PCR techniques . All human CaMKII constructs and their variants were expressed in E . coli and purified as described ( Chao et al . , 2011 ) . Briefly , for the full length CaMKII variants , protein expression was done in Tuner ( DE-3 ) pLysS cells that contained an additional plasmid for λ phosphatase production ( Chao et al . , 2010 ) . For constructs that do not have a kinase domain , as well as for calmodulin , protein expression was carried out in BL21 cells . CaMKII-β was co-expressed in Rosetta2 ( DE3 ) pLysS cells with λ phosphatase ( expressed on a separate plasmid ) . Cells were induced by the addition of 1 mM isopropyl β-D-1-thiogalactopyranoside and grown overnight at 18°C . Cell pellets were resuspended in Buffer A ( 25 mM Tris , pH 8 . 5 , 150 mM potassium chloride , 1 mM DTT , 50 mM imidazole , and 10% glycerol for human CaMKII variants and all hub domains including those from S . rosetta and N . vectensis; 25 mM Tris , pH 8 . 5 , 600 mM sodium chloride , 1 mM DTT , 50 mM imidazole , and 10% glycerol for the S . rosetta CaMKII holoenzyme ) and lysed using a cell disrupter . Filtered lysate was loaded on 5 mL Ni-NTA column , eluted with 0 . 5 M imidazole ( 0 . 76 M imidazole for human CaMKII-α hub ) , desalted using a HiPrep 26/10 desalting column into Buffer A with 10 mM imidazole , and cleaved with Ulp1 or PreScission protease ( overnight at 4°C ) . The cleaved samples were loaded onto the Ni-NTA column and the flow through was loaded onto a Q-FF 5 ml column , and then eluted with a KCl ( or NaCl in case of S . rosetta holoenzyme ) gradient . Eluted proteins were purified further using a Superose 6 gel filtration column equilibrated in gel filtration buffer ( 25 mM Tris , pH 8 . 0 , 150 mM KCl , 1 . 0 mM tris ( 2-carboxyethyl ) phosphine [TCEP] and 10% glycerol for the human CaMKII variants and all hub domains including those from S . rosetta and N . vectensis; 50 mM BisTrisPropane , pH 7 . 2 , 200 mM NaCl , 1 . 0 mM TCEP and 10% glycerol for the S . rosetta CaMKII holoenzyme; 2 mM DTT was added and no glycerol was used in the gel filtration buffer for the human CaMKII-α hub ) . One additional step was added for the human CaMKII-α hub crystallization construct only , where after the gel-filtration column , the protein was diluted into 20 mL of a solubilization buffer ( 4 M urea , 25 mM Tris , pH 8 . 0 , 150 mM KCl , 1 mM TCEP , 5% glycerol ) . The protein was then dialyzed into a final buffer of 25 mM Tris , pH 8 . 0 , 150 mM KCl , 1 . 7 mM urea , 1 mM DTT , 0 . 5 mM TCEP , and 5% glycerol . Fractions with pure protein were pooled , concentrated and frozen at −80°C . All purification steps were carried out at 4°C and all columns were purchased from GE Healthcare ( Piscataway , NJ ) . Mass spectra were acquired on a quadrupole time-of-flight mass spectrometer ( Q-TOF Premier , Waters , Milford , MA , USA ) with the backing pressure increased to ~6 mbar to promote ion desolvation of large protein complexes . Ions were formed by nanoelectrospray ionization from borosilicate capillaries ( 1 . 0 mm o . d . /0 . 78 mm i . d , Sutter Instruments , Novato , CA , USA ) that were pulled to a tip i . d . of ~1 µm with a Flaming/Brown micropipette puller ( Model P-87 , Sutter Instruments , Novato , CA , USA ) . The tip of the capillary was held ~8–10 mm from the mass spectrometer inlet , and nanoelectrospray was initiated by applying ~1 . 2–1 . 5 kV relative to instrument ground to a platinum wire ( 0 . 127 mm diameter , Sigma , St . Louis , MO , USA ) that is in contact with the sample solution . The instrument was calibrated with cesium iodide ( CsI ) clusters formed from solutions consisting of 20 mg/mL CsI in 70:30 Milli-Q water:2-propanol . Protein samples were buffer exchanged into 1 M ammonium acetate , pH 7 . 2 , via ion exchange spin columns ( Bio-Spin 6 , Bio-Rad Laboratories , Inc . , Hercules , CA , USA ) and then diluted to the desired concentration with this same buffer . Collision induced dissociation ( CID ) of protein complexes was performed by applying an accelerating voltage to ions in a collision cell containing argon gas at a pressure of 8 mbar . For non-dissociative conditions , this voltage was held at 5 V , where no complex fragmentation was observed , and for CID conditions , this voltage was increased until dissociation of the complexes was observed . Raw data were smoothed three times using the Waters MassLynx software Savitsky-Golay smoothing algorithm with a smoothing window of 50–100 m/z ( mass-to-charge ratio ) . For negative-stain electron microscopy , a 5 μL sample of CaMKII-α protein ( 15 μg/mL ) in 20 mM Tris pH 8 . 0 , 150 mM KCl and 5% glycerol was placed on the continuous carbon side of a glow-discharged copper grid ( Ted Pella , Redding , CA , USA ) , and the excess sample was removed by wicking with filter paper after 1 min incubation . The bound particles were stained by floating the grids on four consecutive 30 μL drops of 2% uranyl acetate solution and incubating each drop for 10 s . The excess stain was removed by blotting with filter paper and grids were air-dried . Images of stained full-length CaMKII-α were recorded on a 4049x4096 pixel CMOS camera ( TVIPS TemCam-F416 ) using the automated Leginon data collection software ( Suloway et al . , 2005 ) . Samples were imaged using Tecnai 12 transmission electron microscope ( FEI , Hillsboro , OR , USA ) at 120 keV at a nominal magnification of 49 , 000 ( 2 . 18 Å calibrated pixel size at the specimen level ) using a defocus range of -0 . 8 to 1 . 5 μm . All data were acquired under low dose conditions , allowing a dose at around 35e- / A2 . The initial image processing and classification steps were performed using the Appion image-processing environment ( Lander et al . , 2009 ) . Particles were first selected from micrographs using DoG Picker ( Voss et al . , 2009 ) . The contrast transfer functions ( CTFs ) of the micrographs were estimated using the CTFFIND ( Mindell and Grigorieff , 2003 ) . CTF correction of the micrographs was performed by Wiener filter using ACE2 ( Mallick et al . , 2005 ) . A total of 25 , 555 particles were extracted using a 176x176 pixel box size and binned by a factor of 2 . Each particle was normalized to remove pixels whose values were above or below 4 . 5 σ of the mean pixel value using the XMIPP normalization program ( Scheres et al . , 2008 ) . In order to remove incorrectly selected protein aggregates or other artifacts , particles whose mean intensity deviated too much from the mean value of the data set were removed . The remaining 25 , 485 particles were subjected to 2D iterative reference-free alignment and classification using a topology-representing network classification and IMAGIC multi-reference alignment ( MRA ) ( van Heel et al . , 1996; Ogura et al . , 2003 ) . The initial 200 2D class averages were manually inspected to check for classes that appeared to correspond to protein aggregates and contaminants . The 25 , 485 particles were subjected to another two rounds of MRA and classification , to produce the final 50 2D class averages . No symmetry operator was applied at any point in this analysis . For fluorescence polarization experiments , a peptide spanning the CaM-binding element of human CaMKII-α ( Peptide A: 296RRKLKGAILTTMLATR311C ) was synthesized by David King at the HHMI mass spectrometry facility , UC Berkeley . Phosphorylated peptides , corresponding to the sequence 290LKKFNARRKLKGAILTTMLA309C ( phosphorylated at either Thr 305 or Thr 306; Peptides C and D ) were obtained from Elim Biopharm ( Hayward , CA ) . Other peptides corresponding to parts of the autoinhibitory segment of human CaMKII-α and variants thereof ( Peptide B: 290LKKFNARRKLKGAILTTMLA309 , Peptide E: 290LKKFNAERKLKGAILTTMLA309 , Peptide F: 290LKKFNARRKLEGAILTTMLA309 , Peptide G: 290LKEFNAERKLEGAILTTMLA309 , ) were a gift of Leta Nutt , St . Jude Children's Research Hospital . All peptides were purified by HPLC and the purity was assessed by HPLC and/or mass spectrometry . Peptides A , C and D were labeled with BODIPY FL-maleimide ( ThermoFisher Scientific , Waltham , MA ) . BODIPY FL-maleimide was dissolved in DMSO . This was added in 1 . 5 fold molar excess to 500 μM peptide solution ( final concentration ) in Tris buffer at pH 7 . 4 in the presence of 1 mM TCEP . The reaction mixture was incubated for ~10 min at room temperature . The reaction time was optimized using analytical HPLC profiles using ~15–20 μg of peptide samples from the labeling reaction mixture . The BODIPY labeled peptides were purified using HPLC and characterized using mass spectrometry ( data not shown ) . The fluorescence polarization binding experiments were initiated by adding 15 μL of 2 nM BODIPY labeled peptides ( buffer: 25 mM Tris at pH 8 . 0 ) to 15 μL of different concentrations of the human CaMKII-α hub domain ( ranging from 0–1000 μM in 25 mM Tris at pH 8 . 0 , 150 mM KCl , 10% glycerol and 0 . 02% Tween ) . For the competition assays , 10 μL of BODIPY labeled peptides at 2 nM were added to 10 μL of hub at 30 μM . This was followed by the addition of 10 μL of the unlabeled peptides in excess ( concentration ranging between 0–500 μM in 25 mM Tris at pH 8 . 0 ) to these reaction mixtures . The fluorescence polarization was read from 20 μL of these reaction mixtures in opaque black 96-well plates using a Synergy H4 hybrid microplate reader with a 485/20 nm excitation filter , and a 528/20 nm emission filter . Error bars are calculated from the standard error of mean between replicates of experiments . The competition assay requires us to add both the hub and the unlabeled peptide at concentrations that are close to the value of KD . Under these conditions , the determination of the value of KD requires the use of an equation that is cubic in concentration ( Wang , 1995 ) . We made the simplifying assumption that the dissociation constants of the labeled and unlabeled peptides are the same , and took advantage of the fact that the concentration of the labeled peptide is very small compared to the hub concentration . Given these assumptions , the following equation ( that is second-order in concentration ) can be derived:A = Af+ ( A0− Af ) ×KD+[P]tot[P]tot× ( [P]tot+[L]tot+KD ) − ( [P]tot+[L]tot+KD ) 2−4[P]tot[L]tot2[L]tot where , A is the observed fluorescence polarization ( FP ) value , Af is the FP value for free labeled ligand , A0 is the FP value in the absence of unlabeled ligand , KD is the dissociation constant , [P]tot is the total protein concentration , and [L]tot is the total concentration of the unlabeled ligand . We determined the KD values by non-linear fitting using this equation . To optimize labeling and the FRET signal , two surface-exposed cysteine residues in CaMKII-α ( Cys 280 and Cys 289 ) were mutated to serine , and Asp 335 was mutated to cysteine ( the numbering is according to PDB code: 3SOA ) , as described ( Stratton et al . , 2014 ) . Similarly , in S . rosetta CaMKII , Cys 230 was mutated to a serine and Asp 367 was mutated to cysteine . For CaMKII-β , no mutations were made and the wild-type protein was labeled . Labeling with Alexa-488 and Alexa-594 was achieved as described ( Stratton et al . , 2014 ) . Briefly , purified CaMKII variant was mixed with three to five-fold molar excess of Alexa Fluor C5-maleimide dyes ( Alexa-488 and Alexa-594 , Life Technologies ) over CaMKII subunit concentration . This was incubated for 3–4 hr at 25°C . Excess dye at the end of the labeling reactions was removed using PD-25 or PD-10 columns ( GE healthcare ) equilibrated with 25 mM Tris at pH 8 . 0 , 150 mM KCl , 10% glycerol and 1 mM TCEP . Samples were concentrated using Amicon filters and dye incorporation was estimated using spectrophotometric analysis ( Nanodrop , Thermo Scientific , DE ) , as described ( Stratton et al . , 2014 ) . The percentage of labeled cysteine residues varied between different proteins ( ranging from ~30% to 85% ) . For the FRET assays , labeled samples were mixed at a final concentration of ~5 µM and incubated at 25°C or 37°C . At each time point , 15 or 25 μl from the mixed sample was removed and diluted to a final volume of 150 μl . An emission spectrum ( 500–700 nm ) was acquired for each diluted sample excited at 490 nm using a Fluoromax-3 fluorometer ( Horiba Scientific , Edison , NJ ) . The data were analyzed by calculating the FRET ratio ( acceptor emission at 614 nm divided by donor emission at 510 nm ) . For analytical gel filtration chromatography , samples were loaded onto a Superose 6 10/300 column ( 10/300 GL; GE Healthcare ) equilibrated with 25 mM Tris at pH 8 . 0 , 150 mM KCl , 10% glycerol and 1 mM TCEP or 2 mM DTT , at a flow rate of 0 . 3–0 . 5 ml/min ( Prominence UFLC , Shimadzu ) . Beta amylase ( 200 kDa ) , gamma globulin ( 158 kDa ) , alcohol dehydrogenase ( 150 kDa ) , bovine serum albumin ( 66 kDa ) , chicken ovalbumin ( 44 kDa ) , carbonic anhydrase ( 29 kDa ) , gamma phosphatase ( 25 kDa ) , myoglobin ( 17 kDa ) , and cytochrome C ( 12 . 4 kDa ) were used as molecular weight standards to calibrate the column ( see Figure 4—figure supplement 4 for the calibration curve ) . The standard proteins were obtained from Sigma Aldrich , Bio-Rad and New England Biolabs . The crystallization construct for the human CaMKII-α hub consisted of residues 345–475 ( UNIPROT id: Q9UQM7 ) . This sequence was preceded by an expression tag containing hexahistidine followed by a PreScission protease cleavage site ( GSSHHHHHHSSGLEVLFQGPHM ) . This expression tag was left uncleaved for crystallization . The crystallization construct for the N . vectensis hub domains ( CaMKII-B and CaMKII-A ) included residues 335–476 ( UNIPROT id: A7RF52 ) and residues 331–472 ( UNIPROT id: A7T0H5 ) respectively . For S . rosetta CaMKII , the hub domain crystallization construct is comprised of residues 335–479 and the kinase domain construct ranged from residues 1–330 ( UNIPROT id: F2UPG5 ) . The kinase domain construct for S . rosetta had an additional 11 residues , containing a hexahistidine tag on the C-terminus ( AAALEHHHHHH ) . The crystallization construct for the N . vectensis hub domains and that of S . rosetta were similar to that of the mouse CaMKII-α hub ( PDB code: 1HKX ) , and included 12 C-terminal residues of the linker ( 9 C-terminal residues of the linker were included in the construct for 1HKX ) . Crystals were grown at 22°C ( except for the human CaMKII-α hub for which the trays were stored at 4°C for the first 24 hours before moving them to 22°C ) using the sitting drop vapor diffusion technique , using the following compositions for the reservoir - human CaMKII-α hub: 35% ( v/v ) MPD , 0 . 1 M HEPES pH 7 . 3; S . rosetta hub: 0 . 2 M lithium sulphate , 0 . 1 M Tris pH 7 . 0 , 2 . 0 M ammonium sulphate; N . vectensis CaMKII-B hub ( pH 4 . 2 ) : 2 . 5 M sodium chloride , 0 . 1 M acetate pH 4 . 2 , and 0 . 25 M lithium sulphate; N . vectensis CaMKII-A hub ( pH 7 . 0 ) : 0 . 2 M potassium thiocyanate , 20% w/v PEG 3350; S . rosetta kinase: 1 . 8 M sodium phosphate monobasic monohydrate , potassium phosphate dibasic pH 6 . 9 . The crystals were cryoprotected in 25–30% glycerol and X-ray diffraction data were collected at the Advanced Light Source using beamlines 8 . 2 . 1 , 8 . 2 . 2 and 5 . 0 . 2 at 100 K . All CaMKII hub structures were solved by molecular replacement using Phaser ( McCoy et al . , 2007 ) and the structure of mouse CaMKII-α hub domain ( PDB code: 1HKX ) as the search model . The model used for the molecular replacement of the S . rosetta kinase domain was the C . elegans CaMKII-α kinase domain ( PDB code: 2BDW ) . The model building was done using Coot ( Emsley et al . , 2010 ) and refinement was performed in Phenix ( Adams et al . , 2010 ) . Model validation was performed using Molprobity ( Chen et al . , 2010 ) . The details of the data collection and refinement statistics for each structure are given in Supplementary file 1 . For the N . vectensis pH 4 . 2 structure ( CaMKII-B hub ) , there is a crystallographic axis of 2-fold symmetry that runs through the middle of the tetramer formed by the E-F and G-H dimers ( see Figure 6—figure supplement 1 ) . The left-handed lock-washer architecture allows 12 of the 14 subunits in the pH 4 . 2 structure to obey this axis of symmetry . The M-N dimer cannot obey the symmetry , because of the spiral geometry . The symmetry axis generates two copies of the M-N dimer . One copy is “joined” to the A-B dimer , and dislocated from the K-L dimer . The other copy is joined to the K-L dimer , but dislocated from the A-B dimer ( the structural assemblies showing these two alternate conformations of the M-N dimer are available as Supplementary file 2 and 3 ) . The presence of these two copies was apparent upon examination of difference electron density maps , calculated from the initial molecular replacement solution . The alternate conformations of the M-N dimer could be due to crystal twinning , or static disorder . A clear distinction cannot be made between the two at the resolution of the data . We treated this as static disorder , with 50% occupancy for each conformation of the dimeric unit . Alternating model refinement and rebuilding steps were performed using Phenix and Coot respectively . NCS restraints were imposed during refinement along with the use of the Translation-Libration-Screw-rotation model , as implemented in Phenix . For MALS studies on the S . rosetta hub , purified protein at ∼3 . 2 mg/ml was injected into a Superdex 200 10/300 analytical SEC column equilibrated overnight in gel filtration buffer ( 25 mM Tris at pH 8 . 0 , 150 mM KCl , 1 mm TCEP , and 10% glycerol ) . The chromatography system was coupled to an 18-angle light scattering detector ( DAWN HELEOS-II ) and a refractive index detector ( Optilab T-rEX ) ( Wyatt Technology ) . Data were collected every second and the flow rate was set to 0 . 5 ml/min . Data analysis was carried out using the program ASTRA ( Wyatt Technology ) . Monomeric bovine serum albumin ( BSA; Sigma ) was used for calibration of the light scattering detectors and data quality control . Measurement was carried out at 25°C . Molecular dynamics trajectories were generated using the Gromacs 5 . 1 package ( Berendsen et al . , 1995; Páll et al . , 2015 ) and Amber14 ( Case et al . , 2014 ) . The ff99SB-ILDN force field ( Lindorff-Larsen et al . , 2010 ) was used for all the calculations . All simulations were carried out in water using the TIP3P water model and appropriate counter ions ( Na+ and Cl- ) were added to neutralize the net charges . After initial energy minimization , the systems were subjected to 100–500 ps of constant number , volume and temperature ( NVT ) equilibration , during which the system was heated to 300K . This was followed by a short equilibration at constant number , pressure and temperature ( NPT , 100–500 ps ) . The equilibration steps were performed with harmonic positional restraints on all protein or peptide atoms . For the simulation with the tetradecamer and dodecamer bound to the RRKLK motif-containing peptide , additional steps of NVT and NPT equilibrations were added with harmonic positional restraints on the protein but not on the peptide atoms . Finally , the production simulations were performed under NPT conditions , with the Berendsen and v-rescale thermostats in Amber14 and Gromacs 5 . 1 . 0 respectively , in the absence of positional restraints . Periodic boundary conditions were imposed , and particle-mesh Ewald summations were used for long-range electrostatics and the van der Waals cut-off was set at 1 nm . A time step of 2 fs was employed and the structures were stored every 2 ps . The normal mode analysis was performed using the web based program elNémo and the structure of a dimer from the mouse CaMKII-α hub ( PDB code: 1HKX ) ( Suhre and Sanejouand , 2004 ) . Five low frequency modes were obtained using the default settings of the web server and visual inspection of the modes were carried out using Pymol .	How does memory outlast the lifetime of the molecules that encode it ? One enzyme that is found in neurons and has been suggested to help long-term memories to form is called CaMKII . Each CaMKII assembly is typically composed of 12 to 14 protein subunits associated in a ring and can exist in either an “unactivated” or “activated” state . In 2014 , researchers showed that CaMKII assemblies can exchange subunits with each other . Importantly , an active CaMKII can mix with an unactivated CaMKII and share its activation state . CaMKII may use this mechanism to spread information to the next generation of proteins – thereby allowing activation to outlast the lifespan of the initially activated proteins . However the molecular mechanism that underlies this process was not clear . Now , Bhattacharyya et al . – including some of the researchers involved in the 2014 work – address two questions about this mechanism . How do subunits exchange between CaMKII assemblies ? And how does the activation of CaMKII initiate subunit exchange ? A closed-ring hub ties the subunits of CaMKII together , similar to the organization of the segments in an orange . To undergo subunit exchange , the hub must open up to release and accept subunits . Bhattacharyya et al . have now uncovered an intrinsic flexibility in the hub that is triggered by a short peptide segment in CaMKII . This segment , which is exposed in activated CaMKII but not in the unactivated form , can crack open the hub ring by binding between the hub subunits , like a finger separating the segments of an orange . This allows the hub to flex and expand , and once open , the hub’s flexibility allows room for subunits to be released or accepted . Although this subunit exchange mechanism could be a powerful means for spreading the activated state throughout signaling pathways , the biological relevance of this phenomenon has not been clarified . However , the mechanistic framework provided by Bhattacharyya et al . may allow new experiments to be performed that test the consequences of subunit exchange in live cells and organisms . It could also enable investigations into the importance of subunit exchange in long-term memory .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2016	Molecular mechanism of activation-triggered subunit exchange in Ca2+/calmodulin-dependent protein kinase II
Subjective well-being or happiness is often associated with wealth . Recent studies suggest that momentary happiness is associated with reward prediction error , the difference between experienced and predicted reward , a key component of adaptive behaviour . We tested subjects in a reinforcement learning task in which reward size and probability were uncorrelated , allowing us to dissociate between the contributions of reward and learning to happiness . Using computational modelling , we found convergent evidence across stable and volatile learning tasks that happiness , like behaviour , is sensitive to learning-relevant variables ( i . e . probability prediction error ) . Unlike behaviour , happiness is not sensitive to learning-irrelevant variables ( i . e . reward prediction error ) . Increasing volatility reduces how many past trials influence behaviour but not happiness . Finally , depressive symptoms reduce happiness more in volatile than stable environments . Our results suggest that how we learn about our world may be more important for how we feel than the rewards we actually receive . Decisions are guided by beliefs about states of the world . Some states are directly observable , like the potential prize for a bet . Other states , like the probability of winning , may not be directly observable but can be inferred from past events . Thus , learning from experience is essential for adaptive behaviour . In the standard theoretical framework , learning is driven by how unexpected the outcome is ( i . e . by the prediction error ) : the difference between outcome and prediction ( Barto , 1995 ) . Sensitivity to the prediction error ( i . e . the learning rate ) flexibly adapts to environmental statistics . The decisions of both humans and non-human primates are consistent with a higher learning rate in more volatile environments , and subjects are more likely to stay on the same option after positive compared to negative feedback when reward probabilities change more frequently ( Behrens et al . , 2007; Browning et al . , 2015; Donahue and Lee , 2015; Massi et al . , 2018; Mathys et al . , 2011 ) . Emotions are widely believed to play a role in adaptive behaviour ( Fredrickson , 2004 ) , but no computational framework exists to link them . Unexpected outcomes influence affective states , so that bad outcomes feel worse when unexpected than when expected , and good outcomes feel better when unexpected than when expected ( Mellers et al . , 1997; Shepperd and Mcnulty , 2002 ) . It has recently been shown that reward expectations and reward prediction errors ( RPEs ) , the difference between experienced and predicted rewards , can explain changes in affective state in the context of decision-making under uncertainty when learning is not required ( Rutledge et al . , 2014; Rutledge et al . , 2015 ) . A number of studies have found results consistent with the idea that happiness is modulated by past prediction errors ( Otto et al . , 2016 ) including a recent report showing in students that prediction errors due to exam performance influence real-world emotions ( Villano et al . , 2020 ) . Mood has been proposed to represent environmental momentum , whether an environment is getting better or worse , which could be a useful variable for adaptive behaviour ( Eldar et al . , 2016; Eldar and Niv , 2015 ) . Impairments in reward and emotion processing are associated with depression ( Shepperd and Mcnulty , 2002 ) . Learning in depression has been extensively studied , but there is not consistent evidence for a specific deficit ( Blanco et al . , 2013; Cella et al . , 2010; Chase et al . , 2010; Gillan et al . , 2016; Herzallah et al . , 2013; Kunisato et al . , 2012; Mueller et al . , 2015; Pechtel et al . , 2013; Robinson et al . , 2012; Taylor Tavares et al . , 2008; Thoma et al . , 2015; Vrieze et al . , 2013 ) , as reviewed in Huys et al . , 2013; Scholl and Klein-Flügge , 2018 . The ability of individuals to appropriately adjust learning rates to environmental volatility is associated with anxiety symptoms ( Browning et al . , 2015 ) . Individuals with high trait anxiety show reduced ability to adjust updating of outcome expectancies for aversive outcomes to the volatility of the environment compared to individuals with low trait anxiety . Failure to appropriately adjust learning to environmental volatility has not been established in depression . Affective states reflect subjective estimates of uncertainty , which predict the dynamics of subjective and physiological stress responses ( de Berker et al . , 2016 ) . Overall mood during risky decision-making tasks is reduced with increasing depression severity , both in the laboratory and using remote smartphone-based data collection ( Rutledge et al . , 2017 ) . Here , our goal was to quantify the relationship between mood and adaptive behaviour in two common reinforcement learning tasks ( Figure 1 ) : one in which reward probabilities do not change ( stable ) and one in which reward probabilities periodically change ( volatile ) . We addressed the following questions: ( 1 ) Is mood more sensitive to learning-relevant or learning-irrelevant variables in established reinforcement learning models ? ( 2 ) Do mood dynamics adjust to environmental volatility ? ( 3 ) Are mood dynamics affected by depression in the context of learning ? Participants chose the option with the higher expected value ( i . e . choice accuracy ) more often than chance ( see Figure 2A ) in the stable environment ( 82 . 1 ± 1 . 1% ( mean ± SEM ) , z = 7 . 5 , p < 10−13 ) and in the volatile environment ( 73 . 3 ± 1 . 0% , z = 7 . 5 , p < 10−13 ) . Participants chose the higher probability option more often than chance in the stable environment ( 80 . 8 ± 1 . 0% , mean ± SEM , z = 7 . 5 , p < 10−13 ) and in the volatile environment ( 62 . 6 ± 1 . 1% , z = 7 . 2 , p < 10−12 ) . To ensure that participants incorporated information about the magnitudes of potential rewards into their decisions , we considered only trials where the car with the lower outcome probability had the higher expected value . Subjects chose the low probability car in these trials more often than chance ( stable: 61 . 2 ± 3 . 0% , z = 3 . 5 , p < 0 . 001; volatile: 75 . 6 ± 2 . 7% , z = 6 . 4 , p < 10−9 ) . These results are consistent with participants integrating both probability and reward to make their decisions . Multiplicative and additive models that integrate probability and reward in different ways have been widely used to explain behaviour across stable and volatile environments ( Behrens et al . , 2007; Browning et al . , 2015; Donahue and Lee , 2015; Massi et al . , 2018 ) . Both types of models include the same learning component for updating the probability estimate for each car to win ( Equations 1 and 2 ) , based on the probability prediction error ( PPE ) , the difference between the outcome ( 0 or 1 ) and the estimated probability of winning: ( 1 ) Pcar1 wins ( t+1 ) =Pcar1 wins ( t ) +α PPE ( t ) , ( 2 ) PPE ( t ) =Outcome ( t ) −Pcar1 wins ( t ) , where Pcar1 wins ( t ) is the estimated probability for car 1 winning on trial t , and α is the learning rate . For choices to car 2 , a similar equation applies . Multiplicative and additive models differ regarding implementation of choice predictions with probability and reward magnitude either integrated multiplicatively ( Behrens et al . , 2007; Browning et al . , 2015 ) or additively ( Donahue and Lee , 2015; Farashahi et al . , 2017; Massi et al . , 2018 ) . The multiplicative selector resembles the maximisation of expected utility common to economic decision models ( Kahneman and Tversky , 1979 ) : ( 3 ) EUcar1 ( t ) =max[min[η ( ( Pcar1wins ( t ) −0 . 5 ) +0 . 5 ) , 1] , 0]×Rcar1wins ( t ) ( 4 ) Pcar1 chosen ( t ) =11+exp⁡ ( −β ( EUcar1 ( t ) −EUcar2 ( t ) ) ) , where η is a free parameter related to the level of risk aversion and β is the inverse temperature ( i . e . choice stochasticity or precision ) , and Rcar1winst is the reward magnitude if car 1 is chosen and wins . The multiplicative model captured choice data in both stable ( pseudo-r2 = 0 . 54 ± 0 . 02 , mean ± SEM ) and volatile environments ( pseudo-r2 = 0 . 41 ± 0 . 02 ) . In contrast , the additive selector is implemented as follows: ( 5 ) Pcar1chosen ( t ) =11+exp⁡ ( −β ( ϕ ΔProbability ( t ) + ( 1−ϕ ) ΔReward ( t ) ) ) where ∆Probability corresponds to the difference in estimated probability between the options and ∆Reward corresponds to the difference in normalised reward magnitude between the options , and ϕ represents the relative weight of probability and magnitude on choice . The additive model captured choice data in the stable ( pseudo-r2 = 0 . 62 ± 0 . 02 ) and volatile environments ( pseudo-r2 = 0 . 45 ± 0 . 02 , see Figure 2A ) . Model comparison demonstrated that the additive model better explained choice data with the same number of parameters as the multiplicative model in both stable ( ∆BIC = 690 ) and volatile environment ( ∆BIC = 321 , see Table 1 ) . Our results are consistent with similar findings obtained in highly trained non-human primates using a task design in which changes between stable and volatile environments were signalled ( Massi et al . , 2018 ) as in the present study . We next examined whether model fits were consistent with subjects integrating both potential reward magnitudes and probabilities to make decisions . The relative weight of probability and reward magnitude in the additive model ( ϕ ) was balanced on average ( stable: ϕ = 0 . 57 ± 0 . 017 [mean ± SEM]; volatile: ϕ = 0 . 44 ± 0 . 027 ) , suggesting that subjects integrated both probabilities and reward magnitudes to make decisions . Omitting ϕ and evaluating simpler models that considered only probabilities ( α and β ) or reward magnitudes ( β only ) resulted in worse fits ( see Table 1 ) . Lower BIC for additive and multiplicate models compared to the simpler models confirmed that subjects considered both probability and reward magnitude when making decisions . We then asked whether greater environmental volatility was associated with higher learning rates as observed in previous studies ( Behrens et al . , 2007; Browning et al . , 2015; Massi et al . , 2018 ) . A simple prediction for standard reinforcement learning models is that subjects should stay more on the same option after winning than losing ( Figure 2B ) . Subjects did not stay more on the same option after winning than losing for the stable environment ( difference in choice proportion , high probability car: 4 . 1 ± 1 . 9% , mean ± SEM , z = 1 . 7 , p = 0 . 098; low probability car: 1 . 3 ± 3 . 2% , z = -0 . 43 , p = 0 . 67 ) . Subjects stayed on the same option more after winning than losing in the volatile environment ( difference in choice proportion , high probability car: 22 . 1 ± 2 . 3% , z = 6 . 6 , p < 10−10 , low probability car: 21 . 4 ± 3 . 7% , z = 4 . 8 , p < 10−5 ) . Subjects stayed on the same option after winning compared to losing more in volatile than stable environments ( difference volatile – stable , high probability car: 18 . 1 ± 2 . 7% , z = 5 . 7 , p < 10−7 , low probability car: 19 . 8 ± 4 . 1% , z = 4 . 0 , p < 10−4; see Figure 2B ) . We then checked that the predictions generated by a reinforcement learning model fit separately to each environment correspond to observed behavioural patterns described above in model-independent analyses ( predicted difference in choice proportion after winning and losing in stable [high probability car: 8 . 8 ± 1 . 6% , z = 5 . 2 , p < 10−6 , low probability car: 6 . 1 ± 2 . 6% , z = 1 . 5 , p = 0 . 12] and volatile environments [high probability car: 21 . 5 ± 2 . 2% , z = 6 . 9 , p < 10−11 , low probability car: 21 . 1 ± 2 . 7% , z = 5 . 9 , p < 10−8; see Figure 2B] ) . The model predictions were able to capture observed differences in behaviour following wins and losses and also the difference in in choice proportion after winning and losing between volatile and stable environments ( high probability car: 12 . 8 ± 2 . 4% , z = 4 . 6 , p < 10−5; low probability car: 14 . 3 ± 3 . 6 , z = 3 . 4 , p < 0 . 001 ) . We found that learning rates ( Figure 2C ) were substantially higher in volatile than stable environments ( stable α = 0 . 16 ± 0 . 02 , mean ± SEM; volatile α = 0 . 47 ± 0 . 03; difference volatile – stable: 0 . 31 ± 0 . 03 , z = 6 . 9 , p < 10−11 ) . Overall , these results demonstrate that the learning rate increases substantially in the volatile compared to the stable environment , in line with previous studies ( Behrens et al . , 2007; Browning et al . , 2015; Massi et al . , 2018 ) . We next examined how happiness changes over time during the tasks . Subjects varied their happiness ratings in both the stable ( SD = 24 . 2 ± 1 . 1 , mean ± SEM ) and volatile ( SD = 25 . 0 ± 1 . 1 ) environments . They were happier on average after winning than after losing ( stable: 63 . 8 ± 1 . 9 vs 34 . 5 ± 2 . 0 , z = 7 . 5 , p < 10−13; volatile: 61 . 9 ± 1 . 8 vs 33 . 6 ± 2 . 0 , z = 7 . 5 , p < 10−13 , Figure 3A ) . Participants were happier on average in the stable environment than in the volatile environment ( stable: 55 . 0 ± 1 . 7 , volatile: 49 . 5 ± 1 . 6 , z = 3 . 7 , p < 0 . 001 ) . This effect may be at least partly due to lower choice accuracy in volatile environments , and the difference in average happiness between environments was correlated between participants with the difference in choice accuracy in terms of EV maximisation ( Spearman’s ρ ( 73 ) = 0 . 24 , p < 0 . 05 ) . Previous studies have reported that momentary happiness in response to outcomes in a probabilistic reward task were explained by recent RPEs when maximising cumulative reward does not require learning ( Rutledge et al . , 2014; Rutledge et al . , 2015 ) . In our task , maximising cumulative reward requires learning the outcome probability . In this context , PPEs ( depending on whether the outcome was a win or a loss and the subjective probability of that outcome ) are relevant to learning but RPEs ( depending on the magnitude of the reward received and the expected value of the chosen option ) are not relevant to learning or future behaviour . Reward information was choice relevant and choices were driven by the ( additive ) integration of the estimated outcome probability and the magnitude of potential rewards . Therefore , we tested whether happiness was more strongly associated with the PPEs used for learning or alternatively by RPEs that incorporate learning-irrelevant reward magnitudes . We compared two models which both use the subjective probability as estimated in the additive choice model to compute prediction errors: ( 6 ) Happiness ( t ) =w0+wPPE^∑j=1tγt−jPPEj^ , where PPE^ refers to the probability prediction error ( PPE ) , defined as the difference between the outcome ( one for win , 0 for loss ) and the subjective probability estimated from the additive choice model , w0 is a constant term , wPPE is a weight capturing the influence of past PPEs , and 0 ≤ γ ≤1 is a forgetting factor that makes events in more recent trials more influential than those in earlier trials; ( 7 ) Happiness ( t ) =w0+wRPE^∑j=1tγt−jRPEj^ , where RPE^ is the difference between reward magnitude and the expected value of the chosen option computed based on the subjective probability estimated from the additive choice model . Reward magnitudes were rescaled from 0 to 1 . Mood fluctuations were better explained by a model including past PPEs than by a model including past RPEs , both in the stable ( BICPP̂E = −698 , BICRP̂E = −299 , ∆BIC = 399 ) and volatile ( BICPP̂E = −559 , BICRP̂E = −319 , ∆BIC = 240 ) environments ( see Table 2 and Figure 4A ) . This result holds for a broader model space including other definitions of the prediction error terms ( see Table 2 and Figure 4 ) : ( 8 ) Happiness ( t ) =w0+wPPE∑j=1tγt−jPPEj , where PPE refers to the objective PPE defined as the difference between the outcome sign and the objective probability of the chosen option ( 0 . 2 or 0 . 8 ) , and also: ( 9 ) Happiness ( t ) =w0+wRPE∑j=1tγt−jRPEj , where RPE is computed by taking the difference between the reward magnitude and the objective expected value ( potential reward multiplied by the objective probability of the chosen option ) as above . We then tested whether mood fluctuations were additionally sensitive to current expectations , as shown in risky choice tasks that do not require learning ( Rutledge et al . , 2014; Rutledge et al . , 2015 ) . Again , two types of expectations may explain mood fluctuations: a subjective probability relevant to learning , or expected values that incorporate learning-irrelevant reward magnitudes . We compared the following models: ( 10 ) Happiness ( t ) =w0+wP^∑j=1tγt−j ( P^−0 . 5 ) +wPPE^∑j=1tγt−jPPEj^ , where P^ is the probability estimated with the additive choice model , and: ( 11 ) Happiness ( t ) =w0+wEV^∑j=1tγt−j ( EV^−EV− ) +wPPE^∑j=1tγt−jPPEj^ , where EV^ is the product between P^ and the reward magnitude and this term is mean-centred . The model including choice probability better explained happiness ratings ( stable: mean r2 = 0 . 58; volatile: mean r2 = 0 . 62 ) than the model including the expected value in the stable ( BICP̂+PP̂E = −882 , BICEV+PP̂E = −752 , ∆BIC = 130 ) and in the volatile ( BICP̂+PP̂E = −1147 , BICEV+PP̂E = −691 , ∆BIC = 454 ) environments ( see Table 2 and Figure 4B ) . The probability and PPE weights were significantly different from 0 at the group level in both the stable ( wP̂ = 0 . 74 ± 0 . 09 , z = 6 . 2 , p < 10−9; wPP̂E = 1 . 32 ± 0 . 06 , z = 7 . 5 , p < 10−14 ) and the volatile environments ( wP̂ = 0 . 94 ± 0 . 09 , z = 6 . 7 , p < 10−10; wPP̂E = 1 . 14 ± 0 . 05 , p < 10−14 , Figure 3C ) . We next extended the model space with plausible alternative models . We included two models incorporating the history of reward magnitude . In the first model , we centred the reward magnitude regressor for each participant . This model thus predicts that reward magnitudes larger than the averaged reward magnitude will increase happiness , and that the larger the reward magnitude , the greater the happiness . ( 12 ) Happiness ( t ) =w0+wR∑j=1tγt−j ( Rj−R− ) , where Rj is the reward magnitude at trial j and R- is the average reward magnitude . Instead of assuming a reference point of the average reward , we also used a free parameter in a subsequent model above and below which reward magnitudes increase or decrease happiness , respectively: ( 13 ) Happiness ( t ) =w0+wR∑j=1tγt−j ( Rj−RP ) , where RP is a free parameter corresponding to the reference point in an individual subject . If this value is 0 , receipt of rewards always increases happiness in proportion to the reward magnitude , so this model also provides a test of whether failing to obtain reward decreases happiness during reinforcement learning . The average reward magnitude was 25 . 5 ± 2 . 5 ( mean ± SD ) points in the stable environment and 22 . 6 ± 3 . 0 points in the volatile environment . The reference point RP estimated in model 14 was on average 14 . 2 ± 9 . 5 points and greater than 0 in the stable environment ( z = 7 . 0 , p < 10−11 ) and 14 . 8 ± 7 . 3 points and greater than 0 in the volatile environment ( z = 7 . 4 , p < 10−12 ) . This result supports the idea that obtaining 0 points on a trial is aversive: failing to obtain reward decreases happiness in our tasks . We also included two additional models incorporating RPE^ . ( 14 ) Happiness ( t ) =w0+wEV^∑j=1tγt−j ( EVj^−EV− ) +wRPE^∑j=1tγt−jRPEj^ , where EV^ corresponds to the expected value of the chosen option , corresponding to the weighted sum of probability and reward estimated based on each individual participant’s choices with the additive choice model , EV¯ corresponds to the averaged expected reward , and RPE^ corresponds to the difference between the outcome reward magnitude and EV^ . We also included a model combining the estimated probability with the reward prediction error . ( 15 ) Happiness ( t ) =w0+wP^∑j=1tγt−j ( Pj^−0 . 5 ) +wRPE^∑j=1tγt−jRPEj^ , Besides the constant and the forgetting factor , the P^+PPE^ model includes two parameters , one for predictions ( wP̂ ) and one for probability prediction error ( wPP̂E ) . We also asked whether the history of wins ( excluding any information about reward magnitude ) and losses could account for happiness by fitting the following model: ( 16 ) Happinesst=w0+wwin∑j=1tγt-jwinj-wloss∑j=1tγt-jlossj As reported in Table 2 , the model evidence for this new model was similar to the P^+PPE^ model overall . We next used estimated model frequency to compare both models . The P^+PPE^ is preferred to the win-loss model in the stable environment ( EFP̂+PP̂E = 0 . 65 ± 0 . 05 , EFwin-loss = 0 . 35 ± 0 . 05 , exceedance probability = 0 . 99 ) . However , both models performed similarly in the volatile environment ( EFP̂+PP̂E = 0 . 50 ± 0 . 06 , EFwin-loss = 0 . 50 ± 0 . 06 , exceedance probability = 0 . 48; see Figure 4—figure supplement 1 ) . We next asked whether an alternative analysis could test whether happiness was influenced by trial-by-trial probability estimates , the key difference between the models . If the weights for the two terms of the P^+PPE^ model are identical , the equation mathematically reduces to a constant plus an exponentially weighted average of previous wins . However , the wPP̂E parameter was larger than wP̂ in both stable ( z = 6 . 0 , p < 10−8 ) and volatile tasks ( z = 2 . 35 , p < 0 . 05 ) . The difference between wPP̂E and wP̂ was significantly larger in the stable compared to the volatile task ( z = 3 . 7 , p < 0 . 001 ) . Comparison of weights across tasks therefore suggests a reduced impact of expectations on happiness as environmental volatility increases . We next computed the residuals of the win-loss model ( which does not include probability estimates ) and tested for a correlation with trial-by-trial probability estimates . Because prediction errors are equal to outcomes minus expectations and numerically wP̂ is lower than wPP̂E in both environments , the overall influence of probability on happiness should be negative after accounting for the impact of wins and losses . In the stable environment , we found the expected negative correlation between the win-loss model residuals and trial-by-trial probability estimates ( average Spearman’s ρ ( 73 ) = −0 . 06 ± 0 . 03 , z = 2 . 2 , p = 0 . 03 ) . This relationship was not present in the volatile environment ( average Spearman’s ρ ( 73 ) =−0 . 02 ± 0 . 03 , z = 0 . 65 , p = 0 . 51 ) . A potential explanation for this pattern of results is that expectations cannot affect happiness when participants do not have strong predictions , as it is the case immediately after reversals in the volatile condition . This would be consistent with findings from the animal literature showing that dopamine early in training does not represent prediction errors ( Coddington and Dudman , 2018 ) . Finally , we focused on the win-loss model . We asked whether weights from the win-loss model were positively correlated , consistent with similar but opposite impacts . We found instead a negative correlation across participants between win and loss weights ( stable: Spearman’s ρ ( 73 ) = −0 . 56 , p < 10−6; volatile: Spearman’s ρ ( 73 ) = −0 . 68 , p < 10−20 ) , suggesting that individuals that respond to wins tend to respond less to losses and vice versa . Indeed , comparing the weight of wins and losses shows that participants reacted more strongly on average to losses than to wins ( difference in stable: 0 . 69 ± 0 . 12 , z = 5 . 2 , p < 10−6; difference in volatile: 0 . 31 ± 0 . 13 , z = 2 . 7 , p < 0 . 01; see Figure 3—figure supplement 1 ) . Given that participants received positive feedback on average in 81% of trials in the stable environment and 63% of trials in the volatile environment , asymmetric responses to wins and losses are consistent with happiness reflecting knowledge of the underlying structure of both environments . Interestingly , the difference between win and loss weights was not correlated across participants with overall performance including the percentage of trials with positive feedback ( stable: Spearman’s ρ ( 73 ) = −0 . 05 , p = 0 . 65; volatile: Spearman’s ρ ( 73 ) = −0 . 18 , p = 0 . 12 ) or the percentage of trials where the higher expected value option was chosen ( stable: Spearman’s ρ ( 73 ) = 0 . 04 , p = 0 . 74; volatile: Spearman’s ρ ( 73 ) = −0 . 17 , p = 0 . 14 ) . Our results suggest that although reward information influences choice , contrary to what would be predicted from the literature , RPEs and reward magnitudes do not explain happiness when this information is not necessary for participants to learn the structure of the environment . Happiness reflects knowledge of the underlying structure of the environment in a way that cannot be explained by simple performance metrics . RPEs are relevant to learning in many paradigms , and happiness should relate to RPEs in such tasks because of their value for learning the structure of environment . Learning and reward are dissociable in our paradigm , and we find in this context that RPEs and reward magnitudes do not explain happiness . We found that the learning rate ( i . e . , behavioural sensitivity to PPE ) was approximately three times higher in the volatile compared to the stable environment . Furthermore , mood dynamics were highly sensitive to PPE . However , PPE weights were actually higher in stable than volatile environments ( ∆wPP̂E = 0 . 18 ± 0 . 05 , z = 3 . 5 , p < 0 . 001; see Figure 3C ) . Futhermore , the difference between wPP̂E and wP̂ was greater in stable than volatile environments ( stable – volatile: 0 . 39 ± 0 . 11 , z = 3 . 7 , p < 0 . 001 ) , consistent with a greater influence of trial-by-trial probability estimates on happiness in stable environments ( see previous section ) . PPE weights were not correlated between participants with the learning rate in the stable ( Spearman’s ρ ( 73 ) = −0 . 10 , p = 0 . 40 ) and volatile ( Spearman’s ρ ( 73 ) = −0 . 1 , p = 0 . 37 ) environments nor was the difference of PPE weights across environments related to the difference in learning rate ( Spearman’s ρ ( 73 ) = −0 . 02 , p = 0 . 84 ) . Instead , PPE weights were highly consistent across environments ( Spearman’s ρ ( 73 ) = 0 . 44 , p < 0 . 001 , see Figure 5A ) . The happiness model forgetting factor γ determines how many previous trials influence current affective state . When γ is equal to 1 , mood is equally influenced by all previous trials , when γ is equal to 0 , mood is influenced by only the most recent trial . If the change in forgetting factor mirrors behaviour , forgetting factors should be lower in volatile than stable environments , reflecting integration over fewer trials and consistent with the higher learning rates observed in volatile compared to stable environments . Instead , the forgetting factor was slightly higher on average in the volatile environment ( stable: γ = 0 . 59 ± 0 . 04 , volatile γ = 0 . 63 ± 0 . 03 , corresponding to current happiness being influenced by 6–7 previous trials on average in both environments , stable – volatile: ∆γ = 0 . 05 ± 0 . 03 , z = 1 . 9 , p = 0 . 064; see Figure 5B ) . Higher values for γ in volatile environments are not consistent with happiness integrating over fewer trials as behaviour would predict . Furthermore , the change in happiness forgetting factor was not correlated across participants with the learning rate difference between environments ( Spearman’s ρ ( 73 ) = −0 . 08 , p = 0 . 50 ) . A linear regression with ten previous probability prediction errors as independent variables confirmed this model-based result ( see Figure 5—figure supplement 1 ) . To further test for a relationship between the forgetting factor and the learning rate , we switched the learning rates estimated from stable and volatile conditions in each individual before re-fitting happiness data ( i . e . , we used the ‘wrong’ learning rate to estimate probabilities and PPEs before fitting the happiness model and estimating a forgetting factor ) . This did not substantially affect estimates of the happiness forgetting factor ( stable γ = 0 . 58 ± 0 . 03 , volatile γ = 0 . 62 ± 0 . 03 , stable – volatile: ∆γ = 0 . 032 ± 0 . 026 , z = 1 . 4 , p = 0 . 16 ) . The resulting forgetting factor estimates were highly correlated with forgetting factors estimated using the actual learning rates ( stable: Spearman’s ρ ( 73 ) = 0 . 63 , p < 10−8; volatile: Spearman’s ρ ( 73 ) = 0 . 56 , p < 10−6 ) . The happiness forgetting factor was highly consistent across environments ( Spearman’s ρ ( 73 ) = 0 . 41 , p < 0 . 001 , see Figure 5C ) , suggesting that the number of previous trials that affective state depends on may be a trait-like feature of individuals unrelated to environmental volatility . Previous studies have linked learning rates to anxiety ( Browning et al . , 2015; Pulcu and Browning , 2019 ) and individuals with high trait anxiety showed less ability to appropriately adjust updating of outcome expectancies between stable and volatile environments . We found that depressive symptoms ( PHQ ) were uncorrelated across participants with choice accuracy ( stable: Spearman’s ρ ( 73 ) = −0 . 06 , p = 0 . 63; volatile: Spearman’s ρ ( 73 ) = −0 . 09 , p = 0 . 46; volatile – stable: Spearman’s ρ ( 73 ) = 0 . 04 , p = 0 . 71 ) and all parameters estimated in the additive choice model ( α , stable: Spearman’s ρ ( 73 ) = 0 . 08 , p = 0 . 51; volatile: Spearman’s ρ ( 73 ) = 0 . 10 , p = 0 . 38; volatile – stable: Spearman’s ρ ( 73 ) = 0 . 19 , p = 0 . 1; ϕ , stable: Spearman’s ρ ( 73 ) = 0 . 20 , p = 0 . 09; volatile: Spearman’s ρ ( 73 ) = 0 . 01 , p = 0 . 94; volatile – stable: Spearman’s ρ ( 73 ) = −0 . 20 , p = 0 . 09; β , stable: Spearman’s ρ ( 73 ) = −0 . 04 , p = 0 . 76; volatile: Spearman’s ρ ( 73 ) = −0 . 07 , p = 0 . 53; volatile – stable: Spearman’s ρ ( 73 ) = 0 . 04 , p = 0 . 75 ) . In the stable environment , where uncertainty and volatility are low , average happiness did not correlate across participants with depressive symptoms ( Spearman’s ρ ( 73 ) = 0 . 07 , p = 0 . 58; Figure 6A , left panel ) . In the volatile environment , where uncertainty is high and volatility is high , average happiness was correlated with depressive symptoms , with lower happiness associated with higher depressive symptoms ( Spearman’s ρ ( 73 ) = −0 . 23 , p = 0 . 043; Figure 6A , central panel ) . Finally , the difference between average happiness between volatile and stable environments was also correlated with depressive symptoms even after standardising the variables ( Wilcox and Tian , 2008 ) ( volatile – stable: Spearman’s ρ ( 73 ) = −0 . 28 , p = 0 . 014; Figure 6A , right panel ) . Baseline mood parameters estimated using our happiness model fit to non-z-scored happiness ratings showed the same relationship to depressive symptoms ( stable: Spearman’s ρ ( 73 ) = −0 . 07 , p = 0 . 54; volatile: Spearman’s ρ ( 73 ) = −0 . 28 , p = 0 . 017; volatile – stable , standardised: Spearman’s ρ ( 73 ) = −0 . 32 , p = 0 . 0049; see Figure 6B ) . No other happiness model parameters were correlated with depressive symptoms ( wP̂ , stable: Spearman’s ρ ( 73 ) = 0 . 20 , p = 0 . 09; volatile: Spearman’s ρ ( 73 ) = 0 . 15 , p = 0 . 19; volatile – stable: Spearman’s ρ ( 73 ) = −0 . 02 , p = 0 . 84; wPP̂E , stable: Spearman’s ρ ( 73 ) = 0 . 09 , p = 0 . 45; volatile: Spearman’s ρ ( 73 ) = 0 . 07 , p = 0 . 54; volatile – stable: Spearman’s ρ ( 73 ) = −0 . 03 , p = 0 . 81; γ , stable: Spearman’s ρ ( 73 ) = 0 . 06 , p = 0 . 63; volatile: Spearman’s ρ ( 73 ) = −0 . 09 , p = 0 . 44; volatile – stable: Spearman’s ρ ( 73 ) = −0 . 12 , p = 0 . 28 ) . We found that subjects tracked outcome probabilities and made decisions by integrating both learned probability and explicit reward magnitudes . Learning rates adapted to environmental volatility , with a higher learning rate in the more volatile environment consistent with previous studies ( Behrens et al . , 2007; Browning et al . , 2015; Massi et al . , 2018 ) . That behaviour was consistent with an additive choice model ( Donahue and Lee , 2015; Farashahi et al . , 2017; Massi et al . , 2018; Rouault et al . , 2019 ) which is consistent with recent empirical evidence ( Farashahi et al . , 2019; Koechlin , 2020 ) that humans and non-human primates adopt a multiplicative strategy under risk when probabilities are explicit , but both species adopt an additive strategy under uncertainty when probabilities must be learned . Our tasks required learning the probability of getting a reward and the reward magnitude was explicitly given . In such an environment , mood dynamics were more closely related to learning-relevant variables than learning-irrelevant variables and we found convergent evidence that this was the case across both stable and volatile learning tasks . Mood was sensitive to the combined influence of past chosen subjective probabilities and past PPEs . Parameters for PPE and forgetting factors estimated from happiness ratings were correlated across stable and volatile environments . Finally , we found that although choice accuracy and choice model parameters were not affected by depressive symptoms when changes between safe and volatile environments are signalled , the decrease in happiness observed in the volatile relative to the stable environment was correlated with symptom severity . The same pattern was present for the baseline mood parameter in the happiness model . Experiencing a stable environment with low uncertainty and volatility could attenuate the expression of depressive symptoms on mood . Risky decision tasks used in previous studies ( Rutledge et al . , 2017; Rutledge et al . , 2015; Rutledge et al . , 2014 ) maximise irreducible uncertainty ( i . e . , risky options had a 50% probability of each option ) , which is more comparable to the volatile environment in the current study and may explain the previous finding of a link between baseline mood parameters and depressive symptoms . Computational models that capture ecologically relevant learning and decision processes may provide a critical advantage for understanding the mechanisms that underlie psychiatric symptoms ( Scholl and Klein-Flügge , 2018 ) . Our findings suggest that subjective feelings measured during tasks in depression-relevant domains may provide additional information not captured by computational models of learning and decision-making . One reason depression might reduce mood more in volatile than stable environments could be an increase in the number of negative prediction errors experienced . Misestimation of the level of uncertainty may also lead to a tendency for negative events to disproportionally affect depressed individuals , and this uncertainty misestimation is believed to contribute to depression and anxiety ( Pulcu and Browning , 2019 ) . That the learning rate difference between the volatile and the stable environment did not correlate with anxiety symptom severity is consistent with previous findings ( Browning et al . , 2015 ) that anxiety is linked to learning deficits for aversive but not appetitive outcomes . In the aversive domain , anxiety severity might be associated with differences in behavioural adaptation to volatility changes as well as mood . It is not yet established what the neural signal associated with PPEs is . On the one hand , reward prediction errors have been associated with neuromodulator dopamine and are thought to be linked to ventral tegmental area ( Bayer and Glimcher , 2005; Cohen et al . , 2012; Hart et al . , 2014; Montague et al . , 1996; Pessiglione et al . , 2006; Schultz et al . , 1997 ) . In most studies , RPEs and PPEs are equivalent and therefore the specific link between PPEs and dopamine is less documented . The probability of obtaining reward and probability prediction error have been associated with VTA activity correcting for expected value ( Behrens et al . , 2007 ) , suggesting that PPEs may be represented by dopamine . Boosting dopamine levels pharmacologically during risky decision making increases the happiness resulting from smaller rewards to a level similar to that resulting from larger rewards ( Rutledge et al . , 2015 ) . Although happiness in that study was influenced by the history of RPEs , dopamine drug impacts were limited to rewards . Dopamine has also been associated with other signals than prediction errors , for example incentive salience which might relate to ‘wanting’ and might influence choice and action ( Berridge , 2012; Smith et al . , 2011; Zhang et al . , 2009 ) . Some studies suggest that dopaminergic activity in the midbrain is linked to information-seeking to reduce uncertainty about an upcoming reward , even though such information is not instrumental ( Bromberg-Martin and Hikosaka , 2009; Brydevall et al . , 2018; Charpentier et al . , 2018; Gruber and Ranganath , 2019 ) . The intrinsic reward resulting from reducing uncertainty could influence mood , and mood ratings could then be used to quantify the relative subjective weight of extrinsic and intrinsic reward . Previous studies using risky decision tasks where reward and probability were explicitly represented ( Rutledge et al . , 2014; Rutledge et al . , 2015 ) showed that mood dynamics were explained by past expected values and RPEs . The present design allows us to dissociate the impact of learning-relevant and learning-irrelevant information for mood in two different standard learning environments . Our results suggest that when goal attainment requires adaptive behaviour , mood dynamics reflect learning-relevant information . Consistent with the results obtained from risky decision tasks used in previous studies ( Rutledge et al . , 2014; Rutledge et al . , 2015; Rutledge et al . , 2017 ) , potential rewards were a key determinant of behaviour in our task . However , rewards were not a determinant of mood in the current study , in contrast to previous results in risky decision tasks ( Rutledge et al . , 2014; Rutledge et al . , 2015; Rutledge et al . , 2017 ) . Unlike most reinforcement learning experiments , our task design allows dissociating the impacts of PPEs and RPEs on behaviour and mood . Here , our results suggest that happiness does not always depend on reward and preferentially reflects learning about the structure of the environment . However , if learning the structure of the environment requires tracking changing reward magnitudes , we would expect that happiness would track learning-relevant variables ( e . g . , reward magnitudes and RPEs in such an environment ) . This result might imply a role for mood in learning in line with influential proposals ( Eldar et al . , 2016; Eldar and Niv , 2015 ) . However , mood did not reflect all learning-relevant information that influenced behaviour . Happiness forgetting factors corresponding to the number of past trials that influence affective state were highly correlated across environments and acted more as a stable trait that differed between individuals and did not adjust to environmental volatility . Overall , our findings show that mood dynamics are sensitive to depressive symptoms and reflect variables relevant to adaptive behaviour irrespective of environmental volatility . Seventy-five healthy subjects ( age range 18–35 , 24 males ) took part in the experiment . Thirty-seven completed the stable learning task first and the volatile learning task second . Group allocation was randomised . Subjects were paid £10 for their participation . The number of participants recruited for the current cohort was selected to provide >95% power of detecting a similar effect size as that reported in a previous study in which a volatility manipulation was used to influence learning rate ( Browning et al . , 2015 ) . All subjects gave informed consent and the Research Ethics Committee of University College London approved the study ( Committee approval ID Number: 12673/001 ) . Participants were first instructed about the tasks . They performed 20 practice trials before a test ensuring that they understood that both probability and magnitude mattered to maximise the number of points obtained . They were told that they would be exposed to two environments: an environment where one car is more likely to win for the entire session , and an environment where which car is more likely to win changes occasionally . Because we wanted to maximise the efficiency of the behavioural manipulation ( i . e . , the difference in learning rate between environments observed in previous studies [Behrens et al . , 2007; Browning et al . , 2015; Massi et al . , 2018] ) to study how mood dynamics varied with behavioural sensitivity , we explicitly signalled the environment by using different pairs of cars in the two environments . Moreover , before each condition , an instruction screen explained in which environment participants will be placed . Finally , a fixation symbol displayed in the centre of the screen in each trial was specific for each environment: ‘-’ for the stable environment and ‘~’ for the volatile environment . Participants were given no guidance as to how they should use information about environmental volatility . After completion of the task , participants completed three standard clinical questionnaires: Beck Depression Inventory ( BDI-II Beck et al . , 1996 ) , Patient Health Questionnaire ( PHQ-9 , Kroenke et al . , 2001 ) , and the State/Trait Anxiety Inventory ( STAI , Spielberger , 1983 ) . The task was implemented using the Cogent toolbox in MATLAB ( MathWorks , Inc ) . Subjects had to choose between two cars , each associated with a probability ( 20% or 80% ) of winning . If the chosen car won , participants earned the corresponding amount of points . In the stable environment ( 80 trials ) , the probability to win for the best car was 80% . In the volatile environment ( 80 trials ) , reward probabilities switched between 80% for one car and 80% for the other car every 20 trials . The order was counterbalanced between subjects ( n = 38 in stable-volatile order , n = 37 in volatile-stable order ) . The outcomes were locally pseudo-randomised , to ensure that every 10 trials ( i . e . , trials 1–10 , 11–20 ) , the car with the highest outcome probability won on exactly 8 of 10 trials . The possible pairs of rewards were 10–10 , 10–40 , 10–60 , 10–80 , 20–40 , 40–10 , 40–20 , 40–40 , 60–10 , 80–10 . Subjects were primed when the side of the screen for each car was swapped ( every six to ten trials ) with an explicit cue . They were also instructed that the car location was unrelated to the outcome probability and to the change in outcome probabilities in the volatile environment . Every three to four trials , subjects were asked to indicate ‘How happy are you right now ? ’ on a scale from very unhappy to very happy . They were told to consider these extremes within the context of the experiment . Each trial started with a fixation screen for a duration varying between 0 . 9 and 1 . 9 s . Cars were displayed for 1 . 2 s without any information about reward magnitudes , and no choice was allowed in this phase . Subjects were free to choose the option they preferred without any time constraints as soon as the potential reward for each car was displayed . The chosen option was surrounded by a yellow frame for 1 . 5 s . Finally , the outcome was displayed for 2 s . Both the car and the reward magnitude frames were green if the chosen car won . They were red and the car was crossed out if the chosen car lost . Two-sided Wilcoxon signed rank tests were used to compare performance , proportion of win-stay/lose-shift , and model parameters between environments at the group level . Spearman rank correlations across participants were used to test for relationships between parameters and to relate depression scores to behavioural and happiness parameters . All analyses were performed using MATLAB . To test whether the correlation between PHQ score and happiness baseline parameters was higher in the volatile condition than in the stable condition , we correlated the standardised difference between the happiness baseline parameter in the stable and volatile conditions with the standardised PHQ score which quantifies depressive symptoms ( Wilcox and Tian , 2008 ) . The analysis codes were written in MATLAB and are available at Github ( https://github . com/BastienBlain/MSWB_LearningNotReward; copy archived at swh:1:rev:b7c4a0cd761dcf249c72caf809dd81af24c4a49b; Blain , 2020 ) . All models were fitted to experimental data by minimising the negative log likelihood of the predicted choice probability given different model parameters using the fmincon function in MATLAB ( Mathworks Inc ) . More specifically , parameters were treated as random effects that could differ between subjects ( Kreft and De Leeuw , 1998 ) : data were fitted for each participant and statistical tests were performed at the group level . We used standard model comparison techniques ( Burnham and Anderson , 2004; Schwarz , 1978 ) to compare model fits . For each model fit in individual subjects , we computed the Bayesian Information Criterion ( BIC ) , which penalises for model complexity ( i . e . number of parameters ) , and then summed BIC across subjects . The model with the lowest BIC is the preferred model . For all choice models , the learning rate was bounded between 0 and 1 , the inverse temperature between 0 and 50 ( to avoid ceiling effects ) , the probability-magnitude relative weight phi between 0 and 1 , and the gamma risk aversion parameters between 0 and 10 . Note that reward magnitude was normalised between 0 and 1 in the additive model . For the happiness models , we first fitted each happiness model on both environments using the same parameters for both environments . Then each model was fitted for each environment separately , with the starting parameters determined by the joint model fit under standard constraints ( the forgetting factor could vary only between 0 and 1 and the baseline mood parameter could only vary between 0 and 100 ) . Because participants vary in how they use the scale , we z-scored happiness ratings for all the analyses reported in the main text , except in the analyses where we asked how the baseline mood parameters are related to depression . Models were treated as random effects that could differ between subjects and have a fixed ( unknown ) distribution in the population . Model frequency with which any model prevails in the population , as well as exceedance probability ( EP ) , which measures how likely it is that any given model is more frequent than all other models in the comparison set ( Stephan et al . , 2009 ) , were estimated using the VBA Matlab toolbox ( Daunizeau et al . , 2014 ) . See figure supplementary figure 2 for an illustration of the estimated frequency for three different model spaces . An EP greater than 0 . 95 is considered significant . To estimate the influence of the past trials on the current happiness without any assumption regarding the shape of the influence decay , we fitted a general linear model including for each rating the previous 10 probability prediction errors using the Matlab glmfit function . Each value was then tested against 0 at the group level using two-sided Wilcoxon signed rank tests ( see supplementary figure 3 ) .	Many people believe they would be happier if only they had more money . And events such as winning the lottery or receiving a large pay rise do make people happy , at least temporarily . But recent studies suggest that the main factor driving happiness on such occasions is not the size of the reward received . Instead , it is how well that reward matches up with expectations . Receiving a 10% pay rise when you were expecting 1% will make you feel happier than receiving 10% when you had been expecting 20% . This difference between an expected and an actual reward is referred to as a reward prediction error . Reward prediction errors have a key role in learning . They motivate people to repeat behaviours that led to unexpectedly large rewards . But they also enable people to update their beliefs about the world , which is rewarding in itself . Could it be that reward prediction errors are associated with happiness mainly because they help us understand the world a little better than before ? To test this idea , Blain and Rutledge designed a task in which the likelihood of receiving a reward was unrelated to the size of the reward . This study design makes it possible to separate out the contributions of learning versus reward to moment-by-moment happiness . In the task , volunteers had to decide which of two cars would win a race . In the ‘stable’ condition , one of the cars always had an 80% chance of winning . In the ‘volatile’ condition , one car had an 80% chance of winning for the first 20 trials . The other car then had an 80% chance of winning for the next 20 trials . The volunteers were not told these probabilities in advance , but had to work them out by playing the game . However , on every trial , the volunteers were shown the reward they would receive if they chose either of the cars and that car went on to win . The size of the rewards varied at random and was unrelated to the likelihood of a car winning . Every few trials , the volunteers were asked to indicate their current level of happiness on a scale . The results showed that volunteers were happier after winning than after losing . On average they were also happier in the stable condition than in the volatile condition . This was especially true for volunteers with pre-existing symptoms of depression . Moreover , happiness after wins did not depend on how large the reward they got was , but instead simply on how surprised they were to win . These results suggest that how we learn about the world around us can be more important for how we feel than rewards we receive directly . Measuring happiness in various types of environment could help us understand factors affecting mental health . The current results suggest , for example , that uncertain environments may be especially unpleasant for people with depression . Further research is needed to understand why this might be the case . In the real world , rewards are often uncertain and infrequent , but learning may nevertheless have the potential to boost happiness .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Momentary subjective well-being depends on learning and not reward
Climate regions form the basis of many ecological , evolutionary , and conservation studies . However , our understanding of climate regions is limited to how they shape vegetation: they do not account for the distribution of animals . Here , we develop a network-based framework to identify important climates worldwide based on regularities in realized niches of about 26 , 000 tetrapods . We show that high-energy climates , including deserts , tropical savannas , and steppes , are consistent across animal- and plant-derived classifications , indicating similar underlying climatic determinants . Conversely , temperate climates differ across all groups , suggesting that these climates allow for idiosyncratic adaptations . Finally , we show how the integration of niche classifications with geographical information enables the detection of climatic transition zones and the signal of geographic and historical processes . Our results identify the climates shaping the distribution of tetrapods and call for caution when using general climate classifications to study the ecology , evolution , or conservation of specific taxa . Climate governs the basis of life on Earth . Besides historical contingencies and geographical barriers , abiotic conditions determine species ranges ( Woodward , 1987; Hoffmann and Parsons , 1997; Calatayud et al . , 2019b ) and derived diversity patterns ( Hawkins et al . , 2003; Kreft and Jetz , 2007; Mendoza and Araújo , 2019 ) . On a global scale , distinctive climate types impose generalized constraints that shape species pools adapted to particular climates ( Whittaker , 1962 ) . Identifying the boundaries of these climate types is a fundamental challenge to understanding how life organizes on Earth , with acute implications in diverse disciplines given the current climate change . Pythagoras proposed a classification of climate regimes of the known world in the 6th century BC ( Sanderson , 1999 ) . Still , it was not until the 19th century that geographers laid the foundations for such classifications ( Oliver , 1991 ) . By that time , researchers noticed the close relationship between the distribution of various life forms , especially vegetation types , and climate ( Oliver , 1991 ) . For instance , Köppen built his long-standing climate classification from pioneer plant classifications , assuming that vegetation types carry information about climatic conditions ( Kottek et al . , 2006; Thornthwaite , 1943 ) . This assumption has received considerable support ( Rohli et al . , 2015 ) , and researchers and stakeholders use Köppen’s classification system in a wide range of disciplines , including climatology ( Spinoni et al . , 2015 ) , geography ( Gentine et al . , 2012 ) , conservation planning ( Tobin et al . , 2014 ) , and ecology ( Garcia et al . , 2014 ) . But the general climatic conditions derived from plant species may not apply to other organisms . Plants’ and animals’ climatic determinants will likely differ , given their large physiological differences . Although diversity patterns of plants and animals are relatively congruent , the climatic correlates of these patterns vary ( Whittaker et al . , 2007; Qian and Ricklefs , 2008 ) . Hence , if climatic adaptations vary among taxa , so will the boundaries defining their climate types . In Thornthwaite , 1943 words , the ‘truly active factors’ describing a climate type may vary among organisms , and it remains unknown whether Köppen’s climate classification can indicate the active climatic factors for other organisms than plants . Despite several attempts to refine or propose alternative climate types or regions ( Trewartha , 1954; Holdridge , 1947; Thornthwaite , 1948; Netzel and Stepinski , 2016; Gardner et al . , 2020 ) , quantitative studies defining climatic regions for other organisms are still lacking . The current information on species distributions and global climatic variables , together with recent advances in niche modeling and classification techniques , provides an unprecedented opportunity to identify the climatic boundaries shaping the distribution of faunas and floras across the globe . The last decades have witnessed a tremendous collective effort to record occurrences of a large number of species ( GBIF , 2019 ) , which has resulted in comprehensive datasets of the distributional ranges of several groups of species ( IUCN , 2015; BirdLife , 2015; Roll et al . , 2017 ) . Also , data on climatic variables at a global scale have been developed at high spatial resolutions ( Fick and Hijmans , 2017; Trabucco and Zomer , 2009 ) . Combining these two sources of information , we can characterize the realized climatic niches of different taxa and find regularities among them . For example , projecting these realized climatic niches into a climatic space ( Broennimann et al . , 2012 ) should , if climatic boundaries exist , reveal co-occurring groups of species across particular portions of the climatic space . Thus , identifying these niche domains should uncover the main climatic boundaries shaping the organization of life ( Figure 1 ) . In addition to the climate , dispersal barriers and historical contingencies influence the shape of niche domains ( Warren et al . , 2014; Calatayud et al . , 2016; Calatayud et al . , 2019b ) . Therefore , similar climates can have different effects across different geographic regions ( Ricklefs , 1987 ) . For instance , while a given climate in some parts of the Earth may lead to specific species pools , the same climate in other parts of the Earth may not . Such a potential lack of specific species can occur , for example , because the required adaptations have not appeared ( Flohr et al . , 2013 ) , the adapted species have not been able to disperse ( Tuomisto et al . , 2003 ) , or the area is too small to hold large species pools ( Connor and McCoy , 1979 ) . Studying the signature of these historical and geographical processes , known as geographical signals , in niche domains can provide valuable information about the potential mechanisms behind them and their associated climatic regions . We explore the global climate regions of tetrapods by characterizing the climatic niche domains of amphibians , birds , mammals , and reptiles . Tetrapods are a well-suited group for our purpose . First , comprehensive databases are available , including the distributional ranges of most species in the group ( IUCN , 2015; BirdLife , 2015; Roll et al . , 2017 ) . Second , the different classes of tetrapods have diverse capabilities to disperse and withstand abiotic conditions , allowing us to investigate whether various capabilities influence climatic classifications . Third , accumulated evidence about the main climatic factors controlling the distribution of these species simplifies the selection of appropriate climatic variables . In particular , the distribution of tetrapods is strongly determined by the water and energy aspects of climate ( Hawkins et al . , 2003; Currie , 1991; Tingley et al . , 2009; Gouveia et al . , 2014; Pie et al . , 2017; Cooper et al . , 2011 ) . Finally , researchers study tetrapods in several disparate fields – from animal husbandry ( Abecia et al . , 2017 ) to ecological ( Englert Duursma et al . , 2019 ) and evolutionary studies ( Rolland et al . , 2014 ) – where a description of their climatic regions can be especially useful . In our classification approach , we project the realized niche of each tetrapod species onto a binned two-dimensional space representing water and energy , the primary climatic factors impacting their geographic distribution ( Figure 1 ) . After translating this climatic niche space into a weighted bipartite network , we apply a network community detection algorithm to identify climatic niche domains with similar species pools . Finally , by mapping the climatic niche domains back onto the Earth’s surface , we provide the climatic regions . The detected climatic regions support the notion that similar climatic determinants underlie animal and plant distributions in high-energy regions , including deserts , tropical savannas , and steppe regions . However , differences in temperate climates across all groups indicate that specific climatic regions for each group of taxa are required to address ecological , evolutionary , and conservation questions . We first identified the species niche domains by calculating the proportion of observations of each species within each bin of a two-dimensional climatic space defined by potential evapotranspiration ( PET ) and annual precipitation ( AP; Figure 1 , 'Materials and methods' and Appendix 1 ) . We represented this data as a weighted bipartite network , where climatic bins and species form two disjunct sets of nodes and the probabilities of finding the species in the bins form the link weights . Using a hierarchical network clustering algorithm ( Rosvall and Bergstrom , 2008; Rosvall and Bergstrom , 2011 ) , we obtained groups of climatic bins holding similar species , niche domains , and the species associated with them . We found similarities among tetrapods classes in the detected niche domains but also observed some differences ( Figure 2 ) . For instance , the number of major domains with 50 or more species in the lower hierarchical level is similar ( ranging from 13 to 15 ) across tetrapods classes . However , mammals and birds show a domain of low-energy inputs , whereas reptiles present some domains across arid conditions , that is , with high energy and low water availability ( Figure 2 ) . Finally , we classified the climatic space of tetrapods by using all species jointly . The clustering algorithm divided the niche space of tetrapods into 16 main domains that are similar to those of the independent classes , and some of the particularities described above did not appear ( Figure 2 ) . Since uncertainties related to the ranges of species exist , we employed a bootstrap and a significance clustering procedure ( Rosvall and Bergstrom , 2010; Calatayud et al . , 2019a ) to assess the domains’ robustness ( 'Materials and methods' ) . While several domains were well supported , we found that the niche domains corresponding to intermediate energy ( between approximately 1000 and 1500 PET units; EM climates in Figure 2 ) and low to moderate water ( up to approximately 800 ml; WL to WM ) were less robust . This robustness analysis shows that these niche domains are more challenging to classify . While Köppen’s climate classification is based on expert knowledge on vegetation physiognomy and the distribution of vegetation types , its wide use makes it worthwhile to compare the climate regions derived from this classification system to the ones produced here . Hence , we studied the geographic location of the climatic conditions associated to niche domains , the climatic regions shown in Figures 1 and 3 , which allowed for a more precise comparison between groups and Köppen’s plant-based regions . The similarities among the regions of tetrapods classes measured as adjusted mutual information ( AMI ) ranged from 0 . 57 to 0 . 68 , with mean AMI = 0 . 62 ( Table 1 ) . Moreover , the regions based on the niche domains of all tetrapods together were to some extent congruent with the regions of its independent classes ( mean AMI = 0 . 71 , ranging from 0 . 66 to 0 . 77 ) . Köppen’s regions were more dissimilar to the regions of all tetrapods together ( AMI = 0 . 44 ) and the regions of each class of tetrapods independently ( mean AMI = 0 . 44 , ranging from 0 . 40 to 0 . 47 ) . Focusing on particular regions , we saw that climates of high energy ( EH ) were consistent among tetrapod groups and Köppen’s classification . Desert climates ( high energy and low water , EHWL , Figure 3 , and Appendix 1—figure 1 ) were the most similar across all groups . Tropical savanna and steppe climates ( high energy and medium water , EHWM ) were also consistently defined , though both of these Köppen regions were classified together for all groups but reptiles ( Figures 2 and 3 and Appendix 1—figure 1 ) . Similarly , Köppen’s tropical rainforest and tropical monsoon climates were for the most part well recovered ( Appendix 1—figure 1 ) . However , we found three different tropical-humid regions , each one mostly corresponding to one of the three larger masses of tropical rainforests: Amazonian , African , and Southeast Asian rainforests; EHWH1 , EHWH2 , and EHWH3 , respectively ( Figures 2 and 3 ) . Regarding regions of low energy , we found a slightly higher level of disagreement between Köppen’s and tetrapods’ regions ( Figure 3 ) . Finally , temperate climates ( medium energy EM ) were the least congruent between tetrapod groups and Köppen’s regions . These regions of medium energy were at the same time the least congruent among groups and the least supported by the bootstrap analyses , suggesting that these climates could impose less restrictive conditions in general and allow the appearance of idiosyncratic and variable adaptations . A complete understanding of niche domains and their associated climatic regions entails the exploration of whether their boundaries represent abrupt or diffuse transitions . Climatic conditions corresponding to diffuse transitions should present low specificity levels to the domain where they belong ( Figure 1 ) . Hence , we can consider specificity as the opposite of transitivity . Our network approach allows us to calculate this specificity by the dual classification of climatic bins and species into the same niche domains ( Figure 1 ) . We computed the specificity of each climatic bin as the ratio between the link weights of the species classified in the same domain and the total link weights ( Bernardo-Madrid et al . , 2019; Calatayud et al . , 2019b ) . Then , we projected these values geographically . As expected , lower specificity values were in general associated with the boundaries of the climatic regions ( Figure 4A and Appendix 1—figure 2 ) . Beyond boundaries , our results also revealed that harsh conditions , such as desert and continental-polar climates ( EHWL and ELWL ) , present the highest specificity levels , regardless of the group ( Figure 4A and Appendix 1—figure 2 ) , reflecting the difficulty of colonizing these climates . Conversely , temperate regions showed the lowest levels of specificity . These regions were also weakly supported in the bootstrap analyses , and we found that bootstrap p-values and mean specificity were significantly correlated ( standard generalized linear mixed model [GLMM] coeff . 6 . 21; p < 0 . 001 conditional = 0 . 29 , see 'Materials and methods' ) . Together with the higher variability of these regions across groups , this result further supports the idea that these climatic conditions could impose less restrictive conditions to tetrapods . Historical and geographical processes may lead to species pools adapted to a given climate in some regions of the Earth but not in others . To search for this geographical signal , we first compared the distribution of the climatic conditions and species grouped within the same niche domain . A geographic mismatch between species and climate distributions would point to portions of the climatic regions that are defined by species occurring in other geographic areas . Exploring these patterns for each niche domain revealed notable geographic agreement between species and climatic conditions of the same domain ( Figure 4B and Appendix 1—figures 3–7 ) . Nevertheless , we found some differences across groups and regions . More extreme climates showed larger mismatches between the distribution of species and climates . For instance , for all groups but reptiles , desert climate ( EHWL ) was mostly defined by species inhabiting Australia and to a lesser extent by species from the Namibian desert and the Horn of Africa , with few or no species inhabiting the Sahara desert ( Figure 4C and Appendix 1—figures 3–7 ) . Similarly , the northern climatic regions of amphibians and reptiles were defined by species at lower latitudes ( Appendix 1—figures 3 and 4 ) . Approaching the geographical signal more quantitatively ( see 'Materials and methods' ) , we found a stronger signal for the worst dispersers , amphibians and reptiles , than for mammals and birds ( Figure 4D ) , suggesting that dispersal capabilities can contribute to the geographical signal in the niche domains . Finally , all tetrapods together showed the lowest geographical signal , which suggests that , in addition to dispersion , increased evolutionary time can reduce the geographical signal . We detected 16 climatic regions governing the distribution of tetrapods . Despite the substantial physiological and functional differences among the groups , some of their niche domains and climatic regions are consistent . These climatic regions also resemble Köppen’s regions and correspond , in general , with extreme climates , being arid climates specially congruent across groups . These climates also presented overall low levels of transitivity , showing that species adapted to other climates have more difficulty withstanding these conditions . These results suggest that extreme climates impose strong adaptive barriers , producing a filtering of species ( Butterfield , 2015; Cadotte and Tucker , 2017 ) even across distinctive evolutionary lineages . Conversely , milder climatic conditions , especially temperate climates , showed the lowest statistical support and congruence across groups , as well as the highest transitivity . These results indicate that temperate climates are more difficult to classify ( low support and congruence ) due to the overlap in the climatic space of species pools with different climatic optima ( high transitivity ) . Two complementary reasons can explain this ambiguity: first , while we used two variables widely recognized to shape the distributions of tetrapods , alternative variables , such as seasonal changes in energy and precipitation ( Köppen and Geiger , 1930 ) , may also influence species inhabiting temperate regions . Including these variables might help to further separate temperate species pools across the climatic space . Second , the climatic conditions of these domains may not prevent the colonization of species with other realized optima or preferences , which would generate the observed overlap in the climatic space ( i . e . , high transitivity ) across milder conditions . Questions remain about the relative contribution of each process . Similarly , the exact causes and consequences of the climatic transition zones call for future investigation . We also found some domains that were well supported but unique for each group . These differences between groups seem to relate to the particular physiological adaptations of each group . For instance , homeothermic birds and mammals defined a region of low energy , consistent with Köppen’s polar climates , that reptiles and amphibians lacked . Similarly , reptiles , a group that includes several groups adapted to arid environments ( Pie et al . , 2017 ) , defined some regions of low precipitation and high PET . Hence , our results stress that caution is needed when generalizing the climate classification to other groups of organisms , and question the validity of using plant-based classifications for studies dealing with animals . Beyond niche domains , our results also show differences in the geographical signal across groups . That amphibians – the group with the lowest dispersal capacity – showed the highest geographical signal suggests that dispersal processes play an essential role: species that are worse dispersers have more difficulties tracking their preferred climates ( Araújo and Pearson , 2005 ) , limiting the colonization of disjoint areas with similar climates . Moreover , all tetrapods together showed the lowest geographical signal , suggesting that increased evolutionary time can reduce this geographic pattern . That is , an extended evolutionary time that enables the appearance of convergent adaptations to similar climates in different geographic regions ( Mazel et al . , 2017 ) may reduce the geographical signal in niche domains . Alternatively , tetrapods include taxa with varying dispersal strategies – such as the high active dispersal of birds and the elevated survival probabilities of reptiles to passive transoceanic dispersal – which may also influence the geographical signal . In any case , the ultimate causes and consequences of this signal require further attention . Why are some amphibians able to inhabit arid conditions in the Australian desert but not in the Sahara desert ( Appendi 1—figure 3 ) ? Why can some reptiles withstand cold climates around the Himalayan mountains but not in the northern hemisphere ( Appendi 1—figure 4 ) ? These are some examples of the emerging fundamental questions related to the geographical signal . Answering them would require studying the adaptations that allow the species to tolerate particular climatic conditions and the abiotic , biotic , and historical differences between regions ( Moncrieff et al . , 2015 ) that favor these adaptations in some places but not in others . Moreover , the geographical signal idea could also help unravel idiosyncratic processes occurring at smaller scales by comparing local climate regions with global or regional ones . These examples form exciting avenues for future historical biogeographical and evolutionary studies . Our results and data-driven methodology have the potential to bring us closer to a definition of climatic regions that represent active factors for the spatial organization of life . Nevertheless , it would be interesting to improve some aspects in future studies . First , while we used a notable number of species ( about 26 , 000 ) , they are taxonomically biased and only represent a small fraction of terrestrial organisms . Second , we used two climatic variables widely known to affect the distribution and diversity patterns of animals and plants in general ( Currie , 1991; Hawkins et al . , 2003 ) , but other climatic variables might refine some of the least supported regions . Finally , our domains represent portions of the realized climatic niche space , which entails two sources of uncertainty . On the one hand , the estimation of realized niches depends on the data quality , scale , and treatment choices . On the other hand , the realized niche space may be influenced by historical , geographical , and biotic factors beyond pure climate ( Warren et al . , 2014; Calatayud et al . , 2019b; Soberón , 2007 ) . Hence , using fundamental rather than realized niches may also improve the accuracy of defining climatic regions . At the current pace of biological data accumulation and computational development , it is reasonable to expect that some of these limitations will soon be overcome . Meanwhile , the considerable congruence of several climatic regions across the studied groups and Köppen’s system provides confidence in their robustness . Hence , it is likely that using better data would not produce regions substantially different from those presented here . Regardless of how generalizable the results are , the niche domains and their associated species pools and climatic regions can be used as a basis for ecological , evolutionary , and conservation studies concerning tetrapods . Some of the many questions that the results reported here ( data available in source data of Figures 2 and 3 ) can help to answer include: Are the species belonging to different niche domains similarly conserved or protected , or both ( Hanson et al . , 2020 ) ? To what extent do the differences between the four classes of tetrapods reflect phylogenetic and functional differences , and are such differences to be found in other taxonomic groups ? Is the adaptation to niche domains evolutionarily constrained ? Do diversification , extinction , or speciation rates differ among the species associated with different domains ? Moreover , combining niche domains with bioregions based on pools of species ( Bernardo-Madrid et al . , 2019 ) or lineages ( Holt et al . , 2013 ) can also help answer several relevant questions: Are introductions more common between bioregions with the same climate types ? To what extent does the current climate govern bioregions ? Do ecological and evolutionary processes differ between different bioregions with same climate types ( Moncrieff et al . , 2015 ) ? In conclusion , our data-driven climate classification reveals major climatic boundaries organizing the distribution of tetrapods on Earth . Questions remain regarding the mechanism underlying the differences between groups in the climatic regions and the geographical signal . Nevertheless , these differences across groups highlight the need for classifications of climates specific to each group to study their ecology , evolution , or conservation . We obtained the distribution ranges of mammals and amphibians from the IUCN , 2015 , of birds from BirdLife , 2015 , and of reptiles from Roll et al . , 2017 . We included only the native range of terrestrial species in the analyses in all instances . In the case of birds , we only used the breeding ranges . Moreover , since there is a higher uncertainty when determining the realized niches of narrow-ranging species ( Lehmann et al . , 2002 ) , we arbitrarily removed the species whose ranges were less than five grid cells of 0 . 5° . After this cleaning of the data , we used 3657 amphibians , 7204 reptiles , 4574 mammals , and 10 , 684 birds , for a total of 26 , 119 tetrapod species . We approximated the species’ Grinnellian niches ( Soberón , 2007 ) with two climatic variables that represent energy and water inputs . While we could have used several other variables , we chose energy and water as they best explain climatic effects on species distributions ( Hawkins et al . , 2003 ) . As surrogates for energy and water inputs , we used mean annual PET and AP , respectively . Both variables have been shown to be important factors for tetrapod species distributions ( Currie , 1991; Tingley et al . , 2009; Gouveia et al . , 2014 ) . Moreover , they have also been used in previous climate classifications ( Thornthwaite , 1948 ) and are regularly used to derive other drivers of species distributions such as the UNEP aridity index ( Unep and Thomas , 1992; Fuller et al . , 2016 ) . We obtained PET from Trabucco and Zomer , 2009 and AP from Fick and Hijmans , 2017 , both at a 0 . 08° resolution . Finally , we obtained Köppen’s climatic regions from Kottek et al . , 2006 and Rubel et al . , 2017 . We characterized the realized climatic niche of each species using an approach similar to the one proposed in Broennimann et al . , 2012 . We divided the climatic space formed by PET and AP into bins and calculated the proportion of occurrences a given species has in each climatic bin . Both the shape of the divisions and the number of divisions of each climatic axis affect the result . For instance , dividing the axis into regular intervals can destroy critical information if the climatic values more important for the species distributions are skewed toward any extreme of the distribution or if the climatic values are represented non-uniformly across the globe ( as for AP , Appendix 1—figure 8 ) . Similarly , the grain size to divide the climatic space may affect subsequent results ( Levin , 1992; Daru et al . , 2020 ) . Dividing the space into too few intervals destroys information , whereas using too many divisions can generate niche domains with only a few species . To overcome the first issue , we divided the axes into quantiles based on the distribution of climatic values across the Earth . By doing so , we obtained an almost uniformly divided PET axis ( Appendix 1—figure 8 ) . Contrarily , the number of divisions of the AP axis was skewed toward low values , which resulted in a higher resolution over the presumably more relevant low-precipitation conditions ( Appendix 1—figure 8 ) . To solve the second issue , we selected the lowest number of divisions that maximized the gain in information ( see Appendix 1 ) . The optimal number of axis divisions was 17 in all cases but amphibians , where it was 18 . Next we accounted for potential commission errors , which may affect the estimated climates a species experiences . Specifically , range maps can overestimate the area occupied by a species , which directly influences the niche characterization ( Rondinini et al . , 2006 ) . Because range maps typically represent the species’ maximum geographical extent ( Rondinini et al . , 2006; La Sorte and Hawkins , 2007 ) , extracting the climatic values that a species range covers from a well-fitting 0 . 08° high-resolution climatic raster can reduce commission errors at the species range’s borders . But with many pixels inside the species range , the noise from extreme and unrepresentative climatic values can increase the error ( La Sorte and Hawkins , 2007 ) . Conversely , extracting climatic values from a coarser raster can reduce the commission errors inside a range by averaging out extreme values at the cost of increasing commission errors over the borders . To alleviate the effects of these potential errors , we first extracted the climatic values from the high-resolution rasters ( 0 . 08° ) . Then , we computed the average climatic values among selected raster pixels located within cells of 0 . 5° . In this way , we reduced the effects of commission errors both at the borders of and inside species ranges . Finally , mean climatic values may not accurately represent the cells’ climates when there is high climate variability or the values are non-normally distributed . Comparing results obtained from different cell sizes is an indirect way to asses the influence of distorted mean values . Instead , we chose to directly explore the effects of high climatic variability and non-normal distributions with non-parametric bootstrap analysis: we resampled climatic values within 0 . 5° cells with replacement ( see below ) . We employed a network community detection approach to identify the niche domains and the species mainly associated with them . For each group of species , we first generated a weighted bipartite network where species and climatic bins formed the disjoint sets of nodes , and the proportion of occurrences of species in intervals of the climatic values corresponding to the climatic bins formed the weighted links . To identify the niche domains , we used the hierarchical version of the community detection algorithm known as Infomap ( Rosvall and Bergstrom , 2008 ) . Infomap capitalizes on the minimum description length principle of information theory , which equates finding regularities and compression: the model that finds most regularities in a given set of data can compress the data the most ( Rissanen , 1978 ) . In our case , modules of highly interconnected climatic bins and species form the regularities , and describing the network with an optimal set of communities corresponds to minimizing the description length ( Bernardo-Madrid et al . , 2019; Rosvall and Bergstrom , 2008 ) . Among the many community-detection algorithms available , we used Infomap because it can find hierarchically nested communities and is known for its high performance ( Lancichinetti and Fortunato , 2009 ) , also for regionalizations ( Bernardo-Madrid et al . , 2019; Bloomfield et al . , 2018; Vilhena and Antonelli , 2015 ) . We ran the algorithm 1000 times , selecting the network partition with the best quality . To consider the uncertainty associated with both the species ranges and the community detection , we conducted a bootstrap analysis . For each species , we resampled with replacement from the distribution of climatic values within species ranges at a resolution of 0 . 08° . We averaged climatic values occurring within 0 . 5° cells and calculated the proportion of occurrences in each climatic bin . With resampled data from all species , we generated a bootstrapped network and clustered it with Infomap 1000 times . We repeated this procedure for 100 bootstrap networks and followed the community-stability approach proposed in Calatayud et al . , 2019a to calculate the niche domains’ support . For each identified domain , we calculated the proportion of bootstrap networks with a domain more similar to Jaccard index 0 . 5 ( Calatayud et al . , 2019a ) . With obtained niche domains , we detected the climatic regions by identifying areas across the Earth’s surface that hold the climatic conditions grouped within each niche domain . Finally , to compare climatic regions across tetrapod groups and with Köppen’s classification , we calculated the AMI . AMI measures the mutual information between two partitions , the classifications of raster pixels into climatic regions in this case , correcting for the similarity between partitions that are just due to chance ( Vinh et al . , 2010 ) . The index is 1 when the partitions are equal and tends to 0 otherwise . The joint classification of climatic bins and species into domains D allowed us to calculate the bins’ specificity . Though species belong to single domains , typically together with the bins to which they have many and strong links , they may also have links to bins in other domains . Bins that contain species from different domains have low specificity and form a transition zone between domains ( Figure 1; Calatayud et al . , 2019b; Bernardo-Madrid et al . , 2019 ) . We calculated the specificity SiD of a climatic bin i in domain D as the sum of link weights wi , j from bin i to species j assigned to the same domain as the bin , divided by the sum of link weights from bin i to all species j such that ( 1 ) SiD=∑j∈Dwi , j∑jwi , j⁢ for ⁢i∈D . This index is 1 when the bin connects only to species in the same domain and tends to 0 otherwise . We projected the specificity values into the geographic space . To all geographical raster cells q with the climate of climatic bin i , which we call raster cell set Qi , we assigned bin i’s specificity . The projected specificity is SqP . Finally , we explored the relationship between the average SD and bootstrap support . We fitted a logistic GLMM of bootstrap p-values as a function of the mean SD and the taxonomic group . We used the R ( R Development Core Team , 2018 ) package lme4 ( Bates et al . , 2015 ) with the mean SD as a fixed term and the taxonomic group as a random intercept term . To quantify the geographical signal , we compared the geographically projected specificity SP with the actual specificity based on the pool of species that co-occur geographically . Large differences between the species co-occurring in the climatic and geographic spaces indicate a strong geographic signal . For instance , when species assigned to the same domain co-occur only in a portion of the corresponding geographical space , geographical areas with mismatching species contribute to a strong geographical signal . Areas that host most of the species associated with a niche domain have a higher actual specificity than the projected indicates . Areas that are not , or only scarcely , colonized by these species have a lower actual specificity than the projected indicates . Using Equation 1 , we calculated the actual specificity of a geographical raster cell q , whose corresponding climatic bin i is in domain D ( q∈Qi and i∈D ) , as the ratio between the sum of link weights from bin i to species in raster cell q that belong to its associated domain and the total link weight from bin i to all species in q , ( 2 ) SqA=∑j∈D , qwi , j∑j∈qwi , j⁢ for ⁢q∈Qi⁢ and ⁢i∈D . To calculate the geographical signal G , we averaged the absolute difference between the projected and actual specificity of each climatic raster cell q at 0 . 5° resolution , ( 3 ) G=1N⁢∑q=1N\|SqA-SqP\| , where N is the total number of raster cells . This index is 0 when there is no geographical signal and tends to 1 for high signals .	There are many distinct climates on Earth , from tropical savannas and temperate forests to dry deserts . Historically , each region has been defined by how its annual weather patterns shape the type of vegetation present . For example , hot and humid environments support the growth of evergreen forests that would not survive in drier places . Identifying the boundaries between climate regions is key to understanding how life is organized on Earth and predicting how climate change will affect different species . Current climate classifications , however , do not account for where animals can be found or how local conditions , such as precipitation and average temperatures , shape the distribution of different animal species . To bridge this gap , Calatayud et al . analyzed the preferred climate of about 26 , 000 animal species , including amphibians , birds , mammals and reptiles . For each species , Calatayud et al . calculated the annual rainfall and temperature of its local environment , or ‘niche’ , using previously collected data . They then used a computer algorithm to group together climates that had similar species . This identified 16 climate regions which govern the distribution of the animals studied . Calatayud et al . found that these newly defined climatic regions resembled some of the regions classified using plants . This was particularly true for high-energy climates that had lower levels of rainfall and hot temperatures , such as deserts and the tropical savanna . The animals and plant species living in high-energy regions were found to be fairly consistent across both classification systems . Whereas the species present in milder and colder climates , such as temperate forests or Mediterranean climates , were found to be much more varied . This suggests that temperate climates are harder to classify and may affect the distribution of plants and animals differently . It also implies that less extreme conditions support a larger range of species than harsher climates in which only species with certain adaptations are able to survive . These findings build the basis for a better understanding of how climates shape ecosystems . More specific climate classifications , based on such analyses , could be used to inform conservation strategies for animal species in the face of climate change .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology" ]	2021	Regularities in species’ niches reveal the world’s climate regions
Plants generally respond to herbivore attack by increasing resistance and decreasing growth . This prioritization is achieved through the regulation of phytohormonal signaling networks . However , it remains unknown how this prioritization affects resistance against non-target herbivores . In this study , we identify WRKY70 as a specific herbivore-induced , mitogen-activated protein kinase-regulated rice transcription factor that physically interacts with W-box motifs and prioritizes defence over growth by positively regulating jasmonic acid ( JA ) and negatively regulating gibberellin ( GA ) biosynthesis upon attack by the chewing herbivore Chilo suppressalis . WRKY70-dependent JA biosynthesis is required for proteinase inhibitor activation and resistance against C . suppressalis . In contrast , WRKY70 induction increases plant susceptibility against the rice brown planthopper Nilaparvata lugens . Experiments with GA-deficient rice lines identify WRKY70-dependent GA signaling as the causal factor in N . lugens susceptibility . Our study shows that prioritizing defence over growth leads to a significant resistance trade-off with important implications for the evolution and agricultural exploitation of plant immunity . Plants have developed effective defensive systems to minimize herbivore damage . They can specifically perceive attackers and respond to them by activating defence-related signaling pathways , including mitogen-activated protein kinase ( MAPK ) cascades and hormone signaling , leading to the induction of numerous defence-related genes and defence compounds as well as plant resistance ( Wu and Baldwin , 2010; Bonaventure et al . , 2011; Erb et al . , 2012 ) . Jasmonic acid ( JA ) - , salicylic acid ( SA ) - , and ethylene ( ET ) -mediated signaling play a central role in induced resistance to herbivores ( Lu et al . , 2011; Qi et al . , 2011 ) . The induction of defences commonly co-occurs with a reduction of plant growth ( Heinrich et al . , 2013; Attaran et al . , 2014; Huot et al . , 2014 ) . Through silencing defence-related genes , a direct negative link between defence and growth was demonstrated ( Zavala and Baldwin , 2006; Zhang et al . , 2008; Meldau et al . , 2012; Yang et al . , 2012 ) , suggesting that plants actively prioritize defence over growth . Defence prioritization and the associated growth trade-offs are regulated by crosstalk between plant hormones ( Schwachtje et al . , 2006; Stanton et al . , 2013; Huot et al . , 2014 ) . DELLA proteins for instance , which typically suppress gibberellin ( GA ) signaling , can physically interact with the JA pathway repressor JAZ proteins , thereby resulting in mutual suppression ( Hou et al . , 2010; Yang et al . , 2012 ) . Furthermore , SA has been reported to inhibit the expression of TIR1/ABF F-box genes , thereby leading to stabilization of AUX/IAA repressor proteins and decreasing auxin signaling ( Wang et al . , 2007 ) . Conversely , auxin signaling can reduce SA biosynthesis and thereby render plants more susceptible to pathogens ( Robert–Seilaniantz et al . , 2011 ) . Compared to the impact of defence-related hormones , little is known about the impact of growth-related hormones on herbivore resistance and potential resistance trade-offs that may emanate from prioritizing defence over growth through hormonal regulation ( Yang et al . , 2012 ) . Transcription factors ( TFs ) play a potentially important role in herbivore-induced plant reconfiguration and defence prioritization , as they regulate the expression of responses up and downstream of hormonal signaling pathways and thereby influence early and late signaling ( Reymond et al . , 2004; Dombrecht et al . , 2007; Skibbe et al . , 2008; Kaur et al . , 2010; Lu et al . , 2011; Zhou et al . , 2011; Schweizer et al . , 2013 ) . The best-studied TFs involved in plant–insect interactions are MYCs , WRKYs , MYBs , and ERFs . AtMYC2 in Arabidopsis thaliana , for example , was reported to act downstream of JA and to regulate JA-dependent herbivore resistance ( Dombrecht et al . , 2007 ) . Moreover , MYC2 , MYC3 , and MYC4 were shown to regulate the production of toxic glucosinolates via a direct transcriptional activation of glucosinolate biosynthesis genes ( Schweizer et al . , 2013 ) . In Nicotiana attenuata , a R2R3-type MYB TF ( NaMYB8 ) was found to modulate the accumulation of phenylpropanoid–polyamine conjugates , which are essential for defence against herbivores ( Kaur et al . , 2010 ) . In rice , an EAR-motif-containing ERF TF ( OsERF3 ) functions as an early component upstream of MAPK signaling and modulates JA , SA , ET , and H2O2 levels as well as plant resistance to rice herbivores ( Lu et al . , 2011 ) . Also , several WRKYs , such as rice OsWRKY89 , wheat TaWRKY53 , Arabidopsis AtWRKY72 , and tomato SIWRKY70 and SIWRKY72 , have been directly associated with defence against herbivores ( Wang et al . , 2007; Bhattarai et al . , 2010; Van Eck et al . , 2010; Atamian et al . , 2012 ) . NaWRKY3 and NaWRKY6 in N . attenuata have been shown to modulate elicited JA and JA-Ile/-Leu levels and thus mediate herbivory-induced defence responses ( Skibbe et al . , 2008 ) . Identifying and manipulating TFs that are involved in defence prioritization would make it possible to assess the biological impact of herbivore-induced growth suppression . Yet , to date , such an approach has not been taken . Consequently , our understanding of the consequences of defence prioritization for plant resistance has remained limited . To dissect the signaling network that underlies growth defence trade-offs in rice , we identified OsWRKY70 , an herbivory-induced Group I-type WRKY TF from rice , and elucidated its role in herbivore-induced defence prioritization . Through the use of in vivo and in vitro protein assays , molecular characterization and the creation of transgenic OsWRKY70 silenced and overexpressing plants combined with insect bioassays and a variety of phytohormone analyses , we evaluate the resistance benefits and trade-offs of defence prioritization against different herbivores and thereby reveal a new cost of defence prioritization . Using suppressive subtractive hybridization ( SSH ) , we screened rice plants for herbivory-induced TFs . Using this technique , we identified a clone that showed similarity to a WRKY gene . The full-length cDNA of the cloned OsWRKY , including an open reading frame ( ORF ) of 1719 bp , was obtained by reverse transcription PCR ( Figure 1—figure supplement 1 ) . Blast analysis showed that the sequence was 100% identical to the previously identified OsWRKY70 ( TIGR ID Os05g39720 ) . OsWRKY70 has two WRKY domains and belongs to group I ( Rushton et al . , 2010 ) . Phylogenetic analysis of group I-type WRKYs from different species revealed that OsWRKY70 has two homologs in rice , OsWRKY24 and OsWRKY53 , which share 53% and 51% amino acid sequence identity ( Figure 1—figure supplement 2 ) . Quantitative real-time PCR analysis revealed low constitutive expression of OsWRKY70 . Mechanical wounding and infestation by the rice striped stem borer ( SSB ) Chilo suppressalis resulted in a rapid increase in transcript levels ( Figure 1A , B ) . Infestation by the rice brown planthopper ( BPH ) Nilaparvata lugens only slightly increased the transcription levels of ( Figure 1C ) . JA or SA treatment did not induce OsWRKY70 ( Figure 1D ) , suggesting that OsWRKY70 is an early regulator of plant responses to herbivores . 10 . 7554/eLife . 04805 . 003Figure 1 . Expression of OsWRKY70 in rice after different treatments . Mean transcript levels ( +SE , n = 3–4 ) of OsWRKY70 in rice plants that were treated with either rice striped stem borer ( SSB ) ( A ) , mechanically wounded ( B ) , rice brown planthopper ( BPH ) ( C ) , jasmonic acid ( JA ) , salicylic acid ( SA ) , or a buffer ( 50 mM phosphate buffer , pH = 8 . 0 ) ( Buffer ) ( D ) . Controls correspond to non-manipulated plants . Transcript levels were analyzed by QRT-PCR . Asterisks indicate significant differences in transcript levels between treatments and controls ( * , p < 0 . 05; ** , p < 0 . 01; Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 00310 . 7554/eLife . 04805 . 004Figure 1—figure supplement 1 . Nucleotide and amino acid sequence of OsWRKY70 . SP cluster ( green ) , WRKY domain and Zinc fingure ( bold ) were shown . The underline sequence was used for RNAi construction . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 00410 . 7554/eLife . 04805 . 005Figure 1—figure supplement 2 . Phylogenetic relationships of Group Ⅰ type WRKY genes from different species . Neighbor-joining phylogenetic trees were produced using MEGA3 . 1 based on multiple sequence alignments made with ClustalX . Selected proteins accession numbers are as follows: Arabidopsis thaliana ( At , Tigr ID ) : AtWRKY1 ( At02g04880 ) , AtWRKY2 ( At02g30250 ) , AtWRKY3 ( At02g03340 ) , AtWRKY4 ( At01g13960 ) , AtWRKY10 ( At01g55600 ) , AtWRKY20 ( At04g26640 ) , AtWRKY25 ( At02g30250 ) , AtWRKY26 ( At05g07100 ) , AtWRKY32 ( At04g30935 ) , AtWRKY33 ( At02g38470 ) , AtWRKY34 ( At04g26440 ) , AtWRKY44 ( At02g37260 ) , AtWRKY45 ( At03g01970 ) , AtWRKY58 ( At03g01080 ) ; Oryza sativa ( Os , Tigr ID ) : OsWRKY4 ( Os03g55164 ) , OsWRKY24 ( Os01g61080 ) , OsWRKY30 ( Os08g38990 ) , OsWRKY35 . 1 ( Os04g39570 . 1 ) , OsWRKY35 . 2 ( Os04g39570 . 2 ) , OsWRKY41 ( Os11g45924 ) , OsWRKY53 ( 05g27730 ) , OsWRKY61 ( Os11g45850 ) , OsWRKY63 ( Os11g45920 ) , OsWRKY70 ( Os05g39720 ) , OsWRKY78 ( Os07g39480 ) , OsWRKY81 ( Os12g02400 ) ; Nicotiana attenuate ( Na , NCBI ID ) : NaWRKY3 ( AY456271 ) , NaWRKY6 ( AY456272 ) ; Nicotiana benthamiana ( Nb , NCBI ID ) : NbWRKY8 ( AB445392 ) ; Nicotiana tabacum ( Nt , NCBI ID ) : NtWRKY1 ( AF096298 ) , NtWRKY4 ( AF193771 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 005 To clarify the subcellular localization of OsWRKY70 , we constructed an OsWRKY70:GFP fusion gene , driven by a CaMV 35S promoter , and transiently expressed the construct in Nicotiana benthamiana leaves . Fluorescence analysis showed that OsWRKY70 is exclusively localized in the nucleus ( Figure 2—figure supplement 1A ) . To determine the DNA-binding activity of OsWRKY70 , a His-tagged protein was produced in Escherichia coli , and its W-box binding ability was examined by electrophoretic mobility shift assays ( EMSAs ) as described ( Chujo et al . , 2007 ) . In the presence of the oligonucleotide probe BS65 containing two W-box sequences and a WRKY70 recombinant protein , specific protein-DNA complexes with reduced migration were present in the EMSA assays ( Figure 2—figure supplement 1C ) . The DNA-binding specificity was confirmed in a competition experiment using a 250-fold excess of the unlabeled probe BS65 as a competitor , for which no binding complexes were detected . When the W-box in BS65 was mutated from TTGACC to TCCTAC ( mBS65 ) , the binding complexes also disappeared . These results indicated that the recombinant OsWRKY70 protein specifically binds to the conserved W-box in the synthesized probe . To investigate whether OsWRKY70 has transcriptional activation activity , we fused the full-length OsWRKY70 ORF in-frame to the GAL4 DNA-binding domain of the pGBKT7 vector and transformed it into yeast . The yeast transformed with pGBKT7 or pGBKT7-OsWRKY70 was plated on SD medium ( −Trp ) containing X-α-gal . After 12 hr at 30°C , the pGBKT7-OsWRKY70 transformant yeast colonies turned blue . In contrast , the pGBKT7 empty transformant yeast colonies remained white ( Figure 2—figure supplement 1B ) . OsWRKY70 is therefore likely functioning as a transcriptional activator in the yeast system . We found that the promoter region of WRKY70 contains four W-boxes , three reverse W-boxes ( AGTCAA at −82 to −77 , AGTCAA at −56 to −51 , and GGTCAA at −49 to −44 ) , and one forward W-box ( TTGACC at −62 to −57 ) , upstream of the transcription start site ( Figure 2—figure supplement 2A ) . To investigate if WRKY70 regulates its own expression , we first performed an EMSA assay by using the minimal promoter region of WRKY70 ( 86 bp upstream of transcription start site ) as a probe . WRKY70-His can bind to this fragment , while adding 250-fold unlabeled probe resulted in no WRKY70-DNA complex ( Figure 2—figure supplement 2B ) . Using WRKY70 promoter:GUS as a reporter and 35S: WRKY70-GFP , 35S:GFP as effectors expressed transiently in N . benthamiana , we found that WRKY70-GFP significantly increased the GUS activities compared to GFP alone , suggesting that WRKY70 can self-activate its transcription ( Figure 2—figure supplement 2C ) . MAPK proteins can specifically recognize the D domain found in some group I-type WRKYs and specifically phosphorylate the Ser residues of Group I SP clusters ( Ishihama et al . , 2011; Mao et al . , 2011 ) . The D domain is a cluster of basic residues upstream of the LxL motif ( [K/R]1–2-x2–6-[L/I]-x-[L/I] ) and has been reported in some WRKYs to play an important role in determining the selectivity of interacting MAPKs and phosphorylation patterns ( Ishihama et al . , 2011 ) . OsWRKY70 has four SP clusters in the N-terminal region ( Figure 1—figure supplement 1 ) but has no D domains . We hypothesized that the D domain-deficient OsWRKY70 may nevertheless interact with the MAPKs , OsMPK3 and OsMPK6 , the homologs of AtMPK3 and AtMPK6 in Arabidopsis and WIPK and SIPK in Nicotiana tabacum , respectively , all of which are involved in plant defence responses ( Wu et al . , 2007; Lu et al . , 2011 ) . We used a GST pull-down assay to analyse the interaction between OsWRKY70 and OsMPK3 or OsMPK6 in vitro . OsWRKY70-His was pulled down strongly by GST-MPK3 and mildly by GST-MPK6 , suggesting that OsWRKY70 can interact with both OsMPK3 and OsMPK6 , with the OsWRKY70–OsMPK3 interaction being more efficient than the OsWRKY70–OsMPK6 interaction in vitro ( Figure 2B ) . In vivo , we used a bimolecular fluorescence complementation ( BiFC ) assay to confirm the interaction . Fluorescence was observed when nYFP-OsWRKY70 was co-injected with OsMPK3-cYFP or OsMPK6-cYFP , and the signals were in the nuclear compartment according to 4 , 6-diamidino-2-phenylindole ( DAPI ) staining . No fluorescence was observed when nYFP-OsWRKY70 was co-expressed with unfused cYFP ( Figure 2C ) . Taken together , these results strongly suggest that OsMPK3 and OsMPK6 physically interact with OsWRKY70 . In a next step , we investigated if OsWRKY70 is phosphorylated by OsMPK3 or OsMPK6 . We used OsMKK4DD , a constitutively active form of OsMKK4 , to activate recombinant OsMPK3 and OsMPK6 and then exposed them to OsWRKY70 . The result showed that OsWRKY70 can be phosphorylated by both OsMPK3 and OsMPK6 ( Figure 2D ) . Moreover , an EMSA assay revealed that phosphorylation did not alter the W-box-binding activity of OsWRKY70 ( Figure 2E ) . We also investigated if phosphorylation enhances the transactivation activity of OsWRKY70 using N . benthamiana as a transient expression system ( Li et al . , 2014 ) . As the constitutive expression of OsMKK4DD induces HR-like cell death in tobacco plants , we used an estradiol-inducible system of OsMKK4DD ( Ishihama et al . , 2011 ) . Given that OsWRKY70 can auto-activate its promoter ( Figure 2—figure supplement 2C ) , we used WRKY70 promoter:GUS as a reporter and 35S:WRKY70-GFP , 35S:OsMPK3-YFP , 35S:OsMPK6-YFP as effectors ( Figure 2F ) . GUS activity was higher in OsMKK4DD-MPK3-OsWRKY70 or OsMKK4DD-MPK6-OsWRKY70 co-expressed leaves than leaves expressing OsWRKY70 alone ( Figure 2E ) . These results show that phosphorylation of OsWRKY70 can increase transactivation activity , but not W-box binding activity , of OsWRKY70 . 10 . 7554/eLife . 04805 . 006Figure 2 . Interactions between OsWRKY70 and OsMPK3/6 . ( A ) Mean transcript levels ( +SE , n = 5 ) of OsWRKY70 in transgenic lines with silencing of OsMPK3 ( irMPK3 lines , irMPK3-53 , and irMPK3-183 ) or OsMPK6 ( irMPK6 lines , irMPK6-1 , and irMPK6-2 ) after infested by SSB for 1 hr . ( B ) In vitro interaction assays between OsWRKY70 and OsMPK3 or OsMPK6 . GST , GST-MPK3 , and GST-MPK6 purified proteins were incubated with WRKY70-His as indicated . WRKY70-His input and pulled-down fractions were analyzed by immunoblotting using anti-WRKY70 antibody ( top ) . Input proteins were monitored by Coomassie blue staining ( bottom ) . This experiment was repeated 3 times with similar results . ( C ) In vivo bimolecular fluorescence complementation interaction assays between OsWRKY70 and OsMPK3 or OsMPK6 . Fluorescence was observed from complementation of the N-terminal part of the YFP fused with OsWRKY70 ( nYFP-OsWRKY70 ) with OsMPK3 or OsMPK6 fused with the C-terminal part of the YFP ( OsMPK3-cYFP or OsMPK6-cYFP ) and co-localized with DAPI stains in the nuclear compartment of tobacco leaf cells . No fluorescence was observed when nYFP-OsWRKY70 was co-expressed with unfused cYFP . Scale bar , 50 μm . ( D ) In vitro phosphorylation of OsWRKY70 by OsMPK3/6 . The phosphorylated form of OsWRKY70 ( P-WRKY70 ) was detected by using Phos-tag Biotin BTL-104 ( top ) . Input proteins , including OsWRKY70-His ( WRKY70 ) , GST-OsPMK3 ( MPK3 ) , GST-OsPMK6 ( MPK6 ) , and His-OsMKK4DD ( MKK4DD ) were monitored by Coomassie blue staining . ( E ) Assays for W-box binding activity of OsWRKY70 . GST-OsMPK3 or GST-OsMPK6 was activated by a constitutively active form of OsMKK4 , His-OsMKK4DD . BS65 containing two W-boxes was used as the probe . ( F ) Assays for transactivation activity of OsWRKY70 . Leaves of N . benthamiana were agroinfiltrated with the indicated constructs . 24 hr later , leaves were injected with 10 mM 17-β-estradiol and were incubated for 12 hr . Total protein was extracted and GUS activities were subsequently quantified . Eight plants were used for each treatment . Letters indicate significant differences among different lines ( A ) or treatments ( F ) ( p < 0 . 05 , Duncan's multiple range test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 00610 . 7554/eLife . 04805 . 007Figure 2—figure supplement 1 . Subcellular localization , DNA-binding ability , and transcriptional activation activity of OsWRKY70 . ( A ) Subcellular localization of OsWRKY70 . N . benthamiana cells were transformed with GFP and OsWRKY70:GFP . After incubation for 48 hr , the transformed cells were observed under a confocal microscope . The photographs were taken in UV light , visible light , and in combination ( overlay ) , respectively . Scale bar , 25 μm . ( B ) Transcriptional activation activity of OsWRKY70 in yeast cells . Y187 yeast cells containing pGBKT7 and pGBKT7-WRKY70 , respectively , were plated on SD medium ( −Trp ) containing X-α-gal at 30°C for 12 hr and then the color was observed . ( C ) W-box binding ability of OsWRKY70 analyzed by electrophoretic mobility shift assay ( EMSA ) . The recombinant OsWRKY70 protein can bind the W-box sequence BS65 but not to the mutant probe mBS65 . Competition experiments were performed using unlabeled BS65 as a competitor in a 250-fold molar excess . This experiment was repeated twice with similar results . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 00710 . 7554/eLife . 04805 . 008Figure 2—figure supplement 2 . Self-activation of OsWRKY70 . ( A ) Partial sequence of OsWRKY70 promoter . Green highlights represent the translation initiation sites . Red highlights represent the transcription start sites . Blue and orange highlights represent W-boxes in sense and antisense strands , respectively . The framed sequence was used for EMSA assay . ( B ) In vitro EMSA assay for the binding activity of WRKY70 to the minimal WRKY70 promoter ( P70 ) . Competition experiments were performed using unlabeled BS65 as a competitor in a 250-fold molar excess . ( C ) In vivo transient assay for transactivation activity of OsWRKY70 . The WRKY70 promoter ( 1 . 8 kb ) :GUS and 35S:WRKY70-GFP were transiently expressed in N . benthamiana leaf cells . 2 days after infiltration , samples were harvested and GUS activities were quantified . 35S:GFP was used as a control . Eight plants were used for each treatment . Asterisks indicate significant differences in WRKY70-GFP-expressed plants compared with GFP controls ( * , p < 0 . 05; Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 008 To determine whether the two MAPKs regulate OsWRKY70 , we measured transcript levels of OsWRKY70 in MAPK-silenced rice plants and vice versa . OsWRKY70 transcript levels were significantly decreased in OsMPK3 ( Wang et al . , 2013 ) and OsMPK6 silenced lines ( Figure 3A ) after infestation with SSB for 1 hr ( Figure 2A ) . Another group I-type WRKY TF , OsWRKY24 , which has both SP clusters and a D domain , was down-regulated in OsMPK6 silenced plants , but not in OsMPK3 silenced lines ( Figure 3B ) . In WRKY70 silenced lines ( see below ) on the other hand , OsMPK3 and OsMPK6 transcripts were the same as in wild-type ( WT ) plants ( Figure 3C , D ) . These results show that the transcript levels of OsWRKY70 is regulated by OsMPK3 and OsMPK6 , but not vice versa . 10 . 7554/eLife . 04805 . 009Figure 3 . Transcript levels of OsMPK6 , OsWRKY24 , OsMPK3 , and OsMPK6 in different transgenic lines . ( A ) Mean expression levels ( +SE , n = 6 ) of OsMPK6 in OsMPK6 silenced lines ( irMPK6-1 and irMPK6-2 ) . Samples used for QRT-PCR were from plant stems that were infested by SSB for 1 hr . ( B ) Mean transcript levels ( +SE , n = 5 ) of OsWRKY24 in irMPK3 ( irMPK3-53 , irMPK3-183 ) and irMPK6 lines after infestation by SSB for 1 hr . ( C , D ) Mean transcript levels ( +SE , n = 5 ) of OsMPK3 ( C ) and OsMPK6 ( D ) in irWRKY70 lines after infestation by SSB for 1 hr . Letters indicate significant differences among different lines ( p < 0 . 05 , Duncan's multiple range test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 009 To determine the role of OsWRKY70 in herbivore-induced defence responses , we constructed OsWRKY70 overexpression and knockdown lines using Agrobacterium tumefaciens mediated transformation . Through GUS staining and hygromycin resistance selection , we obtained two homozygous , single-insertion OsWRKY70-silenced lines ( irWRKY70-7 and irWRKY70-8 ) and two overexpression lines ( oeWRKY70-8 and oeWRKY70-17; Figure 4—figure supplement 1A ) . OsWRKY70 overexpression resulted in dwarfed plants ( Figure 4B ) , suggesting that OsWRKY70 is a negative growth regulator . To reduce phenotypic effects ( see below ) , we also created hemizygous overexpressing lines ( hemi-oeWRKY70-8 and hemi-oeWRKY70-17 ) whose phenotype was weaker but still visible in both nutrient solution and soil ( Figure 4A , B ) . SSB-induced transcript levels of OsWRKY70 in irWRKY70-7 and irWRKY70-8 were suppressed by more than 80% compared to WT plants at 1 hr after SSB feeding . Conversely , OsWRKY70 transcript levels were increased about 14-fold in hemi-oeWRKY70-8 and hemi-oeWRKY70-17 plants ( Figure 4—figure supplement 1B ) . Transcriptional profiling of the OsWRKY70-homologous genes OsWRKY24 and OsWRKY53 confirmed that gene targeting was specific for OsWRKY70 ( Figure 4—figure supplement 1C , D ) . In soil , the hemi-oeWRKY70 lines displayed dark green leaves and delayed flowering , similar to known GA-deficient mutants ( Sakamoto et al . , 2004 ) . Plant height was reduced by 29% and 27% , and root length by 49% and 30% , respectively ( Figure 4D , E ) . In contrast , the irWRKY70 lines grew similar to WT plants ( Figure 4A , B ) , except for a slight increase in root length . To test whether OsWRKY70 acts as a negative regulator of GA biosynthesis , we profiled GA levels in the different lines using HPLC/MS–MS . The experiment revealed that GA1 , GA7 , GA19 , GA20 , GA24 , and GA53 levels were significantly lower in hemi-oeWRKY70 lines ( hemi-oeWRKY70-8 and hemi-oeWRKY70-17 ) than in WT plants ( Figure 4G ) . Moreover , the growth phenotype of the oeWRKY70 seedlings was successfully restored to WT levels when they were grown on 1/2 MS plates with GA3 at a concentration of 0 . 01 μM ( Figure 4C ) . Consistently , the GA biosynthesis gene GA 20 oxidase ( GA20ox7 ) was significantly down-regulated in the hemi-oeWRKY70-8 lines ( Figure 4F ) . These results suggest that OsWRKY70 regulates plant growth through GA biosynthesis . 10 . 7554/eLife . 04805 . 010Figure 4 . Altering OsWRKY70 expression affects GA levels and plant growth . ( A , B ) Growth phenotypes of OsWRKY70 transgene lines ( irWRKY70 lines , irWRKY70-7 and irWRKY70-8 , and oeWRKY70 and hemi-oeWRKY70 lines , oeWRKY70-8 , hemi-oeWRKY70-8 and hemi-oeWRKY70-17 ) and wild-type ( WT ) plants at tillering stage ( A ) and heading stage ( B ) . ( C ) 10-day-old seedlings of WT and hemi-oeWRKY70-8 lines whose seeds were surface sterilized and placed on 1/2 Murashige and Skoog agar medium containing GA3 ( minimum purity > 99% , Sigma , St Louis , MO ) at various concentrations . This experiment was repeated 3 times with similar results . ( D , E ) Root length ( D ) and plant height ( E ) of transgenic lines with silencing ( irWRKY70 ) or overexpressing ( hemizygous lines , hemi-oeWRKY70 lines ) of OsWRKY70 and WT plants at tillering stage . ( F ) Mean transcript levels ( +SE , n = 5 ) of OsGA20ox7 in hemi-oeWRKY70-8 , hemi-oeWRKY70-17 , and WT plant . ( G ) Mean levels ( +SE , n = 3 ) of gibberellins ( GAs ) , including GA1 , GA3 , GA7 , GA19 , GA20 , GA24 , and GA53 , in hemi-oeWRKY70-8 , hemi-oeWRKY70-17 , and WT plants . Letters indicate significant differences among different lines ( p < 0 . 05 , Duncan's multiple range test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 01010 . 7554/eLife . 04805 . 011Figure 4—figure supplement 1 . OsWRKY70 transgenic lines and levels of OsWRKY70 , OsWRKY24 , and OsWRKY53 transcripts in the transgenic lines and WT plants . ( A ) Southern blot analysis of transgenic lines with silencing ( irWRKY70 lines , irWRKY70-7 , and irWRKY70-8 ) or overexpressing ( hemizygous lines , hemi-oeWRKY70 lines , hemi-oeWRKY70-8 , and hemi-oeWRKY70-17 ) of OsWRKY70 . Genomic DNA was digested with XbaⅠ ( X ) and EcoRⅠ ( E ) . The Blot was hybridized with a probe specific for gus reporter gene . All transgenic lines have a single insertion of the transgene . Lane 1 , irWRKY70-7; lane 2 , irWRKY70-8; lane 3 , hemi-oeWRKY70-8; lane 4 , hemi-oeWRKY70-17 . ( B ) Mean expression levels ( +SE , n = 6 ) of OsWRKY70 in irWRKY70 , hemi-oeWRKY70 , and WT plants at 1 hr after infestation by SSB . ( C , D ) Mean transcript levels ( +SE , n = 5 ) of OsWRKY24 ( C ) and OsWRKY53 ( D ) in irWRKY70 , hemi-oeWRKY70 , and WT plants at different time after infestation by SSB . Asterisks indicate significant differences in irWRKY70 and hemi-oeWRKY70 compared with WT plants ( ** , p < 0 . 01; Student's t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 01110 . 7554/eLife . 04805 . 012Figure 4—figure supplement 2 . Elongation of the second leaf sheath in hemi-oeWRKY70-8 and WT plants in response to GA3 . WT and hemi-oeWRKY70-8 seeds were surface sterilized and placed on 1/2 Murashige and Skoog agar medium containing GA3 ( minimum purity > 99% , Sigma ) at various concentrations . The length of the second leaf sheaths was measured 10 days later . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 012 To understand how OsWRKY70 influences defence signaling in rice , we examined SSB-elicited JA , ET , and SA levels and the expression of biosynthesis genes in OsWRKY70 transgenic lines and compared them to WT plants . JA levels in the irWRKY70 lines were significantly decreased compared with WT plants upon SSB attack , while they were increased in the overexpressing lines ( Figure 5A ) . In accordance with this data OsWRKY70 positively regulated SSB-induced transcript levels of the JA-biosynthesis genes OsHI-LOX ( Zhou et al . , 2009 ) and OsAOS2 ( Mei et al . , 2006 ) ( Figure 5B , C ) . The accumulation of ethylene was similar in the irWRKY70 lines and WT plants , but the levels were significantly elevated in hemi-oeWRKY70 lines after SSB infestation ( Figure 5D ) . Consistent with this result , transcript levels of the ethylene biosynthesis gene OsACS2 were similar in irWRKY70 and WT plants when infested by SSB , but were much higher in the induced hemi-oeWRKY70 lines ( Figure 5E ) . WT plants and irWRKY70 lines had nearly identical constitutive and SSB-induced SA levels , whereas the SA levels in the hemi-oeWRKY70 lines were significantly lower than in WT plants ( Figure 5F ) . Isochorismate synthase ( ICS ) is a key enzyme in plant SA biosynthesis ( Wu and Baldwin , 2010 ) . We examined the OsICS1 gene ( Du et al . , 2009 ) in rice after SSB infestation and found that the OsICS1 transcriptional level was significantly decreased in the hemi-oeWRKY70 lines compared with WT plants ( Figure 5G ) . Taken together , these experiments demonstrate that OsWRKY70 positively regulates SSB-induced JA- and ET levels but negatively regulates SA levels . To explore the notion that OsWRKY70 may be an upstream regulator of the JA and ET pathways , we investigated the expression of OsWRKY70 in transgenic plants with impaired JA and ET signaling . We used an antisense OsHI-LOX line ( as-lox ) , which produces 50% less JA upon SSB infestation than WT plants ( Zhou et al . , 2009 ) , and an antisense-ACS2 line ( as-acs ) , which produces significantly less SSB-elicited ET than WT plants ( Lu et al . , 2011 ) . The experiments revealed that the levels of constitutive and SSB-induced OsWRKY70 transcripts in as-lox and as-acs plants were the same as those in WT plants over the first 60 min of infestation ( Figure 6A , B ) , suggesting that OsWRKY70 functions upstream of JA and ET signaling . To fully demonstrate that OsWRKY70 acts upstream of JA and ET , additional experiments with null-mutants would be required . 10 . 7554/eLife . 04805 . 013Figure 5 . OsWRKY70 mediates SSB-elicited JA , SA , and ET accumulation . Mean levels ( +SE , n = 5–10 ) of JA ( A ) , ET ( D ) , and SA ( F ) , and mean expression levels ( +SE , n = 5 ) of OsHI-LOX ( B ) , OsAOS2 ( C ) , OsACS2 ( E ) , and OsICS1 ( G ) in irWRKY70 , hemi-oeWRKY70 , and WT plants that were individually infested by a third-instar SSB larva . Asterisks indicate significant differences in irWRKY70 , hemi-oeWRKY70 compared with WT plants ( * , p < 0 . 05; ** , p < 0 . 01; Duncan's multiple range test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 01310 . 7554/eLife . 04805 . 014Figure 5—figure supplement 1 . W-box elements in promoter regions of OsHI-LOX , OsICS1 , OsAOS2 , OsACS2 , and OsGA20ox7 . Green highlights represent the translation initiation sites . Red highlights represent the transcription start sites that were predicted by BDGP ( http://www . fruitfly . org/seq_tools/promoter . html ) . Dark gray highlights and gray highlights represent W-box ( C/TTGACT/C ) and W-box like ( TGACT/C ) motif , respectively , which were predicted by PLACE ( http://www . dna . affrc . go . jp/htdocs/PLACE/ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 01410 . 7554/eLife . 04805 . 015Figure 6 . Levels of OsWRKY70 transcripts in WT plants and transgenic lines with impaired JA ( as-lox ) and ethylene ( as-acs ) biosynthesis . Mean transcript levels ( +SE , n = 5 ) of OsWRKY70 in transgenic lines with impaired JA ( A , as-lox ) and ethylene ( B , as-acs ) biosynthesis and WT plants after they were infested by SSB . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 015 Trypsin protease inhibitors ( TrypPIs ) are important direct defence proteins against SSB in rice and their activity is regulated by JA- and ET ( Zhou et al . , 2009; Li et al . , 2012 ) . Thus , we investigated the influence of OsWRKY70 on TrypPI activity and SSB performance . TrypPI activity was suppressed in the irWRKY70 lines and enhanced in the hemi-oeWRKY70 lines compared with WT plants ( Figure 7A ) . As expected , SSB caterpillars gained more mass on irWRKY70-7 and irWRKY70-8 plants and less mass on the overexpressing lines compared to those fed on WT plants ( Figure 7B ) . IrWRKY70 lines were more severely damaged by SSB than the WT plants , whereas the hemi-oeWRKY70 lines were less damaged ( Figure 7C , D ) . To determine whether the impaired SSB resistance and defences in the irWRKY70 lines is due to lower JA levels , we complemented irWRKY70 plants with JA and examined SSB-induced TrypPI production and SSB performance . JA treatment attenuated the difference in TrypPI levels between WT plants and the irWRKY70 lines ( Figure 7E ) . Moreover , SSB larvae fed on JA-treated irWRKY70 lines gained a similar amount of weight to those fed on equally treated WT plants ( Figure 7F ) . The complete restoration of plant resistance to SSB and elicited accumulation of TrypPIs in irWRKY70 by exogenous JA application suggests that OsWRKY70 mediates rice-resistance to SSB through JA signaling . 10 . 7554/eLife . 04805 . 016Figure 7 . OsWRKY70 positively regulates resistance in rice to SSB . ( A ) Mean Trypsin protease inhibitor ( TrypPI ) activities ( +SE , n = 6 ) in irWRKY70 , hemi-oeWRKY70 , and WT plants that were individually infested by a third-instar SSB larva for 3 days . ( B ) Mean larval mass ( +SE , n = 50 ) of SSB that fed on irWRKY70 , hemi-oeWRKY70 , and WT plants for 14 days . ( C , D ) Damaged phenotypes of irWRKY70 ( C ) , hemi-oeWRKY70 ( D ) , and WT plants that were individually infested by a third-instar SSB larva for 8 days ( n = 10 ) . This experiment was repeated twice with similar results . ( E ) Mean activities ( +SE , n = 6 ) of TrypPIs in irWRKY70 and WT plants that were individually treated either 100 μg JA in 20 μl of lanolin paste ( JA ) or with 20 μl of pure lanolin ( insert ) for 24 hr , followed by SSB feeding for 3 days; ( F ) Mean larval mass ( +SE , n = 50 ) of SSB 12 days after fed on irWRKY70 and WT plants that were individually treated either 100 μg JA in 20 μl of lanolin paste ( JA ) or with 20 μl of pure lanolin ( insert ) for 24 hr . Letters indicate significant differences among different lines ( p < 0 . 05 , Duncan's multiple range test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 016 Based on the above results , we investigated whether OsWRKY70 regulation influences plant resistance to a non-target herbivore ( i . e . , a secondary attacker that does not strongly activate OsWRKY70 ) : the piercing sucking rice BPH N . lugens . When irWRKY70 lines and WT plants were exposed to a BPH colony , adult females preferred feeding on the WT rather than the irWRKY70 lines ( Figure 8A , B ) . Similarly , BPH adult females laid more eggs on WT plants than irWRKY70 ( Figure 8A , B , inserts ) . In accordance with these findings , BPH adult females were found more often on hemi-oeWRKY70 lines than on WT plants and laid more eggs on the former than on the latter ( Figure 8C , D ) . Moreover , BPH nymphs fed on the irWRKY70 lines had lower survival rates than those fed on WT plants; in contrast , BPH nymphs fed on the hemi-oeWRKY70 lines had higher survival rates ( Figure 8E , F ) , showing that OsWRKY70 negatively regulates rice BPH resistance . Based on the signaling profiles showing that OsWRKY70 negatively regulates GAs , we hypothesized that the down regulation of GAs may be responsible for the enhanced susceptibility to BPH . We therefore conducted a series of experiments to explore the influence of OsWRKY70 dependent GA on BPH . First , we used a GA-deficient mutant , sd-1 ( Spielmeyer et al . , 2002 ) , and a GA-excessive mutant , eui ( Zhu et al . , 2006 ) , to test the importance of GA for BPH resistance . The BPH female adults preferred feeding and ovipositing on sd-1 rather than the WT ( ZH11 ) ( Figure 9A ) , whereas the eui lines repelled BPH feeding compared with WT plants and did not affect BPH oviposition ( Figure 9B ) . The BPH nymph mortality was significant higher on the eui mutant compared with WT plants , but the sd-1 mutant did not influence BPH nymph performance ( Figure 9C ) . Second , we complemented the hemi-oeWRKY70 lines with GA3 at a concentration of 1 μM . This treatment restored BPH resistance to WT levels: BPH female adults feeding and ovipositing showed no preference between the WT and GA3-treated hemi-oeWRKY70 lines ( Figure 9D , E ) , and the BPH nymph survival rate was the same on GA3-treated hemi-oeWRKY70 and WT plants ( Figure 9F ) . Taken together , these results strongly suggest that OsWRKY70 negatively regulates BPH resistance through GA signaling . 10 . 7554/eLife . 04805 . 017Figure 8 . OsWRKY70 negatively regulates resistance of rice to BPH . ( A–D ) Mean number of female BPH adults per plant ( +SE , n = 8 ) on pairs of plants ( WT vs irWRKY70-7 , irWRKY70-8 , hemi-oeWRKY70-8 , and hemi-oeWRKY70-17 , respectively ) , 1–48 hr after pairs were exposed . Inserts: mean percentage ( +SE , n = 8 ) of BPH eggs per plant on pairs of plants as started above , 48 hr after the release of BPH . ( E , F ) Mean survival rate ( +SE , n = 10 ) of BPH nymphs that fed on irWRKY70 , hemi-oeWRKY70 , or WT plants 1–12 days after the start of feeding . Asterisks indicate significant differences in irWRKY70 , hemi-oeWRKY70 compared with WT plants ( * , p < 0 . 05; ** , p < 0 . 01; Student's t-test [A-D] or Duncan's multiple range test [E , F] ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 01710 . 7554/eLife . 04805 . 018Figure 9 . The GA-signaling pathway positively regulates rice resistance to BPH . ( A , B ) Mean number of adult female BPH per plant ( +SE , n = 8 ) on pairs of plants ( WT ( ZH11 ) vs sd-1 and eui , respectively ) , 1–48 hr after pairs were exposed . Inserts: mean percentage ( +SE , n = 8 ) of BPH eggs per plant on pairs of plants as started above , 48 hr after the release of BPH . ( C ) Mean survival rate ( +SE , n = 10 ) of BPH nymphs that fed on sd-1 , eui lines , or WT ( ZH11 ) plants 1–12 days after the start of feeding . ( D , E ) Mean number of female BPH adults per plant ( +SE , n = 8 ) on pairs of plants , a WT plant that was grown in a nutrient solution without GA3 vs a hemi-oeWRKY70-8 ( D ) or hemi-oeWRKY70-17 ( E ) plant that was grown in a nutrient solution with GA3 at a concentration of 1 μM for 24 hr , 1–48 hr after pairs were exposed . Inserts: mean percentage ( +SE , n = 8 ) of BPH eggs per plant on pairs of plants as started above , 48 hr after the release of BPH . ( F ) Mean survival rate ( +SE , n = 10 ) of BPH nymphs that fed on WT plants that were grown in a nutrient solution without GA3 or hemi-oeWRKY70 lines ( hemi-oeWRKY70-8 and hemi-oeWRKY70-17 ) that had been grown in a nutrient solution with GA3 at a concentration of 1 μM for 24 hr , 1–12 days after the start of feeding . Asterisks indicate significant differences in mutants compared with WT plants ( * , p < 0 . 05; ** , p < 0 . 01; Student's t-test [A , B , D , E] or Duncan's multiple range test [C] ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04805 . 018 Our experiments demonstrate that prioritizing defence over growth in response to a chewing herbivore is linked to a trade-off with resistance against a piercing-sucking herbivore via a WRKY TF . This indirect additional cost of defence may lead to the evolution of divergent , herbivore-community dependent plant resistance strategies in nature and may significantly constrain efforts to breed herbivore-resistant plants . It has been well documented that there are trade-offs between plant growth and defence ( Zavala and Baldwin , 2006; Zhang et al . , 2008; Meldau et al . , 2012; Yang et al . , 2012 ) . Resource availability , competition , plant ontogeny , and herbivory can influence the allocation of resources to growth and defence ( Stamp , 2003; Boege and Marquis , 2005 ) . In nature , defence prioritization is complicated by the fact that plants are often attacked simultaneously by multiple herbivore species , which have different sensitivities to various defence strategies , leading to resistance trade-offs ( Stam et al . , 2014 ) . For example , leaf-chewing caterpillars were found to perform better on Arabidopsis plants that are attacked by phloem-sucking aphids and vice versa ( Soler et al . , 2012 ) . Given that herbivory is an important driving force for the evolution of plant defence ( Agrawal et al . , 2012; Züst et al . , 2012 ) , understanding growth/defence and resistance trade-offs is important to predict and understand selection patterns in nature . Our study reveals that growth/defence and resistance trade-offs can emanate from the same mechanistic basis . From a plant's perspective , this suggests that reducing growth to support defence is even more costly than previously anticipated . From an agricultural point of view , this result indicates that it may be problematic to breed resistant varieties that rely on induced defence , as these plants may suffer from both a depression in growth and increased susceptibility to non-target herbivores . The discovery and manipulation of a TF that directly regulates defence prioritization allows us to draw a detailed picture of the mechanisms that underlie defence prioritization in rice . OsWRKY70 is rapidly induced following mechanical wounding and SSB feeding , but not following attack by a piercing sucking herbivore . Despite a lacking D-domain , OsWRKY70 interacts with and is regulated by two MAP-kinases , OsMPK3 and OsMPK6 ( Figure 2 ) . It has been well documented that Group I-type WRKY TFs can be phosphorylated by MAPKs and that the SP clusters are the phosphorylating sites ( Mao et al . , 2011; Ishihama and Yoshioka , 2012 ) . We found that both OsMPK3 and OsMPK6 phosphorylate OsWRKY70 , which in turn increased the transactivation activity of OsWRKY70 ( Figure 2 ) . Moreover , OsWRKY70 can auto-regulate itself ( Figure 2—figure supplement 2 ) . Given that autoregulation and cross-regulation are common features of WRKY action ( Ishihama and Yoshioka , 2012 ) , OsWRKY70 transcript levels are likely reduced in OsMPK3 and 6 silenced lines because of the reduction in phosphorylated WRKY70 and other WRKYs , which decreases WRKY activity and thereby reduce OsWRKY70 transcript levels . Phytohormones on the other hand do not regulate OsWRKY70 . Combined with its capacity to bind to W-box sequences and to act as transcriptional activator , this places OsWRKY70 at the interface between early recognition and signaling and hormonal regulation ( Rushton et al . , 2010; Wu and Baldwin , 2010; Erb et al . , 2012 ) . Indeed , silencing and over-expression of OsWRKY70 demonstrates its central role in regulating defence and growth through JA , ET , SA , and GA signaling , which enables OsWRKY70 to reduce plant growth and increases defence upon herbivore attack . In other plant systems , WRKYs have also been reported to play important roles in the regulation of transcriptional reprogramming associated with plant growth , development , and stress responses at different levels , including upstream and downstream of protein kinases and hormones ( Rushton et al . , 2010; Ishihama et al . , 2011; Mao et al . , 2011; Zheng et al . , 2013 ) . NaWRKY3 and NaWRKY6 , the homologs of OsWRKY70 in N . attenuata , for instance , have been reported to function downstream of NaSIPK and NaWIPK and upstream of JA biosynthesis ( Skibbe et al . , 2008 ) . In Arabidopsis , AtWRKY33 , the homolog of OsWRKY70 is phosphorylated by AtMPK3/MPK6 and can affect ET synthesis by directly binding promoter region of ACS2 and ACS6 genes ( Li et al . , 2012 ) . Another rice WRKY , OsWRKY24 has been reported to repress GA signaling in rice aleurone cells ( Zhang et al . , 2009 ) . OsWRKY70 may regulate phytohormone signaling via two , not mutually exclusive routes . First , it may directly bind to genes that are involved in hormone biosynthesis and signaling . Consistent with this hypothesis , OsWRKY70 positively modulated the transcript levels of the JA- and ET-synthesis genes OsHI-LOX , OsAOS2 , and OsACS2 , and negatively regulated the transcripts of the SA- and GA-biosynthesis genes OsICS1 and OsGA20ox7 ( Figures 4F , 5B , C , E , G ) . The existence of W-box or W-box like motifs in the promoters of these genes ( Figure 5—figure supplement 1 ) provides additional indirect evidence for their interaction with OsWRKY70 . Second , OsWRKY may regulate growth and defence through indirect hormonal cross-talk . It has been reported that the JA-signaling pathway is connected to the GA-signaling pathway through COI1-JAZ1-DELLA-PIF complexes , resulting in mutual suppression . The activation of JA signaling inhibits GA-mediated plant growth , whereas the activation of the GA pathway inhibits JA-mediated plant defence ( Yang et al . , 2012 ) . Moreover , JA was found to inhibit GA biosynthesis via an unknown mechanism ( Yang et al . , 2012; Heinrich et al . , 2013 ) , and the GA-GID1-DELLA complex was found to positively regulate the production of SA ( Navarro et al . , 2008 ) . Thus , the observed phytohormone levels and associated phenotypes in the transgenic OsWRKY70 lines might be at least in part due to antagonistic and synergistic phytohormone crosstalk . It has also been reported that the absence of JA signaling enhances the sensitivity of plants to GAs ( Yang et al . , 2012 ) . In our experiments , the promotion of JA signaling in oeWRKY70 lines did not decrease the sensitivity of plants to GA3 ( Figure 4—figure supplement 2 ) . Moreover , the dwarf phenotype of oeWRKY70 lines was completely restored by exogenous GA3 at low concentrations of 0 . 01 μM ( Figure 4C ) . This suggests that the dwarf phenotype of oeWRKY70 lines is directly related to the low level of GAs and that the sensitivity of plants to GAs may be influenced by other OsWRKY70-mediated factors other than JA . Interestingly , irWRKY70 lines showed similar growth phenotypes to WT plants , indicating that GA levels are unlikely to be altered in irWRKY70 lines . This suggests an involvement of other factors , such as the homologs of OsWRKY70 , OsWRKY24 , and OsWRKY53 , in the biosynthesis of GAs . Overall , the combination of direct regulation and indirect phytohormone crosstalk may explain how a single TF can act as a node in multiple signaling processes and integrate growth and defence responses . Our experiments do not only connect OsWRKY70 with phytohormone signaling , but also illustrate how hormonal signaling affects resistance responses against different herbivores . In rice , it is well documented that TrypPIs are an effective JA-dependent defence against chewing herbivores , including SSB ( Zhou et al . , 2009; Lu et al . , 2011 ) . Here , we found that OsWRKY70 positively mediated the production of elicited JA and TrypPIs ( Figures 5A , 7A ) , which subsequently modulated resistance in rice to SSB ( Figure 7B–D ) . Moreover , JA complementation of irWRKY70 lines , which had lower herbivore-elicited JA levels , completely restored TrypPI activity and SSB resistance compared to equally treated WT plants ( Figure 7E , F ) . These data suggest that OsWRKY70-mediated resistance in rice to SSB is mainly due to its effect on the JA-signaling pathway . On the other hand , little is known so far about the role of GA in herbivore resistance against piercing-sucking insects . We found that GA3 application restored BPH resistance in hemi-oeWRKY70 mutants ( Figure 9D–F ) . Moreover , the GA-deficient mutant sd-1 improved the performance of BPH , whereas the GA-excessive mutant eui decreased BPH performance ( Figure 9A–C ) . These data demonstrate that the GA-signaling pathway plays an important role in modulating resistance in rice to BPH in addition to its regulation of plant growth . GA-modulated BPH resistance in rice may occur via two mechanisms . One is that GA directly regulates the BPH defence response . It has been reported that GA can positively modulate the pathogen-related PBZ1 gene ( Tanaka et al . , 2006 ) and cell modification ( Yang et al . , 2008 ) . The rigidity of the cell wall is important for resistance to BPH phloem-feeding . Moreover , GA can directly elicit the plant growth , which may enhance the tolerance of rice to BPH . Another possibility is that GA indirectly regulates BPH resistance by eliciting SA and ROS pathways , both of which have been reported to be involved in resistance of rice to BPH ( Zhou et al . , 2009; Lu et al . , 2011 ) , via GA-GID1-DELLA complex ( Navarro et al . , 2008; Alonso-Ramírez et al . , 2009 ) . Thus far , several other elements of the rice defense signaling cascade , including JA ( Zhou et al . , 2009 ) , OsERF3 ( Lu et al . , 2011 ) , and ethylene ( Lu et al . , 2014 ) have also been shown to have similar divergent effects on SSB and BPH , suggesting distinct resistance strategies of rice plants against these two herbivores . The main reason for the divergent signaling , response and resistance to the two herbivores might be their different feeding habits . Resistance mechanisms against phloem-feeders are well documented to differ substantially from mechanisms against chewing herbivores ( Bostock , 2005 ) . In summary , our experiments provide evidence for a key role of OsWRKY70 in defence prioritization and illustrate that reducing growth via GA signaling opens the door to secondary infection by non-target herbivores . When attacked by a chewing herbivore such as SSB , rice plants will recognize signals derived from the herbivore and activate OsMPK3 and OsMPK6 . The activated OsMPK3 and OsMPK6 then elicit OsWRKY70 , which subsequently activates the JA- and ET-signaling pathways , resulting in the production of defence compounds such as TrypPIs and an increase in plant resistance . Simultaneously , the activation of OsWRKY70 decreases the production of GAs , which inhibits plant growth and thus prioritizes defence overgrowth and leads to increased susceptibility to BPH . Through these results , our study illustrates that the transcriptional modulation of hormonal networks allows plants to mount an appropriate defence program . At the same time , however , prioritizing defence over growth leads to significant resistance trade-offs , which may constrain plant resistance breeding and favor the evolution of herbivore-specific responses in plants . The following rice genotypes were used in the present study: ( i ) Xiushui 11 WT plants and the corresponding transgenic lines irWRKY70 , hemi-oeWRKY70 ( see below ) , as-lox ( Zhou et al . , 2009 ) , as-acs ( Lu et al . , 2011 ) , irMPK3 ( Wang et al . , 2013 ) , irMPK6 ( Figure 4A ) and ( ii ) Zhonghua 11 ( ZH11 ) WT plants and the corresponding GA mutants sd-1 ( Spielmeyer et al . , 2002 ) and eui ( Zhu et al . , 2006 ) . Pre-germinated seeds were cultured in plastic bottles ( diameter 8 cm , height 10 cm ) in a greenhouse ( 28 ± 2°C , 14L: 10D ) . 10-day-old seedlings were transferred to 50 L hydroponic boxes with a rice nutrient solution ( Yoshida et al . , 1976 ) . After 30–35 days , seedlings were transferred to individual 500 ml hydroponic plastic pots . Plants were used for experiments 4–5 days after transplantation . A colony of SSB was originally obtained from rice fields in Hangzhou , China and maintained on TN1 rice seedlings using the method described previously ( Zhou et al . , 2009 ) . For BPH , we used a lab population that has been reared on TN1 rice seedlings for more than 20 generations . The full-length cDNA of OsWRKY70 was obtained by RT-PCR from total RNA isolated from WT plants infested by SSB larvae for 24 hr . The primers were designed based on the sequence of the rice OsWRKY70 ( TIGR ID Os05g39720 ) gene ( Supplementary file 1A ) , which showed high homology with the partial sequence of the OsWRKY70 transcript that was cloned by SSH . The PCR-amplified fragments were cloned into the pMD 19-T vector ( TaKaRa , China ) ( pOsWRKY70 ) and sequenced . The full-length cDNA and a 443 bp fragment ( Figure 1—figure supplement 1 ) of OsWRKY70 were cloned into the pCAMBIA1301 and pCAMBIA1301-RNAi vectors , respectively , yielding an over-expression ( oeWRKY70 ) and an inverted-repeat orientation ( irWRKY70 ) vector . Both the oeWRKY70 and irWRKY70 vectors were inserted into the rice variety Xiushui 11 using A . tumefaciens-mediated transformation . Homozygous T2 plants were selected using GUS staining or hygromycin resistance screening ( Zhou et al . , 2009 ) . For most experiments , two irWRKY70 T2 homozygous lines , irWRKY70-7 and irWRKY70-8 , each harboring a single insertion were used . However , oeWRKY70 homozygous lines were severe dwarfing and nearly no seeds , thus we used two hemizygous lines , hemi-oeWRKY70-8 and hemi-oeWRKY70-17 , each also harboring a single insertion to perform the experiments . The full-length ORF of OsWRKY70 was PCR-amplified and cloned into the pET-32a vector ( Novagen , Madison , WI ) . The full-length ORF of OsMKK4 was PCR-amplified and cloned into the pET-28b vector ( Novagen ) , and the two phosphorylation sites ( T and S ) of OsMKK4 were mutant to D ( OsMKK4DD ) by Q5 Site-Directed Mutagenesis Kit ( NEB ) . The full-length ORFs of OsMPK3 and OsMPK6 were PCR-amplified and cloned into the pGEX-4T-3 vector ( GE Healthcare ) . All of primers used for PCR amplification for these genes are listed in Supplementary file 1A . The constructs were transformed into E . coli BL21 ( DE3 ) ( Transgene , China ) . Expression was induced by adding 0 . 4 ( for OsWRKY70 and OsMKK4DD ) or 0 . 2 ( for OsMPK3 or 6 ) mM isopropyl-β-thiogalactopyranoside ( IPTG ) for 20 hr at 20°C ( for OsWRKY70 and OsMKK4DD ) or for 4 hr at 23°C ( for OsMPK3 and 6 ) . Cells were collected and the recombinant protein was purified using His or GST Trap ( GE healthcare , UK ) according to the manufacturer's instructions . The full-length ORF of OsWRKY70 was PCR-amplified and cloned into the GAL4 DNA-binding domain of the pGBKT7 vector ( Clontech , Palo Alto , CA ) . The vector construct was transformed into yeast Y187 ( Clontech ) according to the manufacturer's instructions . Transformants were selected on SD ( −Trp ) plates at 30°C until colonies appeared . The colonies were identified by PCR and transferred into SD ( −Trp ) liquid medium . The transformant yeast with pGBKT7 or pGBKT7-OsWRKY70 was plated on SD ( −Trp ) containing X-α-gal at 30°C for 12 hr until the pGBKT7-OsWRKY70 transformants developed a blue color . The full-length ORF without a stop codon of OsWRKY70 was cloned into the pEGFP vector ( Clontech ) to fuse it with GFP . The fusion gene , OsWRKY70:GFP , was inserted into pCAMBIA1301 , yielding a transformation vector . This vector was used for transient transformation of N . benthamiana leaves as described previously ( Li et al . , 2014 ) . Fluorescence was analyzed by confocal microscopy . Full-length ORFs of OsMPK3 , OsMPK6 , and OsWRKY70 without stop codons were cloned into serial pGreen-pSAT1 vectors containing either amino- or carboxyl terminal EYFP fragments and introduced into Agrobacterium as described previously ( Hou et al . , 2010 ) . 3-week-old N . benthamiana leaves were agroinfiltrated with agrobacterial cells containing the indicated constructs . 2 days after incubation , fluorescence and DAPI staining were analyzed by confocal microscopy . Pull-down assay was performed as described previously ( Ishihama et al . , 2011 ) ; 5 μg of GST-tagged OsMPK3 and OsMPK6 and 2 μg His6-tagged OsWRKY70 were used . The samples were analyzed by SDS-PAGE . After electrophoresis , the gels were stained with Coomassie Brilliant Blue or subjected to immunoblot analysis using anti-WRKY70 antibody . Antibody with specificity to OsWRKY70 was generated by immunizing rabbits with the peptide CVYYASRAKDEPRDD-keyhole limpet hemocyanin-conjugate , and purified by the GenScript Company ( Nanjing , China ) . The full-length ORFs of OsMPK3 , OsMPK6 , and OsMKK4DD without stop codons were cloned into pBA-YFP , pBA-YFP , and 6myc-pBA , respectively . 1 . 8-kb promoter region of OsWRKY70 was PCR-amplified ( primers were listed in Supplementary file 1A ) and cloned into pCAMBIA1391 . All constructs were introduced into AGL1 Agrobacterium . Leaves of N . benthamiana were agroinfiltrated with the indicated constructs ( see details in Figure 2E and Figure 2—figure supplement 2C ) at a ratio of 1:1:1:1 . At 24 hr after agroinfiltration , leaves were injected with 10 mM 17-β-estradiol and were incubated for 12 hr . 2 days after infiltration , leaves were harvested and frozen in liquid nitrogen . Each treatment was repeated 8 times . GUS quantitative assay was performed as described ( Xin et al . , 2012 ) . The probes used in EMSA were BS65 ( 5ʹ-ATCGTTGACCGAGTTGACTTT-3ʹ ) with two W-boxes , P70 ( GCCAGTCAAACCTCGAGGGAGCTTTGACCAGTCAACGGTCAAACGTTCAAAGGTCTATATAATGATCACCGGAGGCGTCGTCGTTG ) and the W-box mutant mBS65 ( 5ʹ-ATCGTCCTACGAGTCCTATTT-3ʹ ) ( Chujo et al . , 2007 ) , all of which were labeled by Biotin . EMSA was performed using a LightShift Chemiluminescent EMSA Kit ( Thermo , Rockford , IL ) according to the manufacturer's instructions . Competition experiments were performed using unlabeled BS65 as a competitor in a 250-fold molar excess . GST-MPK3 ( 1 μg ) or GST-MPK6 ( 1 μg ) with or without His-MKK4DD ( 1 μg ) was incubated in a kinase reaction buffer ( 50 mM Tris–HCl , pH 7 . 5 , 1 mM Dithiothreitol ( DTT ) , 10 mM MgCl2 , 10 mM MnCl2 , 50 μM ATP ) at 30°C for 30 min . After this , recombinant His-WRKY70 ( 1 μg ) was added and the mixture was incubated again at 30°C for 30 min . The reactions were stopped by adding SDS-loading buffer and heated for at 95°C for 5 min . The products were analyzed by SDS-PAGE . Phosphorylated proteins were detected by using Phos-tag Biotin BTL-104 ( Wako , Japan ) according to the manufacturer's instruction . Five independent biological samples were used . Total RNA was isolated using the SV Total RNA Isolation System ( Promega , Madison , WI ) . One μg of each total RNA sample was reverse transcribed using the PrimeScript RT-PCR Kit ( TaKaRa ) . qRT-PCR was performed on a CFX96 Real-Time system ( Bio-RAD , Richmond , CA ) using a Premix Ex Taq Kit ( TaKaRa ) . The primers and probe sequences used for mRNA detection of target genes by qRT-PCR are shown in Supplementary file 1B . A rice actin gene , OsActin ( TIGR ID: Os03g50885 ) , was used as an internal standard to normalize cDNA concentrations . Plants ( one per pot ) were randomly assigned to SSB and control treatments . Two irWRKY70 lines ( irWRKY70-7 and irWRKY70-8 ) , two hemi-oeWRKY70 lines ( hemi-oeWRKY70-8 and hemi-oeWRKY70-17 ) , and one WT line were used . The stems were harvested at 0 , 1 . 5 , and 3 hr after SSB treatment , and JA and SA levels were analyzed by GC–MS using labeled internal standards as described previously ( Lu et al . , 2011 ) . Three plants were covered with a sealed glass cylinder ( diameter 4 cm , height 50 cm ) and ethylene production was determined at 12 and 24 hr after the start of the experiment using the method described previously ( Lu et al . , 2006 ) . Each treatment at each time interval was replicated 10 times . 10-day-old seedlings ( 3 g ) of hemi-oeWRKY70-8 , hemi-oeWRKY70-17 , and WT were frozen in liquid nitrogen , ground to fine powder , and extracted with 15 ml of 80% ( vol/vol ) methanol at 48°C for 12 hr . Different [2H2] labeled GAs were added to plant samples before grinding as internal standards . The extraction and analysis were performed as described previously ( Chen et al . , 2011 ) . Each line was replicated 3 times . Plant stems ( 0 . 2–0 . 3 g per sample ) were harvested at different time after different treatment ( see details in Figure 7 ) . The TrypPI activity was measured using a radial diffusion assay as described previously ( Van Dam et al . , 2001 ) . Each treatment was replicated 5 times . Differences in plant height , root length , herbivore performance , expression levels of genes and JA , SA , GA , ethylene , and H2O2 levels on different treatments , lines , or treatment times were determined by analysis of variance ( ANOVA ) ( Student's t-tests for comparing two treatments ) . All tests were carried out with Statistica ( Statistica , SAS Institute Inc . , http://www . sas . com/ ) . Sequence data from this article can be found in the Rice Annotation Project Database ( RAP-DB ) under the following accession numbers: Os05g39720 ( OsWRKY70 ) , Os05g27730 ( OsWRKY53 ) , Os01g61080 ( OsWRKY24 ) , Os03g17700 ( OsMPK3 ) , Os06g06090 ( OsMPK6 ) , Os02g54600 ( OsMKK4 ) , Os08g44590 ( OsGA20OX7 ) , Os08g39840 ( OsHI-LOX ) , Os03g12500 ( OsAOS2 ) , Os09g19734 ( OsICS1 ) , Os04g48850 ( OsACS2 ) .	Many different animals feed on plants , including almost half of all known insect species . Some herbivores—like caterpillars for example—feed by chewing . Others , such as aphids and planthoppers , use syringe-like mouthparts to pierce plants and then feed on the fluids within . To minimize the damage caused by these herbivores , plants activate specific defenses upon attack , including proteins that can inhibit the insect's digestive enzymes . The inhibitors are effective against chewing herbivores but seem to have little or no effect on some insects that feed by the ‘pierce-and-suck’ method . Investing in defense requires energy , and so plants attacked by herbivores actively slow their growth to meet this demand . Plants achieve this trade-off by changing the levels of different plant hormones . These hormones can control the expression of thousands of genes and have widespread effects throughout the plant . However , little is known about how prioritizing defense overgrowth in response to an attack by one herbivore affects the plant's ability to defend itself against other herbivores . Transcription factors are proteins that control which genes inside a cell are active or inactive . Li et al . searched for a transcription factor in rice plants that was specifically triggered in response to an attack by the caterpillars of a moth called the rice striped stem borer . This search identified a protein called WRKY70 as a transcription factor that prioritizes defense overgrowth . WRKY70 achieves this by increasing the levels of a defensive plant hormone ( called jasmonic acid ) while reducing the levels of a growth hormone ( called gibberellin ) . Further experiments show that the increase in jasmonic acid production is required to activate the enzyme inhibitors and for resistance against these caterpillars . Li et al . then found that increased WRKY70 activity makes rice plants more susceptible to attack by a second herbivore , a piercing-sucking insect called the rice brown planthopper . Further experiments revealed that this is due to the reduced levels of gibberillin . These findings show that while prioritizing defense overgrowth is effective against some insect herbivores , it comes with a cost as it makes the plants more susceptible to attack by other herbivores . This trade-off has important implications for both the evolution of plant immunity , and efforts to exploit plant immunity to help protect crops from herbivore attack .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2015	Prioritizing plant defence over growth through WRKY regulation facilitates infestation by non-target herbivores
Pentameric ligand-gated ion channels ( pLGICs ) mediate fast chemical signaling through global allosteric transitions . Despite the existence of several high-resolution structures of pLGICs , their dynamical properties remain elusive . Using the proton-gated channel GLIC , we engineered multiple fluorescent reporters , each incorporating a bimane and a tryptophan/tyrosine , whose close distance causes fluorescence quenching . We show that proton application causes a global compaction of the extracellular subunit interface , coupled to an outward motion of the M2-M3 loop near the channel gate . These movements are highly similar in lipid vesicles and detergent micelles . These reorganizations are essentially completed within 2 ms and occur without channel opening at low proton concentration , indicating that they report a pre-active intermediate state in the transition pathway toward activation . This provides a template to investigate the gating of eukaryotic neurotransmitter receptors , for which intermediate states also participate in activation . Pentameric ligand-gated ion channels ( pLGICs ) mediate fast chemical signaling between cells . Prominent members of the family include nicotinic acetylcholine ( nAChRs ) , serotonin-type-3 ( 5-HT3Rs ) , glycine ( GlyRs ) and γ-aminobutyric acid-A ( GABAARs ) receptors that are widely expressed in the nervous system . Their mutation can cause congenital myasthenia , epilepsy , hyperekplexia and possibly autistic and schizophrenic syndromes ( Steinlein , 2012 ) . They are the target of major classes of therapeutic substances , including anxiolytics and sedatives , general anesthetics , smoking cessation drugs and antiemetics ( Corringer et al . , 2012 ) . Hence , understanding the molecular mechanisms underlying their chemo-electric conversion is currently the matter of intensive work . Following the medium resolution structure of the Torpedo marmorata nAChR by electron microscopy in lipids ( Miyazawa et al . , 2003 ) , the first X-ray structures of full-length pLGICs were obtained on the bacterial homologs Erwinia chrysanthemi Ligand-gated Ion Channel ( ELIC ) ( Hilf and Dutzler , 2008 ) and Gloeobacter violaceus Ligand-gated Ion Channel ( GLIC ) ( Bocquet et al . , 2009; Hilf and Dutzler , 2009; Sauguet et al . , 2014 ) , followed by structures of eukaryotic members: the GluClαR ( Hibbs and Gouaux , 2011; Althoff et al . , 2014 ) , the β3GABAAR ( Miller et al . , 2014 ) , the 5-HT3R ( Hassaine et al . , 2014 ) , the α1GlyR ( Du et al . , 2015 ) by electron microscopy , the α3GlyR ( Huang et al . , 2015 ) by crystallography and the α4β2 nAChR ( Morales-Perez et al . , 2016 ) showing a highly conserved fold from bacteria to mammals . pLGICs are composed of five identical or homologous subunits arranged pseudosymmetrically around a central ion-conducting channel . They bind neurotransmitters within their extracellular domain ( ECD ) , promoting remote opening of an intrinsic ion channel within their transmembrane domain ( TMD ) . Each subunit’s ECD is composed of a rigid β-sandwich . The orthosteric binding sites for neurotransmitters are located at the subunit interface , halfway between the membrane and the top of the ECD . Each subunit’s TMD is composed of four membrane-spanning α-helices named M1–4 . M2 lines the ion channel , its upper part contributing to the gate that shuts the pore in the closed conformation . The well-conserved loops 2 , 7 and M2–M3 are found at the ECD-TMD interface ( Figure 1A ) . 10 . 7554/eLife . 23955 . 003Figure 1 . Structure of GLIC . ( A ) Positions for fluorophore and quencher insertion . Top panel: side view of GLIC crystalized at pH 7 . Lower panel: Top-down view of GLIC . Insets show particular regions targeted for mBBr labeling . Positions where cysteines are engineered are represented in color and positions for tryptophan/tyrosine mutations are in black . All amino acids are represented by a sphere for the Cß carbon , to show the orientation of the side chain . Endogenous tryptophans and tyrosines are in stick representation . ( B ) X-ray structure of the V135C-Bimane . A top and side view are presented showing the localization of the two Bimane molecules resolved in the structure . Insets show a zoom for both the top and side view of the V135C-Bimane and the putative quencher W72 . In the top zoom , the experimental 2Fo-Fc density map contoured at one sigma is shown in mesh for the Bimane . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 003 pLGICs exist in multiple allosteric states . The major states can be classified as: resting ( with a closed channel and a low affinity for agonists ) , active ( with an open channel ) and desensitized ( with a closed channel and a high affinity for agonists ) ( Changeux and Edelstein , 2005 ) . In addition , ensemble and single-channel kinetic analyses identified short-lived intermediate states during activation of nAChRs and GlyRs . These states , named ‘flip’ and ‘prime’ , display increased agonist affinity as compared to the resting state but still carry a closed channel ( Burzomato et al . , 2004; Lape et al . , 2008; Mukhtasimova et al . , 2009 ) . Finally , desensitization was early described as involving multiple conformations , including fast and slow desensitized states ( Sakmann et al . , 1980 ) . Among pLGICs of known structures , GLIC ( Bocquet et al . , 2009; Prevost et al . , 2012; Sauguet et al . , 2014 ) , GluClα ( Hibbs and Gouaux , 2011; Althoff et al . , 2014 ) and α1GlyR ( Du et al . , 2015 ) were each solved in three different conformations showing multiple tertiary and quaternary conformations that are difficult to assign to a particular allosteric state . Notably , all the structures were solved on detergent-solubilized proteins , most of them constrained in a crystal lattice that appears , in several cases , to be favoring a particular conformation regardless of the nature of bound ligands ( Nury et al . , 2011; Gonzalez-Gutierrez et al . , 2012 ) . Complementary methods are thus required to study the conformational changes of lipid-inserted , unconstrained receptors , in a time-resolved manner . Here , we used the proton-gated ion channel GLIC ( Bocquet et al . , 2007 ) to identify the conformational changes that occur during gating in the family of pLGIC channels . GLIC has been crystallized in three conformations ( called open , closed and locally-closed ) ( Bocquet et al . , 2009; Prevost et al . , 2012; Sauguet et al . , 2014 ) and studied by molecular dynamics , normal mode analysis and EPR spectroscopy ( Sauguet et al . , 2014; Calimet et al . , 2013; Velisetty et al . , 2014 ) . We used the tryptophan-induced ( TrIQ ) ( Mansoor et al . , 2002; Islas and Zagotta , 2006; Mansoor et al . , 2010 ) and tyrosine-induced ( TyrIQ ) ( Semenova et al . , 2009; Jones Brunette et al . , 2014 ) quenching methods to study short-range ( 5–15 Å ) inter-residue , pH-elicited motions of GLIC . The methods consist in covalently linking the protein with a fluorophore , here the small and pH-insensitive bimane ( Figure 2A ) , together with the insertion of a tryptophan or a tyrosine residue that quenches the fluorophore when the inter-residue Cα-Cα distances are less than approximately 15 Å and 10 Å , respectively . The Tr/TyrIQ approach was validated on the model system T4 lysosyme ( Mansoor et al . , 2002 , 2010; Jones Brunette et al . , 2014 ) and was previously used to follow the allosteric transitions of proteins such as the β2-adrenoceptor ( Yao et al . , 2006 ) and the lactose permease ( Smirnova et al . , 2014 ) . Previous fluorescence-based studies on pLGICs have used large fluorophores to follow conformational changes ( Talwar and Lynch , 2015 ) , but the TrIQ method has the unique advantage of allowing the assignment of fluorescence changes to relative motions between a fluorophore and a quencher of small sizes ( Yao et al . , 2006; Smirnova et al . , 2014 ) . 10 . 7554/eLife . 23955 . 004Figure 2 . Bimanes characteristics and labeling of GLIC-expressing cells . ( A ) Left: Structures of the Monobromo bimane ( mBBr ) and the bimane Bunte salt ( BBs ) before and after reaction with cysteines . Right: Excitation and emission spectra of both fluorophores and emission spectra of the mBBr ( 10 µM ) after reaction with cysteines ( 1 mM ) , acidification of the medium ( pH 4 ) or addition of tryptophan ( 25 mM ) . All spectra are normalized to the peak intensity of the mBBr emission spectrum . ( B , C ) Representative electrophysiological recordings of oocytes expressing GLIC Q193C mutant . Unless otherwise indicated , oocytes were perfused with a pH 7 . 3 solution . ( B ) Traces showing no functional inhibition of GLIC Q193C after mBBr exposure with acute ( left ) or prolonged application ( right ) , whereas the same oocytes are inhibited by reaction with MMTS , which can be reversed by application of DTT . ( C ) Left: Trace showing the inhibition of GLIC Q193C after reaction with the MMTS but not the BBs after acute application . Right: Effective inhibition of GLIC Q193C after reaction with the BBs after the oocyte was incubated for 1 hr in presence of the fluorophore , followed by reduction of the Cys-BBs bond by DTT . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 00410 . 7554/eLife . 23955 . 005Figure 2—figure supplement 1 . Bimane labeling of CHO cells . The pictures were taken on a confocal microscope with a 405 nm excitation laser and show GLIC D136C transfected cells , positive for the transfection marker mCherry ( not shown ) . On the left panel , cells were labeled for 1 hr , on ice , with the qBBr , a commercially available ‘non-permeant’ bimane derivative . On the right panel , cells were labeled for 1 hr , on ice , with the BBs . The scale bar represents 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 00510 . 7554/eLife . 23955 . 006Figure 2—figure supplement 2 . 1HNMR ( 300 MHz , D2O ) and 13C NMR ( 75 MHz , D2O/MeOD ) spectra of the BBs . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 00610 . 7554/eLife . 23955 . 007Figure 2—figure supplement 3 . Characterisation of the synthetized BBs . ( A ) RP-HPLC chromatogram and MS spectrum of bimane thiosulfonic acid . ( B ) : HRMS ( ESI ) spectrum of bimane thiosulfonic acid . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 00710 . 7554/eLife . 23955 . 008Figure 2—figure supplement 4 . Typical electrophysiological traces for each major quenching-pairs mutants of GLIC . On the left are recordings made before BBs labeling and on the right are recordings made for the same oocyte after a 1 hr BBs incubation , with the exception of the P250C mutant which showed an important run down ( either with or without labeling ) and for which two different representative oocytes are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 00810 . 7554/eLife . 23955 . 009Figure 2—figure supplement 5 . Immunolabeling of oocytes expressing GLIC WT and loss-of-function mutants . Oocytes were co-injected with GLIC-HA WT or loss-of-function mutants and GFP . Control oocytes were injected with GFP only . Immunolabeling was directed against the HA-tag . Labeling of receptors expressed at the membrane is shown in red and GFP expressed in the cytoplasm in shown in green . The scale bar represents 30 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 009 We focused our analysis on regions known to contribute critically to allosteric transitions of pLGICs and that are solvent-accessible , starting from the most membrane-distal region ( top ) of the ECD . A bimane was introduced at: ( 1 ) the subunit interface at the top of the ECD ( D136 and V135 on loop B ) and the middle of the ECD ( R133 on loop B , at the level of the orthosteric site ) , to monitor the quaternary reorganizations of this domain; and ( 2 ) at both sides of the ECD-TMD interface , at the bottom of the ECD ( K33 on loop 2 ) , on the M2-M3 loop ( the most N-ter proline of the loop P250 ) , and at the top of pore-lining M2 α-helix ( E243 , also termed E19’ , which is nearby the channel gate residues I233/I9’ and I240/I16’ ) ( Figure 1A ) . For each targeted position , a cysteine was engineered on the GLIC Cys-less background ( C27S , which does not produce functional alterations [Table 1] ) , followed by bimane labeling through cysteine modification . Each mutant was first characterized by two-electrode voltage clamp electrophysiology in Xenopus oocytes before and after bimane labeling to verify the functionality of the channel . Mutants were then expressed and purified from E . coli membranes , labeled with bimane and studied by fluorescence in detergent ( n-Dodecyl β-D-maltoside ) or asolectin liposomes . We first developed a labeling procedure in Xenopus oocytes . Indeed , we found that the currently used mBBr ( Monobromo bimane ) does not efficiently label surface receptors , since the reporter mutant GLIC Q193C ( pre-M1 ) , which is fully inhibited by reaction with methyl-methanethiosulfate ( MMTS ) ( Figure 2B ) , is neither affected nor protected from MMTS inhibition by mBBr treatment . Control experiments on chinese hamster ovary ( CHO ) cells expressing GLIC-D136C showed no surface labeling , but clear accumulation of the fluorophore in the cytoplasm ( Figure 2—figure supplement 1 ) . To decrease the hydrophobicity of the reagent , we synthesized the bimane-derived Bunte salt ( BBs ) , introducing a negatively charged SO3− leaving group ( Figure 2A and Figure 2—figure supplements 2 and 3 ) . After reaction with cysteine residues , the mBBr and BBs yield almost the same coupling product , with similar fluorescence and side-chain volume ( Mansoor and Farrens , 2004 ) ( Figure 2A ) . GLIC-expressing CHO cells show weak entry of the BBs in the cytoplasm and strong labeling at the membrane ( Figure 2—figure supplement 1 ) . Accordingly , BBs labeling of Q193C-expressing oocytes led to a severe loss of function indicating an efficient reaction with the surface receptors ( Figure 2C ) . Most mutants investigated herein were functional , generating robust pH-elicited currents , with few exceptions ( Table 1 and Figure 2—figure supplements 4 and 5 ) . For all functional mutants , after BBs labeling , currents showed wild-type like biphasic GLIC activation kinetics , with time constants ranging from 1 to 2 . 8 s ( τ1 ) and 5 to 8 . 9 s ( τ2 ) ( Table 2 and Figure 2—figure supplement 4 ) , except for P250C baring its endogenous quencher Y197 ( termed P250C-Y197 ) which displays slower kinetics of activation ( 5 . 4 s and 50 s ) . pH-current relationship measurements show ∆pH50 ( pH value for which half of the maximal electrophysiological response is recorded ) of less than one between wild-type and BBs-labeled mutants ( Table 1 ) . Thus , the labeling with the small BBs probe weakly affects the gating in most cases . 10 . 7554/eLife . 23955 . 010Table 1 . Dose dependence of currents and fluorescence . The table contains pH50 and nH values obtained through Hill equation fittings of current and fluorescence dose-response curves . Imax values represent the maximal current recorded . NF stands for not functional when maximal currents are smaller than 500 nA . NA stands for non-applicable and was used for mutants that did not elicit fluorescence variations bigger than 10% across the range of pH tested . NM stands for not measured . n represents the number of experiments . For all the data , mean values are presented and error values represent the standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 010Electrophysiology non-labeledElectrophysiology labeledFluorescence detergentFluorescence liposomesMutantpH50nHImaxnpH50nHImaxnpH50nHnpH50nHnWT5 . 3 ± 0 . 21 . 29 ± 0 . 066 ± 14NANANANANMNMNMNMNMNMC27S5 . 31 ± 0 . 071 . 6 ± 0 . 27 . 0 ± 0 . 455 . 3 ± 0 . 11 . 7 ± 0 . 27 ± 23NANA3NANA3R133C5 . 3 ± 0 . 21 . 8 ± 0 . 17 ± 345 . 1 ± 0 . 22 . 14 ± 0 . 058 ± 335 . 4 ± 0 . 31 . 7 ± 0 . 655 . 5 ± 0 . 11 . 3 ± 0 . 23R133C Y23W5 . 2 ± 0 . 42 . 0 ± 0 . 39 . 2 ± 0 . 835 . 1 ± 0 . 22 . 50 ± 0 . 076 ± 13NANA3NMNMNMR133C Q101W4 . 9 ± 0 . 22 . 3 ± 0 . 14 ± 235 . 0 ± 0 . 12 . 49 ± 0 . 034 . 9 ± 0 . 93NANA3NMNMNMR133C L103W5 . 1 ± 0 . 12 . 2 ± 0 . 46 . 8 ± 0 . 835 . 23 ± 0 . 091 . 7 ± 0 . 26 ± 235 . 2 ± 0 . 30 . 7 ± 0 . 245 . 50 ± 0 . 040 . 78 ± 0 . 073V135C W725 . 6 ± 0 . 21 . 7 ± 0 . 37 . 0 ± 0 . 945 . 29 ± 0 . 082 . 0 ± 0 . 17 . 2 ± 0 . 736 . 31 ± 0 . 091 . 7 ± 0 . 276 . 05 ± 0 . 023 ± 14V135C E67Q E75Q D91N5 . 11 . 87 . 315 . 1 ± 0 . 22 . 0 ± 0 . 46 . 5 ± 1 . 036 . 14 ± 0 . 031 . 8 ± 0 . 23NMNMNMD136C5 . 3 ± 0 . 11 . 9 ± 0 . 27 . 0 ± 0 . 945 . 29 ± 0 . 042 . 0 ± 0 . 38 . 3 ± 0 . 93NANA3NANA3D136C S93W4 . 50 ± 0 . 052 . 1 ± 0 . 43 ± 134 . 60 ± 0 . 061 . 5 ± 0 . 12 . 1 ± 0 . 93NANA4NMNMNMD136C Q101W5 . 17 ± 0 . 082 . 3 ± 0 . 18 ± 235 . 4 ± 0 . 22 . 2 ± 0 . 39 ± 335 . 85 ± 0 . 080 . 96 ± 0 . 0845 . 8 ± 0 . 11 . 3 ± 0 . 63D136C D178W51 . 86 . 215 . 31 . 86 . 61NANA3NMNMNMK33C W1605 . 88 ± 0 . 031 . 8 ± 0 . 19 ± 235 . 6 ± 0 . 21 . 8 ± 0 . 48 ± 236 . 22 ± 0 . 032 . 0 ± 0 . 445 . 95 ± 0 . 071 . 25 ± 0 . 096K33C W160F4 . 55 ± 0 . 032 . 6 ± 0 . 23 . 0 ± 0 . 134 . 4 ± 0 . 12 . 2 ± 0 . 61 ± 14NANA3NANA3P250C Y1975 . 1 ± 0 . 11 . 54 ± 0 . 097 . 0 ± 1 . 234 . 8 ± 0 . 22 . 3 ± 0 . 44 . 3 ± 0 . 835 . 88 ± 0 . 092 . 1 ± 0 . 855 . 5 ± 0 . 32 . 2 ± 0 . 64P250C W160FNFNFNF3NFNFNF46 . 06 ± 0 . 061 . 9 ± 0 . 13NMNMNMP250C Y194F5 . 1 ± 0 . 21 . 7 ± 0 . 46 . 4 ± 0 . 834 . 5 ± 0 . 21 . 8 ± 0 . 22 ± 135 . 97 ± 0 . 021 . 4 ± 0 . 44NMNMNMP250C Y197F5 . 2 ± 0 . 11 . 3 ± 0 . 34 . 1 ± 0 . 154 . 73 ± 0 . 091 . 2 ± 0 . 21 . 5 ± 0 . 93NANA3NANA3P250C Y251FNFNFNF7NFNFNF3NANA4NMNMNME243C5 . 1 ± 0 . 11 . 7 ± 0 . 38 ± 144 . 7 ± 0 . 21 . 7 ± 0 . 35 ± 434 . 5 ± 0 . 22 . 0 ± 0 . 244 . 8 ± 0 . 10 . 9 ± 0 . 36E243C K33WNFNFNF4NFNFNF3NANA3NMNMNME243C I201WNFNFNF10NFNFNF35 . 67 ± 0 . 041 . 549 ± 0 . 00435 . 81 . 42E243C F238W5 . 2 ± 0 . 21 . 4 ± 0 . 27 ± 144 . 7 ± 0 . 31 . 5 ± 0 . 55 ± 33NANA3NMNMNME243C L241W5 . 7 ± 0 . 20 . 69 ± 0 . 065 . 09 ± 0 . 0834 . 8 ± 0 . 10 . 61 ± 0 . 032 ± 13NANA3NMNMNME243C V242WNFNFNF4NFNFNF5NANA3NMNMNM10 . 7554/eLife . 23955 . 011Table 2 . Activation kinetics of GLIC mutants labeled with bimane . The τ values were obtained through a double exponential fit to the electrophysiology traces ( see Materials and methods ) . n represents the number of recordings used for the analysis . For all the data , mean values are presented and error values represent the standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 011Mutantτ1 ( s ) τ2 ( s ) τw ( s ) nR133C L103W1 . 0 ± 0 . 56 ± 53 ± 23V135C W721 . 9 ± 0 . 49 ± 53 ± 13D136C Q101W0 . 9 ± 0 . 25 ± 42 ± 23K33C W1601 . 7 ± 0 . 67 ± 55 ± 33P250C Y1975 . 4 ± 0 . 750 ± 830 ± 103E243C2 . 80 ± 0 . 098 . 9 ± 0 . 63 . 57 ± 0 . 063 Emission spectra of DDM-solubilized mutants labeled with bimane were recorded under steady state conditions ( 30 s post proton-application , excitation at 385 nm [see Materials and methods] ) . Fluorescence intensities were measured at the emission peak at various pH values ( from pH 7 . 3 to pH 3 ) ( Figure 3—figure supplement 2 ) and normalized to the intensity of the respective bimane-mutant under denaturing conditions ( 1% SDS ) . GLIC being activated by protons , we additionally performed control experiments to confirm that , as previous studies have shown ( Jones Brunette et al . , 2014 ) , both the bimane fluorescence and its quenching by tryptophans are unaffected by proton concentrations ranging from pH 9 to pH 2 ( Figure 3—figure supplement 1A , B ) ( see Materials and methods ) . Hence , the fluorescence variations of bimane-labeled mutants can be interpreted as reporting local structural reorganizations . When the bimane is introduced on loop B ( top of the ECD ) at positions 133 or 136 ( termed Bimane-133 and Bimane-136 [Figure 1A] ) , the fluorescence shows little variation in the pH 7 . 3–3 range , indicating that nearby putative quenching residues have a weak impact on the pH-dependent bimane fluorescence ( Figure 3A , B ) . Bimane-133 shows a marked blue shift ( 15 nm ) and a fluorescence two times higher than the denatured protein at pH 7 . 3 possibly indicating the probe when reacted to position 133 is located in a confined/hydrophobic environment ( Kosower et al . , 1982; Skjold-Jørgensen et al . , 2015 ) ( Figure 3A and Figure 3—figure supplement 2 ) . To generate quenching pairs , we introduced tryptophan residues on the adjacent subunit β-sandwich ( Figure 1A ) . For mutant Bimane-133 , the introduction of W23 ( β1 strand ) , W44 ( β2 strand ) and W101 ( β6 strand ) has little impact on the fluorescence at all pHs suggesting that these positions are never within quenching distances of the bimane ( Figure 3A ) . For mutant Bimane-136 , the introduction of W93 ( β5 strand ) produces a strong fluorescence decrease at all pHs indicating in this case that this residue is always within quenching distance of the bimane ( Figure 3B ) . In contrast , both the Bimane-133-W103 and Bimane-136-W101 show a marked pH-dependent fluorescence decrease ( Figure 3A , B ) . These data show that the residues at positions 133 and 103 , and those at positions 136 and 101 come closer upon acidification of the receptors . 10 . 7554/eLife . 23955 . 012Figure 3 . Steady-state variations of fluorescence of Bimane-GLIC mutants in detergent . Top: Cartoon view of two subunits of GLIC , all the sites of bimane labeling are represented by a colored sphere . ( A , B , C , D , E , F ) : The data show the fluorescence peak values as a function of the proton concentration , normalized to the peak intensity after SDS treatment . The same color code as the top cartoon was used to indicate the site of bimane labeling for each graph . Within one graph , the major quenching pairs with or without quencher are shown in bold black caption and secondary tryptophan quenchers are captioned in grey . The symbol ø was used for recordings made on receptors baring no other mutation than the one for bimane labeling and for which no endogenous quenchers were identified; otherwise the later are indicated in bold black caption . For all the points , mean values are presented and error bars represent the standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 01210 . 7554/eLife . 23955 . 013Figure 3—figure supplement 1 . Spectral characteristics of mBBr and Bimane-GLIC mutants . ( A ) Excitation spectra of the mBBr ( 10 µM ) ± Cys ( 100 µM ) , ± Trp ( 12 . 5 mM ) and at different pHs . All spectra were normalized on the peak value of the mBBr pH 8 excitation spectrum . ( B ) Variations of peak intensity of Bimane-Cys ( 10 µM ) ( purple ) and Bimane-Cys ( 10 µM ) + Trp ( 12 . 5 mM ) ( orange ) in solutions ranging from pH 9 to pH 2 . All intensities are normalized on the Bimane-Cys pH 7 . 3 peak intensity . All the data presented are mean values and error bars were calculated as standard deviations ( n = 3 ) . ( C ) Background fluorescence when recording Bimane-GLIC . The figure represents superimposed emission spectra of the Cys-less GLIC ( C27S ) and the Bimane-136-W101 mutant at pH 7 . 3 , either in detergent or after reconstitution in asolectin liposomes . For both mutants , proteins were treated with mBBr and subsequently purified by gel filtration to remove unreacted fluorophore . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 01310 . 7554/eLife . 23955 . 014Figure 3—figure supplement 2 . Emission spectra of Bimane-GLIC mutants in detergent . For each mutant , the graph on the left represents representative emission spectra recorded at different pHs and in presence of 1% SDS used as a denaturing agent; the graph on the right represents the same samples after they were individually brought back to pH 7 . 3 ( the 1% SDS spectrum is left for comparison ) . All axis labels are identical between graphs and were shown only on one graph for clarity . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 014 When the bimane is introduced at position 135 ( loop B ) , the pH-fluorescence relationship curve shows an inverted bell-shape , starting after normalization at 0 . 7 , decreasing down to 0 . 3 at pH 5–4 , then increasing up to 0 . 6 at pH 3 ( Figure 3C ) . The GLIC structures show a tryptophan ( W72 on loop 4 ) on the same subunit that could play the role of an endogenous quencher ( Figure 1A ) . W72 is a residue strictly conserved in all pLGICs ( Corringer et al . , 2012 ) . To determine whether W72 has a role in the fluorescence phenotype while avoiding severe mutation-induced structural alteration that would result from the loss of W72 , we solved the X-ray structure of Bimane-135 GLIC at pH 4 at 2 . 6 Å resolution ( Figure 1B and Supplementary file 1A ) . The structure shows a protein conformation quasi-identical to the wild-type GLIC structure ( Bocquet et al . , 2009 ) at pH 4 ( RMSD = 0 . 26 Å ) , with an additional electron density around C135 allowing unambiguous construction of a bimane moiety in two out of the five subunits . The data indeed show that the bimane is at a minimal 3 . 2 Å distance from the indole ring of W72 , which is therefore most likely the major partner in the pH-elicited fluorescence changes of Bimane-135 ( Figure 1B ) . As the 135 side chain projects at the subunit interface , the quenching of Bimane-135 by W72 seen up to pH 4 suggests a contraction of the domains interfaces , driving the bimane moiety toward W72 , although other mechanisms are possible . At very low pH , the bimane appears to move away from the W72 . Finally , we investigated a local tertiary motion of the orthosteric site , by combining the Bimane-136 ( loop B ) mutation with the introduction of a tryptophan at position 178 at the tip of the loop C from the same subunit ( Figure 1A ) . W178 leads to a 50% decrease in fluorescence at all pHs tested , indicating that both positions ( 136 and 178 ) are within quenching distance , and suggesting that their relative distance is unchanged upon increase of proton concentration ( Figure 3B ) , in agreement with EPR measurements that suggest an immobility of this position upon pH drop ( Velisetty and Chakrapani , 2012 ) . It is noteworthy that loop C has been proposed to undergo a closing motion during activation of pLGICs , a feature not seen here with fluorescence . First , we introduced a bimane at the tip of loop 2 in the ECD ( Bimane-33 ) , which lies on top of the M2 helix ( Figure 1A ) . Bimane-33 shows a twofold pH-dependent decrease in fluorescence from pH 7 . 3 to pH 5 ( Figure 3D ) . A putative quencher near Bimane-33 , W160 , is located on the adjacent subunit on the β9 strand at the bottom of the outer beta sandwich ( Figure 1A ) . Its mutation ( W160F ) essentially abolishes the proton-elicited fluorescent decrease , with only a 20% decrease in intensity at pH 6 , followed by a return to pH 7-like values at lower pHs , showing that Bimane-33 and W160 come closer together in the presence of protons . Second , we introduced a bimane at position 250 within the M2-M3 loop ( Figure 1A ) . Bimane-250 shows a 30% increase in intensity as the pH is decreased from 7 . 3 to 3 ( Figure 3E ) . The effect of three surrounding tyrosines and a single tryptophan were investigated through mutation to phenylalanine ( Figure 1A ) . Mutants Bimane-250-W160F ( β9 strand ) and -Y194F ( pre-M1 ) both lead to a modest increase in fluorescence at all pHs suggesting that W160 and Y194 remain at similar quenching distances of the bimane , independently of the protein conformation ( Figure 3E ) . In contrast , Y197F ( M1 helix ) and Y251F ( M2-M3 loop ) reduce the fluorescence quenching at high pHs , yielding a flat pH-dependent curve . The mutant Bimane-250-Y251F is non-functional ( Table 1 ) , which could result in its incapacity to visit different conformations and thus account for this phenotype . In contrast , the mutant Bimane-250-Y197F is functional , the unquenching phenotype showing that positions 250 and 197 move away from each other in the presence of protons ( Figure 3E ) . Third , we introduced a bimane on the top of the M2 α-helices ( Bimane-243 ) ( Figure 1A ) . Upon acidification , the fluorescence decreases slightly at pH 4 and then increases at pH 3 , pointing to complex reorganizations . We introduced tryptophan residues around the probe but failed to engineer new pH-dependent quenching pairs ( Figure 1A ) . Indeed , the mutants were either non-functional ( Bimane-243-W33 , -W201 and -W242 ) or functional but not eliciting pH-dependent fluorescence variations ( Bimane-243-W238 and -W241 ) ( Figure 3F and Table 1 ) . Nonetheless , regardless of the functionality of the mutants , Bimane-243 is robustly quenched by all the introduced tryptophans at pH 7 . 3 . As most introduced tryptophans are located in the TMD , the fluorescence data , in combination with GLIC structure inspection , suggests that the bimane fused bis-heterocycle at position 243 points toward the helix bundle of the adjacent subunit . In addition , the mutant Bimane-243 shows an emission blue shift ( 10 nm ) and a fluorescence 1 . 6 times higher than the denatured protein at pH 7 . 3 possibly reporting a confined/hydrophobic environment of the probe ( Figure 3F and Figure 3—figure supplement 2 ) adding more evidence to its presence inside the bundle of helix . Hence , the unassigned fluorescence variations at position 243 could account for motions of the top of the M2-helix relative to the rest of the subunit’s TMD . Altogether , the TrIQ/TyrIQ analysis generates a series of fluorescent reporters spanning from the apex of the ECD to the top of the pore lining M2 helices . Sensors at positions 136–101 , 135–72 , 133–103 , 33–160 , 250–197 and 243 follow the reorganizations of fully functional channels and have been selected for further analysis . In contrast , some sensors , such as the 250–160 and 243–101 , are of non-functional channels but still show specific quenching signals . This illustrates that particular conformational motions are not necessarily linked to pore opening . Since the molecular mechanisms impairing the function of these mutants are not known , they were not used for further analysis . At each selected position , changes in bimane/quencher distances could be the result of: ( 1 ) local side chains reorganizations , for example due to the protonation of residues surrounding the fluorophore and/or quencher , possibly affecting their orientation , regardless of the allosteric state of the receptor , or ( 2 ) global allosteric protein motions , mainly comprising backbone reorganizations . Two sets of experiments strongly support the latter hypothesis . First , we generated the mutant Bimane-135-W72-E67Q-E75Q-D91N , for which all titratable residues surrounding the Bimane-135-W72 pair were removed ( Figure 4A ) . This mutant shows the same pH-dependence of fluorescence as the simple mutant Bimane-135-W72 ( Figure 4A ) , supporting the hypothesis that the fluorescence variations are independent of the protonation of surrounding titratable residues . Second , we measured fluorescence variations in the presence of propofol , a negative allosteric modulator of GLIC . Propofol binds to the transmembrane domain of GLIC with an IC50 ≈ 25 µM and stabilizes closed channel conformations ( Nury et al . , 2011 ) . In presence of saturating concentrations of propofol , mutants Bimane-136-W101 , Bimane-135-W72 , Bimane-250-Y197 and Bimane-243 show a shift of the pH-fluorescence relationship curve toward higher proton concentrations , in agreement with the effect of an allosteric inhibitor ( Figure 4B ) . The observation that the binding of an effector to the TMD influences the fluorescence variations at the top of the ECD , more than 50 Å away , establishes that the fluorescence sensors indeed report global allosteric motions . 10 . 7554/eLife . 23955 . 015Figure 4 . Effects of amino acid protonation and propofol on the fluorescence variations . ( A ) Top: Fluorescence intensities of the mutants Bimane-135-W72 and Bimane-135-W72-E67Q-E75Q-D91N normalized on each mutant’s intensity at pH 7 . 3 . Bottom , from left to right: Top view of the extracellular domain of GLIC Bimane-C135 showing one subunit in light blue , and the Bimane-C135 ( green ) , E67 , E75 and D91 ( red ) in stick representation; zoom on the subunits interface of the ECD , top view; zoom on the subunits interface viewed from the inside of the ECD . ( B ) For each pH , fluorescence recordings of Bimane-labeled mutants were first made without propofol , followed by addition of propofol at 100 µM final concentration and re-recording of the same samples . All the data were normalized on the value of fluorescence at pH 7 . 3 without propofol . For all the data , mean values are presented and error bars are calculated as standard deviations ( n = 3 to 6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 015 Interestingly , the quenching pairs in the ECD , each located across the subunits’ interface , were mostly introduced on rigid loops and β strands ( e . g . loop B and β6 strand , Figure 1A ) . As they all show a decrease in fluorescence at low pH , the fluorescence variations likely reflect rigid body motions of the ECD’s β-sandwiches , thus coming closer to one another during allosteric transitions . In addition , the increase in fluorescence at low pH for the pair 250–197 reveals that another major allosteric reorganization of GLIC is the separation of loop M2-M3 ( P250 ) from the top of M1 ( Y197 ) . To link these motions to the gating transition of the receptor , we further studied GLIC in lipid bilayers , performing both steady state and real-time fluorescence measurements . To examine the impact of the membrane environment on the receptor reorganizations , we reconstituted selected bimane-GLIC mutants in asolectin liposomes , which are a mixture of lipids that were successfully used to reconstitute GLIC in a functional state ( Labriola et al . , 2013; Velisetty and Chakrapani , 2012 ) . Strikingly , at the top and middle of the ECD ( Bimane-133 ± W103 , Bimane-135-W72 and Bimane-136 ± W101 ) , steady-state fluorescence variations are similar , if not identical , in detergent micelles and liposomes ( Figure 5A , B , C , Table 1 and Table 3 ) , indicating that the corresponding movements are independent of the membrane environment . 10 . 7554/eLife . 23955 . 016Figure 5 . Steady-state fluorescence of Bimane-GLIC in liposomes . ( A , B , C , D , E , F ) The data show the fluorescence peak value for a given pH , normalized to the peak intensity of each mutant at the highest pH recorded ( pH 8 or 7 . 3 ) . Black lines represent the fluorescence values in liposomes and grey lines in detergent , for comparison . Plain lines represent the major fluorescent sensor and dashed lines the corresponding mutant in absence of quencher . The symbol ø was used for recordings made on receptors baring no other mutation than the one for bimane labeling and for which no endogenous quenchers were identified; otherwise the later are indicated in black caption . For all the data , mean values are presented and error bars are calculated as standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 01610 . 7554/eLife . 23955 . 017Table 3 . Fluorescence values at pH 7/8 in detergent or asolectin liposomes . The data presented are the values of fluorescence at the highest tested pH , normalized to the fluorescence after denaturation ( 1% SDS ) . n represents the number of experiments . For all the data , mean values are presented and error values represent the standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 017MutantF/FSDS in detergentnF/FSDS in asolectinnBimane-1331 . 98 ± 0 . 0341 . 91 ± 0 . 023Bimane-133-W1031 . 28 ± 0 . 0341 . 182 ± 0 . 0033Bimane-135-W720 . 66 ± 0 . 0370 . 76 ± 0 . 034Bimane-1360 . 835 ± 0 . 00330 . 87 ± 0 . 033Bimane-136-W1010 . 67 ± 0 . 0240 . 69 ± 0 . 024Bimane-33-W1600 . 83 ± 0 . 0241 . 09 ± 0 . 036Bimane-33-W160F0 . 81 ± 0 . 0830 . 87 ± 0 . 023Bimane-2431 . 68 ± 0 . 0241 . 60 ± 0 . 016Bimane-250-Y1970 . 55 ± 0 . 0540 . 60 ± 0 . 014Bimane-250-Y197F0 . 93 ± 0 . 0131 . 42 ± 0 . 083 At the ECD-TMD interface , liposome reconstitution causes a significant increase in fluorescence of Bimane-33 and Bimane-250-Y197F mutants at pH 8/7 , as compared to detergent conditions ( 30% and 50% , respectively ) ( Table 3 ) . To better visualize the pH-dependent changes , the data were normalized to the pH 8/7 values , showing that the marked pH-dependent changes of Bimane-250 and Bimane-33 are conserved in lipids , although Bimane-33-W160F is significantly brighter in detergent ( Figure 5D , E ) . Additionally , Bimane-243 shows similar pH-dependent variations in fluorescence in detergent and lipids ( Figure 5F ) . The data thus suggest that the relative movements at positions 33–160 and 250–197 are also essentially conserved in lipids . We propose that the differences in fluorescence intensity at certain positions between detergent and lipids are in line with their proximity to the membrane . Indeed , the GLIC structure at pH 4 shows the presence of a bundle of detergent within the pore and of a bound lipid in the upper part of the TMD ( Bocquet et al . , 2009 ) . These bound molecules could contribute to the differences in microenvironment observed here . It is noteworthy that , contrary to GLIC , most pLGICs are highly sensitive to the membrane environment . For instance , CHAPS-solubilized muscle nAChRs are strongly stabilized in a desensitized conformation ( Martinez et al . , 2002 ) and their reconstitution in the absence of anionic lipids or cholesterol ( PC liposomes ) yields an ‘uncoupled’ conformation that binds agonist with resting state–like low affinity , but does not undergo agonist-evoked conformational transitions ( daCosta et al . , 2013 ) . GLIC does not exhibit the same propensity to adopt an uncoupled conformation and , contrary to ELIC , retains the ability to undergo proton-elicited conformational change and to activate in PC liposomes and other lipid mixtures ( Labriola et al . , 2013; Velisetty and Chakrapani , 2012; Dellisanti et al . , 2013 ) . A cluster of Trp residues , strengthening M4 interactions with M1/M3 , was found to critically contribute to this robustness of GLIC function in various lipid environments ( Hénault et al . , 2015; Carswell et al . , 2015 ) . Our data show similar pH-elicited movements in detergent and liposomes , suggesting that this robustness also applies to detergent-solubilized GLIC , at least at the level of the positions herein investigated . We analyzed the time-course of the conformational motions using a stopped-flow apparatus . Bimane-labeled GLIC mutants reconstituted in liposomes at pH 8 were either kept at pH 8 or mixed with acidic solutions buffering the proton concentration to final pH values of 6 , 5 and 4 , and followed by fluorescence . In all cases , the signal recorded 30 s after mixing was in the same range as in steady-state conditions ( Table 4 ) . 10 . 7554/eLife . 23955 . 018Table 4 . Comparison of stopped-flow and steady-state fluorescence variations . The two columns show the maximal fluorescence variation recorded in the steady-state conditions or after 30 s of stopped-flow recordings . ( + ) indicates an increase of fluorescence and ( − ) indicates a decrease . n represents the number of experiments . For all the data , mean values are presented and error values represent the standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 018MutantpHMaximal ∆F steady-state ( % ) nMaximal ∆F stopped-flow ( % ) nBimane-133-W103pH 6 ( − ) 16 ± 13 ( − ) 11 ± 34pH 5 ( − ) 41 ± 13 ( − ) 28 ± 93pH 4 ( − ) 50 ± 13 ( − ) 38 ± 54Bimane-135-W72pH 6 ( − ) 25 ± 34 ( − ) 20 ± 76pH 5 ( − ) 43 ± 24 ( − ) 29 . 9 ± 0 . 96pH 4 ( − ) 29 ± 34 ( − ) 20 ± 34Bimane-136-W101pH 6 ( − ) 20 ± 43 ( − ) 12 ± 34pH 5 ( − ) 49 ± 33 ( − ) 43 ± 44pH 4 ( − ) 53 ± 25 ( − ) 45 ± 44Bimane-33-W160pH 6 ( − ) 31 ± 46 ( − ) 28 ± 53pH 5 ( − ) 61 ± 36 ( − ) 55 ± 44pH 4 ( − ) 64 ± 26 ( − ) 56 ± 33Bimane-250-Y197pH 6 ( + ) 3 ± 74 ( + ) 15 ± 84pH 5 ( + ) 46 ± 94 ( + ) 60 ± 204pH 4 ( + ) 50 ± 64 ( + ) 70 ± 404Bimane-243pH 6 ( − ) 2 ± 26 ( − ) 13 ± 104pH 5 ( − ) 7 ± 26 ( − ) 19 ± 94pH 4 ( − ) 17 . 0 ± 0 . 86 ( − ) 18 ± 94 For most positions , the changes in fluorescence ( as compared to the fluorescence at pH 8 ) are found to mainly occur in the dead-time of the instrument , during the first 2 ms of solution mixing ( Figure 6A ) . While this very fast component is not resolved here , completion of the fluorescence variations in 2 ms indicates an upper value of the time constant below 1 ms . The subsequent variations of fluorescence were subjected to multi-exponential fits , excluding the first 3 ms of recording that show high variability ( see Materials and methods and Figure 6—figure supplement 1 ) . The binning analysis of all multi-exponentials allows the separation of kinetic values in three phases with time constants ranging from 5–24 ms ( fast ) , to 24–966 ms ( intermediate ) and 966 ms–30 s ( slow ) ( Figure 6B and Table 5 ) . At positions 133 , 136 , 33 and 250 , the dead-time 2 ms ( ‘very fast’ ) component , evaluated at the start of the multi-exponential fit at 5 ms , accounts for the majority of the fluorescence variation , especially at pH 4 where 72% to 90% of the fluorescence changes are completed ( Figure 6C and Table 5 ) . Therefore , the motion of positions 136–101 , 133–103 and 33–160 moving closer together , and the separation of 250–197 occur with very fast kinetics . 10 . 7554/eLife . 23955 . 019Figure 6 . Stopped-flow fluorescence measurements . ( A ) Each panel shows representative traces of the major quenching mutants with a zoom on the first 500 ms ( left ) and the entire recording ( right ) . All traces are fitted to a multi-exponential ( see Methods ) represented in red . The color code is identical in each panel with a blue gradient starting from pH 6 ( dark ) to pH 4 ( light ) . The red dash line represents the starting fluorescence at pH 8 after normalization , for the raw pH 8 trace see Figure 6—figure supplement 1 . ( B ) Binning of τ values extracted from the stopped-flow multi-exponential fit ( all pHs and all constructs were used for the plot ) . The log time binning interval is of 0 . 4 and the resulting bins are shown by red dashed lines on the plot . The distribution shows three major clusters for which the according τ value is indicated on the plot . ( C ) Fluorescence variations ( calculated as a % of the overall variation at 30 s ) for each time interval extracted from the binning plot . In this graph , mean values are presented and error bars are calculated as standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 01910 . 7554/eLife . 23955 . 020Figure 6—figure supplement 1 . Residuals of the stopped-flow data fits and pH 8 recordings . Residuals plots: The figure shows an example for each mutant and pH recorded , chosen at random . The time axis ( x ) is represented in log scale for better visualization of the first milliseconds . pH 8 recordings: The figure shows non-treated , raw stopped-flow recordings of GLIC mutants at pH 8 , chosen at random . The axis labels are the same for all the residuals plots and pH 8 raw data , respectively , and were represented only once for clarity . A . U . stands for arbitrary unit . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 02010 . 7554/eLife . 23955 . 021Table 5 . Kinetic and fluorescence parameters of real-time measurements . All the values presented in the table were extracted from multi-exponential fits of the real-time measurements ( see Materials and methods ) . The relative ∆F for each exponential fit were calculated using the maximal F variation of each curve . In the case fluorescence variations were ‘bi-directional’ , the respective amplitudes of each phases were added to determine the maximal F variation value . ( + ) indicates an increase of fluorescence and ( − ) indicates a decrease . NA stands for non-applicable and was used when a single or double exponentials were sufficient to fit the data . For Bimane-133-W103 at pH 6 , no fluorescence variations were measured after the 5 ms non-exploited data; hence no exponential fits were made for this particular condition . n represents the number of experiments . For all the data , mean values are presented and error values are calculated as standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 021Dead-time ( <5 ms ) Exponential 1Exponential 2Exponential 3MutantpHRelative ∆F ( % ) τ1 ( ms ) Relative ∆F ( % ) τ2 ( ms ) Relative ∆F ( % ) τ3 ( ms ) Relative ∆F ( % ) nBimane-133-W103pH 6 ( − ) 94 ± 6NANANANANANA4pH 5 ( − ) 72 ± 110 ± 4 ( − ) 11 . 6 ± 0 . 7290 ± 40 ( + ) 11 ± 521 , 000 ± 6000 ( − ) 5 ± 43pH 4 ( − ) 71 ± 310 ± 7 ( − ) 14 ± 2600 ± 400 ( + ) 6 ± 126 , 000 ± 18 , 000 ( − ) 8 ± 34Bimane-135-W72pH 6 ( − ) 53 ± 950 ± 40 ( − ) 12 ± 4700 ± 400 ( − ) 17 ± 63000 ± 1000 ( − ) 19 ± 46pH 5 ( − ) 50 ± 20100 ± 100 ( − ) 8 ± 51000 ± 1000 ( − ) 19 ± 88000 ± 4000 ( − ) 21 ± 66pH 4 ( − ) 82 ± 6700 ± 400 ( − ) 8 ± 230 , 000 ± 40 , 000 ( − ) 9 ± 6NANA4Bimane-136-W101pH 6 ( − ) 40 ± 309 ± 4 ( − ) 30 ± 10100 ± 10 ( − ) 10 ± 102900 ± 600 ( − ) 14 ± 74pH 5 ( − ) 60 ± 109 ± 3 ( − ) 30 ± 1040 ± 20 ( − ) 5 ± 218 , 500 ± 900 ( − ) 7 ± 24pH 4 ( − ) 86 ± 211 ± 3 ( − ) 12 ± 14000 ± 5000 ( − ) 1 . 6 ± 0 . 9NANA4Bimane-33-W160pH 6 ( − ) 73 ± 99 ± 5 ( − ) 19 ± 9100 ± 100 ( − ) 6 . 1 ± 0 . 73000 ± 4000 ( − ) 3 . 2 ± 0 . 63pH 5 ( − ) 88 . 3 ± 0 . 9100 ± 100 ( − ) 5 ± 21000 ± 1000 ( − ) 2 . 9 ± 0 . 416 , 000 ± 9000 ( − ) 4 ± 24pH 4 ( − ) 90 ± 2100 ± 100 ( − ) 2 . 8 ± 0 . 42000 ± 2000 ( − ) 3 . 6 ± 0 . 810 , 000 ± 2000 ( − ) 4 . 1 ± 0 . 53Bimane-250-Y197pH 6 ( + ) 90 ± 10300 ± 500 ( + ) 10 ± 10NANANANA4pH 5 ( + ) 86 ± 5200 ± 100 ( + ) 5 ± 29000 ± 3000 ( + ) 9 ± 4NANA4pH 4 ( + ) 70 ± 20100 ± 100 ( + ) 6 ± 56000 ± 5000 ( + ) 20 ± 10NANA4Bimane-243pH 6 ( − ) 80 ± 208000 ± 10 , 000 ( − ) 8 ± 712 , 000 ± 3000 ( − ) 21 ± 7NANA4pH 5 ( − ) 60 ± 209 ± 7 ( − ) 11 ± 4270 ± 40 ( + ) 22 ± 816 , 000 ± 6000 ( − ) 12 ± 74pH 4 ( − ) 40 ± 205 ± 1 ( − ) 32 ± 9150 ± 20 ( + ) 19 ± 714 , 000 ± 7’000 ( + ) 9 ± 54 Interestingly , Bimane-243 shows a distinctive kinetic pattern , as its very fast component accounts for only 40% of the fluorescence variation . This is followed by fast and intermediate components accounting for 30% and 20% of the fluorescence variation , respectively ( Figure 6C and Table 5 ) . Although the fluorescence changes at position 243 could not be assigned to particular conformational motions , these data suggest that the upper part of the channel moves over a broad time scale . Finally , at all positions , slow components of fluorescence variations were recorded with time constants ranging from 1 to 30 s , particularly at positions V135 at the top of the ECD , P250 on the M2-M3 loop , and E243 on the top of the M2 α-helix ( Figure 6C ) . As the slow components at all positions ( except E243 ) show fluorescence variations in the same direction as the very fast component , it may suggest that the whole protein follows the previously described motions under prolonged applications of protons . We next investigated how the kinetics of the conformational changes measured above compare with kinetics of activation/desensitization of GLIC . For this , we recorded channel activity from purified GLIC reconstituted in asolectin liposomes , using a fluorescence-based sequential-mixing stopped-flow assay ( Rusinova et al . , 2014; McCoy et al . , 2014; Posson et al . , 2015 ) . GLIC Cys-less and mutant Bimane-136-W101 were reconstituted in liposomes containing the water-soluble fluorophore 8-aminonaphthalene-1 , 3 , 6-trisulfonic acid ( ANTS ) and mixed sequentially , first with protons to activate the channels ( step 1 ) , and then with the channel-permeable ANTS-fluorescence-quencher thallium to assess the flux through open channels ( step 2 ) . By including a variable delay time ( 10–200 ms ) between the two steps , we are able to capture the channels in various levels of activation and/or desensitization and thus populate different functional states . Upon thallium influx into the liposomes via active GLIC the ANTS fluorescence gets quenched , and the quenching rate is proportional to channel activity . Due to intrinsic variations in liposome sizes ( the mean diameter of vesicles is ~ 150 nm [Ingólfsson and Andersen , 2010] ) and different numbers of channels per liposome , the data were fit with a stretched exponential in order to obtain the average quenching rate at 2 ms , a measure of channel activity ( see Materials and methods [Ingólfsson and Andersen , 2010; Rusinova et al . , 2014] ) . Both GLIC Cys-less and mutant Bimane-136-W101 show fast fluorescence decay with a rate of ≈ 80 s−1 after incubation for 15 ms at pH 4 . 5 and 4 . 2 ( Figure 7—figure supplement 1 ) . This indicates robust activation of GLIC under these acidic conditions . The quenching kinetics at both pH values are identical ( Figure 7B ) , suggesting that activation is already maximal at pH 4 . 5 . 10 . 7554/eLife . 23955 . 022Figure 7 . Comparison of GLIC motions and function . ( A ) Fits to the Hill equation of fluorescence values in asolectin liposomes ( black ) and rate values extracted from the thallium flux assay ( pink ) , both normalized to their respective maximum , for the Bimane-136-W101 mutant ( n = 3 ) . For comparison the ΔIf ( thallium flux assay ) curve of GLIC Cys-less is shown in grey ( n = 3 ) . ( B ) Stopped-flow recordings of the thallium fluxes quenching assay for the Bimane-136-W101 mutant . The pre-mix time was 15 ms for the empty and pH 5 . 2 liposomes and of 15 . 5 ms for the pH 4 . 5 and 4 . 2 recordings . ( C ) Fits to the Hill equation for electrophysiological currents ( red ) and fluorescence values in asolectin liposomes ( black ) for all major quenching pair mutants . The current and fluorescence are normalized to their respective maximum; pH 3 . 7 current values were excluded from the fits for mutants Bimane-250-Y197 and Bimane-33-W160 as they were systematically smaller than the maximal current . For all the data except panel B , mean values are presented and error bars are calculated as standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 02210 . 7554/eLife . 23955 . 023Figure 7—figure supplement 1 . Thallium flux assay . ( A ) The figure shows raw , non-treated , recordings of ANTS encapsulated in proteoliposomes carrying either the GLIC Cys-less ( left ) or the GLIC mutant Bimane-136-W101 ( right ) , at different pHs . The pre-mix times ( times for which the liposomes are in presence of proton before addition of thallium ) after which the fluorescence was recorded is indicated on the side of each trace . The black traces show the recordings of ANTS-containing liposomes without any receptors . ( B ) The figure shows plots of ANTS quenching by thallium rates as a function of the pre-mix times , for GLIC Cys-less ( left ) and GLIC Bimane-136-W101 ( right ) . The purple trace shows the rates measured at pH 4 . 2 in presence of the pore blocker picrotoxinin ( PTX ) at 400 µM . In these graphs , mean values are presented , the error bars represent the standard deviation and measurement were done with n = 4 to 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 023 In contrast , very slow kinetics ( ≈ 3 s−1 ) are observed after incubating the channels for 15 ms at pH 5 . 2 ( Figure 7B and Figure 7—figure supplement 1 ) which indicates that these conditions lead to little channel activation . According to these data the pH50 for GLIC Cys-less and GLIC Bimane-136-W101 is estimated at pH 4 . 6 and pH 4 . 7 , respectively ( Figure 7A ) . Increasing the pre-mix time did not increase the rates of fluorescence quenching at any of the pH values tested , suggesting that 15 ms are sufficient to reach the maximal activation of GLIC . On the contrary , prolonged incubation ( above 25 ms ) at low pH reduced the rates of quenching , possibly reflecting GLIC desensitization in asolectin liposomes ( Figure 7—figure supplement 1 ) . For the Bimane-136-W101 mutant in asolectin liposomes , the pH-fluorescence relationship ( ∆F curves ) , which reports on protein motions , and the pH-ion flux relationship ( ∆If curves ) , which reports on activity , are separated by more than one order of magnitude ( Figure 7A ) . Indeed , at pH 5 , the majority of the fluorescence changes are completed , whereas almost no receptors are yet active . This directly reveals an intermediate conformation of the protein , where the quaternary motion of positions 136 and 101 moving closer together has occurred , but where the channel is still closed . Interestingly , a similar scenario is seen at other positions , for which activation curves in oocytes ( ∆I curves ) are significantly shifted to higher proton concentrations as compared to ∆F curves , especially at positions 135 and 250 ( Figure 7C ) . Since ∆I curves are shifted to the left as compared to ∆If curves ( for GLIC Cys-less and Bimane-136-W101 ) , it is expected that ∆If and ∆F would be even more separated for the other mutants . Hence , our data suggest that the intermediate conformation involves global motions of the protein , notably the adjacent subunits moving closer at the level of the ECD ( positions 136–101 and 135–72 ) as well as the separation between positions 250 and 197 and thus the outward motion of the M2-M3 loop . We developed a series of allosteric sensors of the conformation of GLIC using the TrIQ/TyrIQ method . Upon pH drop , GLIC undergoes a cascade of allosteric transitions from the resting to the active and desensitized states . This method allows the investigation , in a time resolved manner , of the global evolution of this mixture of states depending on the proton concentration . While the quenching method follows the fate of mixed populations of receptors , it is informative to compare these ensemble conformational motions to those inferred from the static structures of GLIC solved by crystallography . GLIC was solved at a low proton concentration in a closed-channel conformation ( pH 7 ) ( Sauguet et al . , 2014 ) and at a high proton concentration in an apparently open-channel conformation ( pH 4 ) ( Bocquet et al . , 2009; Hilf and Dutzler , 2009 ) . Interestingly , measurement of the variation of Cβ-Cβ distances between the pH 7 and pH 4 structures parallel the motions inferred from fluorescence measurements at positions 136–101 , 133–103 and 33–160 in the ECD , that are reporting the subunits moving closer together , and positions 250–197 that follow the separation of the M2-M3 loop from the top of M1 ( Supplementary file 1B ) . It is thus likely that the molecular reorganizations inferred from the comparison of the two X-ray structures are contributing to the ensemble motions followed by fluorescence . However , at other positions , namely at pairs 243–238 ( top of the TMD ) and 136–178 ( Loop C ) , no correlation between the structures and the population ensembles can be observed . This suggests that , at these levels , the receptors might undergo different reorganizations when GLIC is outside of the crystal , and/or that additional conformational states are contributing to the overall fluorescence signal . A key finding of the work is the identification of an intermediate state , where the ECD compaction and the outward movement of the M2-M3 loop , monitored by bimane quenching , are not concerted with channel opening . This is directly demonstrated with the 136/101 pair , that reports nearly complete ECD compaction in steady-state conditions , at pH 5 , when no ion fluxes are yet recorded ( Figure 7A ) . We also show that the isomerization toward this intermediate state is very fast , since 60% of the transition is achieved in less than 2 ms at pH 5 ( Table 5 ) . This suggests that the isomerization toward this intermediate state precedes or is at least concomitant with activation . Indeed , previous studies identify GLIC as a slow activating channel within the pLGIC family . For GLIC reconstituted in asolectin liposomes recorded in the inside-out configuration , the activation time constant ( τ ) is in the 10 ms range upon activation by a very high proton concentration ( pH 2 . 5 ) ( Velisetty and Chakrapani , 2012 ) . Likewise , in HEK cells , out-side-out patch clamp recordings under fast perfusion , at pH 4 , reveal activation time constants ranging from 30 to 150 ms ( Laha et al . , 2013 ) . While the activation kinetics of GLIC in liposomes could not be resolved in the present study , it is likely that the very fast transition toward the intermediate state is part of the activation mechanism . Altogether , we propose a kinetic scheme where a global ‘pre-activation’ step first occurs in under 2 ms , involving the whole ECD compaction and the outward motion of the M2-M3 loop , followed by a slower and more localized transition for channel opening ( Figure 8 ) . This latter motion may plausibly be related to the fluorescence changes observed in the 5–966 ms ranges , that are particularly marked for the 243 reporter which is located near the channel gate . 10 . 7554/eLife . 23955 . 024Figure 8 . Conformational motions summary . ( A ) Global and zoomed in view of the GLICpH 7 structure , representing only two subunits , with the left structures viewed from the outside of the pentamer and the right structures from the inside . All major quenching pairs are represented by a sphere , colored for the mutation to cysteine and bimane labeling , and black for the corresponding quencher . All pairs are captioned in the same color and joined by a red line . ( B ) Hypothetic kinetic scheme for the GLIC structural reorganizations . The large arrows represent the major recorded fluorescence variations and the small arrows represent minor variations . DOI: http://dx . doi . org/10 . 7554/eLife . 23955 . 024 Several mutants and a wild-type GLIC C-terminally tagged with 10 histidines , were found to crystallize in different locally-closed ‘LC’ conformations characterized by an overall conformation of the ECD nearly identical to GLICpH 4 , but with a closed channel ( Prevost et al . , 2012; Gonzalez-Gutierrez et al . , 2013 ) . Among these , the structures of GLIC bearing a disulphide bridge C33-C245 , as well as the single mutant GLIC E243P , show a particularity with the M2-M3 loop revolved outwards , identically as for GLICpH 4 , but still with a closed ion channel . Hence , we may speculate that this particular LC conformation could be a candidate for the pre-active state of GLIC , since it shows compaction of the ECD and motion of the M2-M3 loop without channel opening . We also observe molecular reorganizations occurring on longer time scale . First , quenching experiments show small fluorescence changes in the 966 ms–30 s range for all positions . Second , ion flux experiments show a decrease in channel activity when GLIC is pre-incubated with protons for a more than 100 ms . These events could thus be linked to receptor desensitization . Indeed , in HEK cells , out-side-out patch clamp recordings show desensitization to be biphasic , with a fast component observed on half of the patches ( τ around 200 ms ) , followed by a slow component observed in all patches ( τ around 10 s ) ( Laha et al . , 2013 ) . Interestingly , previous analysis of the desensitization mechanism of GLIC by EPR ( Velisetty and Chakrapani , 2012 ) , and of other pLGICs by mutagenesis ( Gielen et al . , 2015 ) and NMR ( Kinde et al . , 2015 ) , support the view that desensitization proceeds through a narrowing of the lower part of the channel . Additionally , recently published X-ray structures of a GABAA receptor ( Miller et al . , 2014 ) and of the α4β2 nAChR ( Morales-Perez et al . , 2016 ) were proposed to correspond to a desensitized conformation and show a channel constriction in the bottom part of the ECD . Therefore , desensitization might occur through reorganizations in the TMD , with comparatively smaller motion at the ECD and particularly at the ECD-TMD interface . In conclusion , the monitoring of electrophysiologically silent states by the TrIQ/TyrIQ method allows us to structurally describe an intermediate state and propose a pre-activation mechanism . Interestingly , available structures of eukaryotic pLGICs , namely the GluClαR ( Hibbs and Gouaux , 2011; Althoff et al . , 2014 ) and the α1GlyR ( Du et al . , 2015 ) , also suggest a quaternary compaction of the ECD upon agonist binding , as well as outward motions of the M2-M3 loop upon channel opening that are similar to that observed on GLIC . In addition , ϕ-value analysis following extensive mutational analysis of the muscle nAChR further suggest that the top of the ECD around the orthosteric pocket , as well as the M2-M3 loop move early in the course of the gating transition toward the active state ( Purohit et al . , 2013 ) . The present study thus not only sheds light on the gating mechanism of the pLGIC model GLIC , but also provides a structural template to investigate the gating of eukaryotic and mammalian receptors . Buffer A consists of 20 mM Tris and 300 mM NaCl , adjusted to pH 7 . 4 unless otherwise stated . Buffer B consists of 300 mM NaCl , 2 . 7 mM KCl , 5 . 3 mM Na2HPO4 and 1 . 5 mM KH2PO4 , adjusted to pH 8 unless otherwise stated . Both buffers were supplemented with 0 . 02% DDM ( Anatrace , Maumee , OH ) when specified . Buffer C consists of 15 mM Na2HPO4 , 150 mM NaNO3 adjusted to pH 7 . Pre-mix buffer consists of 10 mM Na2HPO4 , 140 mM NaNO3 adjusted to pH 7 unless otherwise stated . Quenching buffer consists of 10 mM Na2HPO4 , 90 mM TlNO3 , 50 mM NaNO3 adjusted to pH 7 unless otherwise stated . MBS buffer consists of 88 mM NaCl , 1 mM KCl , 2 . 5 mM NaHCO3 , 5 mM HEPES , 0 . 7 mM CaCl2 and 1 mM MgSO4 . MES buffer consists of 100 mM NaCl , 3 mM KCl , 1 mM CaCl2 , 1 mM MgCl2 and 10 mM MES . CHO labeling buffer consists of 150 mM NaCl , 8 . 1 mM Na2HPO4 , 1 . 9 mM NaH2PO4 , 0 . 1 mM CaCl2 and 1 mM MgCl2 , adjusted to pH 7 . 4 . CHO conservation buffer consists of 160 mM NaCl , 4 . 5 mM KCl , 2 mM CaCl2 , 1 mM MgCl2 , 10 mM HEPES and 8 mM glucose , adjusted to pH 7 . 4 . Monobromo bimane ( ThermoFisher Scientific , Pittsburgh , PA ) was dissolved in 100% DMSO at 10 mM and stored at −20° C . Bimane Bunte salt was dissolved at 50 mM in water and stored at −80° C . Unless otherwise stated , all chemicals were purchased from Sigma Aldrich ( St Louis , MO ) . All GLIC constructs were generated using the molecular probe mutagenesis kit ( ThermoFisher Scientific ) on the Cys-less background ( C27S ) in two vectors previously described ( Bocquet et al . , 2007 ) , constructed as follows: MBP-GLIC was produced as previously described ( Bocquet et al . , 2009 ) , in BL21 E . Coli strains . Briefly , MBP-GLIC was produced at 20° C in BL21 bacteria after IPTG ( 20 mM ) induction . After cell disruption by sonication , membranes were separated through ultracentrifugation ( 40 , 000 rpm ) and MBP-GLIC extracted in buffer A 2% DDM overnight . MBP-GLIC was purified on an amylose resin and eluted by maltose addition in buffer A 0 . 02% DDM . Proteins were run on a gel filtration ( superpose 6 10/300 GL ( GE Healthcare , Chicago , IL ) ) column for removal of remaining maltoporin contaminants . The MBP was subsequently cleaved from GLIC by addition of thrombine ( Merck Millipore , Billerica , MA ) overnight , and GLIC purified again through gel filtration in buffer A 0 . 02% DDM . For receptors with the K33C and E243C mutations , the protein was incubated for 1 hr with 10 mM DTT prior to the gel filtration to reduce potential disulfide bridges . Pentameric GLIC was then incubated with mBBr at a five molar excess ratio ( monomer 1:5 mBBr ) while ensuring the final DMSO concentration did not exceed 1% , overnight , under agitation , at 4°C . Excess fluorophore was removed by gel filtration and GLIC-labeled samples flash frozen for storage at −80°C . Powdered asolectin extracted from soybean ( Sigma Aldrich ) was solubilized in buffer B at a concentration of 10 mg/mL using a potter , and either used fresh or aliquoted and frozen at −20°C for later uses . GLIC was reconstituted in liposomes with a 1:5 GLIC/asolectin w:w ratio . For 200 µg of protein reconstitution , the following procedure was used and all the steps performed at room temperature: 1 mg of asolectin in buffer B was mixed with DDM to reach a final DDM concentration of 0 . 7% and the solution equilibrated for 40 min to solubilize the pre-formed multilayered liposomes . The 200 µg of protein in buffer A 0 . 02% DDM was added to the solution and the overall volume brought to 1 mL to lower the DDM concentration to 0 . 2% , followed by equilibration for 1 . 5 hr under gentle agitation . For liposome formation and gentle inclusion of the protein , detergent was removed by incremented step addition of SM2 Bio-Beads ( 50 mg , 150 mg and 300 mg ( Bio-Rad , Hercules , CA ) ) pre-activated using methanol . Bio-beads were removed through light centrifugation and the proteoliposomes were either used fresh or stored at 4°C for a maximum of 4 days . All fluorescence recordings were done on a Jasco 8200 fluorometer ( MD , USA ) . Buffers and proteins were equilibrated at room temperature before each recording which were made at 20°C . mBBr and Bimane-GLIC samples were excited at 385 nm and their emission spectra were recorded from 420 to 530 nm through 2 . 5 nm slits at excitation and emission . Scan speed , sampling and PMT values were kept constant for all measures for subsequent comparisons . The membrane receptor GLIC being activated by protons ( pH50 = 5 . 3 ) ( Bocquet et al . , 2007 ) , we confirmed that both the mBBr fluorescence and its quenching properties are identical in proton concentrations ranging from pH 9 to pH 2 ( Figure 3—figure supplement 1A , B ) . Control experiment with the Cys-less GLIC treated with mBBr show negligible non-specific fluorescence ( Figure 3—figure supplement 1C ) . All measurements were made on 1 mL of protein sample in disposable UV transparent 2 . 5 mL cuvettes ( Sigma ) . As a consequence to the small volumes used , it was not possible to precisely acidify the protein sample using concentrated acid , or diluted acid without considerably changing the final concentrations of ions and protein . Hence , measurements were made after mixing one volume of protein in buffer A 0 . 02% DDM , with one volume of buffer B 0 . 02% DDM previously acidified by 1 M HCl to reach , after mixing , the desired pH value . The time necessary for the sample mixing and recording of fluorescence was evaluated to be approximately 30 s; this time is sufficient to reach steady-state conditions as shown by the plateau of fluorescence variations recorded on the stopped-flow after 20 s ( Figure 6A ) . The same mixing protocol was used for the protein samples reconstituted in asolectin . For every recording , the fluorescence emission spectra were stable over multiple recordings , not showing signs of fluorophore bleaching . For each pH tested , tryptophan emission spectra were recorded ( excitation at 280 nm ) to ensure the proteins did not suffer denaturation and all fluorescence changes were reversible upon return to pH 7 ( Figure 3—figure supplement 2 ) . For the propofol recordings , the bimane-labeled mutants were recorded , followed by addition of propofol to a final concentration of 100 µM , and re-recording of the sample . In these experiments , the tryptophan fluorescence was not followed as the propofol produces a strong contamination signal at these excitation and emission wavelengths . Recordings were made on a SFM-300 stopped-flow apparatus ( Bio-Logic , Seyssinet-Pariset , France ) . To ensure lamp and temperature stability , the apparatus was turned on and equilibrated at 20°C by temperature-controlled circulating water for at least 1 hr before recordings . Excitation was set at 385 nm through an 8 nm slit and emission fluorescence recorded through a 420 nm high-pass filter . A two-syringe injection system was used where syringe one was loaded with the protein sample in buffer B at pH 8 , and syringe two loaded with buffer B equilibrated at specified pHs . The injection volume was 150 µL for each syringe at an injection speed of 8 . 7 mL/s ( total speed of 17 . 4 mL/s ) . The filling of the FC15 recording flow cell ( 0 . 15 × 0 . 15 cm , 35 µL total ) was thus achieved after a theoretical 2 . 1 ms mixing dead-time ( manufacturer’s information ) . To follow with good time resolution the first fast events , 30 s long recordings were made through three sampling times: the first 500 ms were recorded with a 100 µs sampling time , followed by 0 . 5 to 1 . 5 s with a 1 ms sampling time , and 1 . 5 to 30 s with a 50 ms sampling time . The first 3 ms of recordings were excluded from the analysis because of non-reproducibility , yielding a 5 . 1 ms total ‘non-analyzed’ data ( 3 ms + 2 . 1 ms dead-time ) . For each condition , a mean of 10 recordings was counted as n = 1 and all conditions were recorded with n = 3 or 4 . Each data set at pH 6 , pH 5 and pH 4 was normalized to the integral 30 s recording made at pH 8 on the same day and same batch of proteoliposomes , to correct potential unspecific drifting/bleaching of fluorescence during the recordings . Each mean of 10 was analyzed individually using Datagraph ( Visual data tools ) and fitted over the total 30 s of recording to a maximum three exponentials equation:y ( x ) =F+F1e−k1x+F2e−k2x+F3e−k3x where F represents the final fluorescence intensity , F1 , 2 , 3 represent the ∆fluorescence of a particular exponential phase and k1 , 2 , 3 the kinetic constant for each exponential phase in s−1 . Systematic background fluorescence recordings of buffers were done and subtracted from the data . For residuals of the fits , see Figure 6—figure supplement 1 . A sequential-mixing stopped flow spectrofluorimeter ( SX . 20 , Applied Photophysics ) was employed to assay GLIC channel activity by measuring the Tl+-induced fluorescence quenching of a liposome-encapsulated ANTS ( 8-Aminonaphthalene-1 , 3 , 6-Trisulfonic Acid , Disodium Salt , Life Technologies , NY ) fluorophore via Tl+ influx through channels as previously described ( Rusinova et al . , 2014; McCoy et al . , 2014; Posson et al . , 2015 ) . Functional recordings of GLIC were made on Xenopus oocytes provided by the Centre de Ressources Biologiques Xénopes–Rennes ( France ) . Electrophysiological recordings were made as previously described ( Duret et al . , 2011 ) after 48–96 hr of expression , with the difference that oocytes were clamped at −40 mV for recordings . For all mutants , with the exception of P250C-Y197F ( see below ) , currents were recorded on non-labeled oocytes , followed by BBs labeling and re-recording on the same oocytes . Independent mutants-expressing oocytes were also used for current recordings after labeling without prior non-labeled recordings . Both methods led to similar results and the pH50 and nH values are given for a mix of oocytes recorded with the different methods . The P250C-Y197F mutant showed a notable run-down after activation preventing several consecutive recordings of the same cell . In this case , currents after BBs labeling were recorded using the second method only . BBs labeling of oocytes was obtained after a 1 hr incubation at room temperature , under very gentle agitation , in MBS 1 mM BBs . Oocytes expressing mutants K33C or E243C were treated with 10 mM DTT 10 min prior to BBs labeling . After labeling , all oocytes were rinsed in MBS buffer and recorded within 1 hr post-labeling . Mutants leading to currents smaller than 500 nA at high proton concentrations ( pH 4 ) were categorized as non-functional . For all non-functional mutants , expression tests were performed through immunolabeling of oocytes ( Figure 2—figure supplement 5 ) . Electrophysiological recordings were analyzed using AxoGraph X and ClampFit ( Molecular Devices , Sunnyvale , CA ) . The Hill equation was used for the dose-response fits:y ( x ) =axnHxnH+EC50nH where a represents the maximal current value after normalization , nH represents the hill number and EC50 the proton concentration for which half of the maximal electrophysiological response is recorded . A double exponential equation was used to fit the activation currents of GLIC labeled with bimane:y=A1 ( 1−e−tτ1 ) +A2* ( 1−e−tτ2 ) +C where t is time , A1/A2 represent the current amplitude of the fast and slow activation phases respectively , C accounts for the current value at the end of the fit and τ1/τ2 represent the activation kinetics . A weighted activation constant was also calculated using the following equation:τw= ( ( A1A1+A2 ) τ1 ) + ( ( A2A1+A2 ) τ2 ) Immunolabeling on Xenopus oocytes expressing GLIC WT or mutants was performed as previously described ( Sauguet et al . , 2014 ) . Briefly , oocytes were co-injected in the nucleus with a mix of two separate pmt3 vectors containing either the cDNA of GLIC-HA ( 80 ng/µl ) or GFP ( 10 ng/µl ) . Control oocytes were injected with the GFP alone . After 72 hr of protein expression , GFP positive cells were fixed in 4% paraformaldehyde ( 4°C , O/N ) , blocked in PBS + 4% horse serum ( 30 min , RT [Sigma Aldrich] ) , and immunolabeled in PBS + 2% horse serum using a rabbit anti-HA primary antibody ( 1 . 5 hr ) and an anti-rabbit Cy5 coupled secondary antibody ( 1 hr , RT ( [ThermoFisher Scientific] ) . Oocytes were then re-fixed with 4% paraformaldehyde ( 4°C , O/N ) , placed into 3% low-melting agarose blocks , and subsequently sliced at 40 µm intervals . Slices of three different oocytes per constructs were analyzed using epi-fluorescence microscopy with constant exposure times . Chinese hamster ovary ( CHO-K1 CCL-16 from ATCC , USA ) cells were cultured in Ham’s F-12K Kaighn’s modification medium ( ThermoFisher Scientific ) supplemented with 10% fetal calf serum and containing penicillin ( 100 U/mL ) and streptomycin ( 100 µg/mL ) . As the CHO cells were solely used for the bimane membrane-permeability test , they were not tested for mycoplasma contamination . Cells were plated in 35 mm diameter polystyrene plates ( Corning , Corning , NY ) and transfected with GLIC cDNA and mCherry cDNA , as a transfection control , using a JetPRIME kit ( Polyplus Transfection , Illkirch , France ) according to the manufacturer’s instructions . Protein expression was allowed for 48–72 hr before cell labeling . Throughout the labeling process , cells and buffers were kept at 4°C to avoid endocytosis of the fluorophore . For labeling with BBs , cells were rinsed three times ( 5–10 min , gentle agitation , 4°C ) in labeling buffer and incubated for 1 hr with 1 mM BBs in labeling buffer . They were then rinsed two times with labeling buffer ( 5–10 min , gentle agitation , 4°C ) and kept in conservation buffer at room temperature during the imaging process . Each observation was made at least three times on different cell batches . Confocal laser scanning of fluorescence was performed using an Ultima scanning head ( Bruker Fluorescence Microscopy , Middleton , USA ) mounted on an Olympus BX61W1 microscope and equipped with a 60x ( 1 . 1 NA , Olympus Optical , Tokyo , Japan ) water immersion objective . Bimane and mCherry were excited at 405 nm with a laser power of 1 . 5–2 . 5 µW and at 561 nm with a laser power of 0 . 8–4 µW , respectively . Emitted fluorescence was collected through the same objective lens , and focused on a 150 or 100 μm pinhole ( ≈ 1 to 1 . 5 Airy unit ) , placed on a conjugate image plane ( the confocal pinhole ) . Fluorescent emission from Bimane was filtered with a 525/50 nm band pass filter and detected in gallium arsenide phosphide-based photocathode photomultiplier tube ( H7422P , Hamamatsu Photonics , Hamamatsu , Japan ) . mCherry fluorescence was filtered with a 605/70 band pass filter ( all filters were from Chroma , Taoyuan City , Taiwan ) and detected in a side-on multi-alkali PMT ( 3896 , Hamamatsu Photonics ) . The bimane labeled GLIC V135C was crystallized in the same conditions as WT GLIC ( Sauguet et al . , 2013 ) . The crystals were directly flash-frozen in liquid nitrogen prior to data collection . Data sets were collected on the PROXIMA1 beamline of the SOLEIL synchrotron , Gif-sur-Yvette , France . Reflections were integrated using XDS ( Kabsch , 2010 ) and further processed using programs from the CCP4 suite ( Winn et al . , 2011 ) . As expected , the crystals were isomorphous to the previously described crystal lattice of the open receptor and belonged to space group C121 ( unit-cell parameters: a = 182 . 034 Å , b = 134 . 075 Å , c = 159 . 945 Å , α = γ = 90 . 00° , β = 102 . 51° ) with one pentamer in the asymmetric unit ( see Supplementary file 1A ) . The phases were directly calculated by performing rigid-body refinement with REFMAC5 ( Murshudov et al . , 2011 ) using PDB entry 3EAM ( Bocquet et al . , 2009 ) as a starting model . The structure was then subjected to restrained refinement with REFMAC5 using NCS restraints . As the covalent link between the bimane and the cysteine side chain is non-standard , its description was defined in an additional library and incorporated in the pdb file . The resulting model was subsequently refined by BUSTER ( Blanc et al . , 2004 ) . The final structure was validated using the MolProbity web server ( Chen et al . , 2010 ) . The PDB accession code is 5IUX .	In the nervous system , proteins of the pLGIC family are found in the membrane that surrounds each neuron . These proteins have channels that can allow ions to pass through the membrane and are responsible for transmitting electrical signals from one neuron to the next . Small molecules called neurotransmitters interact with the pLGICs to open or close the ion channel . If the ability of the pLGIC channels to open is altered , it can lead to behavioral changes like addiction , or diseases such as schizophrenia or epilepsy . For a pLGIC channel to switch between the “open” and “closed” states , specific parts of the protein need to move in relation to each other . However , to study these transitions researchers have previously relied on comparing the three-dimensional structures of open and closed pLGICs extracted out of the cell membrane . Different techniques are needed to directly follow these movements within membranes . Bacteria also have proteins belonging to the pLGIC family , and Menny et al . have now investigated one such bacterial protein to understand how pLGICs open . First , a small fluorescent molecule that glows differently if the environment around it changes was attached to various parts of the bacterial channel . These fluorescent markers revealed how several parts of the protein move and they also made it possible to measure how quickly these movements take place . Some of these movements happen before the channel opens , suggesting that the activation of this pLGIC protein happens in stages and involves the protein adopting a temporary intermediate state . The next step will be to better understand the structure of the intermediate state , which could help us to understand how pLGICs work in the nervous systems of animals . In future this may aid the design of new drugs that can modify the activity of these channels in patients with neurological conditions or addictions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "neuroscience" ]	2017	Identification of a pre-active conformation of a pentameric channel receptor
Aerobic glycolysis or the Warburg Effect ( WE ) is characterized by the increased metabolism of glucose to lactate . It remains unknown what quantitative changes to the activity of metabolism are necessary and sufficient for this phenotype . We developed a computational model of glycolysis and an integrated analysis using metabolic control analysis ( MCA ) , metabolomics data , and statistical simulations . We identified and confirmed a novel mode of regulation specific to aerobic glycolysis where flux through GAPDH , the enzyme separating lower and upper glycolysis , is the rate-limiting step in the pathway and the levels of fructose ( 1 , 6 ) bisphosphate ( FBP ) , are predictive of the rate and control points in glycolysis . Strikingly , negative flux control was found and confirmed for several steps thought to be rate-limiting in glycolysis . Together , these findings enumerate the biochemical determinants of the WE and suggest strategies for identifying the contexts in which agents that target glycolysis might be most effective . Proliferating cells increase their glucose consumption and secrete lactate as opposed to completely oxidizing the glucose in the mitochondria ( Warburg et al . , 1927 ) and is known as aerobic glycolysis or the Warburg Effect . Currently , this altered metabolism is exploited for diagnostics and is subjected to multiple drug development efforts ( Koppenol et al . , 2011; Vander Heiden , 2011; Hamanaka and Chandel , 2012 ) . Numerous studies have identified genes such as KRAS , PIK3CA , and cMYC and microenvironments such as hypoxia and hypoglycemia that promote aerobic glycolysis but a complete understanding of the necessary and sufficient biochemical alterations associated with this phenotype is unknown . Furthermore successful translation for biomedical applications is limited by understanding the contexts in which therapies that target glycolysis might be effective . Computational modeling has a successful history in the study of metabolism ( Rapoport et al . , 1976; Fell , 1992; Schilling et al . , 1999; Cascante et al . , 2002 ) . Genome-scale stoichiometric models of metabolism have been developed to study the effects of drug targets in human metabolism and have had success in predicting the WE ( Molenaar et al . , 2009; Vazquez et al . , 2010; Folger et al . , 2011; Shlomi et al . , 2011 ) . However , a comprehensive quantitative understanding of the WE requires knowledge of enzyme activities and metabolic control . Therefore , we collected and integrated multiple forms of data into a modeling framework involving flux balances of glycolysis , detailed chemical kinetics based on reaction mechanisms and parameters measured , physico-chemical constraints from thermodynamics and mass conservation , metabolic control analysis , and Monte Carlo sampling of parameter space . We next use mass spectrometry and isotope tracing to probe concentrations and fluxes through the pathway and their responses to several perturbations . Together , we elaborate the determinants of aerobic glycolysis and identify and confirm novel points of regulation in glycolysis that have remained unidentified for over 50 years since the discovery of the pathway . We investigated the kinetics of the glycolytic pathway from glucose uptake to oxidation of pyruvate in the mitochondria or export of lactate out of the cytosol . We modeled each step of the pathway according to enzymatic mechanism and known modes of allosteric control resulting in a set of differential equations ( Figure 1A , ‘Materials and methods’ , Supplementary file 1 ) . While it is not possible to model every possible interaction explicitly , the aim is to capture enough of the pathway so that a large range of experimentally realized measurements can be obtained and relationships between variables can be observed . 10 . 7554/eLife . 03342 . 003Figure 1 . A quantitative model and statistical simulation method captures the diversity of metabolic states observed in tumor and proliferating cells . ( A ) Schematic of the glycolysis model with chemical reactions and allosteric points of regulation described . Abbreviations: GLC—glucose , G6P—glucose-6-phosphate , F6P—fructose-6-phosphate , FBP—fructose-1 , 6 , -bisphosphate , F26BP—fructose-2 , 6 , -bisphosphate , GAP—glcyceraldehyde-3-phosphate , DHAP—dihydroxyacetone phosphate , BPG—1 , 3 bisphosphoglycerate , 3PG—3-phosphoglycerate , 2PG—2-phosphoglycerate , PEP—phosphoenolpyruvate , PYR—pyruvate , SER—Serine , GLY—glycine , Lac—lactate , MAL—malate , ASP—aspartate , Pi—inorganic phosphate , CI—creatine , PCI—phosphophocreatine , GTR—glucose transporter , HK—hexokinase , PGI—phosphoglucoisomerase , PFK—phosphofructokinase , ALD—aldolase , TPI—triosephosphoisomerase , GAPDH—glyceraldehyde-phosphate dehydrogenase , PGK—phosphoglycerate kinase , PGM—phosphoglycerate mutase , ENO—enolase , PK—pyruvate kinase , LDH—lactate dehydrogenase , MCT—monocarboxylate transporter , PDH—pyruvate dehydrogenase , CK—creatine kinase . ( B ) Overview of the algorithm and simulation method . ( C ) Measured values of the NADH/NAD+ ratio across a population of MCF10A breast epithelial cells . Three values of glucose concentration are considered ( 0 . 5 mM blue , 5 . 5 mM green , and 25 mM red ) . ( D ) Calculated fluxes ( mM/hr ) for glycolysis rate ( Glycolysis ) are defined as the rate of glucose to pyruvate ( per molecule of pyruvate ) , pyruvate to lactate flux ( LDH ) , rate of oxygen consumption ( OxPhos ) , rate of NADH turnover ( NADH ) , and ATP turnover ( ATPase ) . ( E ) Calculated probability density function ( PDF ) of NAD+ concentrations . ( F ) Calculated probability density function ( PDF ) of NADH/NAD+ ratio . ( G ) Calculated probability density function ( PDF ) of ATP concentrations . ( H ) Calculated probability density function ( PDF ) of ATP/ADP ratio . ( I ) Box plots showing the distribution of concentrations computed from the simulation for each intermediate in glycolysis . DOI: http://dx . doi . org/10 . 7554/eLife . 03342 . 003 Since glycolysis is the most extensively studied biochemical pathway , there is a wealth of information on the kinetic parameters and enzyme expression that govern the equations . Nevertheless , it is also not possible to capture cellular physiology in any biochemical model with single values of kinetic parameters ( Daniels et al . , 2008 ) . This difficulty arises from the tremendous amount of heterogeneity within cells at multiple levels . The origins of this heterogeneity vary from genetic variation observed across cancer types , tumor types , differences in signaling mechanisms that affect post-translational modifications in each cell , and the differences in microenvironmental pressures ( e . g . , the oxygen availability ) that each cell within a given tumor experiences , as well as the inherent cell to cell variation common to all cells ( Marusyk et al . , 2012 ) . Therefore , we developed an integrated algorithm to evaluate the statistics of the kinetics of glycolysis that accounts for the possible variation within metabolism ( Figure 1B , ‘Materials and methods’ ) . First the model is constrained using mass conservation constraints that conserve the balance of glucose , redox state , and energy status . Next , thermodynamic constraints are used to constrain fluxes according to the free energy of the reactions determined by Haldane relationships . These physical constraints are combined with the kinetic mechanisms that define each step of glycolysis and the chemical reactions involving the redox-associated metabolites NAD+ , and NADH and energy-associated metabolites ATP , ADP , and AMP . Next , the model is constrained to expression data so that protein concentrations are subsumed in the Vmax values and chosen from typical concentrations in cancer cells , measured kinetic parameters , and measured concentrations of nutrients such as glucose , oxygen , and total intracellular adeno-nucleotide concentration . At this stage , the model is subjected to a thorough statistical analysis . A Monte Carlo simulation is conducted for which the parameters within the model are randomized and resulting differential equations are solved . The distributions for each parameter are chosen to capture observed ranges of variation . After each simulation , numerical stability and thermodynamics are assessed and all simulations that are unstable or appear thermodynamically infeasible ( i . e . , a positive net flux through glycolysis is required ) are rejected . In each simulation , concentrations , fluxes , metabolic control coefficients , and thermodynamic quantities are computed and recorded . This statistical analysis explores the space of glycolysis in the context of the Warburg Effect . By assessing the statistics , inferences can be made on the determinants of the Warburg Effects and its context dependence to pharmacological intervention and nutrient environment . For experimental confirmation of the model , we first utilized a recently developed a NADH/NAD+ fluorescent reporter ( Hung et al . , 2011 ) and measured the ratio of NADH/NAD+ across a population of cells where some experience hypoxia and others glucose deprivation ( Figure 1C ) . This variation in nutrient availability across individual cells occurs in epithelial cell cultures due to differences in diffusion and available cell surface area ( Sheta et al . , 2001 ) . The distribution is plotted for three media concentrations ranging from extreme hypoglycemia to the hyperglycemic conditions typically used in cell culture ( 0 . 5 mM , 5 . 5 mM , and 25 mM ) . Next we noted that the distribution of resulting fluxes ( Figure 1D ) was first found consistent with known measurements . An analysis of the simulation revealed a distribution of NAD+ levels and NADH/NAD+ redox potential consistent with those observed experimentally ( Figure 1E , F ) . ATP levels peaked around 3 mM with little variation ( Figure 1G ) and the ATP/ADP energy state is observed to be bimodal ( Figure 1H ) . Concentrations along the glycolytic pathway varied with means similar to those measured in normal tissues with fructose-1 , 6-bisphosphate ( FBP ) being most variable ( Figure 1I ) . Together , these findings indicate that these simulations capture the range of measured cellular concentrations and provide confidence for further assessment of the behavior within the model . Having developed a model of glycolysis and its regulation , we assessed the relationship of aerobic glycolysis to other characteristics of glycolysis . The distribution of values ( Figure 2A ) of the Warburg Effect ( the ratio of flux to lactate over that entering the mitochondria ) ranged from less than one ( primarily oxidative metabolism ) to over 95% of glucose being converted to lactate . The dynamic range over which the Warburg Effect was observed in our model allowed us to investigate to other variables in metabolism . We correlated the calculated value of the Warburg Effect with metabolite concentrations of intermediates in glycolysis . From an analysis of these correlations , a pattern within glycolysis emerges . The levels of intermediates in the beginning steps in glycolysis positively correlate with the Warburg Effect , the levels leading up to the oxidation of glyceraldehyde-3-phopshate ( GAP ) by GAPDH negatively correlate with the Warburg Effect , and those following GAPDH positively correlate with the Warburg Effect ( Figure 2B ) suggesting together that a bottleneck exists in the pathway that determines the extent of fermentation . An analysis of the correlation of enzyme expression with the Warburg Effect revealed that the enzyme expression for any single step within glycolysis did not completely correlate with the Warburg Effect . However , across glycolysis , GAPDH , the enzyme that carries out oxidative phosphorylation of glyceraldehyde-3-phosphate to yield NADH and 1 , 3-diphosphoglycerate most strongly correlated with aerobic glycolysis ( Figure 2C ) . Notably many of the correlations although significant , appear not very strong indicating that the expression of individual enzymes is not sufficient for induction of aerobic glycolysis . The model also captures many of the activities that are known to correlate with aerobic glycolysis . These include glucose transport , pyruvate kinase , and lactate dehydrogenase activities . Interestingly , the expression of enzymes such as hexokinase and phosphofructokinase were either uncorrelated or negatively correlated with the extent of aerobic glycolysis likely indicating their role in creating bottlenecks at other points along the pathway . Together , these findings identify enzyme expression patterns that determine the extent of the Warburg Effect . 10 . 7554/eLife . 03342 . 004Figure 2 . Evaluation of the statistics of the Warburg Effect and relationships to other variables in metabolism . ( A ) Probability density function ( PDF ) of the Warburg Effect ( WE ) defined as the ratio of flux through LDH to that of flux into the mitochondria . ( B ) Pearson correlations of intermediate metabolite levels in glycolysis with the extent of the Warburg Effect ( WE ) . ( C ) Pearson correlations of the expression levels of glycolytic enzymes with the extent of the Warburg Effect ( WE ) . ( D ) Pearson correlations of coupled metabolic parameters with the extent of the Warburg Effect ( WE ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03342 . 004 We next investigated the relationship between the Warburg Effect and other physiological variables including the NADH/NAD+ redox status , the energy state defined as the ATP/ADP ratio , lactate , oxygen levels , and phosphocreatine levels ( Figure 2D ) . Oxygen concentration , ATP , and phosphocreatine levels positively correlated with the Warburg Effect and NADH/NAD+ redox status and NADH levels in the cytosol negatively correlated with the Warburg Effect suggesting that positive and negative feedback inherent to the circuitry of glycolysis contributes to buffering the Warburg Effect . Together , these results identify multiple relationships between the extent of aerobic glycolysis and measurable variables in metabolism . After assessing how metabolic parameters and concentrations within glycolysis determine the flux to lactate , we next investigated how each node within glycolysis exerts its control on the Warburg Effect . Metabolic control analysis ( MCA ) provides a mathematical framework for evaluating the extent that a change in metabolic activity affects a given flux . In MCA , a pure rate-limiting step occurs when the flux control coefficient ( FCC ) ( Supplementary Information ) is one at that step and zero at all other steps . In most cases , values of FCCs are distributed with gradual values across a pathway . We carried out a metabolic control analysis using MCA within our statistical algorithm to investigate the influence of each node in glycolysis on lactate flux across the ensemble of statistical realizations of glycolysis ( Figure 3A ) . We first computed the distribution of FCCs for lactate production for each step in the pathway ( Figure 3B ) . It was found that for each step within glycolysis , the average control exerted was near zero . Steps early in glycolysis involving enzymes Hexokinase ( HK ) , Phosphoglucoisomerase ( PGI ) , and Phosphofructokinase ( PFK ) exhibit both positive and negative control , and GAPDH on average exhibits the most positive control on the flux through the pathway . In addition to steps within glycolysis , ATPase activity exerts the most influence over lactate production with oxygen consumption having less of an affect . Together , these results both confirm the long-standing hypothesis that ATPase activity is the most prominent rate-determining step in glycolysis but also suggest that GAPDH can often exert a large control on the flux to lactate . In addition , an analysis of the statistics indicates that depending on the context , any point along the pathway can exert large flux control on lactate production with some steps ( e . g . , pyruvate kinase ) less likely to exert control over the Warburg Effect than others . Although the finding that pyruvate kinase typically does not exert substantial control over glycolysis may appear surprising , this finding is consistent with studies that have observed only modest changes in glycolytic flux due to changes in pyruvate kinase activity ( Christofk et al . , 2008; Israelsen et al . , 2013 ) . 10 . 7554/eLife . 03342 . 005Figure 3 . Metabolic control analysis and its relationship to metabolic variables . ( A ) Schematic of workflow for global sensitivity analysis . After the model is constructed and feasible solutions obtained , each realization of glycolysis is subjected to metabolic control analysis ( MCA ) . The resulting analysis is then subject to a statistical evaluation . ( B ) ( left ) Box plots of flux control coefficient ( FCC ) for lactate production for each enzymatic step in glycolysis ( FCC = dlnJlac/dln Ei ) where Jlac is the rate of pyruvate conversion to lactate , and Ei is the ith enzyme in glycolysis for each step of glycolysis . ( right ) Box plots of flux control coefficient ( FCC ) for lactate production for Oxygen consumption ( OxPhos ) and ATP consumption ( ATP ) . ( C ) Pearson correlations between lactate FCC values for each step in glycolysis . Heat map is colored ranging from the minimum value ( green ) to the maximum value ( purple ) . ( D ) Pearson correlations between metabolite concentrations in glycolysis and lactate FCC values for each step in glycolysis . Heat map is colored ranging from the minimum value ( green ) to the maximum value ( purple ) . ( E ) Pearson correlations between metabolic parameters and lactate FCC values for each step in glycolysis . Heat map is colored ranging from the minimum value ( green ) to the maximum value ( purple ) . ( F ) Pearson correlations between ratios and lactate FCC values for each step in glycolysis . DOI: http://dx . doi . org/10 . 7554/eLife . 03342 . 005 To further investigate the contexts in which steps along glycolysis can be limiting , we correlated the FCCs with one another and carried out a hierarchical clustering that revealed modules of co-occurring flux control ( Figure 3C ) . It was found that enzymes residing in proximal regions of upper glycolysis tend to have positively correlated flux control . However , for lower glycolysis the flux control is uncorrelated suggesting that when a step is limiting in lower glycolysis , it can be more readily disrupted . This behavior is in contrast with that of upper glycolysis where limiting steps co-occur . Furthermore , an analysis of ATPase and Oxygen consumption activities reveals their control to be related to the control of enzymes clustered at different points in glycolysis ( Figure 3C ) indicating that flux control exerted by these ancillary fluxes are tied to different regions of glycolysis . Since these simulations yield for each realization of glycolysis , a set of concentrations , fluxes and flux control coefficients , relationships between flux control coefficients and measurable concentrations can be obtained . An analysis of these relationships using hierarchical clustering ( Figure 3D ) revealed a bi-modal relationship in glycolysis where increases in metabolites in upper glycolysis led to steps in lower glycolysis exerting flux control and increases in lower glycolysis resulted in enzymes in upper glycolysis exerting flux control . Further analysis of FCCs and metabolic parameters ( Figure 3E , F ) revealed additional relationships . Together , these findings yield a comprehensive map of the flux control in glycolysis and its connections to metabolic variables . Surprisingly , in several instances , negative flux control is observed for several enzymes that have been previously thought to be the rate-limiting steps . Negative flux control implies that inhibiting the enzyme would actually increase the rate of flux of glucose to lactate . Given the surprising findings on flux control in glycolysis , we sought to experimentally investigate these predictions . We therefore considered systematic perturbations to glycolysis in cancer cells . We first devised a combined flux profiling and metabolite profiling method using 13C isotope tracing from glucose ( Figure 4A ) . Cells were incubated with U–13C glucose and media were extracted at different time points and subjected to high resolution LC-MS analysis ( Liu et al . , 2014 ) . The linearity of the time course allows for an exact measurement of the flux from glucose to lactate . We next carried out direct measurements of glycolysis state and flux control by perturbing glycolysis acutely with pharmacological agents and subsequently measuring metabolite levels and the flux from glucose to lactate . Genetic manipulations such as RNA interference were not possible since the measurement of dose-dependent acute effects was necessary . In each case , the compounds considered have been reported to exhibit direct inhibition of the enzyme in question and to our knowledge do not directly inhibit other enzymes in glycolysis . Nevertheless , the general specificity of these compounds is not established and off target effects that exist are likely reduced by considering acute treatments . 10 . 7554/eLife . 03342 . 006Figure 4 . Experimental flux control coefficients . ( A ) Schematic of experimental flux control analysis . Cells are pre-incubated with 13C glucose and treated with differing concentrations of inhibitors that target glycolysis at different points in the pathway . Media and intracellular metabolites are collected , subjected to ( liquid chromatography high resolution mass spectrometry ) LC-HRMS , and subjected to flux analysis . ( B ) Changes in metabolite levels observed from treatment with 3PO an inhibitor of PFK2 . ( C ) Changes in metabolite levels observed from treatment with IA an inhibitor of GAPDH . ( D ) Changes in metabolite levels observed from treatment with FX11 and inhibitor of LDH . For B–D , the logarithm ( log2 ) of the fold change of treated to vehicle across intermediates in glycolysis is shown for each concentration of compound denoted in the figure legend . Abbreviations are the same as described in Figure 1 except that HP denotes all hexose phosphates that were measured and not distinguished in the current mass spectrometry method . ( E ) Lactate flux from glucose as a function PFK2 inhibition . ( F ) Lactate flux from glucose as a function GAPDH inhibition . ( G ) Lactate flux from glucose as a function LDH inhibition . For E–G , the plot on the left shows the measured glucose to lactate flux as a function of the estimated fraction of enzyme inhibited ( left ) inhibitor concentration ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03342 . 006 We considered inhibiting glycolysis at three separate points along the pathway that are predicted to have widely variable flux control of the pathway . In the beginning of glycolysis we used 3PO , a compound that targets PFK2 thus inhibiting the phosphofructokinase step ( Schoors et al . , 2014 ) . Next , we considered iodoacetate ( IA ) , a compound that targets GADPH ( Campbell-Burk et al . , 1987 ) . Finally FX11 , a compound that targets LDH was considered ( Granchi et al . , 2011 ) . In each case ( Figure 4B–D ) differential , complex , nonlinear responses of metabolite levels to inhibiting glycolysis were observed . The exception was treatment with IA ( Figure 4C ) which exhibited an expected accumulation of intermediates upstream of GAPDH and depletion of intermediates downstream of the target . We next measured directly the flux control coefficients for each compound . An analysis of the flux control ( Figure 4E–G ) revealed several interesting responses . Strikingly , as predicted by the model , a negative flux control was observed for targeting PFK implying that inhibiting this step in these circumstances increases glycolytic flux and Warburg Effect ( Figure 4E ) . The largest flux control was observed with inhibition of GAPDH ( Figure 4F ) . Inhibition of LDH as predicted also resulted in little flux control ( Figure 4G ) . Together , these findings confirm the mechanisms of flux control of glycolysis predicted by the model and demonstrate the novel regulation of glycolytic flux that can be differentially perturbed by pharmacological compounds . We have thus far observed experimentally the dynamic behavior of flux control in glycolysis including both positive and negative flux control in the pathway and a high variability of flux control depending on the point of inhibition in the pathway . Having established the complex relationships between glycolytic flux and susceptibility to specific targeted inhibition of the pathway , we sought to investigate whether there was any predictive capacity and new mechanisms of biochemical regulation related to these findings . We noticed that Fructose- ( 1 , 6 ) -bisphosphate ( FBP ) levels exhibited highly dynamic and counterintuitive behavior with each drug perturbation . An analysis of metabolite levels across a series of 14 conditions in triplicate involving pharmacological perturbations of PFK2 , LDH , and GAPDH at different concentrations and two separate vehicle treatments ( Figure 5A ) revealed large magnitude , dynamic responses in FBP levels . We next investigated the extent that FBP levels could characterize the metabolic state of glycolysis . The simulated PDF of FBP levels exhibited a bimodal distribution ( Figure 5B ) consisting of a state of low FBP where the concentration was in the high micromolar range . In addition , there existed a state of FBP where the concentration was several orders of magnitude high in the millimolar range . When demarcating the experiments into two groups ( high and low FBP ) we correlated the levels with lactate flux experimentally ( Figure 5C , E ) and found the results to match those observed computationally ( Figure 5D , F ) . In addition , correlations in these two states with the remainder of the concentrations of glycolytic intermediates agreed well with experiments . Together these findings suggest in addition to the model being able to capture a diverse set of experimental metabolic conditions , the FBP status in cells was able to determine the state of glycolysis and its rate-limiting steps . 10 . 7554/eLife . 03342 . 007Figure 5 . FBP levels predict distinct mechanisms in glycolysis . ( A ) Variation of metabolite levels across glycolysis over 14 conditions in triplicate resulting in 42 independent experiments involving cells growing in basal conditions and those with differing extents of inhibition of glycolysis from results in Figure 4 . ( B ) Simulated distribution of FBP levels in glycolysis . ( C ) Correlation of lactate flux with measured glycolytic intermediates for low FBP levels . The left panel shows data for FBP and right panel reports the values of the Spearman correlation coefficients for each metabolite . ( D ) Simulated correlation of lactate flux with metabolite levels of glycolytic intermediates in conditions of low FBP levels . ( E ) Correlation of lactate flux with measured glycolytic intermediates for high FBP levels . The left panel shows data for FBP and right panel reports the values of the Spearman correlation coefficients for each metabolite . ( F ) Simulated correlation of lactate flux with metabolite levels of glycolytic intermediates in conditions of low FBP levels . DOI: http://dx . doi . org/10 . 7554/eLife . 03342 . 007 Together , the combined computational model , metabolite profiling , and flux analysis points to a different picture of glycolysis ( Figure 6 ) . When the levels of FBP are low , metabolite levels of glycolytic intermediates tend to be more evenly distributed across the pathway . In this circumstance , flux through the pathway is controlled largely through the initial steps in glycolysis involving hexokinase and phosphofructokinase . Bottlenecks downstream of these canonical rate-limiting steps are not affecting flux through the pathway . Under these conditions , FBP is also not allosterically activating pyruvate kinase . In contrast , when the levels of FBP are high , there is a disconnect in the relative concentration of glycolytic intermediates that is marked by a separation between upper and lower glycolysis at the GAPDH step . In this case , there is an accumulation of intermediates in upper glycolysis , most notably FBP and a depletion of intermediates downstream of GAPDH . Under these circumstances , GAPDH exerts control as the most rate-determining step in the pathway since increased activity through GAPDH will serve then to create a balance along the pathway by pulling the metabolites from upper glycolysis into lower glycolysis . Under these conditions , the beginning steps of glycolysis can exert negative control on the pathway since inhibiting them will result in even greater increases in the intermediates in upper glycolysis . This state has analogies with a recently observed state of glycolysis observed in yeast where an accumulation of FBP leads to cellular toxicity ( van Heerden et al . , 2014 ) . Notably in this case , in order to balance the fluxes in higher and lower glycolysis , an imbalance in the concentrations of metabolites in upper and lower glycolysis results . 10 . 7554/eLife . 03342 . 008Figure 6 . A unified model of aerobic glycolysis . A unified picture of flux control in aerobic glycolysis . ( left ) Under conditions where there is an accumulation of intermediates in upper glycolysis and depletion of intermediates in lower glycolysis a bottleneck exists at the step involving GAPDH . This bottleneck is due to the status of energy and redox metabolism and the thermodynamics of the pathway that together mediate the flux through GAPDH . As a result , inhibiting flux through glycolysis is most sensitive to a perturbation in GAPDH activity . ( right ) Under conditions where the metabolites in glycolysis are distributed more evenly with levels together being either high or low , no such bottleneck exists . Instead flux through glycolysis leading to lactate production is most determined by the canonical pacemaking steps in glycolysis involving PFK and HK . The relative levels of glycolytic intermediates are denoted by the size of the text . DOI: http://dx . doi . org/10 . 7554/eLife . 03342 . 008 Together , our analysis yields a comprehensive , quantitative framework for understanding glycolysis and its regulation in the context of the Warburg Effect . Historically , glycolysis is thought to have a rate-limiting step at several points in the pathway ( Chance and Hess , 1956; Wu , 1965 ) . These points correspond to positions in the pathway where large free energy differences arise including the ATP-coupled enzymes Hexokinase , Phosphofructokinase , and Pyruvate Kinase ( Rose and Warms , 1966; Rapoport et al . , 1976; Hue and Rider , 1987 ) . Surprisingly , we identified a strong context-dependence with both positive and negative control in glycolysis at each of these steps . In the case of inhibiting flux through PFK , the observed negative control in certain conditions implies that inhibiting this point in the pathway can lead to increased rates of fermentation . This finding provides a possible explanation for why its efficacy may be more prevalent in stromal cells ( Schoors et al . , 2014 ) . Unexpectedly , GAPDH was found to be a recurrent rate-controlling step in aerobic glycolysis . This finding , first documented to our knowledge in parasitic bacteria feeding on high glucose ( Bakker et al . , 1999 ) , is in contrast to the longstanding notion that GAPDH is not a rate-determining enzyme in glycolysis with the activity of enzymes such as hexokinase , phosphofructokinase , and pyruvate kinase thought to be more controlling . In the case of GAPDH , the bottleneck occurs due to its unique placement in the pathway where it can be regulated by ATP , NAD+ , and the levels of glucose-derived intermediates in the pathway that affects the thermodynamics of glycolysis . There are multiple mechanisms that lead to this finding . ATP consumption has previously been reported to control the rate of glycolysis and this effect likely occurs to some extent through GAPDH ( Racker , 1976; Locasale and Cantley , 2011; Lunt and Vander Heiden , 2011 ) . NAD+ regeneration that is mediated by the malate–aspartate shuttle and lactate dehydrogenase also affects flux through glycolysis ( DeBerardinis et al . , 2008; Locasale and Cantley , 2011 ) . In addition , the activity of the pathway upstream and downstream of GAPDH changes the balance of the levels of intermediates in glycolysis and results in driving the thermodynamics of the reactions out of equilibrium also can result in greater flux control ( Noor et al . , 2014 ) . Each of these mechanisms separately or together acts to allow for GAPDH to exert flux control over the glycolytic pathway . Enzymes along glycolysis that are believed to control flux have many documented regulatory mechanisms . For example pyruvate kinase and phosphofructokinase have numerous small molecule effectors and post-translational modifications that affect their activities ( Mor et al . , 2011; Chaneton and Gottlieb , 2012; Keller et al . , 2012; Yi , et al . , 2012 ) . It is notable that GAPDH also is subject to multiple forms of regulation including post-translational modifications such as nitrosylation and reactive oxygen species ( ROS ) that interacts with the catalytic cysteine in GAPDH to inhibit its activity ( Gaupels et al . , 2011; Tristan et al . , 2011; Moellering and Cravatt , 2013 ) . In the context of ROS , it is tempting to speculate that alterations in ROS could lead to selective modulation of glycolytic flux as has been suggested to occur with pyruvate kinase ( Anastasiou et al . , 2011 ) . While ROS-mediated inhibition is unlikely in conditions of exponential growth , it may be more apparent in physiological conditions of hypoxia and glucose deprivation with higher concentrations of ROS . Another critical aspect of the flux control that GAPDH exerts over the glycolytic pathway is high expression of GAPDH in cells undergoing aerobic glycolysis . Indeed , reports of quantitative protein abundance in mammalian cells have identified enzymes in the pathway glycolysis as the most highly expressed collective of proteins in cells ( Moghaddas Gholami et al . , 2013 ) . Interestingly , it was found that within glycolysis , GAPDH is often the mostly highly concentrated protein in glycolysis suggesting that the role of this high expression in these cases is to support the increased amount of glycolytic flux in these cells . Nevertheless , although mechanisms that regulate phosphofructokinase and pyruvate kinase for example have been shown to mediate cell growth and proliferation , whether regulatory mechanisms of GAPDH have functional roles in cell growth and proliferation related to aerobic glycolysis is not known . Two states of glycolysis were observed with different extents of flux control , tendencies for aerobic glycolysis , and concentration patterns along glycolysis . Notably , the FBP levels have implications on the activity of pyruvate kinase that is also allosterically activated by FBP . At low FBP concentration , pyruvate kinase is not activated by FBP but this occurs only in the high FBP state . This finding could have implications in understanding the contexts in which pharmacologically activating pyruvate kinase may have efficacy . Furthermore , it remains to be seen if any signature of these metabolite states could manifest in the alterations of peripheral metabolism involving pathways whose fluxes emanate from glycolysis . If this were the case , then a tempting possibility would be that these states of glycolysis could be predicted from measurements of peripheral metabolites that could be excreted into circulation allowing for the possibility of developing serum biomarkers for the status of glycolysis in tumors beyond what can be resolved with positron emission tomography using radioactive glucose . Finally , the surge of interest in metabolism and its contribution to pathogenesis has created an expectation that therapeutics that target glucose metabolism will be clinically successful . Targeting glycolysis in metabolism has raised interest but is limited by the development of biomarkers that could determine the contexts in which targeting glucose metabolism in malignancy would be efficacious ( Vander Heiden , 2011; Galluzzi et al . , 2013; Vander Heiden , 2013 ) . From our model analysis , the consequences of inhibiting glycolysis appear enormously complex that limit biomarker development due to the nonlinear mechanisms that determine the response of the pathway to a drug perturbation . Nevertheless , with the development of these predictive models that can capture diverse behaviors of glycolysis , it is worth considering whether they may have predictive capacity in pre-clinical and clinical settings . The model includes a compartment involving enzymes in glycolysis and additional compartments involving with reactions that are coupled to glycolysis . The following compartments are considered in addition to the enzymes in glycolysis are considered:Glucose uptake through the glucose transporter , Lactate dehydrogenase ( LDH ) activity and lactate transport through the monocarboxylate transporter ( MCT ) , Oxygen consumption , transport and activity of oxidative phosphorylation ( OxPhos ) , ATP-buffering mechanisms involving adenylate Kinase , creatine Kinase , and ATPase activity , NADH/NAD+-mediated mitochondrial shuttles including the Malate–Aspartate Shuttle and the Glycerol-3-Phosphate Shuttle . More generally glycogen metabolism , pentose phosphate pathway and alanine biosynthesis could also be included but are omitted since the conclusions drawn in this study do not depend on their activities . A schematic representation of the kinetic model and resulting network is shown in Figure 2A . We first introduce notation and other conventions . Each symbol indicates the respective concentration . A subscript ‘0’ refers to the steady-state value . Reaction rates are expressed in millimoles per hour per unit intracellular volume . Together with initial state vectors C0∈RN , Vo∈RM , of metabolites and fluxes respectively . The dynamics of the system are therefore formulated as initial value problem for ordinary differential equations ( ODEs ) . Starting parameter values based on published data are defined in Supplementary file 1 . Reaction rates are defined through a consideration of the respective enzyme mechanisms with additional feed forward and feed back regulation giving rise to allosteric activation and inhibition . Therefore , the model is designed to describe: ( a ) steady-state and dynamic behavior of energy metabolism , including the Warburg effect ( W=JLacJOx ) , energy state ( ATP/ADP ratio ) , redox state ( NADH/NAD+ ratio ) together with glucose , lactate , and oxygen supply through exchange fluxes from the intercellular and extracellular compartment , ( b ) steady states and time courses of all variables of the metabolic network , and ( c ) effects of metabolic parameter perturbations on the overall system output . Cellular energy homeostasis through ATP is supported by several mechanisms with different relaxation times and regulation mechanisms: ( a ) directly by glycolysis that converts glucose into intracellular pyruvate , ( b ) mitochondrial respiration through consumption of pyruvate and oxygen via the TCA cycle , and ( c ) the buffering effect of creatine kinase that facilitates the reversible reaction of phosphocreatine ( PCr ) with ADP to produce creatine ( Cr ) and ATP , and ( d ) adenylate kinase activity that catalyzes interconversion of adenine nucleotides: 2ADP ⇔ ATP + AMP , and ( e ) the additional NADH pool produced in cytosol and transported to mitochondria by different shuttle mechanisms . For oxygen consumption , we assumed that mitochondrial respiration ( O2 consumption ) depends on cellular pyruvate and oxygen concentrations and that respiratory chain activity is activated by ADP concentration . The state of the system is represented by the state vector of time-dependent metabolite concentrations Cn , ( n = 1 , … , N ) and includes 27 state variables whose names , balance equations , and steady-state values are presented in Supplementary file 1 . The model also includes three mass conservations laws , that reduce the number of state variables . The temporal profile of the system is governed by the set of ODE based on the model network that distinguishes media and cellular domains . For the metabolite i in the intracellular compartment , the general form is: ( 1 ) dCidt=Jitr+∑jνipjVipj−∑kνiukViuk , where the intracellular concentration of species i is Ci; Jitr is the net transport flux for boundary species i between the media and cell; Vipj , Viuk are normalized reaction fluxes that produce ( j ) or utilize ( k ) cellular species i , νipj , and νiuk are corresponding stoichiometric coefficients . For extracellular boundary metabolites in the media , the dynamic mass balance has the following general form: ( 2 ) dCedt= rie⋅Jitr , where the concentration of extracellular ( e ) species is Ce and rie is the ratio of cell volume to media volume . To ensure that steady states are obtained during perturbations of parameters without loss of generality , we assume that the media boundary metabolites have constant concentrations . The transport of boundary species is taken to be either facilitated ( glucose , lactate , serine , glycine ) or passive ( oxygen ) . For passive diffusion of oxygen the equation for the net transport flux is: ( 3 ) JO2tr=kO2 ( CO2e−CO2i ) , where kO2 is the effective rate constant for passive O2 diffusion , and CeO2 and CiO2 are medium and intracellular oxygen concentrations . For facilitated transport , the equation for the net transport flux is: ( 4 ) Jitr= Jtrmax ( CeiKmitr−CiKmitr ) 1+CeiKmitr+CiKmitr , where Cei is the extracellular concentration for species i , Jtrmax is the maximal transport rate , and Kmitr is the Michaelis–Menten constant for transport . NADH transport from the cytosol to the mitochondria is mediated by the malate–aspartate and glycerol phosphate shuttles . We modeled the NAD+-mediated mitochondrial shuttle flux from NADH to NAD+ by assuming that the total shuttle flux is balanced with the NADH-generation reactions involving LDH and PHGDH . ( 5 ) Vmas=Vox+2Vphgdh When possible , in order to minimize the number of parameters , we utilized a ‘one-step binding enzyme mechanism’ with a rate law of the form ( Segel , 1975 ) : ( 6 ) V= ( Vfmax∏iCsiνiKmf−Vrmax∏jCpjνjKmr1+∏iCsiνiKmf+∏jCpjνjKmr ) χ ( α , ι ) , where χ ( α , ι ) is a control function that accounts for the effects of activation α , or inhibition ι . A general form for the control function is ( Liebermeister and Klipp , 2006 ) : ( 7 ) χ ( α , ι ) ={[A] ( [A]+KA ) in case of activation , α KI ( [I]+KI ) in case of inhibition , ι , where [A] and [I] are the concentrations of the activator or inhibitor , KA and KI are the corresponding activation and inhibition constants . With this convention , there are 4 ( or 5 including the control function ) independent parameters for each metabolic flux with a one-step binding mechanism . For fluxes with no allosteric regulation , χ ( α , ι ) =1 . For all metabolic fluxes involving NAD+/NADH oxidoreductase activity , we utilized a more complex rate law than that of a one step binding reaction ( Cornish-Bowden , 2012 ) . We chose to consider a more complicated rate-law since our initial observations indicated that a one-step binding mechanism produced numerical instabilities in the solutions . We therefore considered a general rate law corresponding to a random bi–bi mechanism ( for LDH and PHGDH ) and a random ter-bi reaction for GAPDH . For example , in its general form the rate law for LDH is: ( 8 ) Vldh=VfldhmaxPYR∗NADHKmldhf−VrldhmaxLACi∗NAD+Kmldhr1+PYRKmpyrldh+NADHKmnadhldh+PYR∗NADHKmldhf+LACiKmlacldh+NAD+Kmnadldh+LACi∗NAD+Kmldhr . However , it has also been suggested that the enzymatic reaction for oxidoreductase activity proceeds in an ordered fashion resulting in a rate law of the form: ( 9 ) Vldh=VfldhmaxPYR∗NADHKmldhf−VrldhmaxLACi∗NAD+Kmldhr1+NADHKmnadhldh+PYR∗NADHKmldhf+NAD+Kmnadldh+LACi∗NAD+Kmldhr When considering this mechanism , it was found that statistical evaluation of the model was robust to changes in the choice of rate law . The rate equations for each metabolic reaction and metabolic parameters values are in Supplementary file 1 . Taking into account additional thermodynamic constraints by invoking the Haldane equation that relates the forward and reverse fluxes and their Vmax values allowed for a further reduction in the number of independent parameters to three parameters per reaction flux . The Haldane equation has the form: ( 10 ) Vrmax=VfmaxKmrKmfKeq , where Keq is the reaction apparent equilibrium constant: Keq=exp ( −ΔG°/RT ) , where ΔG° is standard reaction Gibbs free energy . ΔG° values for glycolytic reactions are listed in Supplementary file 1 . We first used two independent sets of experimental data from a compendium of steady-state metabolite concentrations of liver cells and steady-state fluxes of DB1 melanoma cells evaluated using a cultured bioreactor ( Konig et al . , 2012; Shestov et al . , 2013 ) . The model was further validated by a comparison of measurements of redox status , assessment of ATP , ADP , and AMP concentrations . Together , those data were able to generate a thermodynamically feasible and experimentally observable model of glycolysis that was then used as the starting point for exhaustive Monte Carlo simulations and coupled metabolic control analysis . The specific aim of this work is to enumerate the variables within and coupled to glycolysis that determine the extent of aerobic glycolysis . We therefore developed a novel algorithm to assess these features . The algorithm involves a global sensitivity analysis based on a Monte Carlo method . The Monte Carlo analysis allows for sampling of parameter space over a broad range of simultaneous variations of parameters ( enzyme expression levels ) followed by statistical assessment of the resulting solution . The idea of this method is to inject uncertainty of the parameters in the model by randomly selecting parameter values from uniform probability distributions . This was achieved by a Monte Carlo method using a randomly drawn set of parameters . The range of parameters chosen with uniform sampling was within two orders of magnitudes of each Vmax ( which is proportional to enzyme activity level and was in the range of 10 times less and 10 times greater than the initial Vmax value ) . Such a sampling was chosen to be large enough to cover all feasible solutions of aerobic glycolysis . The sampling was carried out using a Latin Hypercube sampling method ( Oguz et al . , 2013 ) . For each chosen set of random parameter vectors the model was simulated between 2000 and 5000 times and its output was calculated . Briefly , we divided the range of the i-th normalized parameter into n ( n = 2000–5000 ) subintervals of equal size . Then , we randomly sample n values ( e . g . , pi , i = 1 , … , n ) , one from each subinterval , for the i-th parameter . We next randomly permuted the n values for each parameter to get the parameter vector . We then evaluated the following: the Warburg Effect value , energy and redox states , metabolite steady-state concentrations and fluxes and flux control coefficients ( FCCs ) . The flux control coefficient that an enzyme Ei exerts on the lactate flux JLac is defined as ( Heinrich and Rapoport , 1974; Fell , 1992; Shestov et al . , 2013 ) : ( 11 ) FCCi=∂ln⁡JLac∂ln⁡Ei . To compute the FCC values numerically , we considered a perturbation 0 . 01 times of the enzyme value and then evaluated the change in flux . The model was implemented in Matlab as a system of 27 ordinary differential equations ( ODEs ) with an ODE solver designed for stiff ODE systems . HCT116 cells were cultured with a growth medium , which contains RPMI 1640 , 10% Fetal Bovine Serum ( FBS ) , 100 U/ml penicillin , and 100 μg/ml streptomycin . The cells were obtained as a gift from Lewis Cantley's laboratory . The cells were cultured in 37°C with 5% CO2 . For treatments , the cells were seeded in 6-well plate at a density of 2 × 105 to 5 × 105 cells per well . After overnight incubation , full growth media were removed , and cells were washed with 2 ml PBS before the addition of RPMI media ( without glucose ) , supplemented with 10% dialyzed and heat inactivated FBS , 100 U/ml penicillin , 100 μg/ml streptomycin and 5 mM 13C-U-glucose ( Cambridge Isotope Laboratory , Tewksbury , MA ) ( fresh growth media ) . For drug treatments and flux measurements , full growth media were replaced with fresh growth media , containing either 0 . 1% DMSO , or the concentration of the indicated drug . Media were then collected at 30 , 60 , 90 , and 120 min after incubation . From 20 µl media , 80 µl of ice cold H2O was added , together with 400 µl ice cold methanol ( Fisher , Optima LC/MS grade ) . After vigorous vortexing , the solution was then centrifuged at 20 , 000×g at 4°C for 10 min and then the supernatant was dried under vacuum . At 120 min , the media were removed as completely as possible , and then 6-well plates were immediately placed on dry ice , followed by the addition of 1 ml extraction solvent , 80% MeOH/H2O ( Fisher , Optima LC/MS Grade ) , which was pre-cooled in −80°C freezer for at least 1 hr . The dishes were then transferred to the −80°C freezer . The plates were left for 15 min and then cells were scraped into the solvent on dry ice and then transferred to two 1 . 7-ml eppendorf tubes , and centrifuged with the speed of 20 , 000×g at 4°C for 10 min . Metabolite extracts are prepared from three separate wells to make three replicate samples . The supernatant is then transferred to new eppendorf tubes , and dried under vacuum . For analysis , each extract was then re-constituted into water ( 15 μl for cell extract and 50 μl for medium extract ) and then diluted with an equal volume of 50% Methanol/Acetonitrile . Finally , 5 μl was injected into the column for analysis . Absolute concentrations of 13C lactate were first measured by mixing a standard of unlabeled lactate with a defined concentration into the medium . For each calculated flux , an estimation of the flux from glucose to lactate was obtained from the slope of the time course of 13C lactate production using the time points of media collection described above . Slopes were computed using Graphpad Prism with time courses determined to be linear from the goodness of fit ( R2 > 0 . 98 with few exceptions ) . For corresponding plots for different agents , fraction of inhibition was computed from Ki values that were taken based on previous reports ( Foxall et al . , 1984; Le et al . , 2010; Telang et al . , 2012 ) . The Q Exactive Mass Spectrometer ( QE-MS ) is equipped with a heated electrospray ionization probe ( HESI ) , and the relevant parameters are as listed: heater temperature , 120°C; sheath gas , 30; auxiliary gas , 10; sweep gas , 3; spray voltage , 3 . 6 kV for positive mode , and 2 . 5 kV for negative mode . Capillary temperature was set at 320°C , and S-lens was 55 . A full scan range from 60 to 900 ( m/z ) was used . The resolution was set at 70 , 000 . The maximum injection time was 200 ms with typical injection times around 50 ms . These settings resulted in a duty cycle of around 550 ms to carry out scans in both positive and negative mode . Automated gain control ( AGC ) was targeted at 3 , 000 , 000 ions . Liquid chromatography ( Ultimate 3000 UHPLC ) is coupled to the QE-MS for metabolite separation and detection . An Xbridge amide column ( 100 × 2 . 1 mm i . d . , 3 . 5 μm; Waters ) is employed for compound separation . The mobile phase A is 20 mM ammonium acetate and 15 mM ammonium hydroxide in water with 3% acetonitrile , pH 9 . 0 , as described above , and mobile phase B is acetonitrile . A linear gradient was used as follows: 0 min , 85% B; 1 . 5 min , 85% B; 5 . 5 min , 35% B; 10 min , 35% B; 10 . 5 min , 35% B; 14 . 5 min , 35% B; 15 min , 85% B; and 20 min , 85% B . The flow rate was 0 . 15 ml/min from 0 to 10 min and 15 to 20 min , and 0 . 3 ml/min from 10 . 5 to 14 . 5 min . All solvents are LC-MS Optima grade and purchased from Fisher Scientific . Human mammary epithelial MCF-10A cells ( CRL-10317; ATCC ) stably expressing Peredox-NLS were generated and cultured as previously described ( Debnath et al . , 2003; Hung et al . , 2011 ) . 2–4 days prior to imaging , ∼500 cells were plated onto the center of each well of a 96-well plate . On the day of the experiment , cells were placed in custom DMEM/F12 ( Gibco ) containing no glucose and supplemented to the levels mentioned . Fluorescence images were acquired using a Nikon inverted Eclipse Ti microscope , equipped with a Nikon 20×/0 . 75 Plan Apo objective , three different regions of interest chosen , images sequentially acquired every 8 min with 50–100 ms exposure and 2 × 2 binning . Using a custom MATLAB algorithm previously developed , we subtracted background , set a threshold for cell segmentation , and analyzed the data as previously described ( Hung et al . , 2011 ) . The concentration of glucose is maintained by considering a culture system in which cells are seeded at low density and media are present in vast excess . An estimate of glucose maintenance in the media in the low glucose condition can be considered using values the glucose uptake rate of the cells ( ∼100 fmol/cell/hr ) , the cell number ( ∼2000 ) , the volume of media used ( 400 μl ) , and concentration of glucose in the media ( 750 μM , 500 μM from supplementation + 250 μM from the Horse Serum ) . Thus in 24 hr , we estimate that about 5 nmols of glucose are consumed . The media contain roughly 200 nmols of glucose which is in excess of the amount consumed by the cells . Raw data collected from the QE-MS is processed using Sieve 2 . 0 ( Thermo Scientific ) . Peak alignment and detection are performed according to the protocol described by Thermo Scientific . For a targeted metabolite analysis , the method ‘peak alignment and frame extraction’ is applied . An input file of theoretical m/z and detected retention time known metabolites is used for targeted metabolite analysis with data collected in both positive and negative mode . m/z width is set at 10 ppm . The output file including detected m/z and relative intensity in different samples is obtained after data processing . For resulting simulation data , hierarchical clustering was carried out using spearman ranked correlations and the Gene-e software package ( Broad Institute ) . Box-plots ( 25%/75% percentile , mean and median ) was calculated and made with the Graphpad Prism software package .	Cells generate energy from a sugar called glucose via a process called glycolysis . This process involves many enzymes that catalyze 10 different chemical reactions , and it essentially converts glucose step-by-step into a simpler chemical called pyruvate . Pyruvate is then normally transported into structures within the cell called mitochondria , where it is further broken down using oxygen to release more energy . However , in cells that are rapidly dividing , pyruvate is converted into another chemical called lactate—which releases energy more quickly , but releases less energy overall . Cancer cells often convert most of their glucose into lactate , rather than breaking down pyruvate in their mitochondria: an observation known as the ‘Warburg effect’ . And while many factors affect how a cell releases energy from pyruvate , it remains unclear what regulates which of these biochemical processes is most common in a living cell . In this study , Shestov et al . have developed a computational model for the process of glycolysis and used this to investigate the causes of the Warburg Effect . The model was based on the known characteristics of the enzymes and chemical reactions involved at each step . It predicted that the activity of the enzyme called GAPDH , which carries out the sixth step in glycolysis , in many cases affects how much lactate is produced . This suggests that this enzyme represents a bottleneck in the pathway . Next , Shestov et al . performed experiments where they used drugs to block different stages of the glycolysis pathway , and confirmed that the GAPDH enzyme is important for regulating this pathway in living cancer cells too . In these treated cells , the levels of a chemical called fructose-1 , 6-biphosphate ( which is made in a step in the pathway between glucose and pyruvate ) were either very high or very low . Shestov et al . proposed that the flow of chemicals through the glycolysis pathway is controlled by the GAPDH enzyme when the chemicals used by the enzymes upstream of GAPDH in the pathway ( which includes fructose-1 , 6-biphosphate ) are plentiful . However , if these chemicals are limited , other enzymes that are involved in earlier steps of the pathway regulate the process instead . The findings of Shestov et al . reveal that the regulation of glycolysis is more complex than previously thought , and is also very different when cells are undergoing the Warburg Effect . In the future , these findings might help to identify the types of cancer that could be effectively treated using drugs that target the glycolysis process , which are currently being tested in pre-clinical studies .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2014	Quantitative determinants of aerobic glycolysis identify flux through the enzyme GAPDH as a limiting step
Pancreatic cancer has a high mortality rate due to metastasis . Whereas KRAS is mutated in most pancreatic cancer patients , controlling KRAS or its downstream effectors has not been succeeded clinically . ARL4C is a small G protein whose expression is induced by the Wnt and EGF–RAS pathways . In the present study , we found that ARL4C is frequently overexpressed in pancreatic cancer patients and showed that its localization to invasive pseudopods is required for cancer cell invasion . IQGAP1 was identified as a novel interacting protein for ARL4C . ARL4C recruited IQGAP1 and its downstream effector , MMP14 , to invasive pseudopods . Specific localization of ARL4C , IQGAP1 , and MMP14 was the active site of invasion , which induced degradation of the extracellular matrix . Moreover , subcutaneously injected antisense oligonucleotide against ARL4C into tumor-bearing mice suppressed metastasis of pancreatic cancer . These results suggest that ARL4C–IQGAP1–MMP14 signaling is activated at invasive pseudopods of pancreatic cancer cells . Pancreatic cancer is extremely aggressive and exhibits poor prognosis , with a 5 year survival of only 5 % ( Klein , 2013 ) . Most pancreatic cancer-related deaths are due to metastatic disease , and more than 80 % of patients have either locally advanced or metastatic disease ( Hidalgo , 2010; Klein , 2013 ) . Genome sequencing analysis has revealed the mutational landscape of pancreatic cancer and KRAS mutations are considered an initiating event in pancreatic ductal cells ( Collins et al . , 2012; Waddell et al . , 2015 ) . Irrespective of our improved understanding of tumor biology , the treatment outcome has not changed for many years . Therefore , new innovative treatment options need to be tested based on better understanding of the characteristics of pancreatic cancer . ARL4C is a member of the ADP-ribosylation factor ( ARF ) -like protein ( ARL ) family , which belongs to the ARF protein subgroup of the small GTP-binding protein superfamily ( Engel et al . , 2004; Matsumoto et al . , 2017; Wei et al . , 2009 ) . Cytohesin2/ARF nucleotide-binding site opener ( ARNO ) , a GDP/GTP exchange factor of ARF family proteins , has been identified as a direct effector protein ( Hofmann et al . , 2007 ) . ARL4C is expressed through activation of Wnt–β-catenin and EGF–RAS signaling and plays important roles in both epithelial morphogenesis and tumorigenesis ( Matsumoto et al . , 2017; Matsumoto et al . , 2014 ) . Because aberrant activation of the Wnt–β-catenin and/or EGF–RAS pathways are frequently observed in various types of cancers , ARL4C is indeed expressed in a number of cancers ( Fujii et al . , 2015; Fujii et al . , 2016 ) . In colon and lung cancer cells , ARL4C promotes cell proliferation through ARF6 , RAC , RHO , and YAP/TAZ . On the other hand , in liver cancer cells , ARL4C promotes cell proliferation through phosphatidylinositol three kinase δ ( PI3Kδ ) ( Harada et al . , 2019 ) . Thus , ARL4C would activate different downstream pathways in a cancer cell context-dependent manner . These prompted us to study the involvement of ARL4C , as a KRAS downstream molecule , in aggressiveness of pancreatic cancer , and IQ-domain GTPase-activation protein 1 ( IQGAP1 ) was identified as a binding protein of ARL4C . IQGAPs are an evolutionally conserved family of proteins that bind to a diverse array of signaling and structural proteins ( Hedman et al . , 2015 ) . Mammalian IQGAP1 is a well-characterized member of the IQGAP family and a fundamental regulator of cytoskeletal function ( Briggs and Sacks , 2003 ) . IQGAP1 is highly expressed in the tumor lesions and suggested to be involved in cancer cell metastasis ( Johnson et al . , 2009; Sakurai-Yageta et al . , 2008 ) . Here , we show that ARL4C bound to IQGAP1 and recruited IQGAP1 and membrane type1-matrix metalloproteinase ( MT1-MMP , also called MMP14 ) ( Sakurai-Yageta et al . , 2008 ) to invasive pseudopods in a phosphatidylinositol ( 3 , 4 , 5 ) -trisphosphate ( PIP3 ) -dependent manner and accelerated invasion . In addition , ARL4C antisense oligonucleotide ( ASO ) suppressed the lymph node metastases of pancreatic cancer cells orthotopically implanted into the pancreas of immunodeficient mice . These results suggest that the ARL4C–IQGAP1–MMP14 signaling axis promotes pancreatic cancer aggressiveness and that ARL4C is a novel molecular target for the treatment of pancreatic cancer . Whether ARL4C is expressed in pancreatic cancer patients was examined using immunohistochemistry . Fifty-seven pancreatic ductal adenocarcinoma ( PDAC ) patients , who did not receive preoperative chemotherapy , were used in this study ( Supplementary file 1 table 1; Source data 1 ) . ARL4C staining in the tumor lesions was calculated as a continuous variable , and the patients were classified into two groups ( high and low ) , depending on ARL4C expression levels ( Figure 1A ) . ARL4C expression was considered high when the total area of the tumor stained with anti-ARL4C antibody exceeded 5 % . High expression of ARL4C was observed in 47 cases ( 82% ) , but minimally detected in non-tumor regions of pancreatic ducts ( Figure 1A ) . Anti-ARL4C antibody used in this study was validated in western blotting and immunohistochemical assay ( IHC ) ( Figure 1—figure supplement 1A and B ) . A significant difference was observed between low and high ARL4C expression based on perineural invasion ( Supplementary file 1 table 1 ) . Because the perineural invasion is considered as one of the causes of the recurrence and metastasis after pancreatic resection ( Liang et al . , 2016 ) , ARL4C expression may be correlated with the ability of cancer cell invasion . Consistently , ARL4C expression was correlated with decreased overall survival ( Figure 1B ) . Analysis of TCGA and GTEx datasets revealed that ARL4C is highly expressed in tumor tissue than in non-diseased tissue ( Figure 1C ) . In addition , when ARL4C high and low expression groups were separated based on the top 75 % of mRNA values of ARL4C in TCGA dataset , high expression of ARL4C indicated a poor prognosis ( Figure 1D ) . Univariate and multivariate analysis revealed that higher ARL4C expression is an independent prognostic factor ( Table 1 ) . Taken together , these results indicate that high expression of ARL4C is correlated with the aggressiveness and poor prognosis of pancreatic cancer . Pancreatic intraepithelial neoplasia ( PanIN ) lesions were observed in 26 specimens . ARL4C was expressed in 20 of 26 cases ( 77% ) of PanIN , suggesting that ARL4C is expressed in early stages of PDAC ( Figure 1—figure supplement 1C ) . The results are consistent with our recent observations that ARL4C is frequently expressed in atypical adenomatous hyperplasia , which is the possible precursor lesions and develops to lung adenocarcinoma ( Kimura et al . , 2020 ) . In cultured pancreatic cancer cell lines , ARL4C was highly expressed in PANC-1 and S2-CP8 cells and it was barely detected in BxPC-3 cells ( Figure 1E ) . Consistent with the previous results with IEC6 rat intestinal epithelial cells and colorectal and lung cancer cells ( Fujii et al . , 2015; Matsumoto et al . , 2014 ) , the MEK inhibitors PD184161 and U0126 and siRNAs for β-catenin and KRAS decreased ARL4C expression in S2-CP8 and PANC-1 cells ( Figure 1F–H ) . In addition , simultaneous knockdown of KRAS and β-catenin further suppressed ARL4C expression ( Figure 1I ) . Taken together , these results suggest that ARL4C is expressed in pancreatic cancer cells through activated RAS–MAP kinase and Wnt–β-catenin pathways . ARL4C ASO-1316 has been shown to inhibit growth of xenograft tumors induced by colon and lung cancer cells ( Harada et al . , 2019; Kimura et al . , 2020 ) . However , ARL4C ASO-1316 had little effect on sphere formation of pancreatic cancer cell ( Figure 2—figure supplement 1A and B and B ) and did not induce cell death , which is assessed by propidium iodide ( PI ) staining ( Figure 2—figure supplement 1C ) . Since the clinicopathological analysis of human pancreatic cancer specimens indicates that ARL4C expression may be correlated with invasive ability , migratory and invasive abilities of S2-CP8 and PANC-1 cells were studied in Boyden chamber assays . ARL4C ASO-1316 inhibited the migratory and invasive abilities with dominant effects on invasion ( Figure 2A and B; Figure 2—figure supplement 1D ) . Inhibition of migratory and invasive abilities by ARL4C ASO , targeting the non-coding region of ARL4C mRNA , was not observed in the cells expressing ARL4C-GFP ectopically ( Figure 2C and D; Figure 2—figure supplement 1E ) . Thus , ARL4C could be involved in migration and invasion of pancreatic cancer cells . ARL4C has been shown to be localized to membrane protrusions of non-tumor cells , such as IEC6 and Madin-Darby canine kidney ( MDCK ) cells ( Matsumoto et al . , 2014 ) . ARL4C-tdTomato was localized to protrusive structures extending from S2-CP8 cells under Matrigel-coated 2D culture conditions ( Figure 2—figure supplement 1F ) . At the structures , focal adhesion proteins such as paxillin , phosphorylated paxillin , FAK , and phosphorylated FAK were localized with ARL4C-tdTomato , also with F-actin ( Figure 2—figure supplement 1F ) . Therefore , we defined the membrane protrusions as actin-based structures that contain the adhesion sites , of which length is longer than 10 μm and diameter is shorter than 10 μm . ARL4C is unique in that it is locked to the GTP-bound active form , and ARL4CQ72L-GFP , in which the amino acid at the same position in a constitutively active RAS mutant was mutated , showed a similar distribution to ARL4C-GFP . However , ARL4CT27N-GFP , which is an inactive form ( Hofmann et al . , 2007 ) , was not present in the protrusions ( Figure 2—figure supplement 1G ) . These results suggest that ARL4C is present in the tips of membrane protrusions where it is expressed as wild type . Invadopodia are well-known membrane protrusions that localize at the ventral surfaces of cells and are active in extracellular matrix ( ECM ) degradation during cancer invasion ( Murphy and Courtneidge , 2011 ) . To analyze invadopodia , pancreatic cancer cells were grown on gelatin-coated glass coverslips ( Figure 2—figure supplement 2A ) . Dark areas represent gelatinolytic activity of invadopodia and are equal to invadopodia structures . BxPC-3 cells exhibited invadopodia clearly , whereas S2-CP8 and PANC-1 cells did not ( Figure 2—figure supplement 2A ) . Thus , some pancreatic cancer cells do not form typical invadopodia in gelatin surface but can invade into ECM through probably other structures . Meanwhile , components of invadopodia , such as cortactin and ARPC2 , were localized at the tips of protrusions defined above , with ARL4C ( Figure 2—figure supplement 2B ) , suggesting that the protrusions might contribute to invasive phenotypes of pancreatic cancer cells and ARL4C functions there . Therefore , we referred to the protrusive structures as ‘invasive pseudopods’ , because they seem to be analogous to invadopodia ( Jacquemet et al . , 2013; Murphy and Courtneidge , 2011; Yu and Machesky , 2012 ) . ARL4C knockout did decrease numbers of invasive pseudopods but slightly , while knockdown of ARPC2 , which regulates formation of pseudopods as one of the components of Arp2/3 complex , clearly reduced the number of pseudopods ( Figure 2—figure supplement 2C-G ) . Next it was tested whether ARL4C is involved in the presence of invadopodia markers in the tips of pseudopods . ARL4C knockout did not affect ARPC2 staining statistically and reduced the staining of cortactin only modestly ( Figure 2—figure supplement 2H and I ) . It is quite likely that ARL4C contributes to invasive properties through other than the formation of pseudopods . Therefore , ARL4C may be necessary for functions of invasive pseudopods rather than their formation . This prompted us to look further into the invading process . For visualization of cancer cells invading through the ECM ( Poincloux et al . , 2009 ) , a 3D microfluidic cell culture with type I collagen ( Farahat et al . , 2012; Shin et al . , 2012 ) ( 3D gel invasion assay ) was performed ( Figure 2E ) . At 0 time the same numbers of cells treated with control and ARL4C ASO were placed in the starting position ( Figure 2—figure supplement 3A ) , and after 72 hr directional invading ability was compared . Whereas control S2-CP8 cells invaded into type I collagen , ARL4C ASO decreased invasive ability ( Figure 2F ) . ARL4C KO cells also decreased invasive ability ( Figure 2—figure supplement 3B ) . When the collagen concentration was reduced , S2-CP8 cells invaded irrespective of ARL4C knockdown ( Figure 2G ) , suggesting that their invasive ability is not required for cells to move into the ECM when collagen fiber-formed 3D net structures are sparse . Furthermore , in the 3D gel invasion assay ARL4C-tdTomato accumulated in the tips of invasive pseudopods ( Figure 2H; Figure 2—video 1 ) . Fluorescence intensities of ARL4C-tdTomato in the edges of the pseudopods and cytoplasm ( 20 μm away from the tip of pseudopod ) , respectively , were measured over time , and then the intensities were plotted as a function of time . The results indicate that ARL4C dynamically appeared and disappeared in the pseudopods , but it did not accumulate in the cytoplasm ( Figure 2H ) . Using time-lapse imaging the angle of pseudopods to the direction of cell movement towards FBS was observed . Most of them were located in the angle of –45 to +45 degrees in the polar histogram , suggesting that invasive pseudopods play a role in purposeful directional invasion ( Figure 2I; Figure 2—video 2 ) . To visualize the relationship between the localization of ARL4C and matrix degradation , the steady-state activity of cell-derived protease was measured as the dequenched signal emitted from collagen I fibers with dye-quenched ( DQ ) FITC ( DQcollagen I ) ( Wolf et al . , 2007 ) in the 3D gel invasion assay . Protease-induced fluorescence dequenching was detected in the collagen fibers crossing the tips of the pseudopods but not in the cell body ( Figure 2J ) . Protease activity was decreased when ARL4C was depleted ( Figure 2K ) , suggesting that ARL4C is involved in degradation of the ECM through its localization to the tips of invasive pseudopods and plays an important role in the invasion of pancreatic cancer cells . ARL4C recruits cytohesin2 to the plasma membrane through their direct interaction in HeLa cells ( Hofmann et al . , 2007 ) . In S2-CP8 cells , ARL4C did not bind to cytohesin2 ( Figure 3—figure supplement 1A ) , and knockdown of cytohesin2 had no effect on the migratory or invasive ability ( Figure 3—figure supplement 1B ) . Furthermore , cytohesin2 was distributed throughout the cytosol in S2-CP8 cells , whereas it was localized to the cell periphery of HeLaS3 cells ( Figure 3—figure supplement 1C ) . Whereas ARL4C ASO inhibited RAC1 activity in A549 cells ( Fujii et al . , 2015 ) , the ASO did not affect RAC1 activity in S2-CP8 cells ( Figure 3—figure supplement 1D ) . Although ARL4C induces the nuclear import of YAP/TAZ in HCT116 cells ( Harada et al . , 2019 ) , ARL4C knockdown did not inhibit it in pancreatic cancer cells ( Figure 3—figure supplement 1E ) . These results suggest that cytohesin2 neither functions downstream of ARL4C nor is involved in migration or invasion of pancreatic cancer cells and prompted us to explore an uncharacterized effector protein of ARL4C . ARL4C-FLAG-HA–binding proteins were precipitated and the precipitates were analyzed by mass spectrometry ( Figure 3A ) . Among the possible interacting proteins , IQGAP1 was further studied ( Figure 3A; Supplementary file 1 table 2; Source data 2 ) because its expression is associated with the aggressiveness of various types of cancer ( Johnson et al . , 2009 ) . Ectopically expressed and endogenous ARL4C were associated with endogenous IQGAP1 in S2-CP8 cells ( Figure 3B and C ) . ARL4C-FLAG-HA and ARL4CQ72L-FLAG-HA formed a complex with GFP-IQGAP1 to the similar levels , but ARL4CT27N-FLAG-HA showed diminished binding to GFP-IQGAP1 in X293T cells ( Figure 3D ) . Using another anti-ARL4C antibody for the immunocytochemical study ( Figure 3—figure supplement 2A and B ) , ARL4C and IQGAP1 were shown to accumulate to invasive pseudopods at endogenous level in S2-CP8 and PANC-1 cells under Matrigel-coated 2D culture conditions ( Figure 3E; Figure 3—figure supplement 2C ) . Colocalization of ARL4C and IQGAP1 at invasive pseudopods was observed in 94 % of cells with ARL4C accumulation to the pseudopods . In 3D culture conditions , IQGAP1 was found at the tips of invasive pseudopods , similar to ARL4C-tdTomato ( Figure 3F ) . IQGAP1 siRNA inhibited the migratory and invasive abilities in S2-CP8 and PANC-1 cells , and the cells expressing GFP-IQGAP1 were resistant to IQGAP1 siRNA ( Figure 3G and H; Figure 3—figure supplement 2D and E ) . Simultaneous knockdown of ARL4C and IQGAP1 decreased the invasive ability , but the inhibitory degree was similar to that induced by knockdown of either ARL4C or IQGAP1 ( Figure 3I ) . Thus , IQGAP1 and ARL4C regulate invasion in identical signaling pathways . IQGAP1 was highly expressed in 31 of 57 PDAC patients ( 54% ) ( Figure 3J ) . The anti-IQGAP1 antibody was validated by Western blotting and immunocytochemical and immunohistochemical analyses ( Figure 3—figure supplement 2E-G ) . Although higher expression of IQGAP1 was not associated with clinical parameters ( Supplementary file 1 table 3 ) , IQGAP1 expression correlated with decreased overall survival ( Figure 3K ) . Similar results were obtained from the analysis of TCGA and GTEx datasets ( Figure 3—figure supplement 2H and I ) . TCGA dataset revealed that expression of ARL4C mRNA in pancreatic cancer patients is positively correlated with that of IQGAP1 mRNA ( Figure 3L ) . Of 47 PDAC patients with high ARL4C expression , IQGAP1 was highly expressed in 27 patients ( Supplementary file 1 table 4 ) . Higher expression of ARL4C in the patients positive for IQGAP1 was associated with perineural invasion ( Supplementary file 1 table 4 ) . The overall survival of the patients who were double positive for ARL4C and IQGAP1 tended to be worse although it is not statistically significant ( Figure 3—figure supplement 2J ) . Therefore , the relationship between ARL4C and IQGAP1 expression on patient survival using public datasets was analyzed . Overall survival was significantly decreased in the order of low ARL4C/low IQGAP1 , high ARL4C/low IQGAP1 , and high ARL4C/high IQGAP1 , although the result of low ARL4C/high IQGAP1 could not conclude because of the small case numbers ( n = 2 ) ( Figure 3—figure supplement 2K ) . Thus , simultaneous expression of ARL4C and IQGAP1 would be correlated with aggressiveness of pancreatic cancer . ARL4C is modified by myristate at the N terminus and has a polybasic region ( PBR ) , comprising nine Lys or Arg residues , at the C terminus ( Donaldson and Jackson , 2011 ) . ARL4CG2A , whose N-terminal myristoylation site ( Gly2 ) is mutated to Ala , and ARL4CΔPBR were expressed in S2-CP8 cells . In contrast to ARL4C-GFP , ARL4CG2A-GFP and ARL4CΔPBR-GFP were not accumulated at invasive pseudopods where cortactin was present , but distributed throughout the cytosol ( Figure 4A and B; Figure 4—figure supplement 1A-C ) , and both mutants severely decreased the binding activity to GFP-IQGAP1 ( Figure 4C ) . The C-terminal region of KRAS includes the PBR and the CAAX motif , which is farnesylated , and fusion of the KRAS C-terminal region triggers the localization of the proteins to the cell surface membrane ( Hancock et al . , 1990 ) . The KRAS C-terminal region was fused to the ARL4C mutants , which were referred to as ARL4C-GFP-Cterm . Both ARL4CG2A-GFP-Cterm and ARL4CΔPBR-GFP-Cterm were localized to invasive pseudopods where cortactin was present ( Figure 4B; Figure 4—figure supplement 1A ) . However , although ARL4CG2A-FLAG-HA-Cterm formed a complex with GFP-IQGAP1 , ARL4CΔPBR-FLAG-HA-Cterm did not ( Figure 4D ) , suggesting that membrane localization of ARL4C is not sufficient for its binding to IQGAP1 . Taken together , the PBR is necessary for ARL4C to associate with IQGAP1 , as well as for recruiting ARL4C to invasive pseudopods . The localization of IQGAP1 to invasive pseudopods was lost in ARL4C KO cells , but not vice versa ( Figure 4E and F; Figure 2—figure supplement 2C and D; Figure 4—figure supplement 1D-G ) . The similar results were obtained in ARL4C knockdown cells ( Figure 4—figure supplement 1H ) . In ARL4C KO cells , ARL4C-GFP and ARL4CG2A-GFP-Cterm rescued the recruitment of IQGAP1 to the plasma membrane , unlike ARL4CG2A-GFP , ARL4CΔPBR-GFP , and ARL4CΔPBR-GFP-Cterm ( Figure 4E ) . Therefore , for IQGAP1 to be recruited to invasive pseudopods , the localization of ARL4C to the plasma membrane and the binding to IQGAP1 through the PBR might be necessary . In addition , inhibition of invasive ability by ARL4C ASO-1316 was cancelled by expression of ARL4CG2A-GFP-Cterm but not by that of ARL4CG2A-GFP , ARL4CΔPBR-GFP , or ARL4CΔPBR-GFP-Cterm ( Figure 4G; Figure 4—figure supplement 1I ) . Thus , the binding of ARL4C and IQGAP1 in invasive pseudopods could be essential for the invasive ability . PI ( 4 , 5 ) P2 ( PIP2 ) and PI ( 3 , 4 , 5 ) P3 ( PIP3 ) are required for ARL4C membrane targeting ( Heo et al . , 2006 ) . The pleckstrin homology ( PH ) domain functions as a protein- and phospholipid-binding structural protein module ( Maffucci and Falasca , 2001 ) . The PH domains of PLCδ and GRP1 prefer to bind to PIP2 and PIP3 , respectively ( Lemmon , 2008 ) . GFP-PLCδPH was detected throughout the cell surface membrane , whereas GFP-GRP1PH was accumulated in invasive pseudopods ( Figure 5A ) . The levels of PIP2 and PIP3 in the plasma membrane were decreased by a rapamycin-inducible PIP2-specific phosphatase ( Inp54p ) ( Suh et al . , 2006 ) and a PI3 kinase inhibitor LY294002 ( Petrie et al . , 2012 ) , respectively . S2-CP8 cells were treated with rapamycin and LY294002 for 30 min to examine the localization of ARL4C and IQGAP1 , and for 24 hr to test invasive ability . PIP3 depletion decreased the membrane targeting of ARL4C and IQGAP1 and reduced the invasive ability , but PIP2 depletion did not ( Figure 5B and C ) . IQGAP1 and ARL4C-mCherry colocalized with GRP1PH in invasive pseudopods ( Figure 5D ) , suggesting that both proteins accumulate in the cell peripheral regions containing PIP3 and promote invasion . To reveal the importance of PIP3 for the localization area of ARL4C and IQGAP1 , PLCδPH or GRP1PH was fused to the C terminus of ARL4CG2A-GFP ( Figure 5E ) . While both ARL4CG2A-GFP-GRP1PH and ARL4CG2A-GFP-PLCδPH formed a complex with GFP-IQGAP1 , the former construct was localized to invasive pseudopods , but the latter construct was present throughout the cell surface membrane ( Figure 5F; Figure 5—figure supplement 1A ) . Consistently , in ARL4C KO cells extending invasive pseudopods , the localization of IQGAP1 to invasive pseudopods was rescued by ARL4CG2A-GFP-GRP1PH but not by ARL4CG2A-GFP-PLCδPH ( Figure 5G ) . Furthermore , ARL4C ASO-1316 inhibited the invasive ability of S2-CP8 cells expressing ARL4CG2A-GFP-PLCδPH but not those expressing ARL4CG2A-GFP-GRP1PH ( Figure 5H; Figure 5—figure supplement 1B ) . Taken together , these results suggest that PIP3-dependent membrane targeting of ARL4C recruits IQGAP1 to invasive pseudopods and promotes invasion . IQGAP1 is involved in the trafficking of MMP14-containing vesicles to the leading structures of cancer cells ( Sakurai-Yageta et al . , 2008 ) . TCGA dataset showed that expression of MMP14 mRNA in pancreatic cancer patients is positively correlated with that of both ARL4C and IQGAP1 mRNA ( Figure 6—figure supplement 1A ) . In addition , MMP14 expression was associated with poor prognosis ( Figure 6—figure supplement 1B ) . Cell surface MMP14-GFP accumulated in invasive pseudopods containing IQGAP1 and ARL4C-FLAG-HA ( Figure 6A ) . MMP14-GFP extremely disappeared from invasive pseudopods of ARL4C knockdown and KO cells and the phenotype was rescued by expression of ARL4C-FLAG-HA ( Figure 6B; Figure 6—figure supplement 1C and D ) . IQGAP1 KO caused the loss of MMP14-GFP from the pseudopods , and FLAG-HA-IQGAP1 expression rescued this phenotype ( Figure 6C ) . The failure of MMP14 membrane targeting in ARL4C KO cells was rescued by expression of ARL4CG2A-FLAG-HA-Cterm but not by that of ARL4CG2A-FLAG-HA , ARL4CΔPBR-FLAG-HA , or ARL4CΔPBR-FLAG-HA-Cterm ( Figure 6B ) . In addition , PIP3 depletion , but not PIP2 depletion , suppressed the membrane localization of MMP14 ( Figure 6D ) . Therefore , in co-operation with ARL4C and IQGAP1 , MMP14 is likely to be trafficked to invasive pseudopods with PIP3 accumulation . Consistent with these results , the inhibited invasive ability after double knockdown of ARL4C and MMP14 or IQGAP1 and MMP14 was similar to that seen after single knockdown of ARL4C , IQGAP1 , or MMP14 ( Figure 6E; Figure 6—figure supplement 1E and F ) . Knockdown of ARL4C , IQGAP1 , or MMP14 decreased invasive ability in 3D microfluidic cell culture ( Figure 6F ) and the protease activity was also reduced ( Figure 6G ) . Previous work has shown that MMP14ΔC ( Δ563–582 ) lacking the cytoplasmic region fails to be endocytosed ( Jiang et al . , 2001 ) . Here , MMP14ΔC was retained in invasive pseudopods of ARL4C-KO cells ( Figure 6—figure supplement 1G ) , and the ARL4C knockdown-mediated decreases in cell invasion and collagen degradation were rescued by MMP14ΔC ( Figure 6H; Figure 6—figure supplement 1H and I ) . Thus , ARL4C-dependent recruitment of MMP14 to invasive pseudopods is required for cell invasion . Pancreatic cancer tissues were stained with anti-ARL4C , anti-IQGAP1 , and anti-MMP14 antibodies in the serial section . Notably , ARL4C and MMP14 were expressed more highly in invasive cancer cells rather than in PanIN lesions , although IQGAP1 was thoroughly expressed in tumor lesions including PanIN area ( Figure 6I ) . Using triple immunofluorescence imaging assay , it was confirmed that three proteins are simultaneously expressed in PDAC cells invading the surrounding interstitial tissues ( Figure 6—figure supplement 1J ) . Taken together , these results support the idea that the ARL4C–IQGAP1–MMP14 signaling axis participates in pancreatic cancer cell invasion . To show that ARL4C is indeed involved in cancer cell invasion in vivo , the effects of subcutaneous injection of ARL4C ASO-1316 on an orthotopic transplantation model were tested . S2-CP8 cells expressing luciferase were injected into the pancreas of nude mice , and control ASO or ARL4C ASO-1316 was subcutaneously injected from day 3 ( Figure 7A ) . After 2 and 3 weeks , ARL4C ASO-1316 suppressed the luminescence signal compared with control ASO ( Figure 7B ) , and ARL4C expression was decreased immunohistochemically ( Figure 7C ) . Whereas ARL4C ASO-1316 did not reduce the size of the primary tumor in the pancreas , the ASO decreased the numbers of lymph node metastases and tended to improve the survival ( Figure 7D and E; Figure 7—figure supplement 1A ) . When 6-FAM–labeled ARL4C ASO-1316 was subcutaneously injected into tumor-bearing mice , the fluorescence was extremely detected in the pancreas and slightly observed in the kidney which is due to renal excretion ( Figure 7F ) . 6-FAM–labeled ARL4C ASO-1316 was highly accumulated in tumor lesions but not in the neighboring normal tissues ( Figure 7G ) , indicating that ASO was incorporated into tumor lesions after systemic injection . In primary pancreatic tumors , ARL4C ASO-1316 reduced ARL4C expression at protein and mRNA levels ( Figure 7H; Figure 7—figure supplement 1B ) and decreased the localization of IQGAP1 to the cell surface area ( Figure 7I; Figure 7—figure supplement 1C ) . Tumor cells were observed in lymphatic vessels of peritumoral areas of control ASO-treated mice but not in those of ARL4C ASO-treated mice ( Figure 7J ) . In addition , tumor cells surrounding peritumoral lymphatic vessels were also decreased , which is consistent with our hypothesis that ARL4C is required for cell invasive activity . To compare molecular characteristics between pancreatic tumors from mice injected with control ASO and ARL4C ASO-1316 , RNA sequence analysis was performed for primary tumors ( Figure 7—source data 2 ) . Principal component analysis ( PCA ) indicated a clear difference in the gene expression profiles of tumors from control ASO- and ARL4C ASO-1316–treated mice ( Figure 7K ) . Furthermore , hierarchical clustering revealed a drastic change in expression of genes due to ARL4C ASO-1316 injection ( Figure 7L ) . Two hundred and three differentially expressed genes ( DEGs ) were detected , and by subjecting them to Ingenuity Pathway Analysis ( IPA ) , the top five significantly enriched terms of the biological process of molecular function in the inhibition and activation of the pathways were obtained ( Figure 7M ) . In particular , DEGs linked to the inhibition of the pathways in ARL4C ASO-1316–treated mice were predicted to be involved in terms such as cell migration and invasion ( Figure 7M ) . Taken together , these results suggest that ARL4C ASO inhibits the invasion of tumor cells into lymphatic vessels in vivo , and the gene profiles of tumors treated with ARL4C ASO in vivo support the putative functions of ARL4C in pancreatic cancer invasion . Pancreatic cancer represents one of the leading causes of cancer death , despite advances in cancer therapy ( Keleg et al . , 2003 ) . Major problem of pancreatic cancer is uncontrollable invasion and metastasis . In this study , we found that the ARL4C–IQGAP1–MMP14 signaling axis is involved in pancreatic cancer invasion . Because ARL4C expression is induced by Wnt and EGF signaling , it is reasonable that ARL4C would be expressed in a β-catenin- and RAS-dependent manner in pancreatic cancer cells . ARL4C is a unique small G protein because it is constitutively active , regardless of wild-type ( Burd et al . , 2004; Matsumoto et al . , 2017 ) . The long interswitch region of ARL4C may prevent the retractile conformation change in the GDP-bound state ( Burd et al . , 2004; Pasqualato et al . , 2002 ) . ARL4C could be a constitutively active form without active mutations , and its activity may be controlled by transcriptional regulation . ARL4C binds to cytohesin2 ( Hofmann et al . , 2007 ) , leading to activation of ARF6–RAC–RHO–YAP/TAZ signaling in colon and lung cancer cells ( Fujii et al . , 2015; Kimura et al . , 2020 ) . Because ARL4C did not bind to cytohesin2 but to IQGAP1 in pancreatic cancer cells , it is likely that ARL4C regulates different downstream signaling pathways in a cancer cell context-dependent manner . Invadopodia are the unique structures observed at the ventral sites of certain types of cancer cells , such as BxPC-3 , breast cancer MDA-MB-231 cells , and head and neck squamous carcinoma SCC61 cells ( Dalaka et al . , 2020; Murphy and Courtneidge , 2011 ) , but the typical invadopodia are not observed in S2-CP8 and PANC-1 cells . Invasive pseudopods that we defined in these pancreatic cancer cells highly expressing ARL4C consisted of similar molecules , including cortactin , ARPC2 , IQGAP1 , and MMP14 , which are involved in invadopodia functions ( Caswell and Zech , 2018; Jacquemet et al . , 2013; Murphy and Courtneidge , 2011 ) . ARL4C depletion severely suppresses the localization of IQGAP1 and MMP14 to pseudopods and inhibits invasive ability , but does affect the localization of cortactin as well as the structure itself only moderately . Therefore , major function of ARL4C in invasive pseudopods would be to recruit MMP14 by binding to IQGAP1 rather than pseudopod formation . Both myristoylation and the PBR of ARL4C support plasma membrane targeting ( Heo et al . , 2006 ) . In our results , both motifs were necessary for the localization of ARL4C to the plasma membrane , whereas the PBR , rather than myristoylation , was indispensable for the activity of the ARL4C–IQGAP1–MMP14 signaling axis . Phosphoinositides have been implicated in many aspects of cell physiology ( Di Paolo and De Camilli , 2006 ) . PIP3 is localized to the leading edge of migrating cells and invadopodia of cancer cells ( Saykali and El-Sibai , 2014 ) and recruits cytosolic proteins containing lipid-binding domains , such as the PH domain , to the plasma membrane ( Toker and Cantley , 1997 ) . ARL4C in pancreatic cancer cells preferred PIP3 to PIP2 . Because PI3 kinase is one of the direct effector proteins of RAS ( Castellano and Downward , 2011; Rodriguez-Viciana et al . , 1994 ) , RAS-dependent PI3 kinase activation and ARL4C expression could co-operatively function to promote pancreatic cancer invasion . In conclusion , this study clarified that invasion of pancreatic cancer cells is promoted by ARL4C , of which expression is induced by KRAS and Wnt signaling , and that association of ARL4C with IQGAP1 and MMP14 at the tips of invasive pseudopods are essential for the invasive ability . The novel functions of ARL4C were confirmed by the mouse model . The inhibition of ARL4C expression by ARL4C ASO could directly inhibit invasive ability of pancreatic cancer cells and may indirectly affect the genes involved in invasion perhaps through the interaction between tumors and surrounding tissues . Because histological damage to the non-tumor regions was not observed after the administration of ARL4C ASO-1316 ( Harada et al . , 2019 ) , ARL4C might represent an appropriate target for pancreatic cancer therapy . HeLaS3 cells were kindly provided by Dr . K . Matsumoto ( Nagoya University , Aichi , Japan ) in May 2002 . S2-CP8 pancreatic cancer cells were purchased from Cell Resource Center for Biomedical Research , Institute of Development , Aging and Cancer , Tohoku University , in April 2014 . Lenti-X 293T ( X293T ) cells were purchased from Takara Bio Inc ( Shiga , Japan ) in October 2011 . PANC-1 cells were purchased from RIKEN Bioresource Center Cell Bank ( RIKEN BRC , Tsukuba , Japan ) in October 2014 . BxPC-3 cells were purchased from American Type Culture Collection ( ATCC , Manassas , VA , USA ) in May 2018 . HPAF-II cells were purchased from ATCC in July 2017 . S2-CP8 , X293T , HeLaS3 , and HPAF-II cells were grown in Dulbecco’s modified Eagle’s medium ( DMEM ) supplemented with 10% fetal bovine serum ( FBS ) . PANC-1 and BxPC-3 cells were grown in RPMI-1640 supplemented with 10 % FBS . All cell lines were authenticated using short tandem repeat profiling by BEX CO . , LTD ( Tokyo , Japan ) and tested negative for Mycoplasma using e-Myco Mycoplasma PCR Detection Kit ( iNtRON Biotechnology , Inc , Gyeonggi-do , Korea ) . S2-CP8 cells stably expressing GFP , ARL4C-EGFP , ARL4CG2A-EGFP , ARL4CT27N-EGFP , ARL4CQ72L-EGFP , ARL4CΔPBR-EGFP , ARL4CG2A-EGFP-Cterm , ARL4CΔPBR-EGFP-Cterm , ARL4CG2A-EGFP-GRP1PH , ARL4CG2A-EGFP-PLCδPH , ARL4C-mCherry , ARL4C-tdTomato , EGFP-IQGAP1 , and luciferase were generated using lentivirus as described previously ( Kimura et al . , 2016 ) . BxPC-3 cells stably expressing EGFP or ARL4C-EGFP were generated using lentivirus . Lentiviral vector CSII-CMV-MCS-IRES2-Bsd harboring a cDNA was transfected with the packaging vectors pCAG-HIV-gp and pCMV-VSV-G-RSV-Rev into X293T cells using Lipofectamine2000 transfection reagent ( Life Technologies/Thermo Fisher Scientific , Carlsbad , CA , USA ) . To generate S2-CP8 stable cells above , 1 × 105 parental cells/well in a 12-well plate were treated with lentiviruses and 5 μg/mL polybrene , centrifuged at 1200 x g for 30 min , and incubated for 24 h . The cells were selected and maintained in the medium containing 10 μg/mL Blasticidin S . ARL4C or IQGAP1 knockout cells were generated as previously described ( Fujii et al . , 2016 ) . The target sequences for human ARL4C , 5’-CTTCTCGGTGTTGAAGCCGA-3’ , and human IQGAP1 , 5’-CACCGTGGGGTCTACCTTGCCAAAC-3’ were designed with the help of the CRISPR Genome Engineering Resources ( http://www . genome-engineering . org/crispr/ ) . The plasmids expressing hCas9 and single-guide RNA ( sgRNA ) were prepared by ligating oligonucleotides into the BbsI site of pX330 ( addgene #42230 ) . The plasmid pX330 with sgRNA sequences targeting ARL4C , IQGAP1 and Blasticidin resistance was introduced into S2-CP8 cells using Lipofectamine LTX reagent ( Life Technologies/Thermo Fisher Scientific ) according to manufacturer’s instructions and the transfected cells were selected in medium containing 5 μg/mL Blasticidin S for 2 days . Single colonies were picked , mechanically disaggregated , and replated into individual wells of 24-well plates . ARL4C ASO-1316 and 6-carboxyﬂuorescein ( FAM ) -labeled ARL4C ASO-1316 were synthesized by GeneDesign ( Osaka , Japan ) as described ( Harada et al . , 2019 ) . The sequences of the ASOs are listed in Supplementary file 1 table 5 . S2-CP8 cell were transfected with ASOs at 10 nmol/L using RNAiMAX ( Life Technologies/Thermo Fisher Scientific ) in antibiotics-free medium . The transfected cells were then used for experiments conducted at 48 hr after transfection . Anti-ARL4C polyclonal antibody ( SAJ5550275 ) for immunoprecipitation and immunocytochemistry was generated in rabbits by immunization with recombinant human ARL4C . Antibodies used in this study are shown in Supplementary file 1 table 6 . The following drugs were used: PD184161 ( Sigma-Aldrich Co , St . Louis , MO , USA ) ; U0126 ( Promega Corp . , Madison , WI , USA ) ; Rapamycin ( Cell Signaling Technology , Beverly , MA , USA ) ; LY294002 ( Cell Signaling Technology ) ; and VivoGlo luciferin ( Promega Corp . ) . Plasmid construction pEGFPC2-IQGAP1 , pEGFP-mCyth2 , pAcGFP-mPlcd1PH , and CSII-CMV-MCS-IRES2-Bsd were kindly provided by K . Kaibuchi ( Nagoya University , Japan ) , J . Yamauchi ( Tokyo University of Pharmacy and Life Science , Japan ) , M . Matsuda ( Kyoto University , Kyoto , Japan ) , and H . Miyoshi ( RIKEN Bioresource Center , Ibaraki , Japan ) , respectively . To generate plasmid DNA with mutated codons or deletions , site-directed mutagenesis method was performed using PrimeSTAR Max DNA Polymerase ( Takara Bio Inc , Shiga , Japan ) . To generate plasmid DNA with insertions , PCR amplified fragments and linearized vector by restriction enzyme digestion were assembled using In-Fusion HD Cloning Kit ( Takara Bio Inc ) . pEGFPN3-ARL4C was constructed as previously described ( Matsumoto et al . , 2014 ) . Full length cDNAs of GRP1 and MMP14 ORF were reversely transcribed from mRNA extracted from MCF-7 cells and U2OS cells , respectively . Linear double strand oligonucleotides of the C-terminal 22 amino acids of KRAS , which includes the PBR and CAAX motifs , were synthesized , and the oligonucleotides were inserted into C terminal of ARL4C-EGFP or ARL4C-FLAG-HA using In-Fusion HD Cloning Kit ( Takara Bio Inc ) . Standard recombinant DNA techniques mentioned above were used to construct the following plasmids: pEGFPN3-ARL4C , pEGFPN3-ARL4CG2A , pEGFPN3-ARL4CT27N , pEGFPN3-ARL4CQ72L , pEGFPN3-ARL4CΔPBR , pEGFPN3-ARL4CG2A-EGFP-PLCδPH , pEGFPN3-ARL4CG2A-EGFP-GRP1PH , pEGFPN3-ARL4CG2A-EGFP-Cterm , pEGFPN3-ARL4CΔPBR-EGFP-Cterm , pEGFPC1-CHD , pEGFPC1-IQ , pEGFPC1-WW , pEGFPC1-IR , pEGFPC1-GRD , pEGFPC1-RGCT , pcDNA3-ARL4C-FLAG-HA , pcDNA3-ARL4CG2A-FLAG-HA , pcDNA3-ARL4CΔPBR-FLAG-HA , pcDNA3-ARL4CG2A-FLAG-HA-Cterm , pcDNA3-ARL4CΔPBR-FLAG-HA-Cterm , pcDNA3-FLAG-HA-IQGAP1 , pmCherryN1-ARL4C , pmCherryN1-MMP14 , pmCherryN1-MMP14ΔC ( Δ563–582 ) , pCAG-ARL4C-tdTomato . To construct lentiviral vectors harboring EGFP , ARL4C-EGFP , ARL4CG2A-EGFP , ARL4CT27N-EGFP , ARL4CQ72L-EGFP , ARL4CΔPBR-EGFP , ARL4CG2A-EGFP-Cterm , ARL4CΔPBR-EGFP-Cterm , ARL4CG2A-EGFP-PLCδPH , ARL4CG2A-EGFP-GRP1PH , EGFP-IQGAP1 , ARLC-mCherry , MMP14-mCherry , MMP14ΔC-mCherry , ARL4C-tdTomato were cloned into CSII-CMV-MCS-IRES2-Bsd provided by Dr . H . Miyoshi ( RIKEN Bioresource Center , Ibaraki , Japan ) . The present study involved 57 presurgical untreated patients with PDAC and ages ranging from 47 to 87 years ( median , 70 years ) who underwent surgical resection at Osaka University between April 2001 and April 2015 . Tumors were staged according to the Union for International Cancer Control ( UICC ) TNM staging system . Resected specimens were fixed in 10 % ( vol/vol ) formalin , processed for paraffin embedding , and were sectioned at 5 μm thickness and stained with hematoxylin and eosin ( H&E ) or immunoperoxidase for independent evaluations . The protocol for this study was approved by the ethical review board of the Graduate School of Medicine , Osaka University , Japan ( No . 13455 ) , under the Declaration of Helsinki , and written informed consent was obtained from all patients . The study was performed in accordance with Committee guidelines and regulations . Immunohistochemical studies were performed as previously described ( Fujii et al . , 2015 ) with modiﬁcation . Briefly , all tissue sections were stained using a DakoReal EnVision Detection System ( Dako , Carpentaria , CA , USA ) in accordance with the manufacturer’s recommendations . Formalin-fixed , paraffin-embedded tissue specimens for examination were sectioned at 5 μm thickness . Heat-induced epitope retrieval was performed using Decloaking Chamber NxGen ( Biocare Medical , Walnut Creek , CA , USA ) . Tissue peroxidase activity was blocked with Peroxidase-Blocking Solution ( Dako ) for 30 min , and the sections were then incubated with G-Block ( GenoStaff , Tokyo , Japan ) or Blocking One Histo ( nacalai tesque , Kyoto , Japan ) for 30 min or 10 min , respectively , to block nonspecific antibody binding sites . Tissue specimens were treated with anti-ARL4C ( 1:100 ) , anti-IQGAP1 ( 1:800 ) , or anti-MMP14 ( 1:100 ) antibody for 3 hr at room temperature . Then , the specimens were detected by incubating with goat anti-rabbit or anti-rabbit/mouse IgG-HRP for 1 h and subsequently with DAB ( Dako ) . The tissue sections were then counterstained with 0 . 1 % ( wt/vol ) hematoxylin . ARL4C expression was considered high when the total area of the tumor stained with anti-ARL4C antibody exceeded 5 % . IQGAP1 expression was considered high when the total area of the tumor stained with anti-IQGAP1 antibody exceeded 40 % . IQGAP1 staining positivity in PDAC patients was measured using HALO ( Indica Labs , Corrales , NM , USA ) . The threshold for positive or negative staining was based on the optical density of the staining: regions above the positivity threshold were scored according to the optical density threshold set in the module; weakly positive is shown in yellow and strongly positive in red . The samples were viewed and analyzed using NanoZoomer-SQ ( Hamamatsu Photonics K . K . , Shizuoka , Japan ) . The data on ARL4C and IQGAP1 mRNA expression in pancreatic adenocarcinoma were obtained from the UCSC Xena browser ( http://xena . ucsc . edu ) . Tumors and normal samples in the UCSC Xena browser were derived from The Cancer Genome Atlas ( TCGA ) and Genotype-Tissue Expression ( GTEx ) projects . Differential analysis was performed using a two-tailed Student’s t-test . The correlations of overall survival rates with ARL4C , IQGAP1 , and MMP14 expression in pancreatic cancer in TCGA datasets were analyzed using a Kaplan–Meier plotter ( http://www . kmplot . com ) and visualized using GraphPad Prism 8 ( GraphPad Software . San Diego , CA , USA ) . High and low expression groups were classified by auto select best cutoff . p Values and r values were calculated using GraphPad Prism . Collagen gels were made by diluting and neutralizing rat tail type I collagen ( Corning Inc , Corning , NY , USA ) in PBS and 12 . 1 mM NaOH , and were adjusted to 2 mg/mL . DQ-collagen type I ( Life Technologies/Thermo Fisher Scientific , Carlsbad , CA , USA ) was mixed with collagen gels at a final concentration of 25 μg/mL . The gel channel of 3D microfluidic cell culture chip ( AIM Biotech , Biopolis Rd , Singapore ) was filled with collagen solution and incubated at 37 °C for at least 1 hr to polymerize collagen . After hydration of medium channels , a cell suspension ( 1 × 104 cells ) in serum-free cell culture medium with 0 . 2 % BSA was injected into one of the ports at the medium channel . The opposite medium channel was filled with cell culture medium containing 10 % FBS to create a chemoattractant gradient across the collagen gel . The cells were then incubated for 3 days and fixed for 15 min at room temperature in PBS containing 4 % ( w/v ) paraformaldehyde . Then , the cells were permeabilized and blocked in PBS containing 0 . 5 % ( w/v ) Triton X-100 and 40 mg/mL BSA for 30 min and stained with the indicated antibodies . The samples were viewed and analyzed under an LSM880 laser scanning microscope ( Carl Zeiss , Jana , Germany ) . Reconstruction of confocal z-stack images into 3D animations and analysis of 4D images were performed using Imaris ( Bitplane , Belfast , UK ) . QCM Gelatin Invadopodia Assay ( Red ) ( Merck Millipore , Burlington , MA , USA ) was used in accordance with the manufacturer’s protocol . Briefly , poly-L-lysine–coated coverslips were treated with glutaraldehyde . The coverslips were then incubated with Cy3-labeled gelatin , followed by culture medium quenching of free aldehydes . Cells ( 6 × 104 cells ) were seeded onto the gelatin-coated coverslips and incubated for 4 hr . After incubation , the cells were fixed for 20 min at room temperature in phosphate-buffered saline ( PBS ) containing 4 % ( w/v ) paraformaldehyde and permeabilized in PBS containing 0 . 2 % ( w/v ) Triton X-100 for 10 min . After being blocked in PBS containing 0 . 2 % ( w/v ) BSA for 30 min , the cells were immunohistochemically stained . The samples were viewed and analyzed under an LSM880 laser scanning microscope ( Carl Zeiss , Jana , Germany ) . Cells grown on glass coverslips coated with poly-D-lysine ( Sigma-Aldrich ) or Matrigel Growth Factor Reduced ( Corning ) were fixed for 10 min at room temperature in PBS containing 4 % ( w/v ) paraformaldehyde and permeabilized in PBS containing 0 . 1 % ( w/v ) saponin ( Sigma-Aldrich ) or 0 . 2 % ( w/v ) Triton X-100 for 10 min . The cells were then blocked in PBS containing 0 . 2 % ( w/v ) BSA for 30 min . They were then incubated with primary antibodies for 3 hr at room temperature and with secondary antibodies in accordance with the manufacturer’s protocol ( Life Technologies/Thermo Fisher Scientific ) . For cell surface MMP14 staining , samples were incubated with anti-MMP14 antibody for 3 hr at room temperature without permeabilization . The samples were viewed and analyzed under an LSM880 laser scanning microscope ( Carl Zeiss ) . Migration and invasion assays were performed using a modiﬁed Boyden chamber ( 6 . 5 mm Transwell with 8 . 0 µm Pore Polycarbonate Membrane Insert; Corning ) and a Matrigel-coated modiﬁed Boyden chamber ( BioCoat Matrigel Invasion Chambers with 8 . 0 µm PET Membrane; Corning ) , respectively as described previously ( Kurayoshi et al . , 2006; Matsumoto et al . , 2014 ) . In the standard conditions , S2-CP8 cells ( 2 . 5 × 104 cells ) were seeded in the upper side of Boyden Chamber . In GFP-expressing S2-CP8 cells , after 4 h ( migration assay ) or 24 hr ( invasion assay , except for Figure 6E ) incubation with control ASO , 122 cells ( average ) and 126 cells ( average ) , respectively , were observed in the lower side chamber in the one field of view under fluorescence microscope ( BZ-9000 , Keyence , Osaka , Japan ) using a 10 x air objective . In Figure 6E , cells were observed after 20 hr incubation with ASO . Migration and invasion rates of cells expressing ARL4C , IQGAP1 , and MMP14 mutants were calculated as the percentages of the same cells transfected with control ASO or siRNA . Collagen gels were made by diluting and neutralizing rat tail type I collagen ( Corning ) in PBS and 12 . 1 mM NaOH , and were adjusted to 2 mg/mL . Then , 140 µL of cell-embedded collagen gels ( 1 × 106 cells/mL ) were overlaid onto glass coverslips in a 24-well plate and allowed to polymerize for at least 1 hr at 37 °C and 5 % CO2 . After polymerization , growth medium was added on top of the collagen gel . The cells were then incubated for 3 days and fixed for 15 min at room temperature in PBS containing 4 % ( w/v ) paraformaldehyde . Then , the cells were permeabilized and blocked in PBS containing 0 . 5 % ( w/v ) Triton X-100 and 40 mg/mL BSA for 30 min and incubated with primary antibodies for 3 h at room temperature and secondary antibodies in accordance with the manufacturer’s protocol ( Life Technologies/Thermo Fisher Scientific ) . The samples were viewed and analyzed under an LSM880 laser scanning microscope using a 20 x air objective ( Carl Zeiss ) . In the standard conditions ( for Figure 3L ) with BxPC-3/ARL4C-GFP cells treated with control ASO , the number of cells with pseudopods and the total number of cells were 15 ( average ) and 76 ( average ) , respectively , in the one field of view under an LSM880 laser scanning microscope ( Carl-Zeiss ) using a 20 x air objective . The percentages of cells with pseudopods compared with the total number of cells in the presence of control siRNA or IQGAP1 siRNA were calculated . Inducible recruitment of phospholipid phosphatases mRFP-FKBP-5-ptase-dom and PM-FRB-CFP plasmids were obtained from Addgene ( deposited by the laboratory of T . Balla ) . S2-CP8 cells were then transiently transfected with both mRFP-FKBP-5-ptase-dom and PM-FRB-CFP ( 0 . 5 μg/well of a six-well plate for each vector ) with ViaFect ( Promega Corp . ) . After 24 hr culture , the cells were treated with 100 nM rapamycin or 50 μM LY294002 for 30 min before fixation . Cells transfected with control ASO or ARL4C ASO-1316 were cultured on Matrigel coated dish for 3 . 5 days , and dissociated using TrypLE Express ( Thermo Fisher Scientific ) . Suspension of cells was stained with Hoechst 33342 or propidium iodide ( PI ) using Annexin V-FITC Apoptosis Detection Kit ( nacalai tesque ) . The samples were viewed and analyzed under an LSM880 laser scanning microscope ( Carl Zeiss ) , and the number of PI-positive cells was divided by the total number of nuclei stained with Hoechst 33342 . Confluent X293T cells transiently transfected with ARL4C-FLAG-HA in two 10 cm culture dishes were harvested and lysed in 800 μL of lysis buffer ( 25 mM Tris-HCl [pH7 . 5] , 50 mM NaCl , 0 . 5 % TritonX-100 ) with protease inhibitors ( nacalai tesque ) . After 10 min of centrifugation , lysates were incubated with 40 μL of 50 % slurry of anti-FLAG Affinity Gel ( Sigma-Aldrich ) for 30 min , and then add another 40 μL and incubated for 30 min . Beads were washed three times with 1 mL of lysis buffer . Recovered beads were incubated once with FLAG peptide ( 0 . 5 mg/mL ) to elute proteins in 80 μL of PBS for 30 min at 4 °C . Then , the supernatant was precleaned with 40 μL of 50 % slurry of protein A Sepharose beads ( GE Healthcare , Chicago , IL , USA ) for 30 min at 4 °C . The precleaned lysates were incubated with 2 μg of anti-HA antibody ( Santa Cruz , Dallas , TX , USA ) and 50 μL of 50 % slurry of protein A Sepharose beads for 1 hr at 4 °C . Beads were washed three times with 1 mL of lysis buffer , and bound complexes were dissolved in 50 μL of Laemmli’s sample buffer . The ARL4C-FLAG-HA-interacting proteins were detected by Pierce Silver Stain for Mass Spectrometry ( Life Technologies/Thermo Fisher Scientific ) . Six bands ( arrowheads in Figure 3A ) were cut from the gel and analyzed by mass spectrometry . Immunoprecipitation was performed as described previously with modification ( Matsumoto et al . , 2014 ) . For Figure 3C S2-CP8 cells ( 60 mm diameter dish ) were lysed in 300 µL of lysis buffer ( 25 mM Tris–HCl pH 7 . 5 , 50 mM NaCl , 0 . 5 % Triton-X100 ) with protease inhibitors ( nacalai tesque ) for 10 min on ice . After centrifugation , the supernatant was collected and pre-cleaned using 30 µL of Dynabeads Protein G ( Thermo Fisher Scientific ) . After pre-cleaning , lysates were rotated with complex of Dynabeads ( 50 μL ) and antibody ( 3 . 6 μg ) for 10 min at room temperature . The beads were then washed with lysis buffer three times , and finally suspended in Laemmli’s sample buffer . The RAC1 activity assay was performed as described ( Matsumoto et al . , 2014 ) . Briefly , cells were lysed in 400 μL of RAC1 assay buffer ( 20 mM Tris–HCl [pH 7 . 5] , 150 mM NaCl , 1 mM dithiothreitol , 10 mM MgCl2 , 1 % Triton-X100 ) with protease inhibitors ( nacalai tesque ) containing 20 μg of glutathione-S-transferase ( GST ) -CRIB . After the lysates were centrifuged at 20 , 000 g for 10 min , the supernatants were incubated with glutathione-Sepharose ( 20 μl each ) for 2 h at 4 °C . The beads were then washed with RAC1 assay buffer three times , and finally suspended in Laemmli’s sample buffer . The precipitates were probed with the anti-RAC1 antibody . Orthotopic transplantation was performed as described previously ( Kim et al . , 2009 ) . Ten days after the transplantation , 150 μg/animal ( approximately 7 . 5 mg/kg ) of 6-FAM-ARL4C ASO-1316 was subcutaneously administered . Four h after the injection , the fluorescence intensities of various organs were measured ex vivo using the IVIS imaging system ( Xenogen Corp . ) . After ex vivo imaging , unfixed mouse pancreas tissues were frozen in an O . C . T . Compound ( Sakura Finetek , Tokyo , Japan ) /sucrose mixture [1:1 ( v/v ) OCT and 1 x PBS containing 30 % sucrose] . Freshly frozen tissues were sectioned at 10 μm and fixed for 30 min at room temperature in PBS containing 4 % ( w/v ) paraformaldehyde . The cells were then permeabilized and blocked in PBS containing 0 . 5 % ( w/v ) Triton X-100 and 40 mg/mL BSA for 30 min and stained with the indicated antibodies . The samples were viewed and analyzed under an LSM880 laser scanning microscope ( Carl Zeiss ) . An orthotopic transplantation assay was performed as described previously ( Kim et al . , 2009 ) with modification . Ten-week-old male BALB/cAJcl-nu/nu mice ( nude mice; CLEA , Tokyo , Japan ) were anesthetized and received an orthotopic injection of S2-CP8 cells into the mid-body of the pancreas using a 27 G needle ( 5 × 105 cells suspended in 100 μL of HBSS with 50 % Matrigel ) . ASOs ( 50 μg/mouse , approximately 2 . 5 mg/kg ) were administered subcutaneously twice a week from day 3 . To evaluate the knockdown efficiency of ARL4C ASO-1316 , tumor tissues were harvested from tumor-bearing mice 8 days after transplantation . Total RNAs were isolated using NucleoSpin RNA ( MACHEREY-NAGEL GmbH & Co . KG , Dueren , Germany ) , and complementary DNAs were synthesized using ReverTra Ace qPCR RT Master Mix ( TOYOBO , Osaka , Japan ) . For extraction of tissue proteins , tumor samples were lysed in 150 μL of lysis buffer ( 20 mM Tris–HCl [pH 8 . 0] , 137 mM NaCl , 10 % glycerol , 1% NP40 ) and homogenized using Biomasher II ( KANTO CHEMICAL CO . , Inc , Tokyo , Japan ) . Debris was removed by centrifugation and finally suspended in Lammli’s buffer . Protein concentration was determined with Pierce BCA Protein Assay Kit ( Thermo Fisher Scientific ) . The band intensities of western blotting were calculated using Image J ( National Institutes of Health , USA ) . To assess the effect of ARL4C ASO-1316 on tumor progression , tumor burden was measured once a week using the IVIS imaging system ( Xenogen Corp . , Alameda , CA , USA ) . For the in vivo imaging , 100 μL of VivoGlo luciferin ( 30 mg/mL ) was intraperitoneally administered and the bioluminescence imaging was performed 8 min later . The region of interest ( ROI ) was selected and the radiance values were measured with Living Image 4 . 3 . 1 Software ( Caliper Life Sciences , Hopkinton , MA , USA ) . The mice were euthanized 28 days after transplantation . Tumor weights and numbers of mesenteric lymph nodes ( diameter of lymph nodes > 1 mm ) were measured . All protocols used for the animal experiments in this study were approved by the Animal Research Committee of Osaka University , Japan ( No . 26-032-048 ) . Sequenced reads were preprocessed by Trim Galore ! v0 . 6 . 3 and quantified by Salmon v0 . 14 . 0 with the flags gcBias and validateMappings . GENCODE vM21 annotation was used as the transcript reference . The quantified transcript-level scaled TPM was summarized into a gene-level scaled TPM by using the R package tximport v1 . 6 . 0 . All procedures were implemented using the RNAseq pipeline ikra v1 . 2 . 2 [https://zenodo . org/record/3606888 ( Yu et al . , 2019 ) ] with the default parameters . Downstream analysis was conducted with an integrative RNAseq analysis platform , iDEP . 90 . After normalization with VST , principal component analysis was conducted . Hierarchical clustering was performed on the top 1 , 000 genes in terms of their standard deviation . Finally , DEGs were selected with a log2 fold change >1 and false discovery rate <0 . 1 . DEGs identified from RNA sequence data were subjected to Ingenuity Pathway Analysis ( IPA; Qiagen , Hilden , Germany ) . This analysis examines DEGs that are known to affect each biological function and compares their direction of change to what is expected from the literature . To infer the activation states of implicated biological functions , two statistical quantities , Z-score and p value , were used . A positive or negative Z-score value indicates that biological functions are predicted to be activated or inhibited in the ARL4C ASO-1316–treated group relative to the control ASO-treated group . A negative Z-score means that the indicated biological functions are inhibited by ARL4C ASO-1316 . The p value , calculated with the Fisher’s exact test , reflects the enrichment of the DEGs on each pathway . For stringent analysis , only biological functions with a \|Z-score\| > 2 were considered significant . Biological replicates are replicates on independent biological samples versus technical replicates that use the same starting samples . All experiments in this study were repeated using biological replicates . A minimum of three biological replicates were analyzed for all samples , and the results are presented as the mean ± s . d . or s . e . m . The cumulative probabilities of overall survival were determined using the Kaplan–Meier method; a log-rank test was used to assess statistical significance . The Student’s t-test or Mann–Whitney test was used to determine if there was a significant difference between the means of two groups . One-way analysis of variance ( ANOVA ) with Bonferroni tests was used to compare three or more group means . Statistical analysis was performed using Excel Toukei ( ESUMI Co . , Ltd . , Tokyo , Japan ) and GraphPad Prism 8 ( GraphPad Software , La Jolla , CA , USA ) ; p < 0 . 05 was considered statistically significant . In box and whiskers plots , the top and bottom horizontal lines represent the 75th and the 25th percentiles , respectively , and the middle horizontal line represents the median . The size of the box represents the interquartile range and the top and bottom whiskers represent the maximum and the minimum values , respectively . The siRNAs and primers used in these experiments are listed in Supplementary file 1 tables 7 and 8 , respectively . 2 . 5D Matrigel growth assay and quantitative PCR were performed as described previously ( Matsumoto et al . , 2019; Sato et al . , 2010 ) .	Most cases of pancreatic cancer are detected in the later stages when they are difficult to treat and , as a result , survival is low . Over 90% of pancreatic cancers contain genetic changes that increase the activity of a protein called KRAS . This hyperactive KRAS drives cancer growth and progression . Attempts to treat pancreatic cancer using drugs that reduce the activity of KRAS have so far failed . The KRAS protein can accelerate growth in healthy cells as well as in cancer and it does this by activating various other proteins . Drugs that target some of these other proteins could be more effective at treating pancreatic cancer than the drugs that target KRAS . One of these potential targets is called ARL4C . ARL4C is active during fetal development , but it is often not present in adult tissues . Harada et al . investigated whether the protein is important in pancreatic cancer , and what other roles it has in the body , to better understand if it is a good target for cancer treatment . First , Harada et al . used cells grown in the lab to show that ARL4C contributes to the aggressive spread of human pancreatic cancers . Using mice , Harada et al . also showed that blocking the activity of ARL4C in pancreatic cancers helped to slow their progression . Harada et al . ’s results suggest that ARL4C could be a good target for new drugs treating pancreatic cancers . Given that this protein does not seem to have important roles in the cells of adults , targeting it is unlikely to have major side effects . Further investigation of ARL4C in more human-like animal models will help to confirm these results .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "cancer", "biology" ]	2021	Localization of KRAS downstream target ARL4C to invasive pseudopods accelerates pancreatic cancer cell invasion
To better understand smoking cessation , we examined the actions of varenicline ( Chantix ) during long-term nicotine exposure . Varenicline reduced nicotine upregulation of α4β2-type nicotinic receptors ( α4β2Rs ) in live cells and neurons , but not for membrane preparations . Effects on upregulation depended on intracellular pH homeostasis and were not observed if acidic pH in intracellular compartments was neutralized . Varenicline was trapped as a weak base in acidic compartments and slowly released , blocking 125I-epibatidine binding and desensitizing α4β2Rs . Epibatidine itself was trapped; 125I-epibatidine slow release from acidic vesicles was directly measured and required the presence of α4β2Rs . Nicotine exposure increased epibatidine trapping by increasing the numbers of acidic vesicles containing α4β2Rs . We conclude that varenicline as a smoking cessation agent differs from nicotine through trapping in α4β2R-containing acidic vesicles that is selective and nicotine-regulated . Our results provide a new paradigm for how smoking cessation occurs and suggest how more effective smoking cessation reagents can be designed . Tobacco continues to be widely used world-wide , primarily via cigarette smoking , and is the leading cause of preventable deaths in the United States ( National Center for Chronic Disease Prevention and Health Promotion ( US ) Office on Smoking and Health , 2014 ) . The currently approved treatments for smoking cessation are nicotine replacement therapy , bupropion , and varenicline ( Chantix ) . While varenicline is the most effective , successful quit rates only reach ~50% of smokers ( Agboola et al . , 2015 ) . Other therapies or novel approaches are clearly needed to increase rates of smoking cessation , and the design of smoking cessation reagents would be greatly aided with a mechanistic understanding of how the reagents act . Nicotine binds to high-affinity nicotinic acetylcholine receptors ( nAChRs ) in the brain , and binding initiates its addictive effects . nAChRs are members of the Cys-loop family of ligand-gated ion channels , all of which are pentameric neurotransmitter receptors ( Karlin , 2002; Albuquerque et al . , 2009 ) . In the mammalian CNS , these critical binding sites are nAChRs composed of α2 - α6 and β2 - β4 subunits; the most predominant contains α4 and β2 subunits ( Lindstrom , 1996; McGehee and Role , 1995 ) . The α4β2 nAChR subtype ( α4β2R ) is closely linked to nicotine addiction ( Govind et al . , 2009; Vezina et al . , 2007; Lewis and Picciotto , 2013 ) . Loss of either subunit in α4 or β2 subunit knockout mice reduces the pharmacological and behavioral effects of nicotine ( Picciotto et al . , 1998; Marubio et al . , 2003 ) . In addition , targeted expression of β2 subunits in the brain ventral tegmental area ( VTA ) of β2-knockout mice rescues nicotine-seeking behavior and nicotine-induced dopamine release ( Maskos et al . , 2005 ) . Varenicline is a high-affinity partial agonist for α4β2R . The rationale for its design as a smoking cessation reagent was predicated on the idea that a partial agonist could compete for and reduce cell-surface α4β2R activation by nicotine , reduce dopamine overflow in the mesolimbic reward system and thereby reduce the reward sensation of smoking ( Rollema et al . , 2007 ) . Varenicline does indeed reduce the rapid effects of nicotine on α4β2Rs in VTA dopaminergic neurons and acutely reduces nicotine-induced dopamine release in the nucleus accumbens ( Rollema et al . , 2007 ) . How varenicline alters the longer-lasting effects of nicotine is less clear , however . Nicotine upregulation of nAChRs is linked to different processes in nicotine addiction , including sensitization ( Govind et al . , 2009; Vezina et al . , 2007 ) and withdrawal ( De Biasi et al . , 2011 ) and is the only effect of nicotine on nAChRs that lasts longer than a few minutes . Upregulation occurs when nicotine exposure increases high-affinity nicotine-binding sites in brain , measured by radiolabeled agonists such as nicotine ( Marks et al . , 1983; Schwartz and Kellar , 1983; Benwell et al . , 1988; Breese et al . , 1997 ) or epibatidine ( Perry et al . , 1999 ) . Chronic varenicline exposure in mice induces upregulation to the same degree , or even more than , chronic nicotine exposure when measured using 3H-epibatidine binding to membrane preparations from different brain regions ( Hussmann et al . , 2012; Turner et al . , 2011 ) . Recent studies using PET imaging in humans found that reversal of α4β2R upregulation was correlated with less smoking relapse ( Brody et al . , 2013; Staley et al . , 2006; Cosgrove et al . , 2009 ) . Thus , the observation that varenicline induced upregulation like nicotine was surprising and difficult to reconcile with its smoking cessation actions . In this study , we describe novel , long-lasting changes in nicotine upregulation by varenicline and other α4β2R ligands . The weak base nature of varenicline and other α4β2R ligands was fundamental to this action on upregulation . Nicotinic receptor ligands that are weak bases have two states , the membrane permeable uncharged state and the membrane impermeable protonated state . Nicotine accumulates in intracellular acidic compartments of cells and neurons because it is more highly protonated at the lower pH ( Brown and Garthwaite , 1979; Bhagat , 1970; Putney and Borzelleca , 1971 ) , which leads to ‘trapping’ in these compartments in Xenopus oocytes expressing α4β2Rs . After removal of extracellular nicotine , nicotine slowly leaks from the intracellular compartments and causes α4β2R desensitization at the plasma membrane ( Jia et al . , 2003 ) . However , an analogous phenomenon was not observed in mammalian cells heterologously expressing α4β2Rs ( Jia et al . , 2003 ) , raising questions as to its relevance to neuronal adaptations to chronic nicotine exposure . Here we find that varenicline , as well as lobeline and epibatidine , are trapped as weak bases within acidic vesicles of live mammalian cells and neurons; in contrast , nicotine rapidly exits these vesicles . The trapping is selective depending on ligand pKa and affinity for α4β2Rs in acidic vesicles . Using 125I-epibatidine binding , we directly measured trapping and slow release from acidic vesicles and found it required α4β2R expression and increased with nicotine upregulation . Using pH-sensitive , pHluorin-tagged α4 subunits , we found that nicotine increased the numbers of acidic vesicles containing α4β2Rs , thereby promoting accumulation of 125I-epibatidine . Our results provide a new paradigm for how smoking cessation occurs and suggests how more effective smoking cessation reagents can be designed . We determined if the smoking cessation drug and partial agonist varenicline , as well as other nAChR ligands , caused upregulation of nAChRs or altered nicotine upregulation using HEK cells stably expressing α4β2Rs or cortical neurons expressing endogenous α4β2Rs . α4β2R upregulation was assayed with two approaches: 125I-epibatidine binding ( Figure 1 ) or patch-clamp recording of ACh-evoked currents ( Figure 2 ) . Binding and function were assayed following 17–20 hr treatments of cells with control media , nAChR ligand ( e . g . , varenicline ) , nicotine ( 1 or 10 μM ) , and nAChR ligand co-incubated with nicotine . All results were normalized to the amount of binding or mean current amplitude observed with the nicotine-only condition ( i . e . , the upregulated state ) . Values for mean , variance and numbers of measures are given in the figure legends . 10 . 7554/eLife . 25651 . 003Figure 1 . Effect of smoking cessation reagents on α4β2R upregulation . ( A ) Nicotine , varenicline and lobeline chemical structures . ( B ) Varenicline and lobeline reduced nicotine upregulation of α4β2Rs for live cells , but not membranes . 125I-epibatidine binding performed on live α4β2R-expressing HEK cells ( left ) ( n = 3 ) or membranes ( right ) ( n = 3 ) . Cells were treated for 17 hr with 30 μM varenicline or 30 μM lobeline with or without 10 μM nicotine . Specific epibatidine binding was represented as % of binding relative to nicotine upregulated cells ( C ) Model of varenicline ( Var ) trapping in acidic vesicles . Varenicline is trapped when protonated in the acidic vesicle lumen . α4β2Rs on the plasma membrane are depicted in two states: α4β2Rs with and without high-affinity binding sites consistent with the findings of Vallejo et al . ( Benowitz et al . , 2009 ) . ( D ) An intracellular acidic compartment is required for varenicline and lobeline effects on upregulation . 125I-epibatidine binding performed on live α4β2R HEK cells with 20 mM NH4Cl treatment for 10 min . Cells were exposed to 30 μM varenicline or 30 μM lobeline with or without 10 μM nicotine for 17 hr as in ( B ) ( n = 4 ) . ( E ) A distribution plot comparing the reduction in upregulation ( 125I-epibatidine binding ) by varenicline and lobeline to the recovery after disruption of intracellular pH gradient by membrane preparation or various agents that raise pH in intact cells . Each point represents the mean and the standard error of the mean ( s . e . m ) from the indicated columns in Figure 1B , D , and Figure 1—figure supplement 2A , B . ( F ) 125I-epibatidine binding on live cortical neurons without ( left ) or with ( right ) 20 mM NH4Cl treatment . Neurons were exposed to varenicline or lobeline as in B ) in the presence or absence of 1 μM nicotine ( n = 3 ) . ( G ) Varenicline reduced nicotine induced upregulation of human α4β2Rs . 125I-epibatidine binding was performed on live cells either stably expressing rat α4β2R or transiently expressing human α4β2Rs . HEK cells were transfected with human α4 and β2 subunits for 24 hr . Cells were treated with 100 nM or 30 μM varenicline in the presence of 10 μM nicotine for 17 hr prior to 125I-epibatidine binding ( n = 3 ) . In all the column graphs in ( B , D , F , G ) : p<0 . 05; p<0 . 00 one by one-way ANOVA with Bonferroni’s multiple comparison test; n indicates number of independent experiments performed on separate days and cultures . Columns represent group mean and error bars are the standard error of the mean ( s . e . m ) . DOI: http://dx . doi . org/10 . 7554/eLife . 25651 . 00310 . 7554/eLife . 25651 . 004Figure 1—figure supplement 1 . Dose dependence of varenicline and lobeline effects on upregulation . ( A ) Dose dependence of varenicline upregulation . α4β2R-expressing cells were treated with increasing concentration of varenicline for 17 hr . Cells were washed with PBS prior to 125I-epibatidine binding and with ( red-filled circles ) or without 20 mM NH4Cl ( black-filled circles ) treatment for 10 min ( n = 3 ) . ( B ) Dose dependence of varenicline on nicotine upregulation . Cells were treated with increasing concentration of varenicline with 10 μM nicotine for 17 hr and treated with ( red-filled squares ) or without 20 mM NH4Cl ( black-filled squares ) treatment for 10 min as in ( A ) ( n = 3 ) . ( C ) Dose dependence of lobeline upregulation . α4β2R-expressing cells were treated with increasing concentration of lobeline for 17 hr and treated with ( blue-filled circles ) or without 20 mM NH4Cl ( black-filled circles ) treatment for 10 min as in A ( n = 3 ) . ( D ) Dose dependence of lobeline on nicotine upregulation ( n = 3 ) . For all the points in ( A , B , C , D ) : error bars represent mean ± s . e . m . n indicates number of independent experiments performed on separate days and cultures . DOI: http://dx . doi . org/10 . 7554/eLife . 25651 . 00410 . 7554/eLife . 25651 . 005Figure 1—figure supplement 2 . Altering intracellular pH of acidic vesicles alters the effects of varenicline and lobeline . ( A ) 125I-epibatidine binding on α4β2R-expressing HEK cells exposed to varenicline or lobeline 17 hr at 30 μM and then incubated twice ( 5 min each ) with Bafilomycin A ( 50 nM ) , an inhibitor of vacuolar type H+ -ATPase ( proton pump ) , prior to performing radio-ligand binding ( n = 3 ) . ( B ) 125I-epibatidine binding on α4β2R-expressing HEK cells exposed to varenicline or lobeline 17 hr at 30 μM and then incubated twice ( 5 min each ) with a weak base chloroquine ( 150 μM ) prior to performing radio-ligand binding ( n = 3 ) . In column graphs in A and B: *p<0 . 001 by one-way ANOVA with Bonferroni’s multiple comparison test . n indicates number of independent experiments performed on separate days and cultures and error bars indicate mean ± s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 25651 . 00510 . 7554/eLife . 25651 . 006Figure 1—figure supplement 3 . Release by NH4Cl treatment of α4β2R ligands trapped inside the acidic compartment . ( A ) Schematic of the experimental details . ( National Center for Chronic Disease Prevention and Health Promotion ( US ) Office on Smoking and Health , 2014 ) . HEK cells stably expressing α4β2Rs were untreated or treated with 30 μM varenicline or lobeline for 17 hr . Cells incubated in 20 mM NH4Cl/PBS for 10 min . ( Agboola et al . , 2015 ) . Untreated cells receiving NH4Cl/PBS from untreated , varenicline or lobeline pre-treated cells . ( B ) 125I-epibatidine binding was performed on untreated cells receiving NH4Cl/PBS from cells that had been treated as in ( Agboola et al . , 2015 ) . Error bars represent mean ± standard deviation ( s . d ) from triplicate samples . DOI: http://dx . doi . org/10 . 7554/eLife . 25651 . 00610 . 7554/eLife . 25651 . 007Figure 2 . Effect of smoking cessation reagents on functional upregulation . ( A ) In this and other recordings , a 17–20 hr treatment with nicotine ( Nic , 10 μM ) induced a robust , ~5 fold , increase in peak current amplitudes evoked by 1 mM ACh from α4β2R-expressing HEK cells . Varenicline ( Var , 30 μM ) treatment for an equivalent time had no effect on ACh-evoked peak current amplitudes alone but prevented upregulation of nAChR function when co-incubated with nicotine . Traces represent currents evoked by ACh in cells that were untreated or treated for 17 hr with vehicle , nicotine , varenicline , or with varenicline and nicotine . The number of recordings were 16 , 18 , 18 , and 17 , respectively . ( B ) Chronic lobeline ( Lob , 30 μM ) exposure reduced the peak ACh current amplitude relative to control and prevent nicotine upregulation . Traces represent currents evoked by ACh in cells that were untreated or treated for 17 hr with vehicle , nicotine , lobeline , or with lobeline and nicotine . The number of recordings were 13 , 18 , 13 , and 15 , respectively . ( C ) NH4Cl treatment ( two 10 min washes ) partially reverses the suppressive effects of varenicline on nicotine-induced functional upregulation . Traces show representative currents evoked by ACh from cells pretreated for 17–20 hr with vehicle , nicotine , or nicotine and varenicline before exposure to NH4Cl . The varenicline-alone condition is omitted for clarity . The graph is as in ( A ) except that the cells were treated with NH4Cl as noted . The profound reduction in nicotine-induced upregulation caused by varenicline co-incubation was lessened following treatment with NH4Cl . The number of recordings were 21 , 22 , 25 , and 25 , respectively . ( D ) Traces show representative currents evoked by ACh from cells pretreated for 17–20 hr with vehicle , nicotine , or nicotine and lobeline before exposure to NH4Cl . In this dataset , lobeline-induced suppression of nicotine upregulation was modestly attenuated by treatment with NH4Cl . The number of recordings were 14 , 15 , 16 , and 19 , respectively . In all the column graphs: p<0 . 05; p<0 . 01; p<0 . 001 by one-way ANOVA with Tukey’s multiple comparison test . Columns show group mean and error bars are the s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 25651 . 007 Varenicline ( Figure 1A ) at 30 μM did not upregulate 125I-epibatidine binding in α4β2R-expressing HEK cells ( Figure 1B , ‘Var’ , red columns in the left , ‘live cells’ half of the graph ) nor in cortical neurons ( Figure 1F , left ) following 17 hr of exposure . However , a full concentration-response curve with chronic varenicline revealed an inverted U-shaped dependence of upregulation of binding that peaked at 60–70% of that observed with nicotine , which occurred at varenicline concentrations between 100 nM and 1 μM ( Figure 1—figure supplement 1A , black circles ) . When co-applied with nicotine , varenicline ( 30 μM ) prevented nicotine upregulation of binding in both heterologous cells and cortical neurons ( Figure 1B , F ) . The reversal of the actions of nicotine on α4β2R binding also was concentration-dependent , with upregulation decreasing exponentially with increased varenicline concentration ( Figure 1—figure supplement 1B , black squares ) . Thus , chronic varenicline exposure precludes nicotine upregulation of binding when assayed in living cells . Recent studies indicate that varenicline concentrations in the brain are likely to be higher than in plasma because of the presence of high-affinity binding to nAChRs ( Rollema et al . , 2010 ) . Based on this study’s estimates , unbound varenicline concentrations in the brain fall in the range of 32 to 131 nM and plasma concentrations are approximately 4-fold lower . In addition , a previous study found differences in varencilcine binding to human α4β2Rs compared to rat α4β2Rs when expressed in Xenopus oocytes ( Papke et al . , 2010 ) . In order to compare , human and rat α4β2Rs at relevant concentrations , we measured the effect of varenicline on nicotine upregulation at 0 , 100 nM and 30 μM . We found that at 100 nM varenicline nicotine upregulation was significantly reduced to ~60% of the level without varencline ( Figure 1G ) for both rat and human α4β2Rs . These results indicate that varenicline should reduce nicotine upregulation at concentrations expected to exist in brain . In contrast , varenicline is unlikely to act as a partial agonist at these concentrations ( Rollema et al . , 2010 ) . We also found no significant differences between rat and human α4β2Rs at 0 , 100 nM and 30 μM varenicline ( Figure 1G ) . Another nAChR partial agonist , lobeline , reported to have smoking cessation activity ( Damaj et al . , 1997; Miller et al . , 2003 ) had effects similar to those of varenicline ( Figure 1A ) . Chronic exposure to lobeline alone or together with nicotine reduced 125I-epibatidine binding levels below that observed for the untreated cells and neurons ( ‘Lob’ , blue columns in the left , ‘live cells’ half of the graphs in Figure 1B , F ) . The inverted U-shaped concentration-dependence of lobeline effects on upregulation was similar to that of varenicline . 125I-epibatidine binding peaked at ~50% of nicotine-induced upregulation , was maximal at 100 nM – 1 μM , ( Figure 1—figure supplement 1C , black circles ) , and co-incubation with nicotine reduced upregulation exponentially with increasing concentrations of lobeline ( Figure 1—figure supplement 1D , black squares ) . Lobeline therefore has very similar effects as varenicline in preventing nicotine upregulation . Intact HEK cells or neurons were required to observe the block by varenicline and lobeline of nicotine upregulation . If 125I-epibatidine binding was performed on membrane fragments instead of intact cells , varenicline upregulated 125I-epibatidine binding in α4β2R-expressing HEK cells to the same degree as nicotine and did not reduce nicotine upregulation ( Figure 1B , red columns on right , ‘membranes’ half of the graph ) . Lobeline exposure upregulated 125I-epibatidine binding to about 50% of that of observed with nicotine treatment and reduced nicotine upregulation by 50% ( Figure 1B , blue columns on right half of the graph ) . The effects of varenicline and lobeline on nicotine upregulation were therefore distinct in live cells and membrane fragments , which could explain why previous studies found that chronic varenicline exposure upregulated 3H-epibatidine binding to brain tissue to the same degree as nicotine and reported no effect of varenicline exposure on nicotine upregulation ( Hussmann et al . , 2012; Turner et al . , 2011 ) . These experiments were done with membrane preparations , whereas the suppressive activity of varenicline on nicotine upregulation was only observed in intact cells . An attractive hypothesis to explain why intact cells were required for suppression of upregulation is that varenicline and lobeline , as weak bases , are concentrated and trapped in intracellular acidic compartments where they are highly protonated . A schematic of how this might occur is displayed in Figure 1C . This process was described previously for nicotine in Xenopus oocytes expressing α4β2Rs ( Jia et al . , 2003 ) ; furthermore , slow release of nicotine from acidic compartments caused α4β2R desensitization , thereby reducing ACh-induced α4β2R currents . We postulated that similar processes could account for the differential effects of varenicline and lobeline exposure on nicotine-induced upregulation in live cells versus membrane fragments ( Figure 1B ) . That is , varenicline and lobeline also could be trapped due to protonation , slowly leak out after removal from the extracellular solution , and consequently reduce 125I-epibatidine binding and functional upregulation as shown in Figures 1 and 2 . To test this hypothesis , we determined if suppression of nicotine upregulation by varenicline and lobeline was relieved by increasing the pH of intracellular compartments , thereby reducing the protonation of the two nAChR ligands . The pH gradient was made more basic using three different strategies . First , live HEK cells were incubated briefly with ammonium chloride ( NH4Cl; 20 mM ) prior to 125I-epibatidine binding . We reasoned that this should result in 125I-epibatidine binding similar to that observed with the membrane preparation in Figure 1B ( right half ) as a result of shifting the equilibrium to uncharged species of the ligands at more basic pHs . NH4Cl treatment altered the effects of varenicline and lobeline on 125I-epibatidine binding; both ligands produced upregulation of binding ( Figure 1D left half , Figure 1—figure supplement 1A , C ) , in contrast to their actions at physiological pH ( Figure 1B , leftmost trio of columns ) . Moreover , NH4Cl exposure following co-incubation with nicotine relieved the suppression of upregulation by varenicline and lobeline ( Figure 1D , right trio of columns , Figure 1-figure supplement B , D ) , such that 125I-epibatidine binding in these conditions were indeed similar to the results using membrane preparations ( Figure 1B , rightmost trio of columns ) . NH4Cl treatment of cultured cortical neurons had similar effects on 125I-epibatidine binding after varenicline or lobeline exposures ( Figure 1F , right half of graph ) . We also obtained similar 125I-epibatidine binding results with HEK cells using two other methods to reduce the pH gradient - incubation with the proton pump inhibitor bafilomycin A ( Baf , 50 nM; Figure 1E , Figure 1—figure supplement 2A ) or with the weak base chloroquine ( Chlor , 150 μM; Figure 1E , Figure 1—figure supplement 2B ) . In Figure 1E is a comparison of how varenicline ( Var ) or lobeline ( Lob ) affects 125I-epibatidine binding to α4β2R-expressing HEK cells for all of these conditions . When performed on membrane preparations or in conditions that increased the pH of cellular acidic compartments , 125I-epibatidine binding was increased consistent with upregulation of α4β2Rs . This upregulation of binding was not observed for intact cells . The second assay we used for α4β2R upregulation was to measure changes in the mean amplitude of ACh-induced currents in the receptor-expressing HEK cells following exposure to nicotine or other agents ( Figure 2 ) . Cells were treated for 17–20 hr as shown; we then carried out whole-cell voltage recordings while rapidly applying ACh ( 1 mM ) for 1 s . Nicotine upregulation of currents from α4β2R-expressing HEK cells was observed as an increase by ~5 fold in mean peak current amplitudes ( e . g . , Figure 2A , white vs . black column ) . Varenicline and lobeline ( both at 30 μM ) exposure altered the ACh-induced currents in parallel with what we observed in 125I-epibatidine binding assays with live cells ( Figure 2A , B ) . Exposure to varenicline alone did not alter nAChR current amplitudes , whereas lobeline reduced the mean amplitude to below that of control ( untreated ) receptor currents . When co-incubated with nicotine , both varenicline and lobeline occluded functional upregulation; lobeline treatment again reduced nAChR currents to a mean amplitude below even that of the control group . Neither treatment changed the time course of desensitization of the ACh currents . Thus , varenicline and lobeline prevented nicotine-induced functional upregulation of α4β2R nAChR currents . NH4Cl ( 20 mM ) treatment also partly restored nicotine-induced functional upregulation of ACh currents in lobeline and varenicline treated α4β2R-expressing HEK cells ( Figure 2C , D ) , albeit to a lesser degree than as observed with the binding experiments . We performed additional experiments to test whether varenicline and lobeline are released from intracellular acidic compartments after being trapped in the α4β2R-expressing HEK cells . The protocol is displayed schematically in Figure 1—figure supplement 3A . Briefly , HEK cells chronically treated with varenicline or lobeline for 17 hr were washed to remove free ligands and incubated with NH4Cl-containing PBS for 10 min , which we predicted would cause varenicline or lobeline to be rapidly released from intracellular acidic compartments . The supernatant from the treated cells was then added to a different set of receptor-expressing HEK cells that had not been exposed to varenicline or lobeline . 125I-epibatidine binding was performed on the second set of cells to test whether each supernatant contained competitive ligands that reduced binding ( indicative of the presence of varenicline or lobeline ) . Consistent with the release of varenicline and lobeline from intracellular acidic compartments , 125I-epibatidine binding was highly reduced for cells that received the bathing solution from cells chronically treated with varenicline or lobeline as compared to untreated cells ( Figure 1—figure supplement 3B ) . These results strongly suggest that the weak bases re-equilibrate across the plasma membrane from intracellular compartments in NH4Cl , and that they concentrate in sufficiently high concentration to compete for binding to nAChRs . We next tested if nicotine itself was trapped in intracellular compartments of HEK cells expressing α4β2Rs . Nicotine upregulation of 125I-epibatidine binding from live cells was similar to that in membrane preparations ( Figure 1 ) and following neutralization of the pH gradient of acidic compartments ( Figure 1D ) . Functional upregulation of ACh currents by nicotine also did not change if the pH gradient of acidic compartments was diminished ( Figure 2C , D ) . These observations are consistent with previous findings that nicotine is not trapped in the acidic compartment of α4β2R-expressing HEK cells ( Jia et al . , 2003 ) . We tested whether other α4β2R ligand weak bases were trapped in the intracellular acidic compartment like varenicline and lobeline . Dihydro-beta-erythroid ( DHβE , Figure 3A ) is a weak base and a α4β2R competitive antagonist . Despite being a competitive antagonist , DHβE was found to induce upregulation of α4β2R high-affinity binding sites ( Whiteaker et al . , 1998 ) . Consistent with previous findings , we observed a 2-fold upregulation of 125I-epibatidine binding with 17 hr of exposure of the cells to 100 μM DHβE ( Figure 3B ) . At this concentration , acute application of DHβE completely inhibits ACh-induced currents ( data not shown ) , as was shown previously ( Wu et al . , 2006 ) . Conversely , prolonged DHβE exposure upregulated ACh-induced currents , similar to the binding sites ( Figure 3E ) . DHβE did not occlude nicotine upregulation when the two ligands were co-incubated , however , demonstrating that it has a different effect on functional upregulation than varenicline and lobeline . Like nicotine , we see no evidence of DHβE being trapped in binding assays; upregulation of 125I-epibatidine binding was unchanged whether membranes were assayed ( data not shown ) or if the pH gradient of acidic compartments was diminished with NH4Cl treatment ( Figure 3B , green columns in the right half of the graph ) . 10 . 7554/eLife . 25651 . 008Figure 3 . α4β2R weak base ligands exhibit different degrees of intracellular trapping . ( A ) DHβE , mecamylamine and epibatidine chemical structures . ( B ) NH4Cl treatment does not alter DHβE ( DH ) upregulation . 125I-epibatidine binding performed on live α4β2R-expressing cells without ( left ) or with ( right ) NH4Cl treatment as in Figure 1D . Cells were treated for 17 hr with 100 μM DHβE with or without 10 μM nicotine ( n = 6: -NH4Cl; n = 4: +NH4Cl ) ) . The profile was similar to cells washed with PBS , indicating no effect of pH on DHβE . ( C ) Intact cells were treated with 30 μM epibatidine ( Epb ) and 125I-epibatidine binding were performed after incubating the cells with ( left ) or without ( right ) 20 mM NH4Cl/PBS ( n = 3 ) . Column graphs in ( B , C ) : ** , p<0 . 001 by one-way ANOVA with Bonferroni’s multiple comparison test . n indicates number of independent determinations on separate days and cultures . ( D ) A distribution plot comparing upregulation ( 125I-epibatidine binding ) by DHβE , mecamylamine and epibatidine before and after disruption of intracellular pH gradient by various agents that raise pH in intact cells . Each point represents the mean and s . e . m from the indicated columns in Figure 3B , C with the exception of the mecamylamine data , where the data are not displayed elsewhere , and the points are the means and s . e . m ( n = 4 ) where n indicates number of independent determinations on separate days and cultures . ( E ) ACh-evoked currents following 17–20 hr treatment of α4β2-expressing HEK cells with vehicle , nicotine , DHβE ( DH , 100 μM ) , or nicotine and DHβE . The number of recordings were 17 , 21 , 15 and 21 , respectively . DHβE appeared to upregulate ACh function but the currents were variable in their amplitude and the mean was not statistically different from the vehicle group . No attenuation of nicotine upregulation was observed with DHβE . ( F ) Epibatidine ( Epb , 30 μM ) shows effects on ACh currents similar to that of varenicline . Current amplitudes were similar to the vehicle control when cells were incubated with epibatidine alone , but co-incubation with nicotine prevented functional upregulation . The number of recordings were 14 , 22 , 17 and 24 , respectively . In all the column graphs: p<0 . 05; p<0 . 01; p<0 . 001 by one-way ANOVA with Tukey’s multiple comparison test; n indicates the number of experimental repetitions . Columns show group mean and error bars are the s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 25651 . 00810 . 7554/eLife . 25651 . 009Figure 3—figure supplement 1 . Dose dependence of epibatidine effects on upregulation . ( A ) Dose dependence of epibatidine upregulation . α4β2R-expressing cells were treated with increasing concentration of epibatidine for 17 hr . Cells were washed with PBS prior to 125I-epibatidine binding and with ( magenta-filled circles ) or without 20 mM NH4Cl ( black-filled circles ) treatment for 10 min ( n = 3 ) . ( B ) Dose dependence of epibatidine on nicotine upregulation . Cells were treated with increasing concentration of epibatidine with 10 μM nicotine for 17 hr and treated with ( magenta -filled squares ) or without 20 mM NH4Cl ( black-filled squares ) treatment for 10 min as in ( A ) ( n = 3 ) . For all the points in ( A , B ) : error bars represent mean ± s . e . m . n indicates number of independent experiments performed on separate days and cultures . DOI: http://dx . doi . org/10 . 7554/eLife . 25651 . 009 Why do the weak bases nicotine and DHβE differ from varenicline and lobeline with respect to trapping ? One likely factor is their degree of protonation at the pH found in intracellular compartments ( pH 5–6 ) . Protonation of weak bases depends on the acid dissociation constant , pKa ( Trapp et al . , 2008 ) , and the pKa values of nicotine and DHβE are significantly lower than those of varenicline and lobeline ( Table 1 ) . Based on this correlation between their physical properties and our current data , we propose that nicotine and DHβE are less likely than varenicline and lobeline to be trapped because they are less likely to be in their protonated form in an acidic compartment . 10 . 7554/eLife . 25651 . 010Table 1 . pKas and Kis for the studied weak base α4β2R ligandsDOI: http://dx . doi . org/10 . 7554/eLife . 25651 . 010Weak basepKa ( Basic ) KiTrappingVarenicline ( partial agonist ) 9 . 2 ( Unal et al . , 2012 ) 0 . 4 nM ( Rollema et al . , 2010 ) YesLobeline ( partial agonist ) 8 . 8 ( Drugbank , 2016 ) four nM ( Damaj et al . , 1997 ) YesEpibatidine ( agonist ) 9 . 5 ( ChEMBL , 2017 ) 0 . 01–0 . 05 nM ( Whiteaker et al . , 1998; Badio et al . , 1994 ) YesNicotine ( agonist ) 8 . 0 ( Barlow and Hamilton , 1962 ) eight nM ( Whiteaker et al . , 1998 ) NoDHβE ( competitive antagonist ) 7 . 3 ( ChEMBL , 2017 ) 0 . 3 μM ( Whiteaker et al . , 1998 ) NoMecamylamine ( noncompetitive antagonist ) 11 . 2 ( Remington and Beringer , 2006; Nangia et al . , 1996 ) >1 mM ( Whiteaker et al . , 1998 ) No As another test , we examined how the noncompetitive antagonist mecamylamine ( Figure 3A ) , a weak base with a pKa higher than that of varenicline and lobeline ( Table 1 ) , altered binding or functional activity alone and when co-incubated with nicotine . For exposures for 17–18 hr at concentrations up to 100 μM , there was no evidence of upregulation as assayed by 125I-epibatidine binding ( Figure 3D ) or by ACh-induced currents ( data not shown ) . Acute application of mecamylamine ( 100 μM ) inhibited ACh-induced currents ( data not shown ) as previously observed ( Wu et al . , 2006 ) , confirming that the drug had the expected pharmacological activity . Despite its high pKa value , we observed no evidence of mecamylamine being trapped because there was no difference in the binding when performed on membranes ( data not shown ) or if the pH gradient of acidic compartments was diminished with NH4Cl treatment ( Figure 3D ) . We propose that two possible mechanisms could account for the lack mecamylamine trapping and the absence of an effect on nicotine upregulation of nAChR current amplitudes . The ligand has a very high pKa ( 11 . 2 ) and therefore might be in a protonated state and unable to permeate the plasma membrane . However , mecamylamine appears to readily cross the blood-brain barrier and cell membranes ( Shytle et al . , 2002 ) . A more likely possibility is that trapping also requires high-affinity ligand binding to α4β2R located within acidic compartments . The affinity of mecamylamine for its α4β2R binding site is very low ( Table 1 ) and thus accumulation is precluded as discussed below . We predicted that epibatidine itself would be trapped in acidic compartments if indeed a high pKa value ( greater than that of nicotine ) and high-affinity binding to α4β2R sites are important determinants of this phenomenon . The pKa value and affinity for epibatidine are both higher than those for varenicline and lobeline ( Table 1 ) . In intact cells , epibatidine ( 30 μM ) did not promote upregulation of either 125I-epibatidine binding ( Figure 3C left half , D; Figure 3—figure supplement 1A ) or ACh-induced currents ( Figure 3F ) . However , epibatidine reduced nicotine-induced upregulation in both assays ( Figure 3C left half , D; Figure 3—figure supplement 1B ) . These effects of epibatidine on binding sites changed when the cells were treated with NH4Cl ( Figure 3C right half , D; Figure 3—figure supplement 1A , B ) . With NH4Cl treatment , we observed upregulation of 125I-epibatidine binding to 60% of the nicotine induced value and that nicotine upregulation was reduced by 40% . Thus , epibatidine behaves in a qualitatively similar way as varenicline and lobeline with respect to upregulation and therefore is subject to trapping in intracellular compartments , consistent with the phenomenon arising from a combination of high binding affinity for the receptor and a very basic pKa . In the preceding experiments , we inferred that trapping in intracellular compartments produced the agonist-specific effects observed in binding and functional assays . We next attempted to directly measure 125I-epibatidine trapping and release from acidic compartments . We anticipated that intracellular trapping of the radioligand results in two distinct pools of 125I-epibatidine in the cells: one pool represents 125I-epibatidine bound to intracellular and cell-surface α4β2Rs , whereas the second pool would arise from unbound 125I-epibatidine trapped within intracellular acidic compartments . These two pools were differentiated by analysis of the association and dissociation kinetics of 125I-epibatidine binding in living receptor-expressing HEK cells . First , association rates were determined to test if 125I-epibatidine binding to intracellular α4β2Rs is significantly slower than to cell-surface α4β2Rs , which would be revealed as an association rate curve with two or more rates of binding . However , the data are consistent with a single exponential process and an association rate constant of 3 . 8 × 107 ( M-sec ) −1 ( Figure 4—figure supplement 1 ) , similar to previous measurements using 3H-epibatidine on isolated membrane preparations containing α4β2Rs ( Whiteaker et al . , 1998; Gnädisch et al . , 1999; Shafaee et al . , 1999 ) and to intact cells ( Whiteaker et al . , 1998 ) . The single association rate is consistent with 125I-epibatidine binding with very similar rates to all cellular α4β2Rs and indicates that the radioligand crosses cellular membranes rapidly enough that permeation does not significantly contribute to the binding rate ( Figure 4E ) . 10 . 7554/eLife . 25651 . 011Figure 4 . Direct measurements of 125I-epibatidine trapping . ( A ) The biphasic dissociation of 125I-epibatidine from α4β2R HEK cells measured at 37°C . Bound 125I-epibatidine is normalized to % bound at time 0 . The line through the data represents a least-squares fit of a double exponential function: % Bound ( t ) = Af exp ( -kft ) + As exp ( -kst ) where Af ( 38 ± 6% ) and As ( 62 ± 8% ) are the % bound 125I-epibatidine at time ( t ) = 0 for the fast and slow component respectively; kf ( 1 . 5 ± 0 . 6×10−3 sec−1 ) and ks ( 6 . 0 ± 1 . 4×10−6 sec−1 ) are the time constants for the fast and slow component respectively ( n = 4 ) . ( B ) Addition of 100 μM nicotine causes the rapid release of slowly dissociating bound 125I-epibatidine . Dissociation of 125I-epibatidine measured at 37°C as in A except after adding 100 μM nicotine at to start the dissociation measurement . The line through the data with the exception of the initial data point represents a least-squares fit of a single exponential function: % Bound ( t ) = Af exp ( -kft ) where kf was 7 . 0 ± 1 . 3×10−4 sec−1 ( n = 3 ) . ( C ) The rapid dissociation of bound 125I-epibatidine measured after washing the cells with increasing concentrations of nicotine . α4β2R HEK cells were with treated with 10 μM nicotine ( dark circles ) or left untreated ( white squares ) for 17 hr . The cells were washed . Cells were washed with PBS followed by three washes with indicated concentrations of nicotine prior to performing 125I-epibatidine binding ( n = 3 ) . For all the points in A , B , C: error bar represents mean ± s . e . m . ( D ) 125I-epibatidine binding to α4β2R-expressing HEK cells versus HEK cells lacking α4β2Rs . 1 mM nicotine was added during 125I-epibatidine binding to estimate nonspecific binding ( n = 3 ) . ( E ) Altered model of ligand trapping with high-affinity α4β2Rs in the acidic vesicles . ( F ) Nicotine exposure increases trapped 125I-epibatidine . The levels of trapped 125I-epibatidine were determined with the addition of 100 μM nicotine to start the dissociation . Trapped 125I-epibatidine was released within 1 min of the nicotine addition as in Figure 4B and C for cells untreated ( -Nic ) or treated with 10 μM nicotine ( +Nic ) for 17 hr . Plotted is the released 125I-epibatidine in pmoles normalized to untreated cells ( n = 3 ) . For ( D , F ) : error bar represents mean ± s . e . m . p<0 . 001 by Student’s t-test . ( A-D , F ) : n indicates number of independent experiments performed on separate days and cultures . DOI: http://dx . doi . org/10 . 7554/eLife . 25651 . 01110 . 7554/eLife . 25651 . 012Figure 4—figure supplement 1 . Association rate of 125I-epibatidine . Live cells were bound with 0 . 1 nM 125I-epibatidine at room temperature for the indicated times . The line through the points represents a least-squares fit to the normalized 125I-epibatidine uptake for the α4β2R-expressing HEK cells plotted as the fraction [125I-Epb]up ( t ) / [125I-Epb]up ( 0 ) and fit by the equation [125I-Epb]up ( ∞ ) / [125I-Epb]up ( 0 ) ( 1 - ( exp ( -kt ) ) +1 , where [125I-Epb]up ( t ) is the 125I-epibatidine uptake at time t , [125I-Epb]up ( 0 ) at time 0 , [125I-Epb]up ( ∞ ) at saturation and k the association rate . Three independent experiments were performed on separate days and cultures . For all the points in the plot , error bars represent mean ± s . e . m from three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 25651 . 012 Second , analysis of the dissociation of 125I-epibatidine exiting the α4β2R-expressing cells revealed a biphasic time course that was best fit as the sum of two exponential processes ( Figure 4A ) . The faster dissociating component ( τ = 11 . 4 min; 1 . 5 × 10−3 sec−1 ) , ~40% of the total binding , was similar to rates previously measured for unbinding of radio-labeled epibatidine from α4β2Rs in membrane fragments . In contrast , the slower dissociation component ( τ = 47 hr ) had a rate constant of 6 . 0 × 10−6 sec−1 and therefore was >100 fold slower than previous measurements of 125I-epibatidine unbinding from α4β2Rs . We hypothesize that this slow component of 125I-epibatidine dissociation arose from radioligand trapped in intracellular acidic compartments ( illustrated in Figure 4E ) . The slowly dissociating component in the preceding assays could in part arise from rebinding of 125I-epibatidine to receptors within intracellular compartments , because epibatidine is a high-affinity agonist and the measured dissociation rate was slow enough to allow repeated cycles of binding and unbinding . We found that 0 . 10–0 . 20 pmol of 125I-epibatidine dissociated during the faster dissociation component in the assay shown in Figure 4A , which corresponds to a concentration of 0 . 2–0 . 4 nM free 125I-epibatidine in our 0 . 5 ml assay volume of the eppendorf tube used in these assays . We predicted that rebinding of radioligand to the receptor should occur at this concentration; to test this hypothesis , we included a membrane-permeable competitive ligand , nicotine ( 100 μM ) , at the start of the dissociation measurements . Dissociation of 125I-epibatidine in the presence of nicotine occurred with a much faster overall time course , such that the radioligand was released in less than a minute rather than being trapped for hours . The rapid release of radioligand was followed by a slower time course of dissociation ( τ = 24 min at 37°C ) , which is typical of rates ( 7 . 0 × 10−4 sec−1 ) previously measured for the dissociation of 3H-epibatidine from α4β2Rs and dissociation constants for affinity ( KD = 18 pM ) ( Whiteaker et al . , 1998; Gnädisch et al . , 1999; Shafaee et al . , 1999 ) . Thus , rebinding of epibatidine to nAChRs in intracellular compartments contributes to the slowly dissociating pool of radioligand . We used a high nicotine concentration , 100 μM , to prevent rebinding of 125I-epibatidine in the experiments shown in Figure 4B , but the Ki value for nicotine for α4β2R binding sites is much lower ( 8 nM; Table 1 ) ; in principle , even 100 nM nicotine added during the dissociation measurement should block essentially all rebinding of 125I-epibatidine . We tested this prediction in additional dissociation assays and instead found that the addition of 100 nM nicotine only reduced the slow component of dissociation by ~20% ( Figure 4C ) . This result suggests that rebinding alone does not adequately account for the extraordinarily slow dissociation of 125I-epibatidine from living cells expressing α4β2Rs . Rather , two processes are key to the trapping phenomenon: ( i ) 125I-epibatidine rebinding to α4β2Rs and ( ii ) protonation of 125I-epibatidine in those same compartments . Higher concentrations of nicotine ( 100 μM – 10 mM ) added into the dissociation assay act on both components of the trapping phenomenon; that is , increasing nicotine concentrations not only prevented 125I-epibatidine rebinding but also reduced the pH gradient to neutralize acidic compartments , leading to deprotonation ( and release ) of 125I-epibatidine . This was evident in the fact that the proportion of 125I-epibatidine dissociation contributed by the slower component decreased with increasing nicotine concentrations and saturated at a value of ~60% of the total ‘bound’ 125I-epibatidine for both the untreated and nicotine-treated cells ( Figure 4C ) , which is the entirety of the slowly dissociating component . Chloroquine ( 150 μM ) , a weak base with a molecular weight and pKa similar to that of nicotine , also reduces the pH gradient of acidic compartment at similar concentrations and has a similar effect on 125I-epibatidine binding ( Figure 1E ) . Thus , the addition of nicotine during the dissociation assay serves two functions that together cause rapid release of all 125I-epibatidine . One function is to bind to α4β2R high-affinity binding sites and block rebinding of 125I-epibatidine , and the other is to reduce the pH gradient across intracellular compartments in which ligand become trapped . The preceding experiments strongly suggest that high-affinity α4β2R binding sites are required for trapping of weak base ligands . We tested this idea by assaying 125I-epibatidine in HEK cells lacking α4β2Rs . 125I-epibatidine binding to these HEK cells was ~1% of that observed with nicotine upregulation and was equivalent to ‘non-specific’ binding obtained with 1 mM competing nicotine present during 125I-epibatidine binding and washing ( Figure 4D ) . This finding demonstrates that no selective trapping of 125I-epibatidine occurs without α4β2R expression . Without high-affinity binding α4β2Rs , 125I-epibatidine concentrates within acidic vesicles but it rapidly exits the vesicles when its extracellular concentration falls . High-affinity α4β2Rs in intracellular compartments trap 125I-epibatidine even when its extracellular concentration falls , whereas nicotine rapidly exits the acidic compartments . In this way , the targeting of α4β2Rs to acidic vesicles selectively traps certain weak bases like epibatidine . Altogether , our data are consistent with the model shown in Figure 4E , which depicts how 125I-epibatidine is selectively trapped within an intracellular acidic compartment . This model differs from that in Figure 1C by the addition of high-affinity α4β2Rs in acidic vesicles . In Figure 4E , α4β2Rs in the vesicles are depicted in two states , either α4β2Rs with high-affinity binding sites that would contribute to trapping or α4β2Rs without high-affinity binding sites . The two states are consistent with the findings of Vallejo et al . ( Vallejo et al . , 2005 ) for α4β2Rs at the plasma membrane . When 125I-epibatidine enters the vesicle lumen , it is protonated and in this state binds to high-affinity sites on α4β2Rs . The high-affinity binding sites appear to play a significant role in selective trapping of 125I-epibatidine within the vesicle lumen for long periods of time . One pool of radioligand , ~40% of the 125I-epibatidine , is bound to the surface and intracellular α4β2Rs and dissociates with the characteristic unbinding rate analogous to that measured in membrane preparations . The other pool of epibatidine , ~60% , is not bound to α4β2Rs , but instead is trapped and exits from the cells at a rate much slower than the unbinding rate . As depicted in the model , 125I-epibatidine exits from acidic vesicles at the slow dissociation rate ( b ) measured in the experiments shown in Figure 4A . As discussed above , the rate that 125I-epibatidine crosses cellular membranes ( a ) , which is close to diffusion-limited ( Figure 4—figure supplement 1 ) , is orders of magnitude faster than the dissociation rate ( b ) . How do the two different pools measured in 125I-epibatidine dissociation assays change with long-term nicotine exposure ? Surprisingly , nicotine upregulation did not change the relative contribution of either the fast or slow component of radioligand dissociation . The fraction of 125I-epibatidine specifically bound to α4β2Rs remained at ~40% and the fraction of trapped in intracellular compartments and slowly released remained at ~60% . As displayed in Figure 4F , nicotine caused the fraction of trapped 125I-epibatidine to increase by 4–5-fold , the same fold increase observed for total ( bound and trapped ) 125I-epibatidine retained in the cells with nicotine exposure ( Figure 4D ) . The fraction of 125I-epibatidine specifically bound to the α4β2Rs also increased 4–5-fold with nicotine exposure ( data not shown ) . This finding is consistent with the results in Figure 4D that α4β2Rs must be present for selective 125I-epibatidine trapping to occur and further supports the idea that the α4β2Rs are located within intracellular acidic compartments . To address why the trapped 125I-epibatidine fraction increased by 4–5-fold with nicotine exposure ( Figure 4F ) and to confirm that α4β2Rs are located within intracellular acidic compartments , we made use of a version of an α4 subunit with the pH-sensitive fluorescent tag , super ecliptic pHluorin ( SEP ) fused to its C-terminus ( α4SEP; [Richards et al . , 2011; Fox et al . , 2015] ) . SEP is a pH-sensitive variant of GFP that fluoresces at neutral pH but is quenched at pH values lower than 6 . When expressed in α4β2R-expressing HEK cells and in cortical neurons , florescent α4SEP was found throughout the ER , as indicated by co-localization with the ER marker , DsRed-ER ( Figure 5—figure supplement 1A , B , C ) , and on the cell surface ( Figure 5—figure supplement 1B , D ) as previously observed ( Richards et al . , 2011 ) . To specifically test whether α4SEP subunits were found in intracellular acidic compartments , cells and neurons expressing α4SEP were treated with 20 mM NH4Cl ( 5 min ) to neutralize the pH . The distribution of α4SEP subunit fluorescence was compared before and after NH4Cl treatment . As displayed in Figure 5A for the HEK cells and Figure 5B for the cortical neurons , NH4Cl treatment revealed an additional pool of the α4SEP subunits in what appeared to be small vesicles . It is likely that the α4SEP subunits in the acidic vesicles are mature , fully assembled α4β2Rs because receptor assembly occurs in the ER ( Sallette et al . , 2005 ) and entry into any acidic compartment occurs after exit from the ER . The number of α4β2R-containing acidic vesicles in the HEK cells and cortical neurons was greatly increased by the 17 hr nicotine exposure ( Figure 5A , B , C ) and paralleled the effect of nicotine on the number of high-affinity binding sites and the amount of 125I-epibatidine trapping ( Figure 4F ) . In the neurons , α4β2R-containing acidic vesicles were found in the processes , in what appeared to be both dendrites and axons , as well as in the somata ( Figure 5B ) . These results demonstrate that nicotine upregulation occurs in part through an augmentation of the number of receptor-containing acidic vesicles . 10 . 7554/eLife . 25651 . 013Figure 5 . Nicotine exposure increased the number of α4β2R-containing acidic vesicles . ( A ) Imaging α4β2R-containing acidic vesicles and the effect of nicotine exposure . α4SEP was transfected into the α4β2R-expressing HEK cells . Image ( 100 x ) of 3 merged slices near the cell surface from untreated ( top ) and nicotine-treated ( bottom ) cells . Cell was imaged without NH4Cl ( left panel ) and after adding NH4Cl ( 5 min; middle panel ) . In the right panel ( difference ) the total fluorescent intensity in -NH4Cl image was subtracted from +NH4Cl image . Scale bar is 10 μm . ( B ) Imaging α4β2R-containing acidic vesicles in cultured neurons and the effect of nicotine exposure . α4SEP and β2HA subunits was transfected into cortical neurons ( DIV 9 ) . Image as in A of untreated ( top ) and nicotine-treated ( bottom ) neurons . Both soma and dendrites from the same cells are shown . Right panel and inset shows the difference obtained by subtracting -NH4Cl image from +NH4Cl image . ( Scale bar:10 μm ) . ( C ) Quantification of the numbers of acidic vesicles from the difference images . The fluorescent vesicles were counted and plotted for HEK cells ( 22 cells , for both untreated and nic treated , n = 4 ) and cortical neurons ( 10 cells for untreated and 8 cells for nic treated , n = 3 ) . Error bar represents mean ± s . e . m . **p<0 . 001 by student t test . n indicates number of independent experiments performed on separate days and cultures . ( D ) Model illustrating how trapping in acidic vesicles of weak base ligands like varenicline , lobeline and epibatidine is selective and regulated by nicotine . Ligand trapping is a function of weak base pKa and its affinity of the ligands for α4β2Rs in the case of α4β2R weak base ligands . As illustrated in the figure weak base ligands ( Lig ) for other receptors or transporters could also be trapped selectively . DOI: http://dx . doi . org/10 . 7554/eLife . 25651 . 01310 . 7554/eLife . 25651 . 014Figure 5—figure supplement 1 . Subcellular distribution of pHluorin tagged α4 subunit ( α4SEP ) in α4β2R-expressing HEK cells and cortical neurons . ( A ) HEK cells are displayed 36 hr post transfection . Transfected α4SEP subunits ( green ) and DsRed ER ( red ) are imaged in untreated ( top panel ) or nicotine treated ( 10 μM for 17 hr; bottom panel ) cells . Stably transfected β2 subunits have an HA epitope ( blue ) at its C-terminal extracellular tail . Cell-surface receptors were labeled on live cells with anti-HA antibody and Alexa Fluor 647 conjugated anti-mouse secondary antibody . ( B ) Higher magnified images of the α4SEP subcellular distribution with the ER marker and surface labeling . ( C ) Cortical neurons ( DIV9 ) are displayed 36 hr post transfection . Transfected α4SEP subunits ( green ) , DsRed ER ( red ) and cell-surface β2HA ( blue ) are imaged in untreated ( top panel ) or nicotine treated ( 1 μM for 17 hr; bottom panel ) cells . 36 hr post transfection α4SEP ( green ) showed co-localization with DsRed ER ( red ) with little to no surface expression . Scale bar is 10 μm . ( D ) Higher magnified images of the α4SEP subcellular distribution with the ER marker and surface labeling . Cell surface expression of α4SEP in cortical neuron dendrite three days post transfection . The surface expressed α4SEP co-localized with surface labeled β2HA in the dendrites . The α4SEP in the shaft showed typical ER distribution and co-localized with DsRed ER . DOI: http://dx . doi . org/10 . 7554/eLife . 25651 . 014 In this study , we find that the smoking cessation agent varenicline ( Chantix ) has significant effects on α4β2Rs independent of its pharmacological action as a partial agonist . Varenicline exposure reduced nicotine upregulation of α4β2Rs , which in vivo causes long-lasting changes in α4β2Rs linked to different components of nicotine addiction ( Govind et al . , 2009; Vezina et al . , 2007; Lewis and Picciotto , 2013 ) . The effects of varenicline on upregulation required intact intracellular acidic vesicles containing α4β2Rs and were not observed if HEK cell or neuronal membranes were compromised . The integrity of acidic compartments was lost in earlier studies that assayed binding using membrane fragments of brain tissue or autoradiography ( Marks et al . , 1983; Schwartz and Kellar , 1983; Marks et al . , 2015 ) accounting for the different actions of varenicline we observed in living cells . Based on our findings , it is evident that varenicline is selectively trapped as a weak base within intracellular acidic vesicles , and once trapped , it is slowly released from the vesicles and cells . The trapped varenicline competes with 125I-epibatidine binding to α4β2Rs within the acidic vesicles while the slow release competes with 125I-epibatidine binding elsewhere in the cells and desensitizes cell-surface α4β2Rs . These effects cause the apparent suppression of binding site and functional upregulation ( Figures 1 and 2 ) . We observed effects of varenicline on nicotine upregulation at concentrations as low as 100 nM ( Figure 1G ) , a concentration that is estimated to occur in human brain with prescribed varencline doses ( Rollema et al . , 2010 ) . Based on our experimental evidence at higher varencline concentrations , we would expect that the effects of varenicline at 100 nM are caused by the trapping of varenicline in acidic vesicles at this lower concentration . It should also be noted that in order to estimate human brain varenicline concentrations , Rollema et al . ( Rollema et al . , 2010 ) used rat brain lysates to measure the high-affinity binding . As is evident from our study , this approach will not include varenicline trapping in acidic vesicles , which should serve as an additional high-capacity reservoir for varenicline in brain neurons , and is lost in the brain lysate preparation . Therefore , this study likely significantly underestimated the levels of varenicline in brain and the effects of trapping in acidic vesicles should be considered in future such estimates . Lobeline and epibatidine also appeared to be trapped in intracellular compartments and showed effects similar to that of varenicline ( Figure 1—figure supplement 1C , D and Figure 3—figure supplement 1 ) , whereas nicotine and other weak base ligands with lower pKas were not significantly trapped and were rapidly washed from cells and cultured neurons . The dose-dependence of all three weak base ligands that were trapped within intracellular acidic vesicles had an inverted U-shape where the peak of the curve occurs at ~1 μM concentration . The decline in the dose-dependence curve at concentrations greater than 1 μM is consistent with the weak base effect of the ligands at these concentrations increasing the pH within the vesicle lumen and thereby increasing the rate of release of the trapped ligand . Our observation that varenicline , but not nicotine , is trapped in α4β2R-containing acidic vesicles can potentially explain differences in the human pharmacokinetic profiles of these drugs . The decay time ( t1/2 ) for the varenicline plasma concentration is 1 day and the time to reach steady-state levels with repeated dosing is 4 days ( Faessel et al . , 2010 ) . In contrast , the decay time for the nicotine plasma concentration to decay is 2 hr and the time to reach steady-state levels with repeated dosing is 2–3 hr ( Benowitz et al . , 2009 ) . These differences are consistent with the slow exit rate we measured for varenicline from acidic vesicles and the rapid exit of nicotine . The residence of a large amount of varenicline in α4β2R-containing acidic vesicles in neurons is likely also to contribute to the differences in the rate at which varenicline and nicotine are metabolized; nicotine is metabolized with the decay time of 2 hr , whereas less than 10% of varenicline is metabolized over this time . The volume of distribution of varenicline and nicotine differs as well . For varenicline , the volume of distribution ( 5 . 9 L/Kg; [Faessel et al . , 2010] ) is more than twice that for nicotine ( 2 . 6 L/Kg; [Benowitz et al . , 2009] ) . This difference again is consistent with and perhaps caused by trapping of varenicline within acidic compartments in cells and neurons and the absence of trapping for nicotine . How varenicline trapping in intracellular acidic vesicles could be altering its clinical efficacy is shown in our working model of this process ( Figures 4E and 5D ) . Based on our findings , the presence of high-affinity α4β2Rs in the vesicles and the weak base nature of varenicline will influence its partitioning between extracellular , cytoplasmic , and vesicular pools . Varenicline trapping in acidic vesicles would create a high-capacity reservoir in neurons that express α4β2Rs . We predict that this phenomenon maintains relatively high and constant concentrations of varenicline , especially in contrast to nicotine levels that rise during the day and decline rapidly at night . Clinical efficacy could therefore result from sustained varenicline levels within neurons that leak out and desensitize α4β2Rs , and perhaps other nAChRs , and that activity counteracts the functional upregulation of nicotine exposure . As modeled in Figures 4E and 5D , the presence of α4β2R is what provides the selectivity for long-lasting trapping of certain nicotinic weak base ligands in acidic vesicles . Selective trapping also requires a low enough pH in the vesicles , to protonate weak base ligands and slows its exit . Trapping only occurs when there are α4β2Rs in the vesicles and ligand pKa and affinity for α4β2Rs are sufficiently high . Nicotine and DHβE fail to show the trapping phenomenon for two reasons: nicotine and DHβE are not sufficiently protonated at low pH and DHβE also does not bind with high enough affinity to α4β2R . Epibatidine , on the other hand , has a high pKa and affinity , and therefore accumulates and leaks back out over the course of days , as was observed in our dissociation experiments ( Figure 4A ) . As measured by 125I-epibatidine ( Figure 4F ) , ligand trapping was increased by nicotine exposure to the same degree as the binding to the receptors ( Figure 4D ) . Higher trapping levels were caused by a rise in the numbers of vesicles containing α4β2Rs as imaged using pH-sensitive , pHluorin-tagged α4 subunits in the α4β2R-expressing HEK cells and cultured neurons ( Figure 5 ) . As modeled in Figure 5D , the higher numbers of acidic vesicles and the resulting rise in trapping capacity acts as a mechanism to regulate selective trapping of varenicline and other weak base α4β2R ligands . However , it is not clear if the increases are caused by the formation of new vesicles or redistribution of α4β2Rs to preexisting vesicles . Intracellular acidic vesicles exist during the late stages of Golgi trafficking and different pools of endosomal membranes , recycling or late endosomes , or lysosomes ( Paroutis et al . , 2004 ) . In this study , we did not distinguish among these possibilities , but our previous findings and those of others ( Vallejo et al . , 2005; Darsow et al . , 2005 ) indicate that surface α4β2R degradation through late endosomes and lysosomes is not increased by nicotine upregulation; therefore , the acidic vesicles are unlikely to be either of these organelles . Still to be determined are which intracellular compartment the vesicles originate from and the role of the vesicles in nicotine upregulation of α4β2Rs . If the smoking cessation activity of varenicline is indeed caused by its trapping in intracellular acidic vesicles , then our findings should help guide the design of more effective smoking cessation drugs . While we have established that both the pKa of a ligand and its α4β2R affinity are important for trapping , further experiments are needed to assay what specific pKa and α4β2R affinity of a weak base ligand is most effective for smoking cessation . The selective and regulated trapping we observe for varenicline also may occur for weak base ligands that bind with high-affinity to other types of receptors , ion channels or transporters . Most drugs of abuse are weak base ligands that bind with varying affinities to a variety of different membrane proteins ( Sulzer , 2011 ) . In particular , amphetamines are weak bases with pKa values in the range of 8 to 10 that can concentrate in acidic vesicles and affect the catecholamine quantum size in synaptic vesicles and chromaffin granules ( Sulzer et al . , 2005 ) . Certain antipsychotic drugs are weak bases that accumulate in and are released from synaptic vesicles ( Tischbirek et al . , 2012 ) . As shown schematically in Figure 5D , the targeting of other membrane protein high-affinity binding sites to acidic vesicles would allow different classes of weak base ligands to be selectively trapped in those vesicles ( Lig in Figure 5D ) . Weak bases that bind with high-affinity to the binding sites where other drugs of abuse bind could serve as cessation agents similar to varenicline . Previously characterized mouse nAChR α4SEP with a super ecliptic pHluorin incorporated on the C-terminus of α4 was a gift from Dr . Christopher I . Richards ( University of Kentucky , Lexington , Kentucky ) ( Richards et al . , 2011 ) . pDsRed-ER was from Clontech Laboratories . Human α4 and β2 cDNA were gifts from Prof . Steven M . Sine ( Mayo Clinic , Rochester , Minnessota ) . Rat α4 and β2 used for generating the stable cell line were provided by Dr . Jim Boulter , University of California , Los Angeles , CA . The HA epitope , YPYDVPDYA , and a stop codon were inserted after the last codon of the 3′-translated region of the subunit DNA of the β2 using the extension overlap method as described in Vallejo et al ( Vallejo et al . , 2005 ) . The human embryonic kidney ( HEK 293T ) cell line stably expressing the large T antigen ( tSA201 cells ) was from Dr . J . Kyle ( University of Chicago , Chicago , IL ) . This cell line is not in the list of Database of Cross-Contaminated or Misidentified Cell Lines . Using this parent HEK 293 T cells , a stable cell line expressing rat α4β2 nAChRs were generated in our lab and it expresses an untagged α4 and a C-terminal HA epitope tagged β2 subunits ( Vallejo et al . , 2005 ) . Both parent HEK cell line and stable α4β2 HEK cell line were maintained in DMEM ( Gibco , Life technologies ) with 10% calf serum ( Hyclone , GE Healthcare Life Sciences , UT ) at 37°C in the presence of 5% CO2 . DMEM was supplemented with Hygromycin ( Calbiochem , EMD Millipore , MA ) at 0 . 4 mg/ml for maintaining selection of α4β2 HEK cells . Transfection of human α4 and β2 subunits into HEK parent cell line or α4SEP into α4β2 HEK cell line were performed using calcium phosphate method ( Eertmoed et al . , 1998 ) . Stable cells were maintained in hygromycin free DMEM prior to transfection . Hoechst staining and immunofluorescent detection were performed periodically to test for mycoplasma contamination . Fresh batch of cells were thawed and were maintained only upto two months . Primary cultures of rat cortical neurons were prepared as described earlier ( Govind et al . , 2012 ) . Dissociated cortical cells from E18 Sprague Dawley rat pups were plated on plates that were coated with poly-D-lysine ( Sigma , MO ) . For live imaging , neurons were plated in glass bottom dishes ( MatTek , MA ) . DIV eight neurons were transfected with α4SEP , β2HA and DsRed ER cDNAs using lipofectamine 2000 ( Invitrogen , Thermofisher scientific , MA ) reagent . 2–3 days after transfection , neurons were treated with 1 μM nicotine for 17 hr and were live imaged in low fluorescence Hibernate E buffer ( Brain bits , IL ) . Poly-D-Lysine ( P7886 ) , Nicotine ( N3876 ) , Varenicline ( PZ0004 ) , Lobeline ( 141879 ) , Epibatidine ( E1145 ) , Mecamylamine ( M9020 ) , Ammonium Chloride ( A0171 ) , Chloroquine ( C6628 ) and Bafilomycin A1 ( B1793 ) were purchased from Sigma , MO . Dihydro beta-erythroidine ( DHβE ) ( 2349 ) was purchased from Tocris , MN . Neurobasal medium , B27 , HBSS and DMEM were purchased from Life technologies ( Thermofisher scientific , MA ) . α4β2 stable cell line was treated with indicated concentrations of nAChR ligands for 17 hr in the presence or absence of 10 μM nicotine . For intact counts , cells were washed four times with PBS , scraped off the plates , pelleted at 2800 rpm for 3 min , resuspended in 1 ml PBS and aliquots were incubated with 2 . 5 nM 125I-epibatidine ( 125I Epb ) ( 2200 Ci/mmol; Perkin Elmer ) for 20 min at room temperature . At the end of incubation cells are harvested on Whatman GF/B filters presoaked in 0 . 5% polyethyleneimine and washed four times with PBS using 24-channel cell harvester ( Brandel , MD ) . Non-specific binding was estimated by incubating parallel samples in 1 mM nicotine prior to and during incubation with 125I Epb . Radioactivity of bound 125I Epb was determined using gamma counter ( Wallac , Perkin Elmer , MA ) . pmole 125I Epb bound to cells are normalized to nicotine upregulated cells and plotted as % of nicotine treated cells . Primary cultures of cortical neurons were treated with indicated concentrations of varenicline or lobeline with or without 1 μM nicotine . Intact neurons were washed three times with PBS , gently scraped off the plates in PBS , pelleted down at 2800 rpm for 3 min and resuspended in PBS prior to binding with 1 nM 125I Epb . After 20 min of incubation with radioactivity at room temperature , cells are transferred to filters using Brandel cell harvester , filters were washed four times with PBS and radio activity associated with the filters measured as described above . Non-specific binding was calculated from cells incubated with 1 mM nicotine during 125I Epb binding . Intact cells were treated with nAChR ligands for 17 hr and washed twice with PBS . Cells were exposed to two 5 min incubations with PBS containing 20 mM ammonium chloride , 150 μM chloroquine or 50 nM bafilomycin A followed by two PBS washes . Cells were gently scraped off the plates , re-suspended in PBS and subjected to 125I-epibatidine binding as described above . Depending on the number of samples per experiment , the time elapsed between when the drug-containing medium was removed from the cells and the epibatidine-binding initiated ranged from 25 to 45 min . α4β2 HEK cells were treated with indicated concentrations of the drugs with or without 10 μM nicotine for 17 hr . Cells were suspended in hypotonic buffer ( 10 mM Hepes , pH 7 . 9 , 1 mM MgCl2 , 1 mM EDTA ) plus protease inhibitors and incubated in ice for 10 min . Cells were homogenized using a dounce homogenizer ( 10 times ) . Nuclei were removed by centrifugation at 1000 x g for 10 min . The supernatant was subjected to ultra centrifugation at 100 , 000 x g for 1 hr at 4°C using Beckman TLA 100 . 3 rotor . Supernatant was removed and the membrane pellet was resuspended in PBS via sonication . Aliquots of resuspended membrane were bound with 2 . 5 nM 125I Epb for 20 min at room temperature and transferred to filter paper and washed using brandel cell harvester . α4β2 HEK cells and cortical neurons were plated in glass bottom live imaging plates ( MatTek , MA ) coated with poly-D-lysine one day before plating . α4β2 stable cells were transfected with α4SEP with or without pDsRed-ER using calcium phosphate method . After 24 hr of transfection cells were treated 10 μM nicotine for 17 hr . For imaging , cells were incubated with anti-HA ( Mouse monoclonal HA . 11 , Biolegend , CA ) antibody for 40 min in DMEM . Cells were washed three times with DMEM and labeled with Alexa Fluor anti-mouse 647 ( Molecular Probes , Thermo scientific , MA ) secondary antibody for 30 min . Cells were washed three times with Low Fluorescence Hibernate E and imaged in the same buffer . Cortical neurons ( DIV 8 ) were transfected with α4SEP , β2HA and DsRed-ER using lipofectamine 2000 ( Invitrogen , Thermo scientific , MA ) as per manufacturers protocol . Neurons were treated with 1 μM nicotine either 2–3 days after transfection and imaged after 17 hr of nicotine exposure . Neurons were live labeled with anti HA antibody for 40 min , washed three times followed by secondary antibody Alexa Fluor anti-mouse 647 incubation for 30 min . Antibody incubations and washes were done in conditioned neurobasal medium . Neurons were subsequently washed three times with Hibernate E buffer and imaged in the same buffer . SEP is pH sensitive and does not fluoresce at pH below 6 . Under physiological pH , the fluorescence emitted from α4SEP is mainly from surface receptors ( extracellular SEP ) and from receptors in the ER ( luminal SEP ) . The fluorescence of SEP is quenched in intracellular vesicles as their pH is below 6 . In order to visualize α4SEP in acidic vesicles , intracellular pH was increased using ammonium chloride ( NH4Cl ) . Cells were initially imaged in Hibernate E buffer ( pH 7 . 4 ) . NH4Cl was added drop wise to a final concentration of 20 mM with minimum disturbance to the cells . 5–10 min after addition of NH4Cl , confocal images were collected from the same cells again . Three to four consecutive slices of the cells were combined for HEK cells and 7–8 consecutive slices were combined for cortical neurons both for pre and post NH4Cl images . The integrated density of pre-NH4Cl image of the cell was subtracted from that of the post-NH4Cl image . The resultant signal was thresholded so that only puncta that were several folds higher than background were selected . Punctas were evaluated using the Analyze Particles function in ImageJ . HEK cells stably expressing α4β2 receptors maintained as described above were plated at low density on glass coverslips before treating for 15–18 hr with media containing nicotine or other drugs as per the protocol for binding experiments . Following the incubation , cells were live-labeled with a 1:200 dilution of mouse anti-HA antibody ( EMD Millipore #05–904 ) for 1 hr at 37°C in media containing the same nicotinic agents . Cells were washed three times with media lacking primary antibody before addition of a 1:500 dilution of the secondary antibody ( Thermo-Fisher goat anti-mouse Alexa Fluor 488 , catalog # A11029 ) for 1 hr at room temperature in media containing the same nicotinic agents . Cells were washed three times with media before use in physiology experiments . For NH4Cl treatment experiments , following the HA labeling the cells were washed twice with PBS and exposed to two 10 min incubations with 20 mM NH4Cl and several final washes with PBS before use in recordings . Cells were only used for recording a maximum of one hour after completion of the HA labeling or NH4Cl treatment . Cells were voltage clamped in whole-cell configuration at a holding potential of −70 mV using an Axopatch 200B amplifier running pClamp 10 ( Molecular Devices; Sunnyvale , CA ) . Currents were elicited from cells lifted from the coverslip by the fast application of 1 mM ACh using a piezo-ceramic bimorph system with a solution exchange time of ~1 ms . External solution consisted of ( in mM ) : 150 NaCl , 2 . 8 KCl , 1 . 8 CaCl2 , 1 . 0 MgCl2 , 10 glucose , and 10 HEPES , adjusted to pH 7 . 3 . Internal pipette solution was ( in mM ) : 110 CsF , 30 CsCl , 4 NaCl , 0 . 5 CaCl2 , 10 Hepes , and 5 EGTA , adjusted to pH 7 . 3 . Peak amplitudes and time courses of desensitization were determined by post-hoc analysis using Clampfit 10 . Statistical differences were tested for using one-way ANOVAs with Tukey’s Multiple Comparison Test in GraphPad ( La Jolla , CA ) . All statistical analyses were performed using StatPlus software ( AnalystSoft Inc , Walnut , CA ) , unless otherwise noted . Statistical tests used are indicated in each figure legend .	Tobacco continues to be widely used worldwide , primarily via cigarette smoking , and is a leading cause of preventable deaths . Stopping smoking is difficult because the nicotine in tobacco is highly addictive , and so several drugs have been developed to help people break their addiction . Varenicline ( also known by the trade name Chantix ) is a commonly prescribed anti-smoking drug , but it is not fully understood how it works . Nicotine affects the brain by binding to proteins called nicotinic acetylcholine receptors ( nAChRs ) that sit on the surface of neurons . This binding releases a number of chemical signals , including some that produce feelings of pleasure . Over time , the receptors become less sensitive to nicotine and produce more “high-affinity” binding sites for nicotine to bind to . This adaptation is one reason why stopping smoking can produce strong feelings of withdrawal . Previously , varenicline was thought to partially activate nAChRs , preventing nicotine from binding to the receptors . However , this can only explain how varenicline counteracts the rapid-acting effects of nicotine , not nicotine’s longer-term effects . Furthermore , it was not known how nAChR signaling responds to long-term exposure to a combination of both drugs ( as occurs when people try to quit smoking with the aid of varenicline ) . Now , Govind et al . reveal how varenicline reverses the effect of long-term nicotine exposure on nAChR signaling . Both varenicline and nicotine accumulate in acidic compartments – called vesicles – inside cells , where they become charged and less able to move through the cell membrane . When the vesicles also contain high-affinity nAChRs , varenicline becomes trapped inside them and is only slowly released . By contrast , nicotine is not trapped because it exits the vesicles more rapidly . Long-term exposure to nicotine greatly increased the number of vesicles that contained high-affinity nAChRs , thereby trapping more varenicline . One consequence of trapping varenicline was that the activity of the nAChRs on the surface of the neuron was diminished , apparently through the slow release of the trapped varenicline from the acidic vesicles . This slow release causes the receptors to enter a “desensitized” state in which they do not signal . Understanding how varenicline counteracts the long-term effects of nicotine on nAChR signaling will help us to design more effective anti-smoking drugs . Govind et al . also found that compounds similar to varenicline become trapped in vesicles , but it is not clear how the degree of trapping of a compound correlates with how effectively it combats nicotine addiction . The results may also help us to understand and treat addictions to other drugs of abuse , such as opioids , amphetamines and cocaine , which have chemical properties that mean they might also be selectively trapped in acidic vesicles .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Selective and regulated trapping of nicotinic receptor weak base ligands and relevance to smoking cessation
Dietary magnesium ( Mg2+ ) supplementation can enhance memory in young and aged rats . Memory-enhancing capacity was largely ascribed to increases in hippocampal synaptic density and elevated expression of the NR2B subunit of the NMDA-type glutamate receptor . Here we show that Mg2+ feeding also enhances long-term memory in Drosophila . Normal and Mg2+-enhanced fly memory appears independent of NMDA receptors in the mushroom body and instead requires expression of a conserved CNNM-type Mg2+-efflux transporter encoded by the unextended ( uex ) gene . UEX contains a putative cyclic nucleotide-binding homology domain and its mutation separates a vital role for uex from a function in memory . Moreover , UEX localization in mushroom body Kenyon cells ( KCs ) is altered in memory-defective flies harboring mutations in cAMP-related genes . Functional imaging suggests that UEX-dependent efflux is required for slow rhythmic maintenance of KC Mg2+ . We propose that regulated neuronal Mg2+ efflux is critical for normal and Mg2+-enhanced memory . Magnesium ( Mg2+ ) plays a critical role in cellular metabolism and is considered to be an essential co-factor for more than 350 enzymes ( Romani and Scarpa , 2000; Vink and Nechifor , 2011 ) . As a result , alterations of Mg2+ homeostasis are associated with a broad range of clinical conditions , including those affecting the nervous system , such as glaucoma ( DeToma et al . , 2014 ) , Parkinson’s disease ( Hermosura et al . , 2005; Hermosura and Garruto , 2007; Lin et al . , 2014; Shindo et al . , 2016 ) , Alzheimer’s disease ( Andrási et al . , 2000; Andrási et al . , 2005; Cilliler et al . , 2007; Durlach et al . , 1997; Glick , 1990; Lemke , 1995; Chui et al . , 2011; Vural et al . , 2010 ) , anxiety ( Sartori et al . , 2012 ) , depression ( Whittle et al . , 2011; Murck , 2002; Murck , 2013; Rasmussen et al . , 1990; Ghafari et al . , 2015 ) , and intellectual disability ( Arjona et al . , 2014 ) . Perhaps surprisingly , increasing brain Mg2+ through diet can enhance neuronal plasticity and memory performance of young and aged rodents , measured in a variety of behavioral tasks ( Slutsky et al . , 2010; Landfield and Morgan , 1984; Mickley et al . , 2013; Abumaria et al . , 2013 ) . In addition , elevated Mg2+ reduced cognitive deficits in a mouse model of Alzheimer’s disease ( Li et al . , 2013 ) and enhanced the extinction of fear memories ( Abumaria et al . , 2011 ) . These apparently beneficial effects have led to the proposal that dietary Mg2+ may have therapeutic value for patients with a variety of memory-related problems ( Billard , 2011 ) . Despite the large number of potential sites of Mg2+ action in the brain , the memory-enhancing property in rodents has largely been attributed to increases in hippocampal synaptic density and the activity of N-methyl-D-aspartate glutamate receptors ( NMDARs ) . Extracellular Mg2+ blocks the channel pore of the NMDAR and thereby inhibits the passage of other ions ( Mayer et al . , 1984; Bekkers and Stevens , 1993; Jahr and Stevens , 1990; Nowak et al . , 1984 ) . Importantly , prior neuronal depolarization , driven by other transmitter receptors , is required to release the Mg2+ block on the NMDAR and permit glutamate-gated Ca2+ influx . The NMDAR therefore plays an important role in neuronal plasticity as a potential Hebbian coincidence detector . Acute elevation of extracellular Mg2+ concentration ( [Mg2+]e ) within the physiological range ( 0 . 8–1 . 2 mM ) can antagonize induction of NMDAR-dependent long-term potentiation ( Dunwiddie and Lynch , 1979; Malenka et al . , 1992; Malenka and Nicoll , 1993; Slutsky et al . , 2004 ) . In contrast , increasing [Mg2+]e for several hours in neuronal cultures leads to enhancement of NMDAR mediated currents and facilitation of the expression of LTP ( Slutsky et al . , 2004 ) . The enhancing effects of increased [Mg2+]e were also observed in vivo in the brain of rats fed with Mg2+-L-threonate ( Slutsky et al . , 2010 ) . Hippocampal neuronal circuits undergo homeostatic plasticity ( Turrigiano , 2008 ) to accommodate the increased [Mg2+]e by upregulating expression of NR2B subunit containing NMDARs ( Slutsky et al . , 2004; Slutsky et al . , 2010 ) . The higher density of hippocampal synapses with NR2B containing NMDARs are believed to compensate for the chronic increase in [Mg2+]e by enhancing NMDAR currents during burst firing . In support of this model , mice that are genetically engineered to overexpress NR2B exhibit enhanced hippocampal LTP and behavioral memory ( Tang et al . , 1999 ) . Olfactory memory in Drosophila involves a heterosynaptic mechanism driven by reinforcing dopaminergic neurons , which results in presynaptic depression of cholinergic connections between odor-activated mushroom body ( MB ) Kenyon cells ( KCs ) and downstream mushroom body output neurons ( MBONs ) ( Schwaerzel et al . , 2003; Aso et al . , 2010; Aso et al . , 2012; Claridge-Chang et al . , 2009; Burke et al . , 2012; Liu et al . , 2012; Plaçais et al . , 2013; Owald et al . , 2015; Hige et al . , 2015; Barnstedt et al . , 2016; Perisse et al . , 2016; Aso et al . , 2014; Owald and Waddell , 2015 ) . In addition , olfactory information is conveyed to KCs by cholinergic transmission from olfactory projection neurons ( Yasuyama et al . , 2002; Leiss et al . , 2009 ) . Although it is conceivable that glutamate is delivered to the MB network via an as yet to be identified route , there is currently no obvious location for NMDAR-dependent plasticity in the known architecture of the cholinergic input or output layers ( Barnstedt et al . , 2016 ) . The fly therefore provides a potential model to investigate other mechanisms through which dietary Mg2+ might enhance memory . The reinforcing effects of dopamine depend on the Dop1R D1-type dopamine receptor ( Kim et al . , 2007; Qin et al . , 2012; Handler et al . , 2019 ) , which is positively coupled with cAMP production ( Tomchik and Davis , 2009; Boto et al . , 2014 ) . Moreover , early studies in Drosophila identified the dunce and rutabaga encoded cAMP phosphodiesterase and type I Ca2+-stimulated adenylate cyclase , respectively , to be essential for olfactory memory ( Dudai et al . , 1976; Byers et al . , 1981; Dudai and Zvi , 1984; Chen et al . , 1986; Livingstone et al . , 1984; Levin et al . , 1992 ) . Studies in mammalian cells have shown that hormones or agents that increase cellular cAMP level often elicit a significant Na+-dependent extrusion of Mg2+ into the extracellular space ( Romani and Scarpa , 1990b; Romani and Scarpa , 1990a; Romani and Scarpa , 2000; Vink and Nechifor , 2011; Vormann and Günther , 1987 ) . However , it is unclear whether Mg2+ extrusion plays any role in memory processing . Here we demonstrate that Drosophila long-term memory ( LTM ) can be enhanced with dietary Mg2+ supplementation . We find that the unextended ( uex ) ( Maeda , 1984; Coulthard et al . , 2010 ) gene , which encodes a functional fly ortholog of the mammalian Cyclin M2 Mg2+-efflux transporter ( CNNM ) proteins , is critical for the memory enhancing property of Mg2+ . UEX function in MB KCs is required for LTM and functional restoration of uex reveals the MB to be the key site of Mg2+-dependent memory enhancement . Chronically changing cAMP metabolism by introducing mutations in the dnc or rut genes alters the cellular localization of UEX . Moreover , mutating the conserved cyclic nucleotide-binding homology ( CNBH ) domain in UEX uncouples an essential role for uex from its function in memory . UEX-driven Mg2+ efflux is required for slow rhythmic maintenance of KC Mg2+ levels suggesting a potential role for Mg2+ flux in memory processing . Prior studies reported that feeding rats with food containing a high concentration of Mg2+-enhanced their learning and memory capability ( Slutsky et al . , 2010; Landfield and Morgan , 1984; Abumaria et al . , 2011; Mickley et al . , 2013; Abumaria et al . , 2013 ) . We therefore tested whether similar effects exist in flies by feeding them with food containing a high concentration of Mg2+ before training . Surprisingly , wild-type flies fed for 4 days before training with food supplemented with additional magnesium chloride ( MgCl2 ) exhibited significantly enhanced 24 hr memory performance . Memory enhancement depends on concentration and was maximal when food was supplemented with 80 mM MgCl2 ( Figure 1A ) . Immediate memory performance was not obviously enhanced ( Figure 1B ) . The enhancing effect of MgCl2 was also observed in flies fed with magnesium sulfate ( MgSO4 ) but not calcium chloride ( CaCl2 ) ( Figure 1C ) . In addition , feeding flies for 4 days with food containing between 5 and 80 mM strontium chloride ( SrCl2 ) resulted in high levels of mortality and flies that survived 5 mM SrCl2 feeding did not show enhanced immediate or 24 hr memory performance ( data not shown ) . The memory enhancing effects can therefore be specifically attributed to dietary supplementation of divalent Mg2+ . Since magnesium-L-threonate enhanced memory in rats was correlated with an upregulation of hippocampal NR2B subunit-containing NMDARs ( Slutsky et al . , 2010 ) , we tested for changes in glutamate receptor expression in flies fed with MgCl2 . RT-qPCR analyses did not reveal a significant difference in the abundance of mRNAs for the putative NMDA ( Nmdar1 , Nmdar2 ) , AMPA ( GluRIA ) , or kainate-type ( GluRIIA ) receptors in heads taken from flies fed for 4 days with 80 mM MgCl2 versus those fed with 1 mM MgCl2 ( Figure 1D ) . We next directly tested whether Mg2+-enhanced memory required NMDAR function , by knocking down expression of the Nmdar1 or Nmdar2 genes using transgenic UAS-driven RNA interference ( RNAi ) constructs ( Dietzl et al . , 2007; Perkins et al . , 2015 ) . Of the two independent UAS-Nmdar1RNAi and four UAS-Nmdar2RNAi lines we tested , only one Nmdar1RNAi ( BDSC 25941 ) line , when driven in all neurons by neuronal Synaptobrevin ( nSyb ) -GAL4 , exhibited significantly decreased 24 hr memory performance , as compared to that of heterozygous control flies ( Figure 1—figure supplement 1A ) . In contrast , more selective expression of this UAS-Nmdar1RNAi in LTM-relevant αβ KCs using c739-GAL4 did not significantly impair 24 hr memory performance ( Figure 1—figure supplement 1B ) . Moreover , flies expressing Nmdar1RNAi in αβ neurons retained robust Mg2+-enhanced memory ( Figure 1—figure supplement 1C ) . These results suggest that Mg2+-enhanced memory does not alter expression of glutamate receptors , or require NMDAR function in αβ KCs . We used MagFRET , the first genetically encoded fluorescent Mg2+ sensor ( Lindenburg et al . , 2013 ) , to test whether Mg2+ feeding altered the intracellular Mg2+ concentration ( [Mg2+]i ) . We constructed flies harboring a UAS-MagFRET-1 transgene and combined it with c739-GAL4 to express MagFRET-1 in αβ KCs . We compared the FRET signals in fixed brains from c739; UAS-MagFRET-1 flies fed with either 1 mM or 80 mM MgCl2 food for 4 days . The MagFRET signal was significantly higher in both the α and β collaterals of αβ KCs of flies fed with 80 mM , than in those fed with 1 mM ( Figure 1E ) . This result indicates that Mg feeding elevates neuronal [Mg2+]i . Given the affinity of MagFRET-1 ( Kd = 148 µM ) and the ~50% increase in FRET signal upon Mg2+ binding ( Lindenburg et al . , 2013 ) , we estimate that the ~8% enhancement of the MagFRET signal measured in flies fed 80 mM MgCl2 corresponds approximately to a 50 µM increase of αβ KC [Mg2+]i on average . We identified unextended ( uex; Maeda , 1984; Coulthard et al . , 2010 ) as a gene altering appetitive olfactory LTM , reinforced with sucrose reward . Flies with the uexMI01943 MiMIC insertion ( Venken et al . , 2011 ) showed a strong defect in 24 hr memory , but their performance immediately after training was indistinguishable from that of wild-type controls . More detailed analysis of uexMI01943 flies revealed a steady decay of memory that first became significantly different to that of wild-type flies 12 hr after training ( Figure 2A ) . No memory defect was evident in heterozygous uexMI01943/+ flies , demonstrating that this putative uex allele is recessive . uex piqued our attention because it is the single fly ortholog of the four human CNNM genes that encode Mg2+ transporters ( Ishii et al . , 2016 ) , and it also contains a putative CNBH domain that is structurally related to those in cyclic nucleotide-gated channels ( Zagotta et al . , 2003; Flynn et al . , 2007; Kesters et al . , 2015 ) . Alignment of the 834 amino acid UEX sequence with CNNM1-4 reveals particularly high sequence conservation with CNNM2 and CNNM4 in the DUF21 , CBS pair , and CNBH domains ( Figure 2—figure supplement 1A–C ) . We therefore hypothesized that UEX had potential to link the memory-enhancing effects of dietary Mg2+ with cAMP-dependent neuronal plasticity . Although uexMI01943 is assigned to the uex gene , the MiMIC element is annotated to lie 17 kb downstream of the uex coding region ( Venken et al . , 2011; Figure 2B ) . RYa ( Yoon et al . , 2016 ) is the next nearest gene to uexMI01943 but is >230 kb further away . We first confirmed the MiMIC location by inverse PCR ( Attrill et al . , 2016 ) . Importantly , no additional MiMIC insertion was detected in these flies . We next tested whether uexMI01943 was responsible for the memory defect by precisely removing the MiMIC element by Minos transposase-mediated excision ( Arcà et al . , 1997; Figure 2—figure supplement 2A and B ) . MiMIC removal in uexMI01943 . ex1 and uexMI01943 . ex2 flies restored normal 24 hr memory performance , demonstrating that the MiMIC insertion is required for the uexMI01943 memory defect ( Figure 2C ) . Both qRT-PCR of mRNA and western blot analysis of protein extracts from fly heads failed to reveal a significant difference in uex/UEX expression in uexMI01943 flies . We therefore used CRISPR to introduce a stop codon into the fifth coding exon of the uex locus ( Figure 2B and Figure 2—figure supplement 2C ) . Flies homozygous for the resulting uexΔ mutation were not viable as adults , dying at the larval stage . In contrast , heterozygous uexMI01943/uexΔ flies were viable , but their 24 hr appetitive memory was significantly impaired ( Figure 2D ) . These data demonstrate that uex is an essential gene and that uexMI01943 is a viable hypomorphic allele of uex . We also tested the aversive memory performance of uexMI01943 mutant flies . Homozygous uexMI01943 flies exhibited immediate memory that was indistinguishable from that of heterozygous and wild-type controls ( Figure 2E ) . However , their 24 hr memory , formed following either five trials of aversive spaced training ( Tully et al . , 1994; Jacob and Waddell , 2020 ) , or one trial of fasting facilitated training ( Hirano et al . , 2013 ) , was significantly impaired ( Figure 2E ) . These experiments suggest that uexMI01943 flies are more generally compromised in their ability to form LTM . Unless otherwise specified , all subsequent analyses of memory in this study use appetitive sugar-rewarded conditioning . To localize uex in the brain we first took advantage of VT23256-GAL4 transgenic flies , in which GAL4 is driven by an 853 bp sequence from the first intron of uex ( Kvon et al . , 2014 ) . VT23256-driven UAS-EGFP revealed restricted expression in αβ KCs with particularly strong label in αβ core ( αβc ) neurons ( Figure 3A ) . We also used CRISPR to insert a C-terminal HA-epitope tag into the uex open reading frame ( Figure 3—figure supplement 1A ) . These flies were viable as homozygotes indicating that the resulting UEX::HA fusion protein retains function . Immunostaining flies harboring this uex::HA locus with an anti-HA antibody revealed prominent labeling of all the major KC classes in the MB , in addition to lower expression throughout the brain ( Figure 3B ) . This uex expression profile is also supported by single-cell sequencing analyses ( Figure 3—figure supplement 1B; Croset et al . , 2018; Davie et al . , 2018 ) . Given the established role for αβ KCs in olfactory LTM ( Pascual and Préat , 2001; Yu et al . , 2006; Krashes et al . , 2007; Krashes and Waddell , 2008 ) , we reasoned that a mnemonic role for UEX may involve expression in KCs . We next used GAL4-directed expression of RNAi to test whether 24 hr memory performance required uex in the MB . Flies expressing uexRNAi ( Perkins et al . , 2015 ) in all αβ KCs ( c739-GAL4; Yang et al . , 1995; Perisse et al . , 2013 ) or only in αβc KCs ( NP7175-GAL4; Tanaka et al . , 2008 ) showed normal immediate memory but significantly impaired 24 hr memory ( Figure 3C ) . In contrast , uexRNAi expression in αβ surface ( αβs , 0770-GAL4; Perisse et al . , 2013 ) or α′β′ KCs ( c305a-GAL4; Krashes et al . , 2007 ) did not significantly alter immediate or LTM performance . Normal 24 hr appetitive memory performance is therefore particularly sensitive to uex expression in αβc neurons . To reduce the likelihood that the uexRNAi associated memory defect results from a developmental consequence , we also restricted UAS-uexRNAi expression to adulthood using GAL80ts-mediated temporal control ( McGuire et al . , 2003 ) . At permissive 18°C , GAL80ts binds to GAL4 and suppresses its transcriptional activator function . At restrictive 30°C , GAL80ts can no longer bind to GAL4 , which frees GAL4 to direct expression of the UAS-uexRNAi transgene . Flies were raised through development at 18°C and moved to 30°C after eclosion . Restricting UAS-uexRNAi expression to αβ KCs in adult flies using c739-GAL4 with GAL80ts produced a similar 24 hr specific memory defect to that observed when UAS-uexRNAi was expressed without temporal control ( Figure 3D–F ) . We assessed the efficacy of the UAS-uexRNAi knockdown using our tagged uex::HA locus . Brains from heterozygous uex::HA flies expressing uexRNAi in the αβ and γ KCs with MB247-GAL4 ( Zars et al . , 2000 ) were immunostained using anti-HA antibody . Comparing the intensity of immunolabeling in brains from uex::HA; MB247-GAL4/uexRNAi flies with that from uex::HA; MB247-GAL4/+ flies showed that uexRNAi expression significantly reduced anti-HA signal in the αβ and γ KCs ( Figure 3G and H ) . This result demonstrates the efficiency of the uexRNAi transgene and the utility of the CRISPR/Cas9 edited uex::HA locus . We next tested whether expression in specific KCs of an UAS-uex transgene could restore 24 hr memory capacity to uexMI01943 flies . Memory performance of uexMI01943 flies expressing UAS-uex in αβ and γ KCs ( MB247-GAL4; Zars et al . , 2000 ) or only the αβ KCs ( c739-GAL4 ) was significantly improved over that of uexMI01943 flies , and was statistically indistinguishable from that of controls with an intact uex locus ( Figure 4A ) . In contrast , UAS-uex expression in α′β′ , αβc , or αβs KCs did not restore memory performance to uexMI01943 flies and overexpressing uex in αβ KCs of wild-type flies did not augment 24 hr memory ( Figure 4A and B ) . Normal 24 hr memory performance could also be restored to uexMI01943 flies if UAS-uex expression was confined to c739-GAL4 neurons ( all αβ KCs ) in adulthood using GAL80ts-mediated temporal control ( Figure 4C and D ) . Together , these loss-of-function RNAi and restoration experiments establish that UEX plays an important role in adult αβ KCs . Finding that αβc RNAi knockdown of uex produces a memory defect ( Figure 3C ) but UAS-uex expression in αβc does not rescue the uexMI01943 mutant defect ( Figure 4A ) suggests that UEX function in αβc KCs is essential for appetitive LTM , whereas both the αβc and αβs KCs need to have functional UEX to support LTM . In addition , the ability of UAS-uex to restore performance to uexMI0194 flies provides further support that uex is responsible for the memory impairment in uexMI01943 flies . We next investigated whether Mg2+ feeding ( 4 days with 80 mM MgCl2 ) could improve memory performance in flies with compromised uex function . Flies carrying the uexMI01943 allele ( Figure 4F ) or those expressing UAS-uexRNAi in the αβ KCs with c739-GAL4 ( Figure 4E ) did not show enhanced memory when fed with 80 mM MgCl2 , as compared to flies fed with 1 mM MgCl2 . Moreover , the Mg2+-enhanced memory was recovered in uexMI01943 mutant flies when uex expression was restored to the αβ KCs ( Figure 4F ) . All control flies ( c739-GAL4 , UAS-uexRNAi , and UAS-uex ) with unperturbed uex expression exhibited significantly enhanced memory when fed with 80 mM as compared to 1 mM MgCl2 . Overexpressing UAS-uex in αβ KCs with c739-GAL4 in flies with a wild-type genetic background neither enhanced regular 24 hr memory ( Figure 4B ) , or that in flies fed for 4 days with 40 or 80 mM MgCl2 ( Figure 4G ) . We also tested whether 4 days of 80 mM MgCl2 supplementation enhanced 24 hr memory performance following aversive spaced training . Again , memory of wild-type , but not uexMI01943 mutant flies showed enhancement ( Figure 4—figure supplement 1 ) . Together these results indicate that optimal memory enhancement with Mg2+ feeding requires , and can be fully supported by , UEX function in αβ KCs . Given the strong sequence conservation of UEX with mammalian CNNM2/4 we tested whether CNNM2 could functionally substitute for UEX and restore the LTM defect of uexMI01943 flies . Several point mutations in CNNM2 have been identified in human patients with hypomagnesemia , which is associated with brain malformation and intellectual disability ( Arjona et al . , 2014 ) . Introduction of the equivalent mutations into mouse CNNM2 ( CNNM2E357K , CNNM2T568I , CNNM2S269W , and CNNM2E122K ) showed that these patient-derived lesions impair magnesium transport ( Arjona et al . , 2014 ) . We constructed flies carrying wild-type and these mutant variant UAS-CNNM2 transgenes ( Figure 5A ) . Staining for an associated C-terminal HA-tag revealed clear expression of all UAS-CNNM2::HA variants in αβ neurons when driven with c739-GAL4 ( Figure 5—figure supplement 1 ) . However , only expression of wild-type CNNM2 , and not point-mutant forms , in αβ KCs of uexMI01943 mutant flies restored 24 hr memory performance ( Figure 5B ) . We also tested whether UEX can mediate Mg2+ extrusion . UEX expressed in HEK293 cells localized to the plasma membrane and cells loaded with Mg2+ and the Mg2+ indicator Magnesium Green showed rapid Mg2+ efflux ( Figure 5—figure supplement 2 and Video 1 ) , as compared to cells transfected with empty vector . Mg2+ extrusion driven by UEX was noticeably less efficient than in cells expressing Human CNNM4 ( Figure 5—figure supplement 2 ) , which is known to have similar efficiency to CNNM2 ( Hirata et al . , 2014 ) . However , we do not know if UEX and CNNM4 expression is equivalent . Nevertheless , demonstration of cross-species complementation and Mg2+ efflux activity defines UEX as a functional homolog of mammalian CNNM2/4 . Given the established role for cAMP signaling in memory-relevant plasticity in invertebrates and mammals ( Kandel , 2012 ) , we tested the importance of the CNBH domain in UEX . We constructed flies carrying a point-mutated CNBH UAS-uexR622K transgene ( Figure 6A ) . The equivalent R622K amino acid substitution abolishes cAMP binding in the regulatory subunit of cAMP-dependent protein kinase , PKA ( Bubis et al . , 1988 ) . Expressing UAS-uexR622K in αβ neurons with c739-GAL4 did not restore 24 hr memory performance , or alter the immediate memory performance , of uexMI01943 mutant flies ( Figure 6B ) . We also used CRISPR to attempt to introduce the R622K mutation into the CNBH of the native uex locus ( Bassett et al . , 2013; Gratz et al . , 2013; Yu et al . , 2013 ) . Unexpectedly , this approach did not introduce the R622K substitution but instead replaced T626 in the CNBH with NRR . Fortuitously , flies homozygous for this uexT626NRR allele were viable as adults , unlike those homozygous for uexΔ , suggesting that the uexT626NRR encoded UEX retains function . However , flies homozygous for uexT626NRR or heterozygous uexT626NRR/ uexMI01943 flies exhibited a strong 24 hr memory defect ( Figure 6C ) . Immediate memory was also impaired in homozygous uexT626NRR flies , unlike flies carrying all other combinations of uex alleles . In addition , memory of uexT626NRR flies could not be enhanced with Mg2+ feeding ( Figure 6D ) . The uexT626NRR mutation therefore uncouples the essential role for uex from a function in memory and suggests that cyclic nucleotide regulated activity is critical for UEX to support normal and Mg2+-enhanced memory . Although we confirmed using western blotting that a full-length protein is expressed in uexT622NRR flies ( Figure 6E ) , our antibody did not permit us to verify that the UEXT626NRR protein localizes appropriately in the brain . Further work is therefore required to characterize the cellular localization , cAMP binding , and Mg2+ transport function of the protein encoded by this serendipitous uexT626NRR allele . We tested whether cAMP could acutely alter UEX activity by applying forskolin to UEX-expressing HEK293 cells . However , no obvious change in the UEX-dependent Mg2+ efflux dynamic was observed ( data not shown ) . We therefore tested whether KC expression of UEX::HA was altered in flies with chronic alterations of cAMP metabolism , by introducing learning-relevant mutations in the rutabaga-encoded Ca2+-stimulated adenylate cyclase , or the dunce-encoded cAMP-specific phosphodiesterase . Anti-HA immunostaining of brains from rut2080; uex::HA and dnc1; uex::HA flies revealed a striking change in UEX localization ( Figure 7A and B and Videos 2–4 ) . Whereas UEX::HA is usually detected in the lobes of all KCs at a roughly equivalent level in wild-type flies , labeling was lower in the MB γ lobe and more pronounced in the αβc KCs in rut2080 and dnc1 mutant backgrounds ( Figure 7C ) , although the overall MB expression of UEX::HA is similar between wild-type and mutant flies ( Figure 7D ) . In addition , western blot analyses of protein extracted from heads of these flies did not reveal a significant difference in overall UEX::HA expression levels ( data not shown ) . These data are therefore consistent with cAMP regulating UEX function and perhaps its cellular localization in KCs . Although MagFRET can report [Mg2+] it does not respond quickly enough to record stimulus-evoked signals . We therefore constructed flies harboring UAS-transgenes for two newer genetically encoded Mg2+ sensors , MagIC ( non-FRET based; Koldenkova et al . , 2015 ) and MARIO ( FRET based; Maeshima et al . , 2018 ) . We were unable to detect UAS-MARIO expression in the fly brain and therefore could only use UAS-MagIC . MagIC was reported to respond most strongly to Mg2+ but also to a lesser extent to Ca2+ ( Koldenkova et al . , 2015 ) . We therefore first verified the specificity of MagIC responses in a cell-permeabilized ex vivo fly brain preparation . Brains were removed from flies expressing UAS-MagIC in αβ KCs with c739-GAL4 ( Figure 8A ) , incubated in a dish with saline ( Barnstedt et al . , 2016 ) and changes in fluorescence were monitored before and after bath application of chemicals . Whereas application of MgCl2 evoked a dose-dependent increase in the MagIC response , chelation of Mg2+ with EDTA produced a dose-dependent decrease ( Figure 8B and Videos 5 and 6 ) . In comparison , CaCl2 only registered a slight increase at the highest concentrations whereas the more Ca2+-selective chelator EGTA had little effect ( Figure 8B ) . These results demonstrate that UAS-MagIC can monitor [Mg2+]i in the αβ KCs in the fly brain . Increasing intracellular cAMP has been shown to elicit Mg2+ flux from mammalian cells ( Romani and Scarpa , 2000; Vormann and Günther , 1987; Jakob et al . , 1989; Romani and Scarpa , 1990b; Romani and Scarpa , 1990a; Vormann and Günther , 1987; Günther et al . , 1990; Howarth et al . , 1994 ) . Since our experiments also indicated that cAMP might regulate UEX , we next tested whether stimulating cAMP synthesis with forskolin ( FSK ) might alter MagIC signals in αβ KCs . For these experiments we again used an ex vivo brain preparation but this time the cells were not permeabilized . 30 μM FSK has been shown to evoke a peak increase in cAMP in KCs that approximates that observed following appetitive conditioning ( Louis et al . , 2018 ) . Applying 30 μM FSK to c739-GAL4; UAS-MagIC brains evoked a consistent dynamic in MagIC fluorescence . After a sharp initial rise , responses slowly decayed back toward baseline before again rising slowly to a point at which the signal started to fluctuate . ( Figure 8C and D and Video 7 ) . The key signatures of this response were only recorded in the Mg2+-sensitive Venus signal ( Figure 8D ) . In contrast mCherry fluorescence did not fluctuate but steadily decreased across the time course of the recording ( likely a result of photo-bleaching ) , demonstrating that the fluctuation in the Venus signal is not a movement artifact ( Figure 8E ) . Importantly , FSKinduced MagIC responses were greater than those following application of saline ( Figure 8—figure supplement 1A ) . However , a fluctuating response also developed after saline applications ( Figure 8—figure supplement 1B ) suggesting that the rhythmic MagIC signal may be a general response to an increase in [Mg2+]i that follows cellular perturbation . The Drosophila MB has previously been reported to exhibit a slow ( 0 . 004 Hz ) Ca2+ oscillation in ex vivo brains whereas a much faster 20 Hz oscillation is evoked by odors in the locust MB ( Laurent and Naraghi , 1994; Rosay et al . , 2001 ) . Although our initial characterization of MagIC in the fly brain indicated a preferential response to Mg2+ ( Figure 8B ) , we nevertheless explicitly tested whether FSK induced fluctuation of the [Ca2+]i of αβ KCs , using expression of UAS-GCaMP6f ( Chen et al . , 2013 ) . FSK induced a delayed increase in the GCaMP response but no clear oscillatory activity was observed ( Figure 8—figure supplement 1C–E ) . Lastly , we tested whether the observed MagIC responses were sensitive to the status of the uex gene . We generated uexMI01943 flies that also harbored c379-GAL4 and UAS-MagIC and compared their FSK- and saline-induced MagIC responses to those of flies with a wild-type uex locus . The uexMI01943 mutant flies showed an increased FSK response to that of wild-type flies , whereas saline-evoked responses were indistinguishable ( Figure 8F and G ) . Responses evoked by the inactive FSK analogue , ddFSK , were also insensitive to the status of uex ( Figure 8—figure supplement 1F ) . Mutation of uex therefore selectively increases mean FSK-evoked MagIC responses . We also noticed that MagIC traces from uex mutant flies did not exhibit a fluctuating signal ( Figure 8H and Figure 8—figure supplement 1G ) . To quantify this difference we calculated the mean power spectral density ( PSD ) of traces from uexMI01943 and wild-type flies treated with FSK or saline . In both conditions the mean PSD was significantly left-shifted toward lower frequencies in the uexMI01943 mutants compared to the wild-type controls ( Figure 8I ) . Wild-type fly brains had significantly more oscillatory activity centered around 0 . 015 Hz than those from uexMI01943 mutants . These data therefore suggest that UEX is required for slow rhythmic maintenance of KC [Mg2+]i . Importantly , finding that MagIC signals are elevated and altered in uex mutants confirms that the observed MagIC responses are Mg2+-dependent . Moreover , they suggest that the KC expressed UEX limits Mg2+ accumulation , consistent with a role in extrusion . We observed an enhancement of olfactory LTM performance when flies were fed for 4 days before training with food supplemented with 80 mM [Mg2+] . This result resembles that reported in rats , although longer periods of feeding were required to raise brain [Mg2+] to memory-enhancing levels ( Slutsky et al . , 2010 ) . A difference in optimal feeding time may reflect the size of the animal and perhaps the greater bioavailability of dietary Mg2+ in Drosophila . Whereas Mg2+-L-threonate ( MgT ) was a more effective means of delivering Mg2+ than magnesium chloride in rats ( Slutsky et al . , 2010 ) , we observed a similar enhancement of memory performance when flies were fed with magnesium chloride , magnesium sulfate , or MgT ( data not shown ) . Elevating [Mg2+]e in the rat brain leads to a compensatory upregulation of expression of the NR2B subunit of the NMDAR and therefore an increase in the proportion of postsynaptic NR2B-containing NMDARs . This class of NMDARs have a longer opening time ( Chen et al . , 1999; Erreger et al . , 2005 ) suggesting that this switch in subunit composition represents a homeostatic plasticity mechanism ( Turrigiano , 2008 ) to accommodate for the increased NMDAR block imposed by increasing [Mg2+]e . Moreover , overexpression of NR2B in the mouse forebrain can enhance synaptic facilitation and learning and memory performance ( Tang et al . , 1999 ) , supporting an increase in NR2B being an important factor in Mg2+-enhanced memory . However , even in the original in vitro study of Mg2+-enhanced synaptic plasticity ( Slutsky et al . , 2004 ) , it was noted that NMDAR currents were insufficient to fully explain the observed changes . Our NMDAR subunit loss-of-function studies in the Drosophila KCs did not impair regular or Mg2+-enhanced memory . Furthermore , we did not detect an obvious change in the levels of brain-wide expression of glutamate receptor subunits in Mg2+-fed flies . Although NMDAR activity has previously been implicated in Drosophila olfactory memory , the effects were mostly ascribed to function outside the MB ( Xia et al . , 2005; Wu et al . , 2007 ) . In addition , overexpressing Nmdar1 in all neurons , or specifically in all KCs , did not alter STM or LTM . Ectopic overexpression in the MB of an NMDARN631Q version , which cannot be blocked by Mg2+ , impaired LTM ( Miyashita et al . , 2012 ) . However , this mutation permits ligand-gated Ca2+ entry , without the need for correlated neuronal depolarization , which may perturb KC function in unexpected ways . It is perhaps most noteworthy that learning-relevant synaptic depression in the MB can be driven by dopaminergic teaching signals delivered to cholinergic output synapses from odor-responsive KCs to specific MBONs ( Claridge-Chang et al . , 2009; Aso et al . , 2012; Burke et al . , 2012; Liu et al . , 2012; Owald et al . , 2015; Hige et al . , 2015; Barnstedt et al . , 2016; Perisse et al . , 2016; Aso et al . , 2014; Owald and Waddell , 2015; Handler et al . , 2019 ) . It is conceivable that KCs receive glutamate , from a source yet to be identified , but there is currently no obvious place in the MB network for NMDAR-dependent plasticity . Evidence therefore suggests that normal and Mg2+-enhanced Drosophila LTM is independent of NMDAR signaling in KCs . In addition , our MagFRET measurements indicate that Mg2+ feeding also increases the [Mg2+]i of αβ KCs by approximately 50 µM . We identified a role for uex , the single fly ortholog of the evolutionarily conserved family of CNNM-type Mg2+ efflux transporters ( Ishii et al . , 2016 ) . There are four distinct CNNM genes in mice and humans , five in C . elegans , and two in zebrafish ( Ishii et al . , 2016; Arjona et al . , 2013 ) . The uex locus produces four alternatively spliced mRNA transcripts , but all encode the same 834 aa protein . The precise role of CNNM proteins in Mg2+ transport is somewhat contentious ( Funato et al . , 2018a; Arjona and de Baaij , 2018; Funato et al . , 2018b; Giménez-Mascarell et al . , 2019 ) . Some propose that CNNM proteins are direct Mg2+ transporters , whereas others favor that they function as sensors of intracellular Mg2+ concentration [Mg2+]i and/or regulators of other Mg2+ transporters . We found that ectopic expression of Drosophila UEX enhances Mg2+ efflux in HEK293 cells and that endogenous UEX limits [Mg2+]i in αβ KCs in the fly brain . Therefore , if UEX is not itself a Mg2+ transporter , it must be able to interact effectively with human Mg2+ efflux transporters and to influence Mg2+ extrusion in Drosophila . Since UEX is the only CNNM protein in the fly , it may serve all the roles of the four individual mammalian CNNMs . However , the ability of mouse CNNM2 to restore memory capacity to uex mutant flies suggests that the memory-relevant UEX function can be substituted by that of CNNM2 . Interestingly , none of the disease-relevant variants of CNNM2 were able to complement the memory defect of uex mutant flies . The CNNM2 T568I variant substitutes a single amino acid in the second CBS domain ( Arjona et al . , 2014 ) . The oncogenic protein tyrosine phosphatases of the PRL ( phosphatase of regenerating liver ) family bind to the CBS domains of CNNM2 and CNNM3 and can inhibit their Mg2+ transport function ( Hardy et al . , 2015; Giménez-Mascarell et al . , 2017; Zhang et al . , 2017 ) . It will therefore be of interest to test the role of the UEX CBS domains and whether fly PRL-1 regulates UEX activity . RNA-seq analysis reveals that uex is strongly expressed in the larval and adult fly digestive tract and nervous systems , as well as the ovaries ( Gelbart and Emmert , 2010; Croset et al . , 2018; Davie et al . , 2018 ) suggesting that many uex mutations will be pleiotropic . Our uexΔ allele , which deletes 272 amino acids ( including part of the second CBS and the entire CNBH domain ) from the UEX C-terminus , results in developmental lethality when homozygous , demonstrating that uex is an essential gene . Mammalian CNNM4 is localized to the basolateral membrane of intestinal epithelial cells ( Yamazaki et al . , 2013 ) . There it is believed to function in transcellular Mg2+ transport by exchanging intracellular Mg2+ for extracellular Na+ following apical entry through TRPM7 channels . Lethality in Drosophila could therefore arise from an inability to absorb sufficient Mg2+ through the larval gut . However , neuronally restricted expression of uexRNAiwith elav-GAL4 also results in larval lethality ( data not shown ) , suggesting UEX has an additional role in early development of the nervous system , like CNNM2 in humans and zebrafish ( Arjona et al . , 2014; Accogli et al . , 2019 ) . Perhaps surprisingly , flies carrying homozygous or trans-heterozygous combinations of several hypomorphic uex alleles have defective appetitive and aversive memory performance , yet they seem otherwise unaffected . Genetically engineering the uex locus to add a C-terminal HA tag to the UEX protein allowed us to localize its expression in the brain . Labeling is particularly prominent in all major classes of KCs . Restricting knockdown of uex expression to all αβ KCs of adult flies , or even just the αβc subset reproduced the LTM defect . The LTM impairment was evident if uexRNAi expression in αβ neurons was restricted to adult flies , suggesting UEX has a more sustained role in neuronal physiology . In contrast , knocking down uex expression in either the αβs or α′β′ neurons did not impair LTM . Activity of α′β′ neurons is required after training to consolidate appetitive LTM ( Krashes and Waddell , 2008 ) , whereas αβc and αβs KC output , together and separately , is required for its expression ( Krashes and Waddell , 2008; Perisse et al . , 2013 ) . Therefore , observing normal LTM performance in flies with uex loss-of-function in αβs and α′β′ neurons argues against a general deficiency of αβ neuronal function when manipulating uex . Dietary Mg2+ could not enhance the defective LTM performance of flies that were constitutively uex mutant , or harbored αβ KC-restricted uex loss-of-function . However , expressing uex in the αβ KCs of uex mutant flies restored the ability of Mg2+ to enhance performance . Therefore , the αβ KCs are the cellular locus for Mg2+-enhanced memory in the fly . It perhaps seems counterintuitive that UEX-directed magnesium efflux is required in KCs to support the memory-enhancing effects of Mg2+ feeding , when dietary Mg2+ elevates KC [Mg2+]i . At this stage , we can only speculate as to why this is the case . We assume that the brain and αβ KCs , in particular , have to adapt in a balanced way to the higher levels of intracellular and extracellular Mg2+ that result from dietary supplementation . Our live-imaging of KC [Mg2+]i in wild-type and uex mutant brains suggests that UEX-directed efflux is likely to be an essential factor in the active , and perhaps stimulus-evoked , homeostatic maintenance of these elevated levels . A number of mammalian cell-types extrude Mg2+ in a cAMP-dependent manner , a few minutes after being exposed to β-adrenergic stimulation ( Romani and Scarpa , 2000; Vormann and Günther , 1987; Jakob et al . , 1989; Romani and Scarpa , 1990b; Romani and Scarpa , 1990a; Vormann and Günther , 1987; Günther et al . , 1990; Howarth et al . , 1994 ) . The presence of a CNBH domain suggests that UEX and CNNMs could be directly regulated by cAMP . We tested the importance of the CNBH by introducing an R622K amino acid substitution that should block cAMP binding in the UEX CNBH . This subtle mutation abolished the ability of the uexR622K transgene to restore LTM performance to uex mutant flies . We also used CRISPR to mutate the CNBH in the native uex locus . Although deleting the CNBH from CNNM4 abolished Mg2+ efflux activity ( Chen et al . , 2018 ) , flies homozygous for the uexT626NRR lesion were viable , demonstrating that they retain a sufficient level of UEX function . However , these flies exhibited impaired immediate and long-term memory . In addition , the performance of uexT626NRR flies could not be enhanced by Mg2+ feeding . These data demonstrate that an intact CNBH is a critical element of memory-relevant UEX function . Binding of clathrin adaptor proteins to the CNNM4 CNBH has been implicated in basolateral targeting ( Hirata et al . , 2014 ) , suggesting that UEXT626NRR might be inappropriately localized in KCs . Furthermore , KC expression of the CNNM2 E122K mutant variant , which retains residual function but has a trafficking defect ( Arjona et al . , 2014 ) , did not restore the uex LTM defect . Although it has been questioned whether the CNNM2/3 CNBH domains bind cyclic nucleotides ( Chen et al . , 2018 ) , we found that FSK evoked an increase in αβ KC [Mg2+]i that was sensitive to uex mutation , and that UEX::HA was mislocalized in rut2080 adenylate cyclase ( Han et al . , 1992 ) and dnc1 phosphodiesterase ( Dudai et al . , 1976 ) learning defective mutant flies . Whereas UEX::HA label was evenly distributed in γ , αβc , and αβs KCs in wild-type flies , UEX::HA label was diminished in the γ and αβs KCs and was stronger in αβc neurons in rut2080 and dnc1 mutants . The chronic manipulations of cAMP in the mutants are therefore consistent with cAMP impacting UEX localization , perhaps by interacting with the CNBH . In addition , altered UEX localization may contribute to the memory defects of rut2080 and dnc1 flies . Our physiological data using Magnesium Green in mammalian cell culture and the genetically encoded MagIC reporter in αβ KCs demonstrate that fly UEX facilitates Mg2+efflux . Stimulating the fly brain with FSK evoked a greater increase in αβ KC [Mg2+]i in uex mutant brains than in wild-type controls which provides the first evidence that UEX limits a rise in [Mg2+]i in Drosophila KCs . Our MagIC recordings also revealed a slow oscillation ( centered around 0 . 015 Hz , approximately once a minute ) of αβ KC [Mg2+]i that was dependent on UEX . We do not yet understand the physiological function of this [Mg2+]i fluctuation although it likely reflects a homeostatic systems-level property of the cells . Biochemical oscillatory activity plays a crucial role in many aspects of cellular physiology ( Novák and Tyson , 2008 ) . Most notably , circadian timed fluctuation of [Mg2+]i links dynamic cellular energy metabolism to clock-controlled translation through the Mg2+ sensitive mTOR ( mechanistic target of rapamycin ) pathway ( Feeney et al . , 2016 ) . It is therefore possible that slow Mg2+ oscillations could unite roles for cAMP , UEX , energy flux ( Plaçais et al . , 2017 ) , and mTOR-dependent translation underlying LTM-relevant synaptic plasticity ( Casadio et al . , 1999; Huber et al . , 2000; Beaumont et al . , 2001; Hou and Klann , 2004; Hoeffer et al . , 2008 ) . A full list of reagents can be viewed in the Key Resources Table . Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact , Scott Waddell ( scott . waddell@cncb . ox . ac . uk ) . A polyclonal UEX antibody was developed commercially by Eurogentec . Two peptides were synthesized as antigens: Peptide 1 H-CLPKLDDKFESKQSKP-OH ( 16aa ) and Peptide 2 H-CVDNRTKTRRNRYKKA-NH2 ( 16aa ) and injected into rabbits . Only Peptide 2 induced a robust immune response and was processed further . The final serum was purified against Peptide 2 and used for western blot analysis as a 1:2000 dilution . For each sample in western blot , proteins were extracted from 20 fly heads by homogenizing thoroughly in 120 µl of protein sample buffer containing a mixture of 30 μl 2-mercaptoethanol ( BioRad ) , 270 µl 4× Laemmli sample buffer ( BioRad ) , and 900 µl Nuclease Free Water ( Invitrogen ) . Samples were then boiled on a 100°C heat block for 3 min and centrifuged for 10 min before loading . A sample volume equivalent to four heads was loaded into each SDS-PAGE gel lane . Proteins were transferred to PVDF membrane and blocked in 5% skim milk for 1 hr at 25°C with 35 rpm agitation . Membrane was then incubated in anti-UEX solution ( 1:2000 rabbit anti-UEX in 5% skim milk ) overnight at 4°C with 35 rpm agitation . Membrane was washed quickly three times followed by 3 × 10 min washes in TBST solution ( 100 ml of TBS 10× solution , BioRad , diluted in 900 ml of MilliQ water , with 0 . 1% Tween 20 ) and then incubated with HRP-conjugated secondary antibody solution ( 1:5000 of goat anti-rabbit in 5% skim milk ) for 1–2 hr at 25°C with 35 rpm agitation . The membrane was again washed quickly for three times followed by 3 × 10 min washes in TBST . Protein bands were visualized using Pierce ECL western blotting substrate ( Life technologies , 32134 ) . Membrane was then stripped using Millipore ReBlot Plus Mild solution ( Merck , 2502 ) , blocked again in 5% skim milk , and probed with mouse anti-Tubulin primary antibody ( 1:2000 , Sigma , T6199 ) and corresponding HRP conjugated goat anti-mouse secondary antibody ( 1:5000 ) following the protocol detailed above . Immunostaining was performed as described ( Wu and Luo , 2006 ) . Brains from 1- to 5-day-old adult flies were dissected in PBS and fixed for 20 min in PBS with 4% paraformaldehyde at room temperature . They were then washed twice briefly in 0 . 5% PBT ( 2 . 5 ml Triton-X100 in 497 . 5 ml PBS ) and three 20 min washes . Brains were then blocked for 30 min at room temperature in PBT containing 5% normal goat serum and then incubated with primary and secondary antibodies with mild rotation ( 35 rpm ) at 4°C for 1 or 2 days . Primary antibodies were rabbit anti-GFP ( 1:250; Invitrogen A11122 ) and rabbit anti-HA ( 1:250 , NEB 3724T ) . Alexa 488–conjugated goat anti-rabbit ( 1:250; Invitrogen , A11034 ) was the secondary antibody . Before and after the secondary antibody incubation , brains were subjected to two quick washes followed by three 20 min washes in 0 . 5% PBT . Stained brains were mounted on glass slides in Vectashield ( Vector Labs H1000 ) and imaged using a Leica TCS SP5 confocal microscope at 40× magnification ( HCX PL APO 40× , 1 . 3 CS oil immersion objective , Leica ) . Image stacks were collected at 1024 × 1024 resolution with 1 μm steps and processed using Fiji ( Schindelin et al . , 2012 ) . For quantification in Figure 3G and H , rectangular ROIs of approximately 40 × 25 μm for the for γ lobe , or round ROIs with diameter of 15 μm for αβ , α'β' , and EB were manually drawn on a single section of a z-stack scan of the fly brain . Corresponding ROIs were also drawn on the superior medial protocerebrum ( SMP ) as a background control region , and the mean fluorescence was calculated using ImageJ . ROI intensity of the MB lobes and the EB was normalized to that of the respective SMP intensity . An average between left and right brains was used for a single data point . For quantification in Figure 7C and D , ROIs are indicated in the figures and ROI intensity was calculated similar to results in Figure 3H . In Figure 7C , a line was drawn through the widest part of the tip of the α lobe . The intensity profile of this line was obtained through ImageJ . Thirty data points in the middle of such a profile spanning about a 15 μm line were extracted for each line profile . The profile was further normalized to the mean value of the first five data points ( F0 ) and calculated as ( F−F0 ) /F0 . Mean values of these normalized profiles from different brains were plotted ( Figure 7C , middle panel ) . Left and right profiles of brains were calculated and are separately displayed . In Figure 7D , the relative intensities from different ROIs representing different regions are added together to generate a total intensity measure for the MB . The human CNNM4 cDNA expression construct used to investigate Mg2+ efflux in cell culture is that described previously ( Yamazaki et al . , 2013 ) . A construct expressing Drosophila uex was generated by inserting a FLAG tag in front of the STOP codon of the uex CDS . FLAG-tagged CNNM4 and uex cDNAs were subsequently inserted into pCMV tag-4A ( Agilent ) for expression in HEK293 cells . HEK293 cells were cultured in Dulbecco’s modified Eagle medium ( Nissui ) supplemented with 10% Fetal Bovine Serum ( FBS ) and antibiotics . Expression plasmids were transfected with Lipofectamine 2000 ( Invitrogen ) . For immunostaining , cells were fixed with 3 . 7% formaldehyde in PBS for 20 min and then permeabilized with 0 . 2% Triton X-100 in PBS for 5 min , both at room temperature . They were next blocked with PBS containing 3% FBS and 10% bovine serum albumin ( blocking buffer ) for 1 hr at room temperature . Cells were then incubated overnight at 4°C with rabbit anti-FLAG antibody ( F7425 , Sigma-Aldrich ) diluted in blocking buffer , washed 3× with PBS , and incubated for 1 hr at room temperature with Alexa 488-conjugated anti-rabbit IgG ( Invitrogen ) and rhodamine-phalloidin ( for F-actin visualization , Invitrogen ) diluted in blocking buffer . After three washes with PBS , coverslips were mounted on slides and imaged with a confocal microscope ( FluoView FV1000; Olympus ) . Mg2+-imaging with Magnesium Green was performed as described ( Yamazaki et al . , 2013 ) , with slight modifications . To avoid potentially decreasing [Mg2+]i with the expressed proteins , transfected HEK293 cells were cultured in growth media supplemented with 40 mM MgCl2 until imaging . Cells were then incubated with Mg2+-loading buffer ( 78 . 1 mM NaCl , 5 . 4 mM KCl , 1 . 8 mM CaCl2 , 40 mM MgCl2 , 5 . 5 mM glucose , and 5 . 5 mM HEPES-KOH [pH 7 . 4] ) , including 2 μM Magnesium Green-AM ( Invitrogen ) , for 30 min at 37°C . Cells were then rinsed once with loading buffer and viewed with an Olympus IX81 microscope equipped with an ORCA-Flash 4 . 0 CMOS camera ( Hamamatsu ) and a SHI-1300L mercury lamp ( Olympus ) . Fluorescence was measured every 20 s ( excitation at 470–490 nm and emission at 505–545 nm ) under the control of Metamorph software ( Molecular Devices ) . Buffer was then changed to Mg2+free buffer ( MgCl2 in the loading buffer was replaced with 60 mM NaCl ) . Data are presented as line plots ( mean of 10 cells ) . After imaging , cells were fixed with PBS containing 3 . 7% formaldehyde and subjected to immunofluorescence microscopy to confirm protein expression . One- to two-day-old flies with genotype c739; UAS-MagFRET-1 were housed in vials with 1 mM or 80 mM [Mg2+] food for 4 days before being collected . Fly brains were dissected in PBS and fixed for 20 min in PBS with 4% paraformaldehyde at room temperature . They were then washed twice briefly in 0 . 5% PBT ( 2 . 5 ml Triton-X100 in 497 . 5 ml PBS ) and three 10 min washes . Brains were then mounted on glass slides in Vectashield ( Vector Labs H1000 ) and imaged using a wide-field Scientifica Slicescope with a 40× , 0 . 8 NA water-immersion objective and an Andor Zyla sCMOS camera with Andor Solis software ( v4 . 27 ) . In order to get the FRET ratio that indicates the Mg2+ concentration of the αβ neuron , time series were acquired alternatively between the cerulean channel and the citrine channel at 3 Hz with 512 × 512 pixels and 16 bit . The excitation wavelength for both channels is 436 nm , while the emission filter for cerulean is 460–500 nm and that for citrine is 520–550 nm . Series acquisition starts from the cerulean channel and lasts for 5 s , then switches to the citrine channel and last for another 5 s , and this cycle is repeated for two more times . A total of 30 s ( 90 frames ) image stack was therefore acquired for each brain . Image stacks were subsequently analyzed using ImageJ and custom-written Matlab scripts . In brief , rectangle ROIs ( Figure 1E , left panel ) were manually drawn on the αβ lobes ( one on α lobe and one on β lobe for each hemisphere ) , and outside the αβ lobes ( one for each hemisphere ) as background control . Fluorescence intensity from the cerulean channel was calculated by dividing each vertical or horizontal lobe ROI by the background ROI , and averaged between the two hemispheres for each lobe , and averaged over the 15 frames for each cycle . That from the citrine channel was obtained similarly . A FRET ratio was obtained from the above intensities , further averaged among the three cycles of acquisition , depicted as one data point in Figure 1E ( right panel ) . Explant brains expressing c739-GAL4 driven UAS-MagIC were placed at the bottom of a 35 mm glass bottom microwell dish ( Part No . P35G-1 . 5–14 C , MatTek Corporation ) , beneath extracellular saline buffer solution ( 103 mM NaCl , 3 mM KCl , 5 mM N-Tris , 10 mM trehalose , 10 mM glucose , 7 mM sucrose , 26 mM NaHCO3 , 1 mM NaH2PO4 , 1 . 5 mM CaCl2 , 4 mM MgCl2 , osmolarity 275 mOsm [pH 7 . 3] ) following dissection in calcium-free buffer ( Barnstedt et al . , 2016 ) . To determine the Mg2+ sensitivity of UAS-MagIC as well as the response of UAS-MagIC to other chemicals such as EDTA , EGTA , and CaCl2 ( Figure 8B ) , brains were incubated in the saline buffer solution with 20 μg/ml digitonin for 6 min before imaging ( Koldenkova et al . , 2015 ) . To investigate the Mg2+ fluctuation in response to Forskolin ( FSK ) application ( Figure 8C–I ) , brains were put in the saline buffer solution without digitonin or incubation . In both situations , saline refers to the buffer ( either with or without digitonin ) in which the brain is submerged . Imaging was carried out in a LSM780 confocal microscope ( Zeiss ) with a 20× air objective using the ZEN 2011 software . The Venus part of MagIC was excited with a 488 nm laser and its emission was collected in the 520–560 nm range . mCherry was excited with a 561 nm laser and its emission was collected in the 600–640 nm range . Time series were acquired at 0 . 5 Hz with 512 × 512 pixels and 16 bit . Following 60 s of baseline Venus/mCherry measurement , 2–20 µl of saline or other relevant chemical solution was added via a micropipette to the dish with constant image capture . The effects of applied agents on Venus/mCherry emission were then recorded for 15–20 min . Image stacks were subsequently analyzed using ImageJ and custom-written Python scripts . In brief , rectangle ROIs were manually drawn on the αβ neurons ( one for each hemisphere , Figure 8A ) , and another ROI of the same size was drawn in the middle but outside the MBs as background control . Fluorescence intensity from the Venus ( or mCherry ) channel was calculated by subtracting the background ROI from the calyx ROIs , respectively , and averaged between the two hemispheres . This is referred as ‘Rel . Intensity ( a . u . ) ' in Figure 8D and E . The ratio between Venus and mCherry intensity was calculated as ‘MagIC Ratio’ in Figure 8B and C and Figure 8F and G . For Figure 8H , the intensity for the two channels was calculated separately . In this case , ‘Rel . Intensity ( ΔF/F0 ) ’ refers to the relative fluorescence intensity normalized to the mean intensity from the baseline period F0 , calculated as ( F−F0 ) /F0 . The relative intensity ΔF/F0 of Venus was used to calculate the PSD ( Figure 8I ) through python function psd ( under matplotlib . pyplot ) , which adopted a Welch’s average periodogram method ( Bendat et al . , 2000 ) . For each sample , 120 flies were frozen in liquid nitrogen and their heads were homogenized completely in TRIzol reagent ( Invitrogen ) . Total RNA was extracted using Direct-zol RNA MiniPrep ( R2050 ) kit following the manufacturer’s instructions . cDNA was synthesized using SuperScript III First-Strand synthesis System ( Invitrogen ) . Five independent samples were prepared for each different treatment or genotype . Quantitative PCR was performed in triplicate for each cDNA sample on a LightCycler 480 Instrument ( Roche ) using SYBR Green I Master Mix ( Roche ) . Melting curves were analyzed after amplification , and amplicons were visualized by agarose gel electrophoresis to confirm primer specificity . Relative transcript levels were calculated by the 2-ΔΔCt method ( Livak and Schmittgen , 2001 ) , and the geometric mean of the Ct values of three reference genes ( Gapdh , Tbp , and Ef1α100E ) was used for normalization . Primers are detailed in the Resource Table . Inverse PCR was used to map the MiMIC insertion position in uexMI01943 flies . Genomic DNA was prepared from 15 adult flies . DNA equivalent to two flies was then digested in a 25 μl restriction reaction with Mbo I and 10 μl of the product was ligated overnight at 4°C overnight to circularize the fragments; 5 μl of the ligation product was used for inverse PCR . PCR product was purified using Exo/SAP reaction ( Thermo Fisher , 78201 ) before being sequenced . Sequence was compared to the D . melanogaster genome ( FlyBase , Release 6 ) by BLAST and matched uniformly to the region 3 , 882 , 886 . 3 , 882 , 641 on 2R , consistent with the reported uexMI01943 insertion on FlyBase . Primers detailed in the Resource Table . Protein sequence alignment was carried out using Geneious R10 . 2 . 2 . Protein domain prediction was performed with InterPro ( Finn et al . , 2017; Jones et al . , 2014 ) and Phyre2 ( Kelley et al . , 2015 ) . Protein domain and structure alignment was performed using TM-align ( Zhang and Skolnick , 2005 ) . Protein structure visualization was rendered in Chimera 1 . 11 . 2 ( Pettersen et al . , 2004 ) . Behavior data were analyzed using Excel and Prism 6 . Imaging data were analyzed using ImageJ and custom-written MATLAB or Python scripts . Unpaired two-tailed t-tests were used for comparing two groups , and one-way ANOVA followed by a Tukey’s post-hoc test was used for comparing multiple groups . Threshold of statistical significance was set at p<0 . 05 .	The proverbial saying ‘you are what you eat’ perfectly summarizes the concept that our diet can influence both our mental and physical health . We know that foods that are good for the heart , such as nuts , oily fish and berries , are also good for the brain . We know too that vitamins and minerals are essential for overall good health . But is there any evidence that increasing your intake of specific vitamins or minerals could help boost your brain power ? While it might sound almost too good to be true , there is some evidence that this is the case for at least one mineral , magnesium . Studies in rodents have shown that adding magnesium supplements to food improves how well the animals perform on memory tasks . Both young and old animals benefit from additional magnesium . Even elderly rodents with a condition similar to Alzheimer’s disease show less memory loss when given magnesium supplements . But what about other species ? Wu et al . now show that magnesium supplements also boost memory performance in fruit flies . One group of flies was fed with standard cornmeal for several days , while the other group received cornmeal supplemented with magnesium . Both groups were then trained to associate an odor with a food reward . Flies that had received the extra magnesium showed better memory for the odor when tested 24 hours after training . Wu et al . show that magnesium improves memory in the flies via a different mechanism to that reported previously for rodents . In rodents , magnesium increased levels of a receptor protein for a brain chemical called glutamate . In fruit flies , by contrast , the memory boost depended on a protein that transports magnesium out of neurons . Mutant flies that lacked this transporter showed memory impairments . Unlike normal flies , those without the transporter showed no memory improvement after eating magnesium-enriched food . The results suggest that the transporter may help adjust magnesium levels inside brain cells in response to neural activity . Humans produce four variants of this magnesium transporter , each encoded by a different gene . One of these transporters has already been implicated in brain development . The findings of Wu et al . suggest that the transporters may also act in the adult brain to influence cognition . Further studies are needed to test whether targeting the magnesium transporter could ultimately hold promise for treating memory impairments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Magnesium efflux from Drosophila Kenyon cells is critical for normal and diet-enhanced long-term memory
Altered insulin signaling has been linked to widespread nervous system dysfunction including cognitive dysfunction , neuropathy and susceptibility to neurodegenerative disease . However , knowledge of the cellular mechanisms underlying the effects of insulin on neuronal function is incomplete . Here , we show that cell autonomous insulin signaling within the Drosophila CM9 motor neuron regulates the release of neurotransmitter via alteration of the synaptic vesicle fusion machinery . This effect of insulin utilizes the FOXO-dependent regulation of the thor gene , which encodes the Drosophila homologue of the eif-4e binding protein ( 4eBP ) . A critical target of this regulatory mechanism is Complexin , a synaptic protein known to regulate synaptic vesicle exocytosis . We find that the amounts of Complexin protein observed at the synapse is regulated by insulin and genetic manipulations of Complexin levels support the model that increased synaptic Complexin reduces neurotransmission in response to insulin signaling . Metabolic disorders such as diabetes are associated with widespread declines in neuronal function including peripheral and proximal neuropathy , retinopathy , reduced cognition , impaired motor functions and increased risk of developing neurodegenerative disease including Alzheimer’s disease ( Deak and Sonntag , 2012; Gispen and Biessels , 2000; Luchsinger , 2012; Park , 2001; Plum et al . , 2005 ) . The loss of normal synapse function is believed to be an important contributor to all these disorders suggesting that changes in insulin signaling can influence synaptic connectivity throughout the nervous system . For example , analysis of human patients with type II diabetes ( T2DM ) reveals changes in brain structures , including synapse numbers , which correlate with decreased cognitive performance ( Qiu et al . , 2014 ) . In addition , numerous rodent studies have demonstrated that changes in peripheral and cerebral insulin result in changes to synapse function and plasticity in both the hippocampus and retinae ( Gispen and Biessels , 2000; Hombrebueno et al . , 2014 ) . Rodent and human studies have also demonstrated that changes in normal insulin signaling can alter peripheral synapses including neuromuscular junctions ( NMJs ) ( Allen et al . , 2015a , 2015b; Fahim et al . , 1998; Francis et al . , 2011; Garcia et al . , 2012; Ramji et al . , 2007 ) . Despite the wide-spread effects of altered insulin signaling on synapse function , the cellular mechanisms underlying the effects insulin signaling on synapse function , especially the control of neurotransmitter release , are poorly understood . There exist well-established evolutionarily conserved targets of insulin signaling that have been implicated in the effects of insulin on synapse function ( Kleinridders et al . , 2014; Park , 2001; Plum et al . , 2005 ) . This includes the mammalian target of rapamycin ( mTOR ) complex that is positively regulated by insulin signaling . In the postsynaptic compartment , TOR signaling has been directly implicated in the regulation of post-synaptic function including the formation of new synapses and the generation of retrograde signaling during homeostatic synaptic plasticity ( Penney et al . , 2012; Stoica et al . , 2011; Takei and Nawa , 2014; Weston et al . , 2012 ) . The role of TOR signaling within the presynaptic nerve terminal is less clear . Another important target of insulin signaling is the FOXO family of transcription factors . Insulin negatively regulates FOXO via phosphorylation by Akt in both flies and rodents ( Puig et al . , 2003; Teleman et al . , 2005; Yamamoto and Tatar , 2011 ) . Previous studies have established that FOXO is required in Drosophila larval motor neurons for synapse growth , synaptic vesicle recycling , and for the control of neuronal excitability downstream of PI3K signaling ( Howlett et al . , 2008; Nechipurenko and Broihier , 2012 ) . In mammals , recent studies have revealed a requirement for FOXO6 , a FOXO family member highly expressed in the hippocampus , during learning and memory ( Salih et al . , 2012 ) . It was shown in these studies that FOXO6 was required for the expression of genes involved in neurotransmission supporting a direct role for FOXO in the regulation of synapse function ( Salih et al . , 2012 ) . It is unclear whether insulin signaling regulates FOXO activity in neurons in any system . In the present study , we present evidence that in adult Drosophila motor neurons , insulin signaling negatively regulates the presynaptic release of neurotransmitter via the FOXO-dependent regulation of the translational inhibitor the eukaryotic initiation factor 4e binding protein ( 4eBP ) . The translational target of this signaling system appears to be the Complexin protein , which is known to regulate the exocytosis of synaptic vesicles providing direct link between neuronal insulin signaling and neurotransmitter release . Electrical recordings from the CM9 muscle group on the adult fly proboscis previously revealed a decrease in the amount of neurotransmitter released from the CM9 NMJ in flies raised on a high-calorie diet compared to flies raised on a low-calorie diet ( Rawson et al . , 2012 ) . The CM9 NMJ is ideal for this research since it combines motor neuron-specific genetic manipulations with a robust synaptic recording preparation allowing us to interrogate the cell autonomous effects of diet ( Figure 1A ) . Using this system , we have extended our initial observations by first determining that changing only the yeast component of the diet is sufficient to alter neurotransmission . Animals raised for 21 days on food containing 100 mg/ml of yeast ( 1X ) release nearly twice as much neurotransmitter , represented as quantal content ( the number of quanta per action potential ) ( Fatt and Katz , 1951 ) , compared to flies raised on food containing 200 mg/ml of yeast ( 2X ) ( Figure 1B–F; see Table 1 for all electrophysiological recording data ) . We further find that shifting 20-day-old flies from the 1X diet to the 2X diet ( Figure 1B ) resulted in a gradual reduction in the quantal content ( QC ) that reaches the level of neurotransmission observed in 2X animals within 24 hr of diet shift ( Figure 1C and F ) supporting that the effects of high-protein diet on release are likely not due to the accumulation of diet-related pathologies that result in reduced neurotransmitter release . In these recordings , we observed no effect of diet on the amplitude of the spontaneous release events ( mEPSPs ) , the resting membrane potential , or the resistance of the muscle demonstrating that the effects of diet on presynaptic function are not due to changes in the excitability of the post-synaptic CM9 muscle fibers ( Figure 1E; Table 1 ) ( Davis , 2013 ) . Paired pulse analysis at the CM9 NMJ reveals that flies raised on 1X diet show pronounced synaptic depression when EPSPs were evoked with a 50-ms interpulse interval that was absent at CM9 NMJs in flies raised on the 2X diet ( Figure 1G and H ) . Under our recording conditions , this result is consistent with a reduction in the probability of release at CM9 NMJs in animals subjected to the 2X diet ( Zucker and Regehr , 2002 ) . Hypertonic challenge of NMJs with sucrose has been used to estimate the size of the readily-releasable pool at Drosophila NMJs ( Mahoney et al . , 2014; Müller and Davis , 2012; Yoshihara et al . , 2010 ) . We find that there is no significant difference in the size of the sucrose-sensitive pool of synaptic vesicles ( SVs ) at the CM9 NMJs in flies raised on a 1X diet compared to flies raised on a 2X diet ( Figure 1I–K ) ( Rosenmund and Stevens , 1996 ) . Combined with our previous observations of a lack of effect of diet on synaptic area ( Rawson et al . , 2012 ) , these data support that increases in the protein content of the diet reduces the probability of SV release at the CM9 NMJ . 10 . 7554/eLife . 16807 . 003Figure 1 . Effects of dietary protein concentrations on neurotransmission at the CM9 NMJ . ( A ) Diagram of Drosophila head indicating the location of the Cibarial Muscle 9 ( CM9 ) . ( B ) 21-day feeding paradigm used for the analysis of dietary effects on neurotransmission . Animals were raised for 21 days on a low-protein diet ( 1X = light gray ) , a high-protein diet ( 2X = dark gray ) , or subjected to a shift from a low-protein diet ( 1X ) to a high-protein diet ( 2X ) on day 20 ( 1X-2X ) . ( C ) Representative traces of evoked CM9 EPSPs and spontaneous miniature EPSPs ( CM9 mEPSPs ) from electrophysiological recordings of CM9 muscle fibers from flies subjected to the indicated dietary conditions . Scale = 1 mv , 10 ms . ( D–F ) Graphs represents the mean values for evoked EPSPs ( D ) , mEPSPs ( E ) , and quantal content ( F ) determined from recordings of CM9 muscle fibers from flies subjected to the indicated dietary condition or subjected to a diet shift ( 1X-2X ) for 12 or 24 hr . Error bars = s . e . m . p<0 . 05 determined using ANOVA . ( G ) Example traces of evoked EPSPs from paired-pulse experiments utilizing an inter-pulse interval ( IPI ) of 50 ms . ( H ) Graph represent the mean percent depression at indicated IPI . Error bars = s . e . m . p<0 . 01 , Student’s t-test . ( I ) Representative traces of electrophysiological recordings of hyperosmotic-induced spontaneous release events from CM9 NMJs in animals raised for 21 days on indicated diet conditions incubated in hyperosmotic recording saline . Inserts represent broader timescale of boxed regions from traces . Scale = 1 s . ( J ) Histogram representing the spontaneous event frequency and K , the average number of total spontaneous release events observed during 1 min of hyperosmotic recordings . DOI: http://dx . doi . org/10 . 7554/eLife . 16807 . 00310 . 7554/eLife . 16807 . 004Figure 1—source data 1 . File contains the values represent the average value for the spontaneous release events per second determined in 5 s increments during the hypertonic stimulation of synaptic vesicle fusion at CM9 NMJs in animals raised on a 1X or 2X diet presented in Figure 1J . Values for each animal ( n = 7 ) are shown with time bins indicating time relative to the application of hypertonic recording solution . The results of Student’s t-tests for each 5-s time bin and the Kolmogorov-Smirnov test of the distributions are presented with data set . DOI: http://dx . doi . org/10 . 7554/eLife . 16807 . 00410 . 7554/eLife . 16807 . 005Table 1 . Quantal analysis of neurotransmission at the CM9 NMJ . DOI: http://dx . doi . org/10 . 7554/eLife . 16807 . 005Genotype ( condition ) DietNmEPSP ( mV ) EPSP ( mV ) QCRMP ( mV ) IR ( MΩ ) w11181X80 . 94 ± 0 . 043 . 46 ± 0 . 303 . 66 ± 0 . 28−40 . 89 ± 1 . 377 . 56 ± 0 . 80w11182X80 . 83 ± 0 . 041 . 65 ± 0 . 082 . 01 ± 0 . 12−39 . 67 ± 0 . 577 . 00 ± 0 . 80w1118 ( 12 hr shift ) 1-2X80 . 88 ± 0 . 022 . 74 ± 0 . 213 . 12 ± 0 . 24−38 . 40 ± 2 . 178 . 13 ± 1 . 01w1118 ( 24-hr shift ) 1-2X80 . 89 ± 0 . 032 . 25 ± 0 . 272 . 58 ± 0 . 27−35 . 65 ± 1 . 538 . 75 ± 0 . 62E49-Gal4/+1X80 . 96 ± 0 . 033 . 40 ± 0 . 163 . 55 ± 0 . 16−35 . 53 ± 3 . 247 . 48 ± 0 . 55E49-Gal4/+2X80 . 92 ± 0 . 022 . 06 ± 0 . 092 . 23 ± 0 . 07−32 . 24 ± 0 . 837 . 12 ± 0 . 58UAS-4eBPRNAi/+1X80 . 94 ± 0 . 043 . 31 ± 0 . 333 . 50 ± 0 . 30−41 . 02 ± 1 . 407 . 88 ± 0 . 79UAS-4eBPRNAi/+2X80 . 83 ± 0 . 021 . 62 ± 0 . 121 . 98 ± 0 . 17−39 . 67 ± 0 . 578 . 25 ± 0 . 82UAS-4eBPRNAi/+1-2X80 . 93 ± 0 . 011 . 90 ± 0 . 052 . 05 ± 0 . 07−37 . 96 ± 0 . 548 . 25 ± 0 . 62E49-Gal4/+; UAS-4eBPRNAi/+1X70 . 92 ± 0 . 021 . 50 ± 0 . 061 . 64 ± 0 . 07−38 . 88 ± 0 . 689 . 14 ± 0 . 77E49-Gal4/+; UAS-4eBPRNAi/+2X80 . 91 ± 0 . 031 . 67 ± 0 . 131 . 86 ± 0 . 19−38 . 13 ± 0 . 508 . 00 ± 1 . 00E49-Gal4/+; UAS-4eBPRNAi/+1-2X80 . 94 ± 0 . 021 . 79 ± 0 . 101 . 90 ± 0 . 11−38 . 68 ± 0 . 678 . 75 ± 0 . 62UAS-chicoRNAi/+1X80 . 90 ± 0 . 043 . 56 ± 0 . 333 . 96 ± 0 . 30−37 . 31 ± 1 . 498 . 63 ± 0 . 30UAS-chicoRNAi/+2X80 . 83 ± 0 . 021 . 74 ± 0 . 102 . 10 ± 0 . 15−39 . 45 ± 0 . 478 . 25 ± 0 . 73E49-Gal4/UAS-chicoRNAi1X80 . 92 ± 0 . 034 . 60 ± 0 . 305 . 09 ± 0 . 43−34 . 75 ± 1 . 078 . 31 ± 0 . 47E49-Gal4/UAS-chicoRNAi2X80 . 95 ± 0 . 054 . 09 ± 0 . 284 . 34 ± 0 . 24−38 . 52 ± 4 . 498 . 40 ± 0 . 77E49-Gal4/UAS-chicoRNAi1-2X80 . 91 ± 0 . 044 . 02 ± 0 . 234 . 48 ± 0 . 28−35 . 47 ± 2 . 828 . 75 ± 0 . 68E49-Gal4/UAS-chicoRNAi; UAS-4eBPRNAi/+2X80 . 84 ± 0 . 052 . 18 ± 0 . 122 . 60 ± 0 . 13−39 . 51 ± 1 . 967 . 75 ± 0 . 85UAS-InRDN/+2X80 . 86 ± 0 . 012 . 07 ± 0 . 172 . 42 ± 0 . 21−32 . 19 ± 1 . 558 . 06 ± 0 . 79E49-Gal4/UAS-InRDN2X80 . 85 ± 0 . 032 . 87 ± 0 . 153 . 43 ± 0 . 24−35 . 92 ± 2 . 208 . 69 ± 0 . 54w11181X80 . 92 ± 0 . 023 . 25 ± 0 . 253 . 53 ± 0 . 26−34 . 61 ± 1 . 777 . 88 ± 0 . 69w11182X80 . 84 ± 0 . 031 . 82 ± 0 . 102 . 18 ± 0 . 14−36 . 62 ± 1 . 148 . 50 ± 0 . 80w1118 ( +CXM ) 1-2X80 . 99 ± 0 . 044 . 31 ± 0 . 204 . 39 ± 0 . 26−40 . 58 ± 1 . 847 . 88 ± 0 . 69w1118 ( +Veh ( CMX ) ) 1-2X80 . 95 ± 0 . 022 . 50 ± 0 . 112 . 63 ± 0 . 11−40 . 01 ± 2 . 568 . 00 ± 0 . 68w1118 ( +CXM ) 1X81 . 04 ± 0 . 034 . 23 ± 0 . 234 . 09 ± 0 . 25−39 . 09 ± 0 . 899 . 00 ± 0 . 82w1118 ( +rapamycin ) 1-2X80 . 86 ± 0 . 032 . 11 ± 0 . 132 . 48 ± 0 . 22−41 . 29 ± 1 . 197 . 63 ± 0 . 78w1118 ( +Veh ( rapa ) ) 1-2X80 . 82 ± 0 . 032 . 05 ± 0 . 202 . 54 ± 0 . 29−39 . 87 ± 1 . 846 . 88 ± 0 . 61w11181X80 . 89 ± 0 . 033 . 37 ± 0 . 203 . 81 ± 0 . 27−31 . 25 ± 1 . 478 . 25 ± 0 . 75w11182X50 . 95 ± 0 . 031 . 91 ± 0 . 162 . 00 ± 0 . 13−34 . 05 ± 1 . 487 . 80 ± 0 . 97dFOXOdel94/dFOXO211X80 . 94 ± 0 . 021 . 98 ± 0 . 132 . 10 ± 0 . 13−34 . 60 ± 1 . 358 . 44 ± 0 . 48dFOXOdel94/dFOXO212X80 . 96 ± 0 . 031 . 69 ± 0 . 091 . 77 ± 0 . 13−38 . 21 ± 1 . 527 . 75 ± 0 . 75dFOXOdel94/dFOXO211-2X80 . 94 ± 0 . 011 . 58 ± 0 . 041 . 68 ± 0 . 03−34 . 54 ± 2 . 068 . 69 ± 0 . 74dFOXOdel94 / dFOXO21 , UAS-4eBP1X80 . 94 ± 0 . 041 . 95 ± 0 . 232 . 06 ± 0 . 19−31 . 36 ± 2 . 837 . 38 ± 0 . 74E49-Gal4/+; dFOXOdel94 / dFOXO21 , UAS-4eBP1X80 . 88 ± 0 . 043 . 28 ± 0 . 223 . 85 ± 0 . 43−31 . 83 ± 2 . 656 . 88 ± 0 . 75UAS-4eBP/+1X80 . 96 ± 0 . 043 . 30 ± 0 . 163 . 47 ± 0 . 16−33 . 08 ± 1 . 037 . 79 ± 0 . 38E49-Gal4/+;UAS-4eBP/+1X80 . 92 ± 0 . 034 . 79 ± 0 . 385 . 25 ± 0 . 46−34 . 77 ± 2 . 128 . 08 ± 0 . 58E49-Gal4/UAS-stauenRNAi1X80 . 89 ± 0 . 033 . 37 ± 0 . 203 . 81 ± 0 . 27−36 . 32 ± 1 . 228 . 32 ± 0 . 66E49-Gal4/UAS-staufenRNAi1-2X80 . 95 ± 0 . 043 . 45 ± 0 . 213 . 65 ± 0 . 16−39 . 64 ± 2 . 427 . 55 ± 0 . 32+/UAS-staufenRNAi1-2X90 . 90 ± 0 . 042 . 35 ± 0 . 132 . 68 ± 0 . 20−35 . 51 ± 1 . 218 . 02 ± 0 . 73W11181X80 . 93 ± 0 . 023 . 21 ± 0 . 193 . 47 ± 0 . 22−30 . 56 ± 1 . 498 . 31 ± 0 . 09W11182X80 . 95 ± 0 . 021 . 96 ± 0 . 152 . 07 ± 0 . 17−30 . 43 ± 1 . 208 . 06 ± 0 . 67+/+ , cpxSH1/+1X80 . 91 ± 0 . 014 . 23 ± 0 . 484 . 66 ± 0 . 51−32 . 32 ± 1 . 407 . 88 ± 0 . 74+/+ , cpxSH1/+2X80 . 96 ± 0 . 012 . 65 ± 0 . 302 . 75 ± 0 . 32−31 . 26 ± 2 . 877 . 94 ± 0 . 83UAS-Complexin/+1X90 . 99 ± 0 . 054 . 15 ± 0 . 464 . 37 ± 0 . 56−36 . 12 ± 1 . 656 . 67 ± 0 . 67UAS-Complexin/+2X90 . 95 ± 0 . 022 . 39 ± 0 . 192 . 54 ± 0 . 24−31 . 27 ± 1 . 997 . 22 ± 0 . 80E49-Gal4/UAS-Complexin1X90 . 90 ± 0 . 052 . 22 ± 0 . 282 . 48 ± 0 . 28−30 . 59 ± 1 . 976 . 79 ± 0 . 73E49-Gal4/UAS-Complexin2X90 . 98 ± 0 . 052 . 58 ± 0 . 212 . 63 ± 0 . 18−34 . 86 ± 2 . 427 . 72 ± 0 . 52Table contents ordered by order of appearance in body of text . All values represent the average value ± sem ( N = animals , 1 recording per animal ) . For each recording , the EPSP value represents the average of 60 evoked responses and the value for mEPSP represents the average of 30 events . All stocks were backcrossed five generations and re-established in the w1118 background . Quantal content ( QC ) is determined for each NMJ by dividing the amplitude of the EPSP by the amplitude of the mEPSP for each recording . RMP = resting membrane potential of CM9 muscle fiber . IR = depolarizing input resistance of CM9 muscle fiber . To determine what signaling systems within the motor neuron are responsible for the effects of diet on neurotransmission , we used motor-neuron-specific RNAi to screen important nutrient-sensing pathways using viability in a diet sensitive glued mutant fly background ( Rawson et al . , 2012 ) followed by analysis of promising candidates using the proboscis extension reflex ( PER ) ( Figure 2A ) ( Gordon and Scott , 2009; Kimura et al . , 1986 ) . This motor reflex requires the CM9 motor neuron and provides a simple assay for investigating CM9 motor neuron function by analyzing the velocity of proboscis extension using particle-tracking software to track bristle paths during the PER ( Figure 2A , panels i-iv ) ( Rawson et al . , 2012 ) . For these analyses , we combine the E49-Gal4 driver , which is expressed in a few number of neurons in the adult including the CM9 motor neuron , with gene-specific RNAi allowing us to focus on the cell autonomous effects of insulin signaling without grossly altering whole animal insulin signaling ( Gordon and Scott , 2009; Rawson et al . , 2012 ) . This approach identified thor , the Drosophila homologue of eukaryotic initiation factor 4e binding protein ( 4eBP ) , as a critical presynaptic mediator of the positive effects of the 1X diet on motor function ( Figure 2B and C ) . Analysis of neurotransmission in these animals found that knockdown of 4eBP in the CM9 motor neuron reduced the presynaptic release of neurotransmitter in animals raised on a 1X diet compared to controls ( Figure 2F–I; see Table 1 for E49-Gal4/+ control values ) . We also observe no difference between neurotransmitter release in 4eBP knockdown animals ( 4eBPRNAi ) raised on 1X , 2X , or 1-2X diet conditions consistent with the effects of 4eBP knockdown being specific to diet regulation of neurotransmission and not basal release ( Figure 2F–I ) . The effectiveness and specificity of all RNAi constructs were determined using quantitative RT-PCR and finds that the reductions in 4eBP mRNA levels are approximately 60% in control experiments ( data not shown ) . All values for the electrophysiological analyses are listed in Table 1 . 10 . 7554/eLife . 16807 . 006Figure 2 . Insulin/DILP signaling negatively regulates presynaptic release at the CM9 NMJ . ( A ) Images from a proboscis extension reflex ( PER ) in response to tarsal stimulation with 0 . 5 M sucrose . Circle indicates location of sensory bristles tracked during the extension event resulting in an extension path ( red line in panel iv ) that is used for analysis of velocity . ( B and C ) Graphs represent the mean values for ( B ) average velocity and ( C ) max velocity for indicated genotypes and dietary conditions . All RNAi knock-downs utilize the Gal4:UAS binary expression system by combining the transgenic UAS construct ( i . e . UAS-4eBPRNAi ) with the CM9 motor neuron-specific E49-Gal4 driver . p<0 . 05 versus 1X controls determined using ANOVA . Error bars = s . e . m . ( D ) Pathway represents the putative effects of insulin signaling on SV exocytosis . ( E , F ) CM9 EPSP traces of indicated genotype and dietary condition demonstrating the requirement for 4eBP on neurotransmission in animals raised on the 1X and 2X diets . Scale bar = 1 mV , 10 ms . ( G–I ) Graphs represent the average values for CM9 EPSPs ( G ) , mEPSPs ( H ) , and quantal content ( I ) determined from CM9 recordings from 21-day-old flies of indicated genotypes raised on indicated dietary conditions . Error bars = s . e . m . * indicates values significantly different from all other values determined using ANOVA ( p<0 . 01 ) . ( J ) CM9 EPSP traces of indicated genotype and dietary condition demonstrating the requirement for chico on neurotransmission in animals raised on the 1X and 2X diets . The effect of chico knock-down ( chicoRNAi ) on neurotransmission in animals raised on a 2X diet is suppressed by knockdown of 4eBP consistent with 4eBP functioning downstream of Chico . Scale bar = 1 mV , 10 ms . ( K and L ) Graphs represent the average values of EPSPs ( K ) and quantal content ( L ) determined from CM9 recordings from 21-day old flies of indicated genotypes raised on indicated dietary conditions . Error bars = s . e . m . p<0 . 05 versus 2X controls determined using ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 16807 . 00610 . 7554/eLife . 16807 . 007Figure 2—figure supplement 1 . Effects of diet and neuronal insulin signaling on SV exocytosis at larval NMJ . ( A ) Representative evoked EPSP and spontaneous mEPSP traces from larval muscle 6 in larval of indicated genotype and diet conditions . In these experiments , larvae were on indicated diets for the entire larval stage of development ( ~3 days ) . Genotypes: ctrl = OK6-Gal4/+; 4eBP RNAi = OK6-Gal4/+; UAS-4eBPRNAi/+; chico RNAi = OK6-Gal4/+; UAS-chicoRNAi/+ . Scale bar = 10 mV , 250 ms ( EPSPs ) ; 2 mV , 50 ms ( mEPSPs ) . ( B and C ) Graphs represent the mean values for EPSPs ( B ) and quantal content ( C ) recorded from the NMJs on muscle 6 in control and 4eBP knockdown larvae raised on 1X diet . Error bars = s . e . m . ( D and E ) Graphs represent the mean values for EPSPs ( D ) and quantal content ( E ) determined at the NMJs on muscle 6 in control and chico knockdown larvae raised on 2X diet . N = 8 animals ( 1 recording per animal ) for each average . Error bars = s . e . m . DOI: http://dx . doi . org/10 . 7554/eLife . 16807 . 00710 . 7554/eLife . 16807 . 008Figure 2—figure supplement 2 . Cycloheximide blocks the effects of diet switch on SV exocytosis . ( A ) Model of the translational regulation of SV exocytosis by insulin signaling via 4eBP . ( B ) 21-day diet shift paradigm for testing the effects of cycloheximide on neurotransmission . Flies were exposed to cycloheximide for 2 hr on 1X diet prior to switching to the 2X diet condition for 24 hr . Control flies consisted of cylcoheximide treated flies that are not subjected to diet switch or treated with vehicle . ( C–D ) Graphs represent the mean values for EPSPs ( C ) , mEPSPs ( D ) , and quantal content ( E ) recorded from CM9 NMJs in flies from indicated treatment groups . Error bars = s . e . m . p<0 . 05 versus 1X +cmx condition determined using ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 16807 . 008 The activity of 4eBP is negatively regulated by the insulin signaling system ( Figure 2D ) ( Laplante and Sabatini , 2012 ) , a signaling system critical for integrating the nutritional status of the organism with cellular metabolism and organ function . Previous studies in Drosophila using similar dietary conditions have shown that changes in diet can alter insulin signaling ( Grönke et al . , 2010; Morris et al . , 2012 ) . To confirm that our diet conditions resulted in changes in insulin signaling , we performed immunoblot analysis of both phosphorylated Insulin receptor ( InR ) and phosphorylated Akt , reporters of increased insulin signaling ( Figure 2E ) ( Bai et al . , 2015; Bjedov et al . , 2010 ) . This analysis found that both the 2X diet and the 1-2X diet shift conditions resulted in increased phosphorylation of InR and Akt compared to the 1X diet condition supporting that our 2X and 1-2X diet conditions result in increased insulin signaling in our flies . Consistent with insulin signaling in the CM9 motor neuron being responsible for the effects of diet on SV release , we observe that CM9-specific knock-down of the Drosophila IRS-1 homologue chico ( chicoRNAi ) in flies raised on the 2X and 1-2X shift diets resulted in a significant increase SV release compared to the 2X controls ( Figure 2J–L ) . This effect of chicoRNAion SV release in flies raised on the 2X diet was phenocopied by the overexpression of a dominant negative insulin receptor ( InRDN ) in the CM9 motor neuron ( Figure 2K and L ) ( Peru Y Colón de Portugal et al . , 2012 ) . We also observe that there is no difference in neurotransmitter release between chicoRNAi animals raised on 1X , 2X or 1-2X diets versus 1X controls ( Figure 2K and L ) , except that chicoRNAi animals raised on 1X have slightly increased EPSPs ( p=0 . 032 ) compared to 1X controls consistent with low-level insulin signaling even in animals raised on a 1X diet . Finally , the increase in neurotransmitter release observed in the chicoRNAi flies raised on the 2X diet was suppressed by the simultaneous knock-down of 4eBP ( Figure 2K–L ) , consistent with 4eBP functioning downstream of Chico during the regulation of neurotransmission in response to diet . This epistasis between chico and 4eBP is similar to what has been recently observed for Drosophila lifespan ( Bai et al . , 2015 ) . The regulation of neurotransmission by insulin signaling appears to be specific to the adult life stage since we do not observe the same effects of diet or presynaptic knockdown of 4eBP or chico on SV release from larval NMJs ( Figure 2—figure supplement 1 ) . Because of the role of 4eBP in the inhibition of translation , our data suggests that insulin signaling results in the translation of a negative regulator ( s ) of SV release ( Figure 2—figure supplement 2A ) . To investigate this model , flies were raised on a 1X diet were fed the protein translation inhibitor cycloheximide ( cmx ) for 1 hr prior to being shifted to 2X diet supplemented with cmx for 24 hr ( Figure 2—figure supplement 2B ) . We predict that this 1-2X shift diet results in an increase in insulin signaling within the CM9 motor neuron resulting in increased protein translation , which is supported by our immunoblot analysis ( Figure 2F ) . We find that cycloheximide effectively inhibits the reduction in SV release in response to a shift from 1X to 2X diet conditions ( Figure 2—figure supplement 2C and E ) without significant effects on the amplitudes of the mEPSPs ( Figure 2—figure supplement 2D ) . Taken together , these data are consistent with increased insulin signaling resulting in the translation of a negative regulator ( s ) of SV release . The activity of Drosophila 4eBP can be positively regulated transcriptionally by the Drosophila forkhead transcription factor dFOXO ( Figure 3A ) ( Puig et al . , 2003; Teleman et al . , 2005 ) . Analysis of 4eBP mRNA levels in thoracic motor neurons purified by FACS from flies raised on a 1X diet , a 2X diet , or subjected to a diet switch from a 1X diet to a 2X diet reveals that 4eBP mRNA levels are sensitive to diet and that during diet shift the declines in 4eBP mRNA levels ( Figure 3B ) correlate with our observed declines in SV release ( Figure 1F ) . This suggested that the effects of diet on SV release are due to the transcriptional regulation of 4eBP . Previous studies have identified dFOXO-binding sites near the 5’ end of the 4eBP gene ( Puig et al . , 2003 ) . Using chromatin immunoprecipitation ( ChIP ) with anti-FOXO antibodies , we found that these dFOXO-binding sites in the 4eBP promoter region were enriched in our dFOXO ChIP of thoracic ganglion isolated from animals raised on a 1X diet as compared to animals raised on a 2X diet ( Figure 3C ) . This difference was not due to changes in dFOXO protein levels under our diet conditions ( Figure 3D ) . This supports that dFOXO binding to the 4eBP gene is increased under our 1X diet condition compared to the 2X diet condition consistent with dFOXO driving the expression of 4eBP under 1X diet conditions . Importantly , these molecular data suggest that the regulation of 4eBP mRNA levels by diet is conserved among all motor neurons and not specific to CM9 MNs . Electrophysiological recordings from CM9 NMJs in dFOXO mutants raised on 1X , 2X or 1-2X shift diets found that neurotransmission is reduced in dFOXO mutants compared to 1X controls but are not different than the 2X controls , similar to what we observed in the 4eBPRNAi flies ( Figure 3E–G , Table 1 ) . Further , this deficit in neurotransmission at the CM9 NMJs is reversed by the over-expression of 4eBP in the CM9 motor neuron in dFOXO mutants ( Figure 3E–G ) . We also observe that overexpression of 4eBP ( 4eBP OE ) increases neurotransmission compared to 1X controls consistent with persistent insulin signaling in animals on the 1X diet ( Figure 3F and G ) . These results support the model that the negative regulation of neurotransmitter release by insulin signaling involves repression of the dFOXO-dependent gene transcription of the 4eBP locus . 10 . 7554/eLife . 16807 . 009Figure 3 . Effects of diet on the release of neurotransmitter requires FOXO . ( A ) Diagram depicts the regulation of 4eBP by either FOXO-dependent transcription or dTOR-dependent phosphorylation . ( B ) Relative mRNA expression levels of 4eBP in purified motor neurons from 21-day-old animals raised on the indicated diet conditions . ( C ) Graphs represent the average relative fold enrichment of 4eBP DNA in anti-dFOXO chromatin immunoprecipitations ( ChIPs ) from thoracic ganglions isolated from animals raised on 1X or 2X diets . ( D ) Graphs represent average dFOXO protein levels estimated from flies used for ChIP . Values were normalized to actin . Error bars = s . e . m . Immunoblot of dFOXO is shown below . ( E ) Representative CM9 EPSP traces from 14-day-old flies raised on 1X diet of the indicated genotypes . In these genotypes , the overexpression of 4eBP is restricted to the CM9 MN using the E49-Gal4 driver . ( F and G ) Graphs represent the mean value for EPSPs ( F ) and quantal content ( G ) for indicated genotypes raised for 14 days on indicated diets . Error bars = s . e . m . p<0 . 05 versus 1X wild-type controls determined using ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 16807 . 009 The phosphorylation , and subsequent inhibition , of 4eBP by mTOR is an established mechanism for regulating protein translation in response to changes in diet ( Figure 4A ) ( Gingras et al . , 1999 ) ( Ma and Blenis , 2009 ) . Previous studies have established that postsynaptic TOR signaling can influence synapse function ( Penney et al . , 2012; Weston et al . , 2012 ) . Furthermore , TOR signaling has been linked to a number of important neuronal processes including the regulation of synapse structure and function ( Bidinosti et al . , 2010; Costa-Mattioli et al . , 2009; Kelleher et al . , 2004; Stoica et al . , 2011 ) . Thus , we wanted to investigate if Drosophila TOR signaling also played a role in the effects of diet on SV release . To test this possibility , flies raised for 14 days on 1X diet were placed on 1X food supplemented with the potent TOR inhibitor Rapamycin for 6 days prior to switching to a 2X diet also supplemented with Rapamycin ( Figure 4B ) . This treatment paradigm was sufficient to reduce the phosphorylation of S6 kinase ( P-S6K ) and 4eBP ( P-4eBP ) supporting successful inhibition of dTOR under these feeding paradigm ( Figure 4C and D ) . Despite the change in phosphorylation of 4eBP , we observed no effect of the rapamycin treatment on the reduction of neurotransmitter release observed in response to the 1X to 2X diet shift ( Figure 4E–G ) . These data are consistent with the effects of insulin signaling on SV exocytosis being largely independent of dTOR signaling . 10 . 7554/eLife . 16807 . 010Figure 4 . Effects of diet on the release of neurotransmitter is independent of dTOR . ( A ) Diagram depicts the regulation of 4eBP by either FOXO-dependent transcription or dTOR-dependent phosphorylation indicating the effects of rapamycin . ( B ) To investigate the effect of rapamycin ( Rapa ) on diet-regulated SV exocytosis , animals were fed for 14 days on 1X food and then switched to a 1X food supplemented with either 200 µM rapamycin or vehicle for 6 more days . On day 20 , animals were switched from a 1X to a 2X diet supplemented with rapamycin or vehicle for 24 hr prior to electrophysiological analyses . ( C ) Immunoblots of phosphorylated S6 kinase ( P-S6K ) or 4eBP ( P-4eBP ) from animals subjected to above rapamycin treatment demonstrating effective inhibition of dTOR kinase activity under these dietary conditions . Actin signals serves as protein loading control . ( D ) Quantification of intensity of P-S6K determined from immunoblots and normalized for loading . p<0 . 05 determined using Student’s T-test . ( E–G ) Graphs represent the mean values for EPSPs ( E ) , mEPSP ( F ) and quantal content ( G ) recorded from CM9 NMJs of 21-day-old wild-type flies of indicated dietary condition . Error bars = s . e . m . p<0 . 05 versus 1X controls determined using ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 16807 . 010 Our data suggest that ultimately insulin signaling was controlling the translation of an existing mRNA ( s ) that altered neurotransmitter release . An analogous process is the translational control of post-synaptic function by BDNF , which utilizes RNA particles consisting of target mRNAs , translational machinery , transport proteins and RNA-binding proteins such as the Staufen-family of RNA-binding proteins ( Leal et al . , 2014; Takei et al . , 2004 ) . Therefore , we investigated whether Drosophila Staufen was involved in the effects of diet on neurotransmitter release at the CM9 NMJ . Similar to what we observed with cycloheximide , we found the RNAi knockdown of Staufen ( staufenRNAi ) in CM9 motor neurons blocked the reduction in neurotransmitter in response to diet shift ( Figure 5A , C and D ) without altering muscle sensitivity to glutamate ( Figure 5B; Table 1 ) . These results support the model that a Staufen bound mRNA is involved in the effects of diet on neurotransmitter release . Recently , the collection of mRNAs bound to Staufen2 in mammalian brain tissues was defined and a number of important presynaptic regulators of synapse function were identified ( Heraud-Farlow et al . , 2013 ) . One of the mRNAs identified in this study was the mRNA encoding complexin , a small peptide that functions as both a facilitator and an inhibitor of SV exocytosis ( Südhof , 2012; Trimbuch and Rosenmund , 2016 ) . In addition , Drosophila complexin mRNA contains predicted Staufen target sequences ( STSs ) ( Laver et al . , 2013 ) . Therefore , we investigated if complexin mRNA was bound to Staufen in motor neurons in adult Drosophila . FACs sorted motor neurons expressing a GFP-tagged Staufen were subjected to an RNA immunoprecipitation ( RIP ) with anti-GFP-coated beads ( Laver et al . , 2013 ) . Enrichment of complexin mRNA in anti-GFP RIPs from experimental versus control motor neurons was determined using quantitative RT-PCR . These analyses revealed that complexin mRNA was highly enriched in Staufen RIPs from motor neurons ( Figure 5E ) . Surprisingly , we also observed that 4eBP mRNA was also highly enriched even though 4eBP mRNAs do not contain a predicted STS ( Figure 5E ) . The RIP of complexin and 4eBP mRNA was specific since a number of other neuronal mRNAs were not enriched in these RIP experiments including mRNAs for tubulin , syntaxin and the cacophony voltage-gated calcium channel ( Figure 5E ) . 10 . 7554/eLife . 16807 . 011Figure 5 . The role of Staufen during the regulation of neurotransmission by diet . ( A ) Representative traces of EPSPs from CM9 NMJs from 21 day old staufenRNAi raised on a 1X diet or staufenRNAi and control flies subjected to a diet switch from 1X to 2X diet on day 20 and recorded on day 21 . Scale bar = 1 mv , 10 ms . ( B–D ) Graphs represent the mean values for EPSPs ( B ) , mEPSPs ( C ) , and quantal content ( D ) recorded from indicated genotypes . p<0 . 05 determined using ANOVA . ( E ) Graphs represent the fold enrichment of Staufen-bound mRNAs immunoprecipitated from FACS sorted motor neurons . p<0 . 01 versus tubulin control determined using ANOVA . ( F ) Diagram depicts the putative regulation of complexin mRNA translation in response to insulin signaling in the CM9 motor neuron . DOI: http://dx . doi . org/10 . 7554/eLife . 16807 . 011 Based on these results , we investigated if synaptic Complexin levels are sensitive to diet conditions . Immunofluorescent microscopy of CM9 NMJs in flies raised on 1X and 2X diets found that Complexin staining was increased at CM9 NMJs in animals on a 2X diet compared to animals on the 1X diet ( Figure 6A and B ) . We used deconvolution microscopy to quantify the intensity of Complexin staining at the CM9 NMJ and find that the average maximum pixel intensity of Complexin staining is significantly increased at NMJs from animals raised on the 2X diet compared to the 1X diet ( Figure 6C ) and that the distribution of m . p . i values is also significantly different ( Figure 6D; see source data file for Kolmogorov-Smirmov test results ) . Furthermore , chicoRNAi animals raised on a 2X diet had significantly reduced Complexin levels ( Figure 6C ) . Identical results were observed for median and mean pixel intensity as well ( data not shown ) . Because our molecular analyses suggested that this signaling system also functioned in motor neurons located within the thoracic ganglion , we investigate synaptic Complexin levels at the NMJs formed on the lateral abdominal muscles ( LAMs ) , which are innervated by motor neurons found within the thoracic ganglion ( Hebbar et al . , 2006; Krzemien et al . , 2012 ) . Using an identical approach to quantifying Complexin at the CM9 NMJs , we find that there is significantly more Complexin at the LAM NMJs in flies raised on a 2X diet compared to flies raised on a 1X diet supporting that this signaling system is present in LAM motor neurons as well ( Figure 6—figure supplement 1B–D ) . Functionally , we find that reduction of one copy of the complexin gene ( cpxSH1/+ ) significantly increases quantal content at CM9 NMJs in animals raised on both the 1X and 2X diets compared to controls ( Figure 6E and F , Table 1 ) . We also observe a significant increase in the frequency of spontaneous SV fusion events in cpxSH1/+ animals consistent with previous observations at the Drosophila larval NMJ ( Figure 6H ) ( Jorquera et al . , 2012 ) . Fluorescent microscopic quantification of synaptic complexin levels reveals a significant reduction in the amount of Complexin at the synapse in the cpxSH1/+ animals compared to wild type ( Figure 6J ) . Conversely , we observe a reduction in quantal content if we overexpress Complexin in the CM9 motor neuron of animals raised on a 1X diet compared to 1X diet controls ( Figure 6K–M ) . We also find that overexpression of Complexin in animals on 2X diets does not reduce neurotransmission further compared to 2X diet controls similar to our data with 4eBP , chico and dFOXO ( Figure 6L and M ) . These data are consistent with the model that diet can influence neurotransmission at the adult NMJ by regulating the synaptic levels of Complexin . 10 . 7554/eLife . 16807 . 012Figure 6 . Complexin levels regulate SV release in response to diet . ( A ) Immunofluorescent images of CM9 NMJs from animals raised on a 1X ( left panels ) or 2X ( right panels ) diet co-stained for Complexin ( red-upper panels and lower panels ) , Discs-large ( Dlg , green upper panels ) and Dapi ( blue-upper panels ) . ( B ) High magnification of Complexin ( Cpx-upper panels ) and Discs-large ( Dlg-lower panels ) from CM9 NMJ boutons in animals raised on a 1X ( left panels ) or 2X ( right panels ) diet conditions . Images for Cpx have been deconvolved . ( C and D ) Graphs represent the average value ( C ) and frequency histogram ( D ) for the maximum pixel intensity ( m . p . i . ) of synaptic Complexin from indicated diet conditions and genotype . p<0 . 05 determined using ANOVA comparison of mean values ( C ) or using a Kolmogorov-Smirmov test ( D ) . ( E ) Representative traces of EPSPs from CM9 NMJs from wild type ( wt ) or complexin heterozygotes ( cpxSH1/+ ) raised for 21 days on the indicated diets . Scale bar = 1 mV , 10 ms . ( F and G ) Graphs represent the mean values for mEPSPs ( F ) and quantal content ( G ) from CM9 NMJs from 21-day-old animals of indicated genotypes raised on the indicated diet . p<0 . 05 versus 1X condition determined using ANOVA . ( H ) Representative traces from wt and cpxSH1/+ animals raised on 2X diet for 21 days . Graphs below traces represent the quantification of the events per second from CM9 NMJs of the indicated genotypes raised on the indicated diets . p<0 . 05 versus 1X control determined using ANOVA . ( I ) Representative traces of EPSPs from CM9 NMJs in 21-day-old animals on the indicated genotypes raised on the indicated diet . Scale bar = 1 mV , 10 ms . ( J and K ) Graphs represent the mean values for mEPSPs ( J ) and quantal content ( K ) from CM9 NMJs from 21-day-old animals of indicated genotypes raised on the indicated diets . p<0 . 05 versus 1X control determined using ANOVA . DOI: http://dx . doi . org/10 . 7554/eLife . 16807 . 01210 . 7554/eLife . 16807 . 013Figure 6—source data 1 . File contains background-corrected values of max pixel intensity from complexin ( Cpx ) staining at the CM9 NMJ from indicated genotypes and diet conditions . Included are data for quantification max pixel intensities for the Cpx staining presented in Figure 6C , D , J , and Figure 6—figure supplement 1 . Data are presented in separate sheets as labeled . The results of statistical analyses ( Student’s t-test , Kolmogorov-Smirnov test ) are presented with each data set . DOI: http://dx . doi . org/10 . 7554/eLife . 16807 . 01310 . 7554/eLife . 16807 . 014Figure 6—figure supplement 1 . Diet effects on synaptic complexin levels at the lateral abdominal muscle NMJs . ( A ) Diagram represents the area of the abdomen that is being analyzed ( grey box ) . Image shows the NMJs on the lateral abdominal muscles ( LAMs ) used in these analyses . Staining for Dlg is shown and dashed line indicates the ventral mid line . ( B ) Non-deconvolved immunofluorescent images of LAM NMJs from animals raised on a 1X ( left panels ) or 2X ( right panels ) diet co-stained for Complexin ( Cpx; upper panels ) , Discs-large ( Dlg , middle panels ) . A merged image is shown in the bottom panels ( Cpx = green channel; Dlg = red channel ) . Scale bar = 10 µm . ( C and D ) Graphs represent the average value ( C ) and frequency histogram ( D ) for the maximum pixel intensity ( m . p . i . ) of synaptic Complexin from indicated diet conditions . p<0 . 05 determined using ANOVA comparison of mean values ( C ) or using a Kolmogorov-Smirmov test ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16807 . 014 Here , we have revealed that insulin signaling in adult Drosophila motor neurons can negatively regulate the release of neurotransmitter from the NMJ . This control of neurotransmission by insulin signaling utilizes the FOXO transcription factor to transcriptionally regulate the eukaryotic initiation factor 4e binding protein ( 4eBP , also known as Thor in Drosophila ) , a negative regulator of cap-dependent translation ( Gingras et al . , 1999 ) . Importantly , our data suggest that the control of neurotransmitter release by insulin signaling is dependent on the diet conditions and likely does not reflect a role for insulin in basal neurotransmitter release . Our data supports the model that repression of FOXO activity due to insulin signaling results in reduced levels of 4eBP mRNA , subsequent increased protein translation , and reduced SV release . The activity of 4eBP is also regulated by phosphorylation via the actions of the target of rapamycin complex ( TOR ) ( Beretta et al . , 1996 ) and numerous studies have implicated the TOR complex in the regulation of synapse function ( Costa-Mattioli et al . , 2009; Hoeffer and Klann , 2010; Penney et al . , 2012; Takei et al . , 2004; Weston et al . , 2012 ) . The regulation of neurotransmission by diet at the CM9 NMJ appears to be largely independent of TOR since the effect of diet on neurotransmission is not affected by rapamycin , a potent inhibitor of the TOR . Importantly , we observe that our rapamycin treatment condition does result in the predicted change in the phosphorylation state of 4eBP demonstrating that TOR can regulate 4eBP in adult Drosophila motor neurons . Because most of the data on the effects of TOR on synapse function suggest a post-synaptic role for this complex , these data suggest that the regulation of 4eBP within the CM9 motor neuron is compartmentalized with the presynaptic pool regulated specifically by FOXO and the post-synaptic pool regulated by TOR . The localization of the TOR complex in neurons is unknown , but it presumably is localized within the cytoplasm and lysosomes ( Betz and Hall , 2013 ) . Whether TOR is excluded from the presynaptic terminal or enriched within the postsynaptic compartment remains to be investigated . There exist three members of the 4eBP family in mammals with 4eBP2 being the most highly expressed family member in the brain ( Banko et al . , 2005 ) . Analysis of 4eBP2 knock-out mice has revealed that this protein is required for a broad range of cognitive and motor behaviors ( Banko et al . , 2007; Gkogkas et al . , 2013 ) . The changes in behavior observed in 4eBP2 knock-out mice correlate with changes in synapse function that are highlighted by changes in post-synaptic glutamate receptor function ( Banko et al . , 2005; Bidinosti et al . , 2010; Gkogkas et al . , 2013; Ran et al . , 2013 ) . To date , there is no evidence from these studies of an effect of the 4eBP2 knock-out on presynaptic function . In addition to effects on glutamate receptor function , 4eBP2 has also been implicated in the regulation of neuroligin levels ( Gkogkas et al . , 2013; Khoutorsky et al . , 2015 ) , a post-synaptic scaffolding protein that functions to regulate synaptogenesis and neurotransmission ( Craig and Kang , 2007 ) . The regulation of neurotransmission by neuroligin is likely due to its trans-synaptic interaction with the presynaptic binding protein neurexin , a cell adhesion molecule known to regulate synaptic vesicle exocytosis ( Südhof , 2008 ) . Thus , changes in post-synaptic neuroligin levels can result in increased presynaptic function , although this would represent a non-autonomous role for 4eBP on neurotransmitter release . It is unclear if altered neuroligin-neurexin signaling contributes to the neurotransmission phenotypes observed in the 4eBP2 knock-out mice . We find that phosphorylation of 4eBP by dTOR has no effect on the regulation of neurotransmitter release by insulin signaling in the CM9 MN . This result suggests that there might exist separate pools of 4eBP within the neuron that specify the effects of TOR versus FOXO on synapse function . Currently , it is unclear how the compartmentalization of 4eBP activity is achieved within the pre- versus postsynaptic compartments . We have found that Staufen binds to 4eBP mRNA in motor neurons and is required for the effects of diet on neurotransmission . In addition to mRNA transport , Staufen is also known to bind nascent mRNAs and mediate their nuclear export ( Elvira et al . , 2006; Jansen and Niessing , 2012; Liu et al . , 2006; Macchi et al . , 2004; Miki and Yoneda , 2004; Miki et al . , 2005 ) . Perhaps , the association of Staufen with nascent 4eBP mRNAs driven by FOXO differentiates the dendritic from axonal populations of 4eBP . In addition to 4eBP mRNA , we also find that Drosophila Staufen binds strongly to complexin mRNA . This suggested that diet might control neurotransmission via the regulation of Complexin . In support of this model , we find that Complexin levels at the CM9 NMJ is increased in animals raised on a 2X diet compared to a 1X diet and that these levels are sensitive to changes in insulin signaling . This suggests that the increased levels of Complexin in animals raised on the 2X diet inhibit the SV release . We also find that genetically altering complexin levels can influence neurotransmitter release from the CM9 NMJ in a diet-dependent manner similar to what is observed with 4eBP , chico and dFOXO mutants supporting the model that Complexin is an important target for the regulation of neuronal function by insulin signaling . Because both complexin and 4eBP mRNAs are bound to Staufen , perhaps diet controls neurotransmitter release by altering the relative amounts of bound 4eBP to complexin mRNAs . The effects of our diet switch on neurotransmission support that an acute increase in Complexin levels can inhibit neurotransmitter release at the CM9 NMJ . It is clear from knock-out studies in mice , worms and Drosophila larvae that Complexin is required for normal calcium-dependent SV exocytosis and supports a facilitatory , not inhibitory , role for Complexin during neurotransmission ( Cho et al . , 2010; Jorquera et al . , 2012; Radoff et al . , 2014; Reim et al . , 2001 ) . But other studies , including acute injections and vesicle targeting studies , have indicated that Complexin can also have an inhibitory role on evoked release ( Archer et al . , 2002; Giraudo et al . , 2006; Liu et al . , 2007; Ono et al . , 1998; Tang et al . , 2006; Tokumaru et al . , 2001 ) . Further , comparison of complexin knock-down to knock-out in different neuronal cell types suggest that the effects of Complexin on SV exocytosis can be sensitive to chronic versus acute manipulations and dependent upon neuronal cell type ( Yang et al . , 2013 ) . Although we find that diet has no effect on SV release from Drosophila larval NMJs , further studies will be needed to determine if this is due to differences between adult and larval motor neurons or to differences in the manipulations of Complexin . In addition , it is likely that the effects of Complexin that we observe require the co-translation of other exocytotic components . Regardless our data support the model that increases in synaptic Complexin levels resulting from insulin signaling can reduce neurotransmitter release . These results have broad implications for the effects of insulin signaling on the nervous system . All analyses were performed on virgin female flies that were flipped to freshly made food vials every other day and kept at 50% humidity on a 12 hr light/dark cycle ( Rawson et al . , 2012 ) . All foods were made fresh every week and flies flipped every 2 days to minimize water loss for all diet conditions . For CM9 motor neuron expression , we used the E49-Gal4 line that was obtained from the Kristin Scott lab ( Gordon and Scott , 2009 ) . The UAS-4eBP line was obtained from the Rolf Bodmer lab ( Birse et al . , 2010 ) . Fly lines harboring the UAS-chicoRNAi and the UAS-4eBPRNAi transgenes were obtained from the Vienna Drosophila RNAi Center ( Vienna Drosophila Resource Center , RRID:SCR_013805 , stocks 101329 and 35439 , respectively ) . The dFOXO94 line was obtained from Bloomington stock center ( RRID:BDSC_42220 ) and the dFOXO21 line was obtained from the Marc Tatar lab ( Min et al . , 2008 ) . The UAS-InRDN fly line was obtained from the Adrian Rothenfluh lab ( Peru Y Colón de Portugal et al . , 2012 ) . Staufen-GFP knock-in flies ( GFP 311 ) were obtained from the Lipshitz lab ( Laver et al . , 2013 ) . The UAS-staufenRNAi line was obtained from the Bloomington stock center ( RRID:BDSC_31247 ) . All transgenes used in this study were backcrossed at lease five generations to the w1118 strain and rebalanced in our w1118 background . The following genotypes were abbreviated in the text: wt = w1118 . 4eBPRNAi = E49-Gal4/UAS-4eBPRNAi . chicoRNAi= E49-Gal4/UAS-chicoRNAi . staufenRNAi = E49-Gal4/UAS-staufenRNAi; dFOXO = w1118; dFOXO94/21 . dFOXO , 4eBP OE = E49-Gal4/UAS-4eBP; dFOXO94/21 . 4eBP OE = E49-Gal4/UAS-4eBP . The low-protein ( 1X ) and high-protein ( 2X ) diets are exactly the same except for the amount of active yeast added and consisted of the following composition per 500 mls of food as per Bass et al . ( 2007 ) : 5 g agar ( Genesee ) , 50 g active yeast ( 1X = 5% ) , or 100 g active yeast ( 2X = 10% ) ( Red Star ) , 25 g corn meal ( Quaker ) , 25 g sucrose ( Speckles ) , 1 . 5 ml propionic acid ( Sigma ) , and 1 . 5 g tegosept ( Sigma ) . For all experiments , newly hatched flies were kept on standard lab food for 5 days prior to being split to indicated diet conditions . All recordings were performed in 21-day-old virgin females except of dFOXO mutants , which were feeble and died within 21 days of eclosion and therefore assayed at 14 days . Dissections and recordings were performed in a modified HL3 solution ( containing , in mM: 70 NaCl , 5 KCl , 10 NaHCO3 , 5 trehalose , 115 sucrose , 5 HEPES , 0 . 5 CaCl2 , 3 MgCl2 ) . Flies were suctioned into a Pasteur pipette and placed on top of ice for 15–20 s until the fly lost postural control . The fly was then quickly transferred to a small Sylgard dissection surface where it was decapitated . The head was moved onto its flat posterior surface , and the proboscis was then pinned into the extended position , and the entire head was covered in ice-cold dissection solution . The anterior head cuticle containing the antennae was dissected from the preparation . The proboscis was then re-pinned in the retracted position to put tension on the CM9 muscles . A loop of the lateral pharyngeal nerve was drawn into a suction electrode filled with modified HL3 ( pulled glass capillary tube with a fire-polished tip , ~15 μm opening ) and stimulated at 0 . 5–5 V for 300 μs ( Digitimer Ltd . , Model DS2A ) . The presence of a presynaptic action potential-based EPSP was verified by the presence of a distinct voltage threshold for EPSP appearance . Intracellular recordings were made on the most cranial CM9 muscle fiber accessible from the anterior side with a sharp recording electrode ( ~30 MΩ , filled with 3 m potassium chloride ) . The overall organization of the fibers is highly stereotyped from animal to animal and across age , so it is likely we are interrogating the same fiber in each recording , which is supported by our low variance . A Neuroprobe Amplifier Model 1600 ( A-M Systems ) was used in combination with a PowerLab 4/30 ( ADInstruments , Colorado Springs , CO ) to amplify and digitize the data . LabChart7 ( ADInstruments , Colorado Springs , CO ) was used to record the data and MiniAnalysis ( Synaptosoft , Fort Lee , NJ ) was used to measure both miniature EPSP ( mEPSP ) and EPSP events . Muscle membrane resistance was calculated using the change in muscle potential in response to current injection . Instantaneous resting membrane potential was determined by measuring the initial potential reading when the recording electrode first penetrated the muscle membrane . For hypertonic stimulation of readily releasable vesicle pools , normal recording saline was initially applied to the preparation to record baseline spontaneous activity before being replaced with recording saline supplemented with sucrose to a total final concentration of 315 mm , and recordings continued for 60 s in hypertonic saline . For diet shift experiments , animals were fed on a 1X diet from day 5-post eclosion until day 20 when half of the animals were switched to a 2X diet . For Rapamycin ( Sigma , St . Louis , MO ) experiments , animals were either fed a 1X diet containing either 200 µM rapamycin or vehicle control ( 200 µM ethanol ) from 14 days post-eclosion until animals were switched onto a 2X plus rapamycin food or vehicle control for 24 hr at 20 days post-eclosion . For cycloheximide ( Sigma , St . Louis , MO ) experiments , animals were either fed a 1X diet containing either 35 µM cycloheximide or vehicle control ( 35 µM ethanol ) from 19 days post-eclosion to condition the animals until they were switched onto a 2X plus cycloheximide food or vehicle control for 24 hr at 20 days post-eclosion . For all electrophysiology analyses , 7–9 animals were assayed with only one recording performed per animal ( see Table 1 ) . For larval analyses , eight animals were assayed with only one recording performed per animal . Virgin flies of the appropriate genotype and dietary conditions were starved and deprived of water for 4–6 hr prior to PER analysis . Flies were anesthetized under carbon dioxide , loaded into pipet tips and allowed to recover for 30 min . For bristle tracking , digital videos of individual PERs from animals subjected to tarsal stimulation with 0 . 5 M sucrose were captured at 10–15 frames per second using a Zeiss MRc digital camera and Slidebook software ( Intelligent Imaging Innovations , Denver CO ) . The Slidebook particle tracking feature was used to manually track bristles on the tip of the proboscis during PER and values for maximum velocity and average velocity were determined for each bristle path . Mean values for PER values consisted of seven animals were assayed with two PER events per animals included in analysis ( 14 events total ) . Actin reference antibody was mouse monoclonal sc-8432 used at a 1:250 dilution ( Santa Cruz Biotechnology , Dallas , TX ) . Rabbit polyclonal antibodies against Phopho-S6K ( Cell Signaling Technology Cat# 9209S RRID:AB_2269804 , Danvers , MA ) and Phospho-4eBP ( Cell Signaling Technology Cat# 2855S RRID:AB_560835 ) were used at a 1:1000 dilution . For immunoblot analysis of insulin signaling , the rabbit antibody against phospho-Insulin Receptor was used at 1:1000 ( Cell Signaling Technology Cat# 3021S RRID:AB_331578 ) , the rabbit antibody against Drosophila Akt was used at 1:1000 ( Cell Signaling Technology Cat# 9272 RRID:AB_329827 ) , and the rabbit antibody against Drosophila phospho-AKT ( Ser505 ) was used at 1:1000 ( Cell Signaling Technology Cat# 4054S RRID:AB_331414 ) . For dFOXO immunoblots , rabbit anti-dFOXO ( gift of Oscar Puig Lab ) was used at a 1:500 dilution . Proteins were extracted from whole flies by homogenizing them in 2x SDS sample buffer with Complete Mini protease inhibitor tablets ( Roche , Indianapolis , IN ) and Halt phosphatase inhibitor cocktail ( Thermo Scientific , Rockford , IL ) . About 30 µg of denatured protein was separated on 4–15% Mini-Protean TGX Gels ( Bio-Rad , Hercules , CA ) until the desired band range was resolved sufficiently and transferred to nitrocellulose membranes using 350 mA with sodium tetraborate/boric acid buffer . Blocking was performed with 3% BSA for 30 min . Following incubation with primary and HRP-conjugated secondary antibodies , the blots were visualized with Novex ECL ( Invitrogen , Grand Island , NY ) . Band intensity was quantified with ImageJ ( NIH , Bethesda , MD ) . For motor neuron-specific analysis , thoracic ganglion from flies expressing GFP in all motor neurons ( D42-Gal4 , UAS-10X-GFP ) were removed and dissociated using 1 mg/ml collagenase ( Sigma , St . Louis , MO ) and sorted from non-fluorescent cells on a Beckman Dickinson Aria FACS unit using a 70 µm tip . Total RNA was extracted by FACS sorting ~50 , 000 Drosophila neurons into RTL buffer from an RNAEasy kit , and subsequently extracted via the same kit using the standard protocol ( Qiagen , Hilden , Germany ) . First strand cDNA was generated by reverse transcription with SuperScript III enzyme ( Invitrogen , Waltham , MA ) . Quantitative PCR was performed on an ABI 7500 Fast Real-Time PCR system , using exon-spanning primers and SYBR green PCR premix ( Applied Biosystems , Warrington , UK ) . The following primer pairs ( Forward/Reverse , 5’-3’ ) were used for these analyses: 4eBP: CACTCCTGGAGGCACCA/ GAGTTCCCCTCAGCAAGCAA , complexin: CGCGAGAAGATGAGGCAAGA/ CATCAGGGGATTGGGCTCTT , tubulin: ACAACTTCGTGTACGGACAGT/ CACCACCGAGTAGGTGTTCA , syntaxin: CCACAAACGGACGAGAAGACC/ CGCCGACGACTTATTCTGCT , cacophony ( cac ) : TTCGGGCGCACTGCATAAG/ GGTGGCCTTTTCCAGGATGT . Technical replicates were performed in triplicate for all target and control genes . Transcript quantification was performed by the ∆∆Ct method . For 4EBP quantification under diet conditions , the experiment was repeated on biologically independent samples four times . Chromatin immunoprecipitation ( ChIP ) was carried out according to Tran et al . ( Tran et al . , 2012 ) . The thoracic ganglions from wild type flies raised on either 1X or 2X diets were dissected ( n = 20 per experimental replicate ) , fixed , and sonicated ( average fragment size ~500 bp ) . Magnetic protein G beads were incubated with an anti-dFOXO polyclonal antibody ( Puig et al . , 2003 ) and then incubated overnight with the sonicated cell lysate . Beads incubated with rabbit pre-immune serum was used as a ChIP control . Quantitative RT-PCR was performed via the ∆∆ Ct method with primers targeting established dFOXO response elements ( FREs ) in the promoter region of d4eBP ( Forward 4EBP Promoter Primer: 5’- CAC CTC TTG ACT CCC AGA CAG -3’; Reverse 4EBP Promoter Primer: 5’- ATG ATA AGG GGT GTA GCG ATG -3’ ) . Primers to a gene desert in chromosome 3 were used as reference values for normalization ( Active Motif , Carlsbad CA , #71028 Drosophila Negative Control Primer Set 1 ) . Staufen-GFP knock-in flies were crossed to D42 , UAS-mCherry flies . The thoracic ganglia from 10 Staufen-GFP/D42 , UAS-mCherry flies and control D42 , UAS-mCherry flies were dissected and pooled separately . The dissected tissue was triturated at 4°C in standard homogenization buffer ( 250 mM sucrose , 10 mM HEPES , 1 mM EGTA , 1 mM EDTA , 0 . 1% NP40 ) with added Complete Mini Protease Inhibitor Cocktail Tablets ( Roche , Penzberg , Germany ) and RNAseOut RNAse inhibitors ( Themo Scientific , Waltham ) at 1% ( v/v ) concentration . This was then further ruptured by vigorous pipetting and briefly centrifuged in a standard tabletop centrifuge to pellet cell debris . The supernatant was then mixed with magnetic beads ( Immunoprecipitation Kit Dynabeads Protein G , LifeTechnologies , Waltham , MA ) that had previously been incubated with anti-GFP antibody ( UC Davis/NIH NeuroMab Facility Cat# N86/8 RRID:AB_2313651 ) according to kit instructions . This was then incubated with rotation at 4°C for 4 hr . The beads were then pelleted and washed according to kit instructions . Buffer RTL Plus ( with 1:100 2-mercaptoethanol ) from an RNAEasy Micro Kit ( Qiagen , Hilden , Germany ) was added to the magnetic beads , then vortexed vigorously for 30 s . Beads were then separated via a magnetic stand , and the supernatant was used as the input to the RNAeasy kit . RNA was then isolated according to the manufacturer’s instructions . RNA was stored at −80°C until used for making cDNA for qRT-PCR . Levels of RNAs bound to Staufen were compared across Staufen-GFP and control flies , normalizing to tubulin transcript . For these analyses , the diet and genotypes of all samples are blinded prior to the procedure . Adult Drosophila fly heads were pinned and dissected as previously described ( Rawson et al . , 2012 ) except for differences noted below . All CM9 muscles were dissected , fixed , and stained on the same day . After primary dissection , all proboscises were fixed in 4% paraformaldehyde for 15 min . After three 5-min washes in PBT , the CM9 was dissected and placed into a microcentrifuge tube containing PBT . After fixation , CM9s were blocked in ImageIt FX Signal Enhance ( LifeTechnologies , Waltham , MA ) for 30 min . CM9s were washed three times with PBT and subsequently placed into a new microfuge tubes for staining and incubated overnight with indicated primary antibodies . For Lateral abdominal muscles ( LAMs ) , adult abdomens were pinned dorsal side up , filleted open , internal organs removed , and pinned down in cold dissecting saline . LAM preps were fixed in 4% paraformaldehyde for 15 min after which the LAM dissections transferred to a microcentrifuge tube and washed three times for 5 min in PBT . LAMs and CM9 preps were processed for immunofluorescence identically from this point . Both anti-cpx ( rabbit polyclonal IgG , gift from Dr . Troy Littleton ) and anti-Dlg ( DSHB Cat# DLG1 RRID:AB_2314322 ) were used at a dilution of 1:500 . After primary antibody incubation , preps were washed three times with PBT and placed into separate Eppendorf tubes containing a 1:500 dilution of secondary antibodies and incubated at room temperature for 1 hr . The preps were then washed three times with PBT and mounted in Vectashield containing DAPI ( Vector Labs , Burlingame , CA ) . Dlg staining was visually inspected and any preps showing inconsistency or poor quality of Dlg staining were removed from analysis . Mounted CM9 preparations were captured using back-cooled Orca digital camera ( Hamamatsu ) attached to a Zeiss Axiovert immunofluorescent microscope using Slidebook software ( Intelligent Imaging Innovations , Denver , CO ) . For intensity analysis , images were subjected to nearest neighbor deconvolution and Complexin signal intensity data acquired from sub masks generated using automated segmentation of Complexin signals provided by the Slidebook software . For the immunofluorescent analysis of Complexin , 7–11 synapses from 4 to 8 animals were analyzed . Each CM9 image yielded between 30 and 150 Complexin data points . Background masks were also generated for each synapse and values for average max pixel intensity for background subtracted from Complexin values . A Student’s t-test was used for all pair-wise comparisons . A one-way ANOVA using a Tukey multiple comparisons test ( alpha = 0 . 05 ) was used to compare all multiple values . Significance for distributions in Figures 1 and Figure 6 were determined using non-parametric pair-wise comparison using a Kolmogorov-Smirmov test . For all statistical analyses a confidence interval of 95% was assumed . Statistical analysis was performed using Prism6 software ( Graphpad Prism , RRID:SCR_002798 ) . The results of the statistical analyses of source data are presented in source data file .	The rates of obesity and diabetes are increasing worldwide . Both conditions produce a wide range of detrimental effects on health , including an increased risk of developing neurodegenerative diseases such as Alzheimer’s disease . Obesity and diabetes reduce how well many of the body’s cells can respond to a hormone called insulin . Insulin signaling is believed to influence how the brain works , but this had not been studied in detail . Mahoney et al . have now studied the fruit fly Drosophila melanogaster to investigate whether insulin signaling within neurons can directly alter neurotransmission – the process by which neurons communicate with each other by releasing chemicals called neurotransmitters . The fruit flies were fed a high protein diet , which increased their insulin signaling and reduced the activity of a protein called FOXO in the neurons . This resulted in the reduced transcription of the translational inhibitor 4eBP and ultimately caused an increase in the amount of the Complexin protein . This protein in turn reduced the release of neurotransmitters . Thus , the results of the experiments demonstrate that insulin signaling within adult fruit fly neurons decreases neurotransmission . Future experiments will be needed to study these mechanisms in more detail . One of the remaining open questions is where proteins such as Complexin are being made in the neuron .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Insulin signaling controls neurotransmission via the 4eBP-dependent modification of the exocytotic machinery
H3K9 methylation ( H3K9me ) specifies the establishment and maintenance of transcriptionally silent epigenetic states or heterochromatin . The enzymatic erasure of histone modifications is widely assumed to be the primary mechanism that reverses epigenetic silencing . Here , we reveal an inversion of this paradigm where a putative histone demethylase Epe1 in fission yeast , has a non-enzymatic function that opposes heterochromatin assembly . Mutations within the putative catalytic JmjC domain of Epe1 disrupt its interaction with Swi6HP1 suggesting that this domain might have other functions besides enzymatic activity . The C-terminus of Epe1 directly interacts with Swi6HP1 , and H3K9 methylation stimulates this protein-protein interaction in vitro and in vivo . Expressing the Epe1 C-terminus is sufficient to disrupt heterochromatin by outcompeting the histone deacetylase , Clr3 from sites of heterochromatin formation . Our results underscore how histone modifying proteins that resemble enzymes have non-catalytic functions that regulate the assembly of epigenetic complexes in cells . The covalent and reversible modification of histones allows cells to establish stable and heritable patterns of gene expression without any changes to their genetic blueprint . Histone H3 lysine nine methylation ( H3K9me ) is a conserved mark of transcriptional silencing that is associated with the formation of specialized domains called heterochromatin . Heterochromatin formation is critical for centromere and telomere function , silencing of transposons and repetitive DNA elements , and for maintaining lineage-specific patterns of gene expression ( Grewal and Jia , 2007; Nicetto et al . , 2019 ) . It is widely assumed that a delicate interplay between histone modification readers , writers , and erasers regulates the establishment and maintenance of epigenetic states ( Allis and Jenuwein , 2016 ) . In fission yeast , the establishment of H3K9 methylation requires the enzymatic activity of a conserved methyltransferase , Clr4Suv39h ( Rea et al . , 2000 ) . H3K9 methylation acts as a platform to recruit chromatin effector proteins with reader domains . Two HP1 homologs , Swi6HP1 and Chp2HP1 , are involved in H3K9 methylation binding and have distinct , non-overlapping functions during transcriptional silencing ( Motamedi et al . , 2008 ) . Ultimately , factors such as Epe1 , a putative histone demethylase , promote epigenetic erasure , thus reversing H3K9 methylation and transcriptional gene silencing ( Ayoub et al . , 2003; Trewick et al . , 2005 ) . The sequence-specific recruitment of Clr4Suv39h restricts heterochromatin establishment to distinct sites in the genome , such as the centromeres , telomeres , and the mating-type locus ( Bayne et al . , 2010; Hall et al . , 2002; Verdel et al . , 2004; Volpe et al . , 2002 ) . Once established , H3K9 methylation spreads to silence genes that are distant from heterochromatin nucleation centers . H3K9me spreading depends on the read-write activity of Clr4Suv39h ( Al-Sady et al . , 2013; Zhang et al . , 2008 ) . Two conserved , structural attributes of Clr4Suv39h mediate this process: a conserved chromodomain that recognizes and binds to H3K9me and an enzymatic SET domain that is involved in catalysis ( Ivanova et al . , 1998 ) . The ability of Clr4Suv39h to bind to the product of its enzymatic activity enables H3K9 methylated histones to serve as carriers of epigenetic information ( Audergon et al . , 2015; Ragunathan et al . , 2015 ) . Following DNA replication , H3K9 methylated histones that are partitioned between daughter DNA strands serve as templates to mark newly deposited histones . H3K9 methylation acts as a multivalent platform that recruits the HP1 homolog , Swi6HP1 to sites of constitutive heterochromatin ( Ekwall et al . , 1995; Ekwall et al . , 1996 ) . HP1 proteins have a conserved architecture consisting of a chromodomain ( CD ) that is involved in H3K9 methylation binding and a dimerization domain called the chromoshadow domain ( CSD ) that mediates protein-protein interactions ( Bannister et al . , 2001; Cowieson et al . , 2000 ) . The oligomerization of HP1 proteins promotes formation of higher-order chromatin complexes that exhibit phase separated , liquid-like properties in vitro and in cells ( Larson et al . , 2017; Sanulli et al . , 2018; Strom et al . , 2017 ) . Swi6HP1 interacts with a broad spectrum of agonists and antagonists that influence epigenetic silencing . Most notably , recruitment of Epe1 , a putative histone demethylase that opposes heterochromatin formation , is dependent on Swi6HP1 ( Ayoub et al . , 2003; Zofall and Grewal , 2006 ) . Loss of Epe1 leads to increased heterochromatin spreading beyond normal boundary sequences , inheritance of H3K9 methylation via sequence-independent pathways , increased cell to cell variation in H3K9 methylation patterns , and acquisition of adaptive epigenetic traits ( Audergon et al . , 2015; Ayoub et al . , 2003; Ragunathan et al . , 2015; Sorida et al . , 2019; Trewick et al . , 2007; Wang et al . , 2015; Zofall and Grewal , 2006; Zofall et al . , 2012 ) . Despite its critical role in heterochromatin regulation , how Epe1 exerts its anti-silencing function in cells remains mysterious ( Trewick et al . , 2007 ) . Histone demethylases have prominent roles in regulating the reversibility of epigenetic states ( Iwase et al . , 2007; Whetstine et al . , 2006 ) . Changes in their expression lead to widespread chromatin reorganization , which alters both the prognosis and treatment of diseases such as cancer ( Liau et al . , 2017 ) . Epe1 most closely resembles JumonjiC ( JmjC ) domain containing proteins that use Fe ( II ) and α-ketoglutarate as co-factors to catalyze histone demethylation ( Tsukada et al . , 2006 ) . Unlike active histone demethylases , where an HxD/E… . H motif is involved in Fe ( II ) coordination , Epe1 harbors a non-canonical HXE… . Y motif ( Tsukada et al . , 2006 ) . Although Epe1 shares conserved features with other histone demethylases at the amino acid level , there is no biochemical evidence to support the notion that Epe1 has any in vitro enzymatic activity . Epe1 purified from fission yeast cells , or a recombinant source ( insect cells ) , exhibits no H3K9 demethylase activity in vitro ( Tsukada et al . , 2006; Zofall and Grewal , 2006 ) . However , point mutations of amino acid residues involved in Fe ( II ) or α-ketoglutarate binding affect Epe1 activity in cells and lead to heterochromatin spreading beyond normal boundary sequences ( Trewick et al . , 2007 ) . These conflicting lines of biochemical and genetic data have prompted several alternative explanations for how Epe1 might fulfill its anti-silencing role . These models include the possibility that Epe1 acts as a protein hydroxylase which targets non-histone proteins such as Swi6HP1 , regulates the activity of the multi-subunit H3K9 methyltransferase CLRC complex , or functions as an H3K9 demethylase when in complex with Swi6HP1 ( Aygün et al . , 2013; Iglesias et al . , 2018; Trewick et al . , 2007; Zofall and Grewal , 2006 ) . Although these models represent attractive possibilities for how Epe1 regulates heterochromatin , there is no direct evidence to suggest that any of these proteins represent bonafide enzymatic targets . An alternative hypothesis is that Epe1 has a non-enzymatic function that regulates heterochromatin spreading and epigenetic inheritance . In support of this hypothesis , the overexpression of Fe ( II ) and α-ketoglutarate binding mutants of Epe1 suppresses heterochromatin spreading defects observed in epe1Δ strains ( Trewick et al . , 2007; Zofall and Grewal , 2006 ) . Hence , co-factor binding mutants of Epe1 can act as multi-copy suppressors of epigenetic silencing despite the presumptive loss of enzymatic activity . In this study , we discovered that the putative catalytic JmjC domain of Epe1 is , at least in part , dispensable for its anti-silencing function in cells . The C-terminus of Epe1 directly interacts with Swi6HP1 and its interaction in the context of full-length Epe1 is regulated by H3K9 methylation . Expressing the Epe1 C-terminus alone is sufficient to reverse heterochromatin establishment and attenuate epigenetic inheritance . We propose that a cis interaction between the Epe1 N- and C-terminus inhibits Swi6HP1 binding . H3K9 methylation binding attenuates this intramolecular interaction and promotes Swi6HP1 binding . A requirement for H3K9 methylation to stabilize a complex comprising Epe1 and Swi6HP1 restricts their interaction to a heterochromatin-specific context . Our work highlights the versatile , non-canonical ways in which histone demethylases can oppose establishment and maintenance of epigenetic states . JmjC domain-containing proteins require Fe ( II ) and α-ketoglutarate as co-factors to catalyze histone demethylation . Aligning the primary amino acid sequences of active histone demethylases with Epe1 reveals a naturally occurring histidine to tyrosine substitution ( Y370 ) within a conserved triad of amino acid residues that coordinate iron ( Figure 1—figure supplement 1A ) . We tested whether the activity of Epe1 in cells is dependent on this non-conserved tyrosine residue ( Y370 ) . To measure Epe1 activity , we used a reporter gene assay that provides a direct read-out of epigenetic inheritance . In this system , an H3K9 methyltransferase , Clr4Suv39h is fused to a DNA binding protein , TetR . This fusion protein is recruited to an ectopic site where ten Tet operator sites ( 10X TetO ) are placed upstream of a reporter gene , ade6+ ( Figure 1A ) . Establishment in the absence of tetracycline results in the appearance of red colonies . The sequence-specific initiator , TetR-Clr4-I , dissociates in the presence of tetracycline , enabling us to test whether cells can maintain silencing in the absence of continuous initiation . Wild-type cells are initially red in medium not containing tetracycline ( −tetracycline medium ) , indicating that the reporter gene is initially silenced ( establishment ) . Cells that have a functional copy of Epe1 turn white and exhibit no maintenance when plated on +tetracycline-containing medium . The ability of fission yeast cells to autonomously propagate epigenetic silencing is exquisitely sensitive to Epe1 activity . We observed epigenetic maintenance in cells where Epe1 is either deleted or inactivated , resulting in red or sectored colonies on +tetracycline-containing medium ( Audergon et al . , 2015; Ragunathan et al . , 2015 ) . Alanine substitutions of amino acid residues involved in Fe ( II ) or α-ketoglutarate binding ( epe1 H297A and epe1 Y307A , respectively ) disrupt co-factor binding , resulting in the concomitant loss of Epe1 activity . When expressed at endogenous levels , these mutants form red or sectored colonies on +tetracycline-containing medium and resemble epe1Δ cells ( Figure 1B ) . Replacing the non-conserved tyrosine residue in Epe1 with alanine ( epe1 Y370A ) leads to a similar loss of function phenotype . Hence , despite the lack of conservation , a natural tyrosine substitution within the JmjC domain of Epe1 is essential for its anti-silencing function in cells . We used chromatin immunoprecipitation assays followed by qPCR to measure H3K9me2 levels associated with the reporter gene locus before and after tetracycline addition . Both wild-type and Epe1 mutant strains exhibit high levels of H3K9me2 during establishment ( Figure 1C ) . However , wild-type cells lose H3K9me2 approximately 24 h after tetracycline addition . In contrast , Epe1 mutants that exhibit a red or sectored phenotype upon +tetracycline addition retain high levels of H3K9 methylation at the ectopic site ( Figure 1D ) . We verified that the expression level of all Epe1 mutant proteins is equal relative to an actin loading control . Hence , neither overexpression artifacts nor changes in protein stability contribute to the maintenance-specific phenotype we observed in our genetic assays ( Figure 1—figure supplement 1B ) . Epe1 is localized at sites of constitutive heterochromatin through its interactions with Swi6HP1 ( Ayoub et al . , 2003; Zofall and Grewal , 2006 ) . We imaged Epe1 and Swi6HP1 in live fission yeast cells using fluorescent protein fusions . We labeled Epe1 with mNeonGreen and Swi6HP1 with mCherry . This labeling scheme allows Epe1 and Swi6HP1 to be visualized in separate green and red emission channels , respectively . Both fusion proteins were expressed from their endogenous promoters to discount any possible overexpression artifacts . mCherry-Swi6HP1 typically exhibits two or three bright foci in individual cells , corresponding to sites of constitutive heterochromatin ( centromeres and telomeres ) . mNeonGreen-Epe1 co-localizes with mCherry-Swi6HP1 as evidenced by the significant overlap between the bright foci that appear green and red emission channels ( see white arrows ) . The overlay also reveals that clusters of Epe1 and Swi6 are co-localized ( Figure 1E ) . Surprisingly , an Epe1 co-factor binding mutant , mNeonGreen-Epe1 Y307A , fails to co-localize with mCherry-Swi6HP1 . Instead , the mutant protein exhibits a diffuse green signal within the nucleus and a complete lack of nuclear foci that co-localize with Swi6HP1 ( Figure 1F ) . The elevated signal in the cytoplasm could not be attributed to a defect in nuclear localization as cells that do not express any mNeonGreen-Epe1 also exhibit high levels of autofluorescence in the green channel ( Figure 1—figure supplement 1C ) . Hence , in addition to affecting any putative enzymatic functions , a co-factor binding mutation within the JmjC domain of Epe1 eliminates protein localization at sites of constitutive heterochromatin . We hypothesized that the absence of heterochromatin localization in the Epe1 JmjC mutant could reflect a loss of Swi6HP1 binding . We used a co-immunoprecipitation assay to compare the interaction between Epe1 and Swi6HP1 in wild-type and Epe1 mutant cells . We expressed an Epe1-3X FLAG fusion protein at endogenous levels . Using a FLAG antibody , we pulled-down Epe1 and detected its interaction with Swi6HP1 using a primary antibody . Swi6HP1 is enriched in pull-down experiments in wild-type cells relative to an untagged control ( Figure 2A ) . However , mutations in residues that affect Fe ( II ) or α-ketoglutarate binding ( H297A , Y307A , and Y370A ) significantly attenuate this interaction ( Figure 2A ) . Altering the position of the FLAG epitope tag did not alter the conclusions of our experiments . Epe1 fused to an N-terminal FLAG tag interacts with Swi6HP1 , whereas a mutation within the JmjC domain ( H297A ) compromises its binding ( Figure 2—figure supplement 1A ) . Epe1 is enriched at sites of constitutive heterochromatin , which include the pericentromeric dg and dh repeats , the mating type locus , and the telomeres ( Zofall and Grewal , 2006 ) . We used formaldehyde to crosslink cells followed by chromatin immunoprecipitation to compare heterochromatin occupancy differences between Epe1 wild-type and Epe1 co-factor binding mutants . After crosslinking , we used a FLAG antibody to pull-down the chromatin-bound fraction of Epe1 . We used qPCR to measure Epe1 occupancy at the pericentromeric dg repeats . Epe1 is enriched within dg repeats in wild-type cells and this heterochromatin-specific occupancy pattern is disrupted both in swi6Δ and clr4Δ cells ( Figure 2B ) . Mutations within the JmjC domain that disrupt co-factor binding lead to a substantial reduction in Epe1 occupancy at sites of heterochromatin formation . Consistent with our co-immunoprecipitation studies , all co-factor binding mutants of Epe1 exhibit a significant reduction or completely fail to localize at the pericentromeric dg repeats ( Figure 2B ) . We altered our fixation conditions using additional reactive crosslinkers and extended the time for formaldehyde crosslinking . Altering crosslinking conditions did not lead to a significant increase in chromatin occupancy amongst Epe1 mutants ( Figure 2—figure supplement 1B ) . Based on these results , we concluded that Epe1 co-factor binding mutants exhibit significant defects in their ability to interact with Swi6HP1 and a complete inability to localize at sites of heterochromatin formation . Our co-immunoprecipitation , imaging , and ChIP experiments preclude us from making any conclusions as to whether the interaction between Epe1 and Swi6HP1 is mutation-dependent or requires the putative catalytic functions of Epe1 in vivo . To address this concern , we purified MBP fusions of wild-type Epe1 and Epe1 H297A from insect cells . Swi6HP1 was purified from E . coli . We used TEV protease to cleave the MBP tag and confirmed that recombinant Epe1 remains soluble , but preserved the tag in subsequent purifications for our binding assays ( Figure 2—figure supplement 1C ) . We compared the thermal stability of the wild-type and mutant Epe1 protein ( Epe1 H297A ) using isothermal calorimetry measurements ( Figure 2—figure supplement 1D ) . Wild-type Epe1 and Epe1 H297A exhibit similar denaturation temperatures , implying that the mutation within the JmjC domain does not destabilize the protein or cause substantial alterations in protein structure . The difference in peak intensities in the isothermal calorimetry ( ITC ) profile reflects differences in protein amounts in this assay . These results are consistent with structural studies of JmjC domain-containing proteins where the loss of co-factor binding within the active site does not alter protein structure ( Horton et al . , 2011 ) . To perform in vitro binding assays , we immobilized Swi6HP1 on FLAG beads and added three different concentrations of Epe1 . Epe1 was detected in these binding assays using an MBP antibody and the total amount of Swi6HP1 was measured using a FLAG antibody . Through a series of titration measurements , we found that using an MBP antibody and a chemiluminescence based readout produces a very limited linear response , which precludes us from reporting an apparent Kd . We verified that the Epe1 protein we purified from insect cells preferentially interacts with Swi6HP1 as opposed to a second HP1 homolog in S . pombe , Chp2HP1 ( Figure 2C ) . Hence , the recombinant Epe1 protein we purified from insect cells recapitulates a known binding preference of Epe1 towards Swi6HP1 ( Sadaie et al . , 2008 ) . The CSD domain of Swi6HP1 mediates protein dimerization and regulates Swi6HP1-dependent protein-protein interactions ( Canzio et al . , 2013 ) . We expressed and purified a dimerization deficient mutant of Swi6 from E . coli ( 3XFLAG-Swi6HP1 L315E ) . Our binding assays using the mutant Swi6HP1 protein ( Swi6HP1 L315E ) reveal a significant reduction in its ability to interact with Epe1 ( Figure 2D ) . Therefore , Epe1 interacts with Swi6HP1 through a conserved mechanism that is shared across other heterochromatin associated factors . To test whether a mutation within the JmjC domain leads to a loss of interaction between Epe1 and Swi6HP1 , we compared binding assays between the recombinant wild-type Epe1 protein and the Fe ( II ) binding deficient mutant , Epe1 H297A . Adding increasing quantities of the wild-type MBP-Epe1 leads to a corresponding increase in the amount of protein that interacts with Swi6HP1 . However , this type of interaction and increase in binding is not observed in the case of MBP-Epe1 H297A ( Figure 2E ) . We compared binding assays performed in the presence and absence of EDTA to rule out any potential contributions that may arise from divalent metals ions that bind to the JmjC domain . The interaction between Epe1 and Swi6HP1 is nearly identical in the presence or absence of EDTA ( Figure 2E ) . These results suggest that the interaction between Epe1 and Swi6HP1 is direct but disrupted by mutations that map to the putative catalytic JmjC domain . To test whether the enzymatic activity of Epe1 may enhance its interaction with Swi6HP1 , we added Fe ( II ) , α-ketoglutarate and ascorbate to mimic ‘histone-demethylase reaction conditions’ in our in vitro binding assays ( Tsukada and Nakayama , 2010 ) . The addition of co-factors required for histone demethylation did not alter the extent of interaction between Epe1 and Swi6HP1 ( Figure 2—figure supplement 1E ) . Hence , our in vitro assays fail to capture any effect that co-factor binding itself may have on the interaction between Epe1 and Swi6HP1 . We also tested whether the interaction between S . pombe Swi6HP1 and Epe1 might be different from the data we obtained using Swi6HP1 purified from E . coli ( Figure 2—figure supplement 1F ) . Swi6HP1 phosphorylation remains intact during our purification as indicated by an upward mobility shift in S . pombe Swi6HP1 compared to E . coli Swi6HP1 ( Figure 2—figure supplement 1G ) . However , we observed no differences in the interaction profile with Epe1 suggesting that the loss of binding we detected in the Epe1 H297A mutant is independent of Swi6HP1 post-translational modifications ( Shimada et al . , 2009 ) . These results suggest that mutations within the JmjC domain of Epe1 may induce a conformational change that attenuates Swi6HP1 binding ( Sorida et al . , 2019 ) . Next , we tested whether Swi6HP1 binding to Epe1 may activate its latent enzymatic properties . We performed histone demethylase assays using recombinant Epe1 in the presence and absence of Swi6HP1 . We used histone H3 tail peptides with a di-methyl or a tri-methyl modification at the lysine nine position ( H3K9me2 or H3K9me3 peptides ) as substrates . We were unable to detect a mass shift corresponding to the removal of one or more methyl groups in reactions that we performed with Epe1 alone or Epe1 in complex with a five-fold molar excess of Swi6HP1 ( Figure 2—figure supplement 2A ) . In contrast , JMJD2A , an active demethylase , is fully capable of demethylating an H3K9me3 peptide substrate ( Figure 2—figure supplement 2B ) . Hence , Swi6HP1 binding to Epe1 is not sufficient to activate its putative enzymatic functions . To map the Swi6HP1 interaction site within Epe1 , we used an in vitro translation ( IVT ) assay where we expressed fragments of Epe1 and tested their ability to interact with Swi6HP1 . We used a computational disorder prediction program to define ordered and disordered regions within the protein ( Figure 3—figure supplement 1A ) . The JmjC domain emerges as one of two ordered regions extending from amino acids 233–434 . The second ordered domain that is located within the C-terminus of the protein has no known similarity to existing protein structures and does not have any ascribed function . We designed and expressed partial fragments of Epe1 using rabbit reticulocyte lysates including the full-length protein as a positive control . We added FLAG beads that were pre-incubated with 3XFLAG-Swi6HP1 to the IVT extract . A C-terminal fragment of Epe1 spanning 434–948 amino acids and an N-terminal fragment of Epe1 encompassing 1–600 amino acids ( Epe1-ΔC ) emerged as putative Swi6HP1 interaction candidates ( Figure 3—figure supplement 1B ) . We hypothesized that the binding region would lie somewhere between amino acid positions 434 and 600 . Importantly , this region is proximal to but non-overlapping with the predicted JmjC domain of Epe1 . To validate the conclusions of our IVT binding assay , we expressed and purified two C-terminal fragments of Epe1 from E . coli . The first fragment encompasses the entire C-terminus of Epe1 from 434 to 948 amino acids ( Epe1434-948 ) . The second fragment corresponds to only the minimal Swi6HP1 interaction site extending from 434 to 600 amino acids ( Epe1434-600 ) . We performed binding assays comparing the interaction between the full-length Epe1 protein and the Epe1 C-terminal fragment ( Epe1434-948 ) with Swi6HP1 . The Epe1 C-terminal fragment exhibits an increase in its interaction with Swi6HP1 relative to full-length Epe1 ( Figure 3A ) . Hence , the C-terminal domain of Epe1 , when placed in the context of the full-length protein , is less accessible to Swi6HP1 . We obtained similar results when we tested the interaction between Epe1434-600 and Swi6HP1 ( Figure 3—figure supplement 1C ) . These observations raise the possibility that the JmjC domain has a steric function and its presence in the context of the full-length protein may impede Swi6HP1 binding . We also performed co-immunoprecipitation experiments in cells expressing 3XFLAG-Epe1434-948 and detected the same pattern of interaction with Swi6HP1 as measured in our in vitro assay ( Figure 3—figure supplement 1D ) . We noted that the expression level of the 3X FLAG-Epe1434-948 protein expressed from the endogenous epe1 locus is at least four-fold lower compared to the full-length protein ( Figure 3—figure supplement 1E ) . Although the Swi6HP1 binding site lies outside the confines of the JmjC domain of Epe1 , point mutations within the putative catalytic JmjC domain perturb a direct interaction between the two proteins ( Figure 2A ) . We hypothesized the existence of an interaction in cis where the N-terminus half of the protein ( Epe1-N , 1–434 amino acids ) interacts with its C-terminal portion ( Epe1-C , 434–948 amino acids ) to interrupt Swi6HP1 binding . To test this model , we expressed and purified the N-terminal half of Epe1 containing the JmjC domain ( 1–434 amino acids ) fused to a 3X-FLAG epitope tag in fission yeast cells ( 3XFLAG-Epe1-N ) . The purified protein was retained on beads without elution and immediately used for subsequent binding assays . S . pombe cells express limiting amounts of the Epe1 N-terminal fragment , which are sufficient for the binding assays described here . We subsequently added a defined amount of the recombinant C-terminal Epe1 fragment ( MBP-Epe1434-948 ) . We detected a direct interaction between the Epe1-N and Epe1-C terminal fragments ( Figure 3B , lane 1 ) . Next , we supplemented our binding assays with a two-fold molar excess of recombinant Swi6HP1 relative to MBP-Epe1434-948 . The addition of recombinant Swi6HP1 is sufficient to compete with and disrupt a trans interaction between the Epe1-N and Epe1-C fragments ( Figure 3B , lane 2 ) . We also performed experiments where we used lysates derived from swi6+ or swi6Δ cells expressing an Epe1-N fragment ( wild-type or Fe ( II ) binding mutant allele , Epe1 H297A ) . We incubated these lysates with a recombinant C-terminal Epe1 fragment , MBP-Epe1434-948 immobilized on an amylose resin . Compared to bead-only controls where the amylose resin was incubated with cell lysates , we discovered that the C-terminal Epe1 fragment , MBP-Epe1434-948 is able to interact with and pull-down Epe1-N ( 1-434 ) in trans from a complex mixture of proteins ( Figure 3—figure supplement 1F ) . We also observed this trans interaction in lysates derived from swi6Δ cells further supporting the notion that the interaction between the N and C terminal halves of Epe1 is direct and not mediated by Swi6HP1 ( Figure 3—figure supplement 1G ) . One prediction emerging from our biochemical analyses is that expressing the Epe1 C-terminus ( Epe1434-948 ) alone might oppose heterochromatin assembly through its direct interaction with Swi6HP1 . As previously described , we used a reporter gene assay where a TetR-Clr4-I fusion protein initiates heterochromatin establishment in an inducible manner . We expressed Epe1434-948 protein in this reporter strain . Surprisingly , this mutant protein , which completely lacks the JmjC domain , can reverse epigenetic maintenance . This reversal leads to a greater proportion of cells that turn white upon +tetracycline addition ( Figure 3C ) . The expression of the Epe1 C-terminal fragment , Epe1434-948 results in a phenotype that is substantially different compared to Epe1 co-factor binding mutants or epe1Δ cells . We quantified the number of red or sectored colonies in the Epe1 JmjCΔ mutant compared to epe1Δ and wild-type cells . Cells that express a full-length wild-type copy of Epe1 turn white on +tetracycline medium and show no trace of red , pink , or sectored colonies . We observed a three-fold reduction in the number of red or sectored colonies in the Epe1 JmjCΔ mutant compared to Epe1 null cells ( Figure 3D ) . Therefore , the Epe1434-948 mutant is a hypomorphic allele that partially retains wild-type levels of Epe1 anti-silencing activity . The non-enzymatic function of Epe1 revolves around its dominant mode of interaction with Swi6HP1 . Therefore , we devised a new ectopic silencing approach where the TetR DNA binding protein was fused to the CSD domain of Swi6HP1 . We hypothesized that the ability of Epe1 to impede heterochromatin establishment might be more pronounced in this reporter strain . We expressed TetR-Swi6-CSD protein in cells where 10X TetO repeats were placed upstream of an ade6+ reporter . Tethering TetR-Swi6-CSD in wild-type cells fails to establish epigenetic silencing . Cells remain white in the presence or absence of +tetracycline ( Figure 3—figure supplement 2A ) . We discovered that Epe1 is the rate-limiting factor that prevents heterochromatin establishment in this ectopic paradigm . Cells turn red and establish epigenetic silencing upon deleting epe1Δ . Furthermore , cells remain red even after the addition of +tetracycline consistent with robust sequence-independent epigenetic inheritance in this mutant background ( Figure 3—figure supplement 2A ) . Next , we expressed only the C-terminal fragment of Epe1 ( Epe1434-948 ) . Remarkably , this C-terminal fragment , which is devoid of the JmjC domain , completely blocks heterochromatin establishment . Cells remain white in the presence or absence of +tetracycline ( Figure 3E ) . We occasionally observed clonal populations that exhibit red colonies ( 25% of transformants ) consistent with the notion that other domains within Epe1 exert additional enzymatic or non-enzymatic functions ( Bao et al . , 2019; Sorida et al . , 2019 ) . To test whether the phenotypes we observed depend on H3K9 methylation , we used chromatin immunoprecipitation assays followed by qPCR to measure H3K9me2 levels associated with the ade6+ reporter gene before and after tetracycline addition . H3K9me2 was observed in cells that turned red in an epe1Δ background in the presence or absence of tetracycline ( Figure 3F , G ) . Wild-type cells and cells expressing the Epe1-C fragment fail to exhibit any significant enrichment in H3K9me2 . In addition , epe1Δ clr4Δ cells exhibit a complete loss of epigenetic silencing , further supporting our observations that H3K9 methylation has a causal role in TetR-Swi6-CSD initiated silencing ( Figure 3—figure supplement 2B ) . We speculated that a heterochromatin-specific trigger might regulate the interaction between the N- and C-terminus of Epe1 to stabilize its interaction with Swi6HP1 . We expressed Epe1 fused to a 3XFLAG epitope tag in cells that lack the H3K9 methyltransferase , Clr4Suv39h or strains where the wild-type H3 allele is replaced with an H3K9R mutant . We performed a co-immunoprecipitation assay where we pull-down Epe1 with a FLAG antibody and measured its interaction with Swi6HP1 . Although Epe1 interacts with Swi6HP1 in wild-type cells , this interaction is obliterated in both of the H3K9 methylation deficient mutant strains , clr4Δ and H3K9R mutants ( Figure 4A ) . Furthermore , deleting histone deacetylases Sir2 or Clr3 , both of which affect heterochromatin formation and Swi6HP1 localization , also resulted in a substantial decrease in the interaction between Epe1 and Swi6HP1 ( Figure 4B ) . In contrast , deleting Mst2 , a histone acetyltransferase that enhances heterochromatin formation leads to no change in the interaction pattern between Epe1 and Swi6HP1 ( Figure 4—figure supplement 1A; Reddy et al . , 2011 ) . Hence , our results indicate that H3K9 methylation and functional heterochromatin are pre-requisites for complex formation between Epe1 and Swi6HP1 in cells . We reconstituted a requirement for H3K9 methylation in stabilizing the interaction between Epe1 and Swi6HP1 using binding assays as previously described . We supplemented our in vitro binding assays with an unmethylated histone H3 peptide ( H3K9me0 , H3 1–15 amino acids ) or an H3K9 tri-methylated peptide ( H3K9me3 , H3 1–15 amino acids ) . Compared to reactions where no-peptide ( lanes 1–3 ) or an unmethylated H3 peptide was added ( lanes 4–6 ) , we observed a substantial increase in the interaction between Epe1 and Swi6HP1 specifically in the presence of an H3K9me3 peptide ( lanes 7–9 ) ( Figure 4C ) . An H3K9me2 peptide was also capable of stimulating the interaction between Epe1 and Swi6HP1 compared to reactions where no peptide was added or assays where an unmodified peptide was used ( Figure 4—figure supplement 1B ) . To test whether the stimulation in the interaction between Epe1 and Swi6HP1 is specific to H3K9 methylation , we carried out binding assays in the presence of an H3K4 tri-methylated peptide ( H3K4me3 ) . The addition of an H3K4me3 peptide fails to enhance complex formation between Epe1 and Swi6HP1 , unlike the significant enhancement in binding we observed upon addition of an H3K9me3 peptide ( Figure 4D ) . Hence , the stimulatory effect we observed in our binding assays is specific to either H3K9me2 or H3K9me3 peptides . Our previous results reveal a severe reduction in the interaction between Epe1 H297A and Swi6HP1 compared to wild-type Epe1 ( Figure 2E ) . We compared binding assays between MBP-Epe1 and MBP-Epe1 H297A with Swi6HP1 in the presence of an H3K9me3 peptide or an H3K9me0 peptide . Although we observed a strong stimulation in the interaction between wild-type Epe1 and Swi6HP1 ( compare lanes 1–3 with 7–9 ) , this stimulatory effect was significantly reduced in the Epe1 H297A mutant ( compare lanes 4–6 with 10–12 ) ( Figure 4—figure supplement 1C ) . These results suggest that the JmjC mutant of Epe1 remains refractory to any interaction with Swi6HP1 in the presence or absence of H3K9 methylation . One possibility is that Swi6HP1 undergoes a conformational change that reverses auto-inhibition upon interaction with an H3K9me3 peptide ( Canzio et al . , 2013 ) . To test this hypothesis , we purified a chromodomain mutant Swi6HP1 protein from E . coli . A tryptophan to alanine substitution ( W104A ) within the Swi6HP1 chromodomain causes a significant reduction in H3K9 methylation binding ( Jacobs and Khorasanizadeh , 2002 ) . We used peptide binding assays to confirm that the Swi6HP1 W104A mutant indeed exhibits a substantial defect in H3K9me3 peptide binding in comparison to the intact Swi6HP1 protein ( Figure 4—figure supplement 1D ) . We tested whether Swi6HP1 W104A protein binding to Epe1 can also be stimulated in the presence of an H3K9me3 peptide . We added increasing amounts of MBP-Epe1 while maintaining a fixed amount of 3X-FLAG Swi6HP1 ( W104A ) on beads . Despite Swi6HP1 being unable to bind to an H3K9me3 tail peptide , we observed a stimulation in the interaction between Epe1 and the Swi6HP1 ( W104A ) ( Figure 4E ) . In addition , we purified a Swi6HP1 Loop-X mutant that abolishes auto-inhibition and prevents chromodomain-dependent dimerization ( Canzio et al . , 2013 ) . We tested whether the wild-type Epe1 protein and a Swi6HP1 Loop-X mutant that is constitutively released from auto-inhibition is sensitive to the presence of an H3K9me3 peptide . We purified 3X-FLAG Swi6HP1 Loop-X mutant from E . coli and immobilized the protein on FLAG beads . We added increasing amounts of Epe1 in the presence of an H3K9me0 or an H3K9me3 peptide . Although the H3K9me0 peptide has no effect on the interaction between Epe1 and Swi6HP1 Loop-X , the presence of an H3K9me3 peptide substantially enhances their interaction . Therefore , Epe1 remains responsive to the presence of an H3K9 methylated peptide even in a context where Swi6HP1 is constitutively released from auto-inhibition ( Figure 4—figure supplement 1E ) . Our previous results suggest that Epe1 might have a latent capacity to bind to H3K9 methylated peptides or histones . We tested whether Epe1 directly binds to an H3K9me3 peptide and specifically interacts with H3K9 methylated histones . We performed peptide binding assays where a biotinylated H3K9me3 peptide was immobilized on streptavidin beads . Our binding assays detect a direct interaction between Epe1 and an H3K9me3 peptide as opposed to an unmodified H3K9me0 peptide . Furthermore , Epe1 selectively interacts with H3K9 methylated histones as opposed to H3K4 methylated histones . ( Figure 5A , B ) . Next , we expressed and purified a C-terminal truncation mutant of Epe1 , MBP-Epe1-ΔC from Sf9 insect cells , which includes amino acids 1–600 and includes the putative catalytic JmjC domain . We found that Epe1-ΔC can also directly bind to an H3K9me3 peptide and specifically interacts with H3K9 methylated histones ( Figure 5—figure supplement 1A , B ) . Based on these observations , we hypothesize that the JmjC domain of Epe1 ( amino acids 233–434 ) might be primarily responsible for H3K9 methylation recognition and binding . We previously demonstrated that the addition of Swi6HP1 disrupts an interaction in trans between the Epe1-N and Epe1-C terminus ( Figure 3B ) . The stimulation in Swi6HP1 binding that we observed prompted us to test whether Epe1 binding to an H3K9 methylated peptide might also have a similar function and interrupt a trans interaction between the Epe1 N- and C- terminal fragments . Disrupting their interaction would enable Swi6HP1 to gain access to the C-terminus of Epe1 . To test this model , we measured a trans interaction between the Epe1 N- and C-terminal halves in the presence of an H3K9me0 peptide or an H3K9me3 peptide ( Figure 5C ) . We purified an Epe1-N fragment fused to a 3X FLAG epitope tag . Next , we added a recombinant Epe1 C-terminal fragment ( 434–948 amino acids ) in the presence of an H3K9me0 peptide or an H3K9me3 peptide . We observed an interaction in trans between the N- and C-terminal halves of Epe1 in binding assays with no peptide or an H3K9me0 peptide . However , the addition of an H3K9me3 peptide eliminates the interaction between the N- and C-terminal fragments of Epe1 ( Figure 5C ) . Based on our observations , we hypothesized that the C-terminus of Epe1 might have a regulatory function in enforcing an H3K9 methylation-dependent mode of interaction between Epe1 and Swi6HP1 . To test this model , we expressed a C-terminal truncation of Epe1 ( Epe1-ΔC ) fused to a 3XFLAG epitope tag in fission yeast cells . We performed a co-IP experiment to test the interaction between Epe1-ΔC , and Swi6HP1 compared to the full-length protein . We also expressed an N-terminal fragment of Epe1 ( Epe1-N ) which lacks the Swi6HP1 binding site . Our co-IP assays detect a substantial increase in the interaction between Epe1-ΔC and Swi6HP1 , compared to a weak interaction in the case of the full-length protein and no interaction in the case of Epe1-N ( Figure 5—figure supplement 1C ) . Next , we performed in vitro binding assays to test the extent of interaction between recombinant MBP-Epe1-ΔC and FLAG-Swi6HP1 relative to the full-length Epe1 protein . Epe1-ΔC significantly outperforms the full-length protein in its ability to interact with Swi6HP1 ( compare lanes 4–6 with lanes 1–3 ) ( Figure 5D ) . We performed binding assays where we added increasing amounts of Epe1-ΔC while maintaining a fixed concentration of Swi6HP1 on beads in the presence of an H3K9me0 or an H3K9me3 peptide . Surprisingly , the addition of a modified peptide had no effect on the interaction between Swi6HP1 and Epe1-ΔC ( Figure 5E ) . This is not because of saturation of the chemiluminescent signal as the addition of increasing amounts of Epe1-ΔC continues to produce a concomitant increase in its interaction with Swi6HP1 . Previously , we demonstrated that the interaction between full-length Epe1 and Swi6HP1 requires functional heterochromatin . We performed a co-IP measurement in clr4+ and clr4Δ cells expressing FLAG-Epe1-ΔC . Consistent with our in vitro binding assays and unlike the full-length Epe1 protein , we observed an interaction between Epe1-ΔC and Swi6HP1 in both clr4+ and clr4Δ cells ( Figure 5—figure supplement 1D ) . Therefore , Epe1-ΔC remains unresponsive to the presence of H3K9 methylation and does not undergo any further stimulation in its ability to interact with Swi6HP1 . We hypothesized that Epe1 might outcompete other heterochromatin associated proteins that localize to sites of heterochromatin formation through a dominant interaction with Swi6HP1 . Genetic studies reveal that Epe1 and a histone deacetylase Clr3 have opposing effects on nucleosome turnover ( Aygün et al . , 2013 ) . One possibility is that the interaction between Epe1 and Swi6HP1 excludes heterochromatin agonists , such as Clr3 from sites of heterochromatin formation . We used a co-immunoprecipitation assay to measure the extent of interaction between Clr3 and Swi6HP1 . Clr3 was fused to a 3X V5 epitope tag and expressed in a wild-type Epe1 and an Epe1 H297A background . We used a V5 antibody to pull-down Clr3 , after which we detected Swi6HP1 using a primary antibody . We measured a weak interaction between Swi6HP1 and Clr3 in a wild-type Epe1 background that substantially increases in an Epe1 H297A strain background ( Figure 6A ) . This positive change in the interaction between Clr3 and Swi6HP1 occurs in the absence of any increase in Swi6HP1 occupancy at the pericentromeric repeats ( Figure 6B ) . Although Clr3 is also recruited to sites of heterochromatin formation by interacting with Chp2HP1 , the levels of Chp2HP1 in fission yeast cells are approximately 100-fold lower compared to Swi6HP1 ( Sadaie et al . , 2008 ) . This difference in stoichiometry between Swi6HP1 and Chp2HP1 could explain how Epe1 may have an outsized role in interfering with heterochromatin assembly by selectively disrupting Swi6HP1 associated protein complexes . We also measured Epe1 localization in clr3Δ mutant cells . Consistent with our model , we observed a three-fold increase in Epe1 localization at the pericentromeric dg repeats and the mating type locus ( mat ) ( Figure 6C , D ) . In contrast , the deletionof clr3Δ results in the complete loss of Epe1 localization at the telomeres ( Figure 6E ) . We hypothesized that directly tethering Clr3 at an ectopic site would eliminate the requirement for Swi6HP1 for its recruitment at sites of heterochromatin formation . In this genetic context , we rationalized that Epe1 would be unlikely to block Clr3 recruitment . One caveat of this experimental strategy is that it does not rule out the possibility that Epe1 and Clr3 have antagonistic catalytic functions . However , this experiment directly tests whether the Clr3 level itself might be rate-limiting at sites of heterochromatin formation . To test this model , we engineered synthetic heterochromatin domains where 10X Tet operator ( 10XTetO ) and 10X Gal4 DNA binding sites ( 10X Gal4 ) are placed next to each other and inserted upstream of an ade6+ reporter gene . This enables inducible heterochromatin formation via TetR-Clr4-I recruitment while the Gal4 DNA binding domain allows for the orthogonal recruitment of additional chromatin effectors to the ectopic site ( Figure 7A ) . In the absence of any additional chromatin modifiers , TetR-Clr4-I recruitment to the newly engineered ectopic site ( 10X Gal4-10X TetO- ade6+ ) leads to the appearance of red colonies on –tetracycline medium . Upon the addition of tetracycline , cells turn white consistent with a causal role for Epe1 in resetting epigenetic inheritance ( Figure 1B ) . We fused the Gal4 DNA binding domain ( Gal4 DBD ) to two histone deacetylases , Clr3 ( a class II histone deacetylase ) and Sir2 ( a NAD-dependent histone deacetylase ) enabling these proteins to be constitutively tethered at the Gal4 DNA binding sites . The TetR-Clr4-I fusion initiates heterochromatin formation and cells turn red on medium lacking tetracycline in the presence of Gal4-Clr3 or Gal4-Sir2 ( Figure 7A ) . However , upon exposure to tetracycline , the constitutive tethering of Clr3 but not Sir2 promotes epigenetic inheritance . Cells exhibit a red and sectored phenotype in cells where Clr3 is artificially tethered , despite Epe1 still being present ( Figure 7B ) . The phenotypes in cells where Gal4-Clr3 is tethered are remarkably similar to cells that lack Epe1 ( epe1Δ ) . We measured changes in H3K9me2 levels associated with the ade6+ reporter gene using ChIP-qPCR . Cells maintain high H3K9me2 levels in the presence of Gal4-Clr3 before and after tetracycline ( Figure 7C-D ) . These results suggest that constitutively tethering Clr3 to sites of heterochromatin formation is sufficient to oppose Epe1 activity resulting in maintenance of H3K9 methylation even after +tetracycline addition . Tethering Gal4-Clr3 in the absence of TetR-Clr4-I causes no change in reporter gene silencing or H3K9me2 levels at the ectopic site ( Figure 7—figure supplement 1A , B , C ) . Hence , Gal4-Clr3 cannot initiate H3K9 methylation de novo in fission yeast ( Figure 7—figure supplement 1B , C ) . This lack of de novo silencing is consistent with the notion that HDAC proteins in fission yeast collaborate with H3K9 methyltransferases to establish epigenetic silencing . Furthermore , expressing Clr3 minus the Gal4 DBD fusion ( Clr3 ΔGal4 ) leads to a loss of epigenetic maintenance on +tetracycline-containing medium . Therefore , Clr3 must be recruited in cis to oppose the anti-silencing effects of Epe1 ( Figure 7—figure supplement 1A ) . Consistent with the phenotypes that we observed , H3K9me2 levels are high during establishment but completely absent during maintenance in cells expressing diffusible Clr3 protein ( Figure 7—figure supplement 1B , C ) . Hence , it is the sequence-specific recruitment of Clr3 , rather than protein dosage , that facilitates H3K9 methylation maintenance . Therefore , Clr3 recruitment in cis is required to maintain silent epigenetic states and oppose Epe1 activity . This property of heterochromatin maintenance is RNAi independent as cells continue to exhibit a red or sectored appearance in a Dicer deficient background ( dcr1Δ ) ( Figure 7—figure supplement 1D ) . The read-write activity of Clr4Suv39h is essential for the inheritance of silent epigenetic states in a sequence-independent manner ( Audergon et al . , 2015; Ragunathan et al . , 2015 ) . This H3K9 methylation-dependent positive feedback loop is disrupted in a Clr4Suv39h chromodomain mutant ( Zhang et al . , 2008 ) . To test whether the chromodomain is essential for maintenance when Clr3 is tethered , we replaced the wild-type allele of Clr4Suv39 with a Clr4 mutant that lacks the chromodomain ( clr4ΔCD ) . Cells that are initially red in –tetracycline medium turn white on +tetracycline medium in a clr4ΔCD expressing mutant ( Figure 7E ) . H3K9me2 levels in clr4ΔCD mutants are similar to those of wild-type cells during establishment . However , H3K9 methylation is absent upon +tetracycline addition in clr4ΔCD expressing strains . Hence , the inheritance of H3K9 methylation depends on the read-write activity of Clr4Suv39h despite Clr3 being constitutively tethered ( Figure 7F , G ) . The establishment and maintenance of epigenetic states is primarily thought to depend on a balance of enzymatic activities between readers , writers , and erasers of histone modifications ( Allis and Jenuwein , 2016 ) . However , the replication-dependent and independent turnover of histones serve as a major mechanism that shapes genome-wide patterns of histone methylation ( Chory et al . , 2019 ) . In Drosophila , modified histones are turned over more than once during each cell-cycle , which limits their capacity to serve as carriers of epigenetic information ( Coleman and Struhl , 2017; Deal et al . , 2010; Laprell et al . , 2017 ) . In principle , passive genome-wide nucleosome exchange can compete with histone modification-dependent read-write mechanisms to oppose epigenetic inheritance . Our data support the notion that enzymatic erasure is at least in part , dispensable for regulating the inheritance of transcriptionally silent epigenetic states in fission yeast . Our observations are fully consistent with earlier work demonstrating an antagonistic relationship between Epe1 , which enhances nucleosome turnover at sites of heterochromatin formation , and Clr3 , which suppresses this process ( Aygün et al . , 2013 ) . In addition , Epe1 also has newly defined roles in recruiting members of the SAGA complex which mediate histone acetylation , a chromatin feature that is also associated with increased nucleosome turnover ( Bao et al . , 2019 ) . Recent work demonstrated a surprising and unexpected function for the N-terminus of Epe1 in transcriptional activation , which suppresses stochastic , epigenetic silencing ( Sorida et al . , 2019 ) . Our findings are complementary to these recent studies and expand the gamut of anti-silencing functions associated with Epe1 . Auto-inhibition is a widely used strategy that proteins use to activate their latent functions in response to specific cellular or environmental signals ( Pufall and Graves , 2002 ) . The enzymatic and non-enzymatic properties of proteins can be regulated by internal control mechanisms that act in cis . For example , in the case of the chromatin remodeler ALC1 , poly ADP ribosylation in response to DNA damage releases the macrodomain of the protein from an auto-inhibited state and activates its latent ATPase-dependent nucleosome remodeling function ( Lehmann et al . , 2017; Singh et al . , 2017 ) . The non-catalytic heterochromatin associated protein Swi6HP1 is auto-inhibited by a histone H3 mimic sequence ( Canzio et al . , 2013 ) . Release from auto-inhibition switches the protein to a spreading competent state . We discovered that a weak interaction between Epe1 and Swi6HP1 can be bolstered by H3K9 methylation . We propose that a cis interaction between the N- and C-terminal halves of Epe1 opposes Swi6HP1 binding . These results suggest that Epe1 may be preserved in an auto-inhibited state until it interacts with H3K9 methylation ( Figure 8 ) . In the absence of H3K9 methylation , the two proteins fail to form a stable complex in vivo despite their ability to interact directly with each other in vitro . The ability of Epe1 to bind to H3K9 methylation raises the possibility that it competes with Swi6HP1 for a shared binding site . However , heterochromatin consists of dense networks of H3K9 methylated nucleosomes . It is also possible that Epe1 while interacting with Swi6HP1 , can also be stimulated via an interaction with an adjacent or a distal nucleosome . Our data do not lend direct support to a structural change that occurs within Epe1 upon H3K9me3 peptide binding but strongly suggest that modified histones can influence the stability of chromatin associated complexes in living cells . Swi6HP1 is dynamic and undergoes rapid exchange on the millisecond timescale between the free and H3K9 methylation-bound state ( Cheutin et al . , 2004; Cheutin et al . , 2003 ) . Given the relative abundance of Swi6HP1 in cells , a simple protein-protein interaction between Epe1 and Swi6HP1 would effectively titrate Epe1 away from sites of heterochromatin formation . By enforcing an H3K9 methylation-dependent mode of interaction , Epe1 selectively targets a sub-population of Swi6HP1 molecules that are bound to sites of H3K9 methylation and have a causal role in heterochromatin formation . Therefore , one outcome of this H3K9 methylation-dependent mode of interaction is that Epe1 selectively interacts with heterochromatin-bound Swi6HP1 molecules as opposed to freely diffusing Swi6HP1 proteins . It is likely that this balance of protein-protein interactions is tunable via post-translational modifications . For example , Swi6HP1 phosphorylation compromises Epe1 binding and promotes histone deacetylase recruitment to sites of heterochromatin formation ( Shimada et al . , 2009 ) . Our studies demonstrate that altering the balance of Swi6HP1-dependent protein-protein interactions profoundly affects the stability and heritability of silent epigenetic states . Tethering Clr3 at an ectopic site renders heterochromatin refractory to the anti-silencing effects of Epe1 . Our observations are in part , similar to previous findings relating to the fission yeast mating type locus where DNA binding proteins , Atf1 and Pcr1 recruit Clr3 to maintain epigenetic silencing following an RNAi-dependent initiation mechanism ( Wang and Moazed , 2017; Yamada et al . , 2005 ) . We also recapitulate a unique requirement for the sequence-dependent recruitment of HDAC proteins in cis to facilitate the inheritance of silent epigenetic states despite the presence of anti-silencing factors such as Epe1 . We favor a model where histone deacetylases such as Clr3 create a chromatin environment that promotes the read-write enzymatic function of Clr4Suv39h . In essence , Epe1 prevents Clr3 mediated histone hypoacetylation by forming an inhibitory complex with Swi6HP1 within a heterochromatin restricted genomic context . The question of whether Epe1 harbors any histone demethylase activity remains unanswered . Inspired by earlier studies on the Drosophila dKDM4A protein where HP1 binding stimulates demethylase activity , we attempted to reconstitute the enzymatic function of Epe1 in the presence of Swi6HP1 ( Lin et al . , 2008 ) . However , our studies are unable to detect Epe1 mediated enzymatic demethylation in the presence or absence of Swi6HP1 ( Figure 2—figure supplement 2A ) . It is noteworthy that a histone demethylase like protein in Neurospora , DMM-1 shares many similarities with Epe1 and also surprisingly lacks any in vitro enzymatic activity ( Honda et al . , 2010 ) . DMM-1 also interacts with the HP1 homolog in Neurospora leading us to speculate that this surprising mode of heterochromatin regulation we uncovered in the case of Epe1 might also extend to other fungal systems . Given that expressing the C-terminus of Epe1 only partially restores wild-type levels of Epe1 activity ( Figure 3C , D ) , we speculate that there could potentially be unique substrates or conditions required to reconstitute the enzymatic activity of Epe1 . We favor a model in which heterochromatin associated proteins could act as positive or negative allosteric regulators of Epe1 . For example , Bdf2 , a BRD4 homolog that localizes to heterochromatin boundaries along with Epe1 , is a possible candidate that may activate the latent enzymatic functions of Epe1 as H3K9 methylation levels are negligible at the heterochromatin-euchromatin boundary ( Wang et al . , 2013 ) . Our observations add to an expanding list of proteins that mimic histone modifiers but have built-in non-enzymatic functions that regulate the establishment and maintenance of epigenetic states . The Drosophila protein dKDM4A is a prominent example of a histone demethylase with enzymatic and non-enzymatic functions ( Colmenares et al . , 2017 ) . More extreme examples of histone-modifying enzyme mimicry are observed in the case of JARID2 , a subunit of the PRC2 complex that shares several characteristics with JmjC domain-containing proteins but has a structural role in regulating PRC2 complex assembly ( Kasinath et al . , 2018; Son et al . , 2013 ) . Proteins involved in the Arabidopsis RNA-dependent DNA methylation pathway , SUVH2 and SUVH9 , resemble SET domain methyltransferases but lack enzymatic activity . Instead , both proteins recognize methylated DNA and are involved in RNA pol V recruitment in plants to establish epigenetic silencing ( Johnson et al . , 2008; Law et al . , 2013 ) . Although Epe1 has a JmjC domain , it is possible that its putative catalytic function has been repurposed to regulate its interaction with Swi6HP1 . Our data suggest that H3K9 methylation promotes the cooperative assembly of complexes between Epe1 and Swi6HP1 at sites of heterochromatin formation . We speculate that the enzymatic and non-enzymatic functions of Epe1 are likely to oppose heterochromatin assembly on different timescales . As a first approximation , the stabilization of Epe1 at sites of heterochromatin formation is sufficient to displace histone deacetylases and hasten the loss of transcriptional silencing . Hence , the non-enzymatic function of Epe1 serves as the first line of opposition to heterochromatin establishment . This transient inhibition of heterochromatin assembly is likely to be followed by a slower enzymatic step where Epe1 might demethylate H3K9 methylated histones or activate transcription ( Sorida et al . , 2019 ) . Our work provides a biochemical basis for how chromatin associated factors that are thought to purely function as enzymes have non-enzymatic properties that regulate heterochromatin assembly and epigenetic inheritance . Plasmids containing Epe1 wild-type and point mutants were constructed by modifying existing pFA6a C-terminal tagging plasmids . Point mutations were introduced by designing primers using guidelines described in Quick Change mutagenesis protocols . A ligation independent cloning approach was used to construct pFastBac vectors containing wild-type Epe1 and Epe1 H297A mutant for recombinant protein expression and also for other MBP fusion constructs for E . coli expression . 3X FLAG Swi6HP1 and 3X FLAG Chp2HP1 were cloned into existing pGEX vectors downstream of the Prescission protease cleavage site using Gibson assembly . The construction of the 10X gal4-10X tetO-ade6+ plasmid involved modifying plasmids containing a 10X tetO sequence and sub-cloning Gal4 UAS sequences derived from a Drosophila pVALIUM 10X UAS vector . Vectors containing Gal4-Clr3 or Gal4-Sir2 were made using a modified pDual vector with an nmt1 promoter that enables facile integration of DNA sequences at the leu1 locus in fission yeast ( Matsuyama et al . , 2004 ) . Further details regarding plasmid construction are readily available upon request . All strains were constructed using a PCR-based gene targeting approach ( Bähler et al . , 1998 ) . In cases where we generated point mutations of epe1 , we reintroduced the full length wild-type or mutant gene in epe1Δ strains . All strains were genotyped using colony PCR assays . We subsequently verified protein expression using western blots for each of the mutant strains . Strains with 10X gal4-10X tetO-ade6+ were constructed using a 5-Fluoroorotic Acid ( FOA ) selection strategy based on disrupting the endogenous ura4 locus . Strains with Gal4-Clr3 or Gal4-Sir2 were made by digesting pDual vectors with a Not1 restriction enzyme followed by transformations and -LEU based selection . Other deletions of heterochromatin associated factors were achieved either by PCR-based gene targeting approaches or by a cross followed by random spore analysis and PCR based screening to select for colonies that harbored the reporter gene . All strains used in this study are listed in Supplementary file 1 , Table S1 . Further details regarding strain construction are available upon request . 1 . 5 L of fission yeast cells cells were grown in YEA medium at 32°C to an OD600 = 3 . 5 and harvested by centrifugation . The cell pellets were washed with 10 ml TBS pH 7 . 5 , re-suspended in 1 . 5 ml lysis buffer ( 30 mM HEPES pH 7 . 5 , 100 mM NaCl , 0 . 25% Triton X-100 , 5 mM MgCl2 , 1 mM DTT ) , and the cell suspension was snap-frozen into liquid nitrogen to form yeast ‘balls’ and cryogenically ground using a SPEX 6875D Freezer/Mill . The frozen cell powder was thawed at room temperature and re-suspended in an additional 10 ml of lysis buffer with protease inhibitor cocktail and 1 mM PMSF . The cell lysates were subjected to two rounds of centrifugation at 18000 rpm for 5 and 30 mins in a JA-25 . 50 rotor ( Beckman ) . Bradford assay was used to normalize protein levels for co-immunoprecipitation and immunoblot analysis . Protein G Magnetic Beads were pre-incubated with antibody for 4 h and crosslinked with 10 volumes of crosslinking buffer containing 20 mM DMP ( 3 mg DMP/ml of 0 . 2 M Boric Acid pH 9 ) for 30 min at room temperature by rotating . Crosslinking was quenched by washing twice and incubated with 0 . 2 M ethanolamine pH 8 for 2 h at room temperature by rotating . The cell lysates were then incubated with antibody crosslinked beads for 3 h at 4°C . Beads were washed three times in 1 ml lysis buffer for 5 mins each , then eluted with 500 μl of 10 mM ammonium hydroxide . The ammonium hydroxide was evaporated using speed vac ( SPC-100H ) for 5 h and re-suspended in SDS sample buffer . Samples were resolved on SDS–polyacrylamide gel electrophoresis ( SDS-PAGE ) and transferred to PVDF membranes . Immunoblotting was performed by blocking PVDF membrane in Tris-buffered saline ( TBS ) pH 7 . 5 with 0 . 1% Tween-20 ( TBST ) containing 5% non-fat dry milk and subsequently probed with desired primary antibodies and secondary antibodies . Blots were developed by enhanced chemiluminescence ( ECL ) method and detected with Bio-Rad ChemiDoc Imaging System . All co-IP experiments were reproduced N = 2 . Cells were grown till late log phase ( OD 600–1 . 3 to 1 . 8 ) in yeast extract supplemented with adenine ( YEA ) or YEA containing tetracycline ( 2 . 5 μg/ml ) medium and fixed with 1% formaldehyde for 15 min at room temperature ( RT ) . 130 mM glycine was then added to quench the reaction and incubated for 5 min at RT . The cells were harvested by centrifugation , and washed twice with TBS ( 50 mM Tris , pH 7 . 6 , 500 mM NaCl ) . Cell pellets were resuspended in 300 μl lysis buffer ( 50 mM HEPES-KOH , pH 7 . 5 , 100 mM NaCl , 1 mM EDTA , 1% Triton X-100 , 0 . 1% SDS , and protease inhibitors ) to which 500 μl 0 . 5 mm glass beads were added and cell lysis was carried out by bead beating using Omni Bead Ruptor at 3000 rpm × 30 s × 10 cycles . Tubes were punctured and the flow-through was collected in a new tube by centrifugation which was subjected to sonication to obtain fragment sizes of roughly 100–500 bp long . After sonication the extract was centrifuged for 15 min at 13000 rpm at 4°C . The soluble chromatin was then transferred to a fresh tube and normalized for protein concentration by the Bradford assay . For each normalized sample , 25 μl lysate was saved as input , to which 225 μl of 1xTE/1% SDS were added ( TE: 50 mM Tris pH 8 . 0 , 1 mM EDTA ) . Dynabeads Protein A were preincubated with Anti-H3K9me2 antibody ( PRID:AB_449854 ) . For each immunoprecipitation , 2 μg antibody coupled to 30 μl beads was added to 400 μl soluble chromatin , and the final volume of 500 μl was achieved by adding lysis buffer . Samples were incubated for 2 h at 4°C , the beads were collected on magnetic stands , and washed three times with 1 ml lysis buffer and once with 1 ml TE . For eluting bound chromatin , 100 μL elution buffer I ( 50 mM Tris pH 8 . 0 , 10 mM EDTA , 1% SDS ) was added and the samples were incubated at 65°C for 5 min . The eluate was collected and incubated with 150 μl 1xTE/0 . 67% SDS in the same way . Input and immunoprecipitated samples were finally incubated overnight at 65°C to reverse crosslink for more than 6 h . 60 μg glycogen , 100 μg proteinase K ( Roche ) , 44 μl of 5M LiCl , and 250 μl of 1xTE was added to each sample and incubation was continued at 55°C for 1 h . Phenol/chloroform extraction was carried out for all the samples followed by ethanol precipitation . Immuno-precipitated DNA was resuspended in 100 μl of 10 mM Tris pH 7 . 5 and 50 mM NaCl and was used for qPCR ( SYBR Green ) using an Eppendorf Mastercycler Realplex . For extra crosslinking , prior to fixing with 1% formaldehyde , the cultures were incubated at 18°C for 2 h in a shaking incubator . The cells were pelleted and resuspended in 4 . 5 ml of 1x PBS . To this 1 . 5 mM EGS ( ethylene glycol bis[succinimidylsuccinate] ) , Pierce ( Fisher ) was added and the samples were incubated at RT for 20 min with mild shaking before adding 1% formaldehyde . The samples were then processed as mentioned above . All ChIP experiments were reproduced N = 2 . MBP-His-TEV-Epe1 and Epe1-ΔC were cloned into a pFastBac vector ( Thermo Fisher Scientific ) and used for Bacmid generation . Low-titer baculoviruses were produced by transfecting Bacmid into Sf21 cells using Cellfectin II reagent ( Gibco ) . Full-length S . pombe Epe1 protein ( wild-type and mutant ) was expressed in Hi5 cells infected by high titer baculovirus which was amplified from Sf21 cells . After 44 h of infection , Hi5 cells were harvested and lysed in buffer A ( 30 mM Tris-HCl ( pH 8 . 0 ) , 500 mM NaCl , 5 mM EDTA , 5 mM β-mercaptoethanol with protease inhibitor cocktails ) using Emulsiflex-C3 ( Avestin ) . The cleared cell lysate was applied to Amylose resin ( New England Biolabs ) followed by washing with buffer A and elution with buffer A containing 10 mM maltose . The N-terminal His-MBP tag can be removed by TEV protease cleavage , which was used to evaluate protein solubility . Proteins were further purified using a Superdex 200 ( GE Healthcare ) size exclusion column . The protein was concentrated in a storage buffer containing 30 mM Tris-HCl ( pH 8 . 0 ) , 500 mM NaCl , 30% glycerol , and 1 mM TCEP . Proteins were expressed in BL21 ( DE3 ) cells . Cells were grown to log phase at 37°C , cooled on ice , and induced with 0 . 3 mM IPTG before incubation for 18 h at 18°C . Pellets were suspended in tris buffered saline ( TBS ) and frozen at −80°C until further use . For purification , cell pellets were thawed in lysis buffer ( 500 mM NaCl , 50 mM Tris pH 7 . 5 , 10% glycerol ) supplemented with protease inhibitor and cells were ruptured using sonicator . Cell debris was removed by centrifugation and the supernatant was incubated with appropriate beads for each purification for 3 h at 4°C . We used a GST tag and glutathione beads ( GST ) beads for 3X FLAG Swi6HP1 , Swi6HP1 W104A , Swi6HP1 loop-X mutant and 3X FLAG-Chp2HP1 purifications . Swi6HP1 and Chp2HP1 were subject to overnight cleavage with Prescission protease . We used an MBP tag and amylose resin for the purification of Epe1434-948 and Epe1434-600 . After washing , Epe1434-948 and Epe1434-600 were eluted with elution buffer ( lysis buffer + 20 mM maltose + 5 mM EDTA ) . To purify Swi6HP1 used for in vitro binding assays , we used a hexahistidine tag and Nickel resin . After elution , the N-terminal 6X His tag was removed using a SUMO protease followed by addition purification using an anion exchange column . In vitro binding assays were performed by immobilizing recombinant 3X FLAG- Swi6HP1 or 3X FLAG- Chp2HP1 on 25 μl of FLAG M2 beads , which were incubated with three different concentrations of recombinant MBP fusion proteins in 600 μl binding buffer containing 20 mM HEPES pH 7 . 5 , 150 mM NaCl , 5 mM MgCl2 , 10% glycerol , 0 . 25% Triton -X 100 , 1 mM DTT . Reactions were incubated at 4°C for 2 h and washed three times in 1 ml washing buffer ( 20 mM HEPES pH 7 . 5 , 150 mM NaCl , 5 mM MgCl2 , 10% glycerol , 0 . 25% Triton -X 100 , 1 mM DTT ) for 5 min each , then 30μl of SDS sample buffer was added followed by incubation at 95°C for 5 min . Proteins were separated through SDS-PAGE and transferred to PVDF membrane followed by incubation with anti-MBP monoclonal antibody ( E8032S , NEB ) and M2 Flag antibody ( A8592 , Sigma ) . Depending on the experiment , we added co-factors 100μM ammonium iron ( II ) sulfate hexahydrate and 1 mM α-ketoglutarate or 5μg of H3 peptides ( 1–21 amino acids ) with or without modifications . Western blot data for in vitro binding assays were analyzed using ImageJ software . The exposure times for the interaction assays were chosen and differ in each experiment to capture differences in the interaction between Epe1 and Swi6 depending on the assay conditions . Assays performed on different blots cannot be compared but samples loaded on the same blot can be readily compared to each other . All in vitro binding experiments were reproduced N ≥ 3 . Mass spectrometry-based demethylase assays were performed using 5μg MBP-Epe1 , 10μg Swi6 , and 20μM peptide ( either H3K9me3 or H3K9me2 ) . The peptide sequences used in these assays were as follows: 1 ) NH2-ARTKQTAR ( K9me3 ) STGGKA-amide ( H3K9me3 , 1–15 amino acids ) . 2 ) H- ARTKQTARK ( K9me2 ) STGGKAPRKQLA - OH ) ( H3K9me2 , 1–21 amino acids ) . The demethylase assay reaction buffer consists of 50 mM HEPES ( pH 7 . 5 ) , 50 mM NaCl , 100 μM ammonium iron ( II ) sulfate hexahydrate , 1 mM L-ascorbic acid , and 1 mM α-ketoglutarate . Reaction mixtures were incubated at 37°C for 3 h , quenched with an equal volume of 1% trifluoroacetic acid , and stored at −20°C . In parallel , we also performed demethylase assays using equivalent amounts of purified JMJD2A ( protein amounts equalized using SDS-PAGE gels ) . Samples were thawed and desalted using a ZipTip ( Millipore ) . The ZipTip was first equilibrated twice with wetting solution ( 50% acetonitrile ) and twice with equilibration solution ( 0 . 1% trifluoroacetic acid ) . Samples ( 10μl ) following the demethylase assay were washed with washing solution ( 0 . 1% trifluoroacetic acid ) before elution with 4μl of 0 . 1% trifluoroacetic acid/50% acetonitrile . Matrix-assisted laser desorption ionization ( MALDI ) mass spectrometry was performed using a Waters Tofspec-2E in reflectron mode with delayed extraction ( Department of Chemistry , University of Michigan ) . All demethylase experiments were reproduced N = 2 . To identify minimal Epe1 fragments that bind to Swi6HP1 , Epe1 fragments were translated in vitro using TNT T7-coupled reticulocyte lysate ( Promega ) with 35S-labeled methionine ( Roche ) . In vitro translated target proteins were incubated with Flag-tagged Swi6 at 4°C for 20 min . M2 FLAG beads pre-equilibrated with buffer B containing 30 mM Tris-HCl ( pH 8 . 0 ) , 50 mM NaCl , 1 mM DTT , and 0 . 1% NP-40 ( w/v ) were mixed and incubated at 4°C for 45 min with rotation . The beads were washed three times with buffer B , and bead-bound proteins were separated by SDS-PAGE . Dried gels were analyzed by overnight exposure of a phosphor imager plate . To generate fission yeast cell lysates , we grew 100 ml of fission yeast cells in YEA medium at 32°C to an OD600 = 3–3 . 5 and harvested cells by centrifugation . The cell pellets were washed with 1 ml TBS pH 7 . 5 and resuspended in lysis buffer ( 30 mM HEPES pH 7 . 5 , 100 mM NaCl , 0 . 25% Triton X-100 , 5 mM MgCl2 , 1 mM DTT ) . 0 . 5 mm glass beads were added and cell lysis was carried out by bead beating using Omni Bead Ruptor at 3000 rpm ( 30 s x eight cycles ) . The cell extract was centrifuged for 20 min at 15000 rpm at 4°C and the lysates were incubated with beads pre-bound with recombinant MBP-Epe1434-948 protein for 3 h at 4°C . Beads were washed three times with 1 ml lysis buffer and proteins were eluted by boiling the beads in SDS sample buffer . Proteins were resolved by SDS-PAGE and analyzed by immunoblotting with appropriate primary and secondary antibodies . To test whether the N- and C-terminal fragments of Epe1 binds in trans , we used the following protocol: 1 . 5 l of fission yeast cells expressing 3XFLAG- Epe1-N ( 1-434 ) were grown in YEA medium at 32°C to an OD600 = 3 . 5 and harvested by centrifugation . The cell pellets were washed with 10 ml TBS pH 7 . 5 , re-suspended in 1 . 5 ml lysis buffer ( 30 mM HEPES pH 7 . 5 , 100 mM NaCl , 0 . 25% Triton X-100 , 5 mM MgCl2 , 1 mM DTT ) , and the cell suspension was snap-frozen into liquid nitrogen to form yeast ‘balls’ and cryogenically ground using a SPEX 6875D Freezer/Mill . The frozen cell powder was thawed at room temperature and re-suspended in an additional 10 ml of lysis buffer ( 30 mM HEPES pH 7 . 5 , 100 mM NaCl , 0 . 25% Triton X-100 , 5 mM MgCl2 , 1 mM DTT ) with protease inhibitor cocktail and 1 mM PMSF . The cell lysates were subjected to two rounds of centrifugation at 18000 rpm in a JA-25 . 50 rotor ( Beckman ) . Protein G Magnetic Beads were pre-incubated with an M2 FLAG antibody and the antibodies were crosslinked to the beads prior to usage . The cell lysates expressing 3X FLAG-Epe1-N ( 1-434 ) were then incubated with the M2 FLAG antibody conjugated beads overnight at 4°C . Beads were washed once in 1 ml lysis buffer . Importantly , 3XFLAG-Epe1-N ( 1-434 ) was retained on beads for subsequent binding assays . Recombinant Swi6HP1 ( 1 μg ) was pre-incubated with recombinant MBP-Epe1434-948 for 1 h in lysis buffer , then this was added to beads pre-bound with the 3XFLAG-Epe1-N ( 1-434 ) fragment . In assays where we tested the effect of modified peptides on the trans interaction between the N- and C-terminal halves of Epe1 , we added 1 μg of either H3K9me0 or K9-trimethyl H3 ( H3K9me3 ) peptides . Immunoblotting was performed by blocking PVDF membrane in TBS pH 7 . 5 with 0 . 1% Tween-20 ( TBST ) containing 5% non-fat dry milk and subsequently probed with desired primary antibodies and secondary antibodies . Blots were developed with the ECL method and detected with Bio-Rad ChemiDoc Imaging System . H3K9me0 or H3K9me3 biotinylated peptides ( 50 nM ) were pre-incubated with either recombinant MBP-Epe1 or 3XFLAG-Swi6 in binding buffer ( 30 mM Tris-HCI [pH 7 . 5] , 600 mM NaCl , 1% Triton X-100 , 5% glycerol , 2 . 5% BSA ) for 1 h at 4°C . Then streptavidin M280 beads ( Invitrogen ) were added to the pre-mixed protein-peptide mixture and incubated for an additional 2 h at 4°C . The beads were then rinsed three times with wash buffer ( 30 mM Tris-HCI [pH 7 . 5] , 600 mM NaCl , 1% Triton X-100 , 5% glycerol , 2 . 5% BSA ) and bound proteins were eluted by boiling the beads in SDS sample buffer . The input and bound proteins were resolved by SDS-PAGE and analyzed by immunoblotting with MBP antibody ( E8032S , NEB ) . MBP pull-down assays were performed using Calf thymus histones ( Sigma ) to evaluate Epe1 binding specificity . 1 μg of Calf thymus histones ( Sigma ) were pre-incubated with recombinant MBP-Epe1 in binding buffer ( 30 mM Tris-HCI [pH 7 . 5] , 600 mM NaCl , 1% Triton X-100 , 5% glycerol , 2 . 5% BSA ) for 1 h at 4°C . Amylose resin was added to the binding assay and incubated for an additional 2 h at 4°C . Beads were rinsed three times with wash buffer ( 30 mM Tris-HCI [pH 7 . 5] , 600 mM NaCl , 1% Triton X-100 , 5% glycerol , 2 . 5% BSA ) and bound proteins were eluted by boiling the beads in SDS sample buffer . The input and bound proteins were resolved by SDS-PAGE and analyzed by immunoblotting with MBP antibody ( E8032S , NEB ) .	A cell’s identity depends on which of its genes are active . One way for cells to control this process is to change how accessible their genes are to the molecular machinery that switches them on and off . Special proteins called histones determine how accessible genes are by altering how loosely or tightly DNA is packed together . Histones can be modified by enzymes , which are proteins that add or remove specific chemical ‘tags’ . These tags regulate how accessible genes are and provide cells with a memory of gene activity . For example , a protein found in yeast called Epe1 helps reactivate large groups of genes after cell division , effectively ‘re-setting’ the yeast’s genome and eliminating past memories of the genes being inactive . For a long time , Epe1 was thought to do this by removing methyl groups , a ‘tag’ that indicates a gene is inactive , from histones – that is , by acting like an enzyme . However , no direct evidence to support this hypothesis has been found . Raiymbek et al . therefore set out to determine exactly how Epe1 worked , and whether or not it did indeed behave like an enzyme . Initial experiments testing mutant versions of Epe1 in yeast cells showed that the changes expected to stop Epe1 from removing methyl groups instead prevented the protein from ‘homing’ to the sections of DNA it normally activates . Detailed microscope imaging , using live yeast cells engineered to produce proteins with fluorescent markers , revealed that this inability to ‘home’ was due to a loss of interaction with Epe1’s main partner , a protein called Swi6 . This protein recognizes and binds histones that have methyl tags . Swi6 also acts as a docking site for proteins involved in deactivating genes in close proximity to these histones . Further biochemical studies revealed how the interaction between Epe1 and Swi6 can help in gene reactivation . The methyl tag on histones in inactive regions of the genome inadvertently helps Epe1 interact more efficiently with Swi6 . Then , Epe1 can simply block every other protein that binds to Swi6 from participating in gene deactivation . This observation contrasts with the prevailing view where the active removal of methyl tags by proteins such as Epe1 switches genes from an inactive to an active state . This work shows for the first time that Epe1 influences the state of the genome through a process that does not involve enzyme activity . In other words , although the protein may ‘moonlight’ as an enzyme , its main job uses a completely different mechanism . More broadly , these results increase the understanding of the many different ways that gene activity , and ultimately cell identity , can be controlled .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "biochemistry", "and", "chemical", "biology" ]	2020	An H3K9 methylation-dependent protein interaction regulates the non-enzymatic functions of a putative histone demethylase
Pancreatic islet β-cell insufficiency underlies pathogenesis of diabetes mellitus; thus , functional β-cell replacement from renewable sources is the focus of intensive worldwide effort . However , in vitro production of progeny that secrete insulin in response to physiological cues from primary human cells has proven elusive . Here we describe fractionation , expansion and conversion of primary adult human pancreatic ductal cells into progeny resembling native β-cells . FACS-sorted adult human ductal cells clonally expanded as spheres in culture , while retaining ductal characteristics . Expression of the cardinal islet developmental regulators Neurog3 , MafA , Pdx1 and Pax6 converted exocrine duct cells into endocrine progeny with hallmark β-cell properties , including the ability to synthesize , process and store insulin , and secrete it in response to glucose or other depolarizing stimuli . These studies provide evidence that genetic reprogramming of expandable human pancreatic cells with defined factors may serve as a general strategy for islet replacement in diabetes . The pancreas is a vital organ with exocrine and endocrine cell functions , and a root of lethal human diseases including diabetes mellitus , pancreatitis , and pancreatic ductal adenocarcinoma . Exocrine acinar cells produce digestive zymogens that are delivered to the intestines by a branching network of exocrine ductal cells that secrete bicarbonate and other products . Pancreatic endocrine functions derive from clusters of epithelial cells ( islets of Langerhans ) called α- , β- , δ- , and PP-cells that respectively synthesize , store , and secrete the hormones Glucagon , Insulin , Somatostatin , and Pancreatic polypeptide ( Benitez et al . , 2012 ) . Insulin production by islet β-cells is highly regulated: key features of mature β-cells include preproinsulin ( INS ) transcription , proinsulin processing by endo- and exo-peptidases and storage of the proinsulin cleavage products insulin and C-peptide in dense core vesicles . Likewise , cardinal β-cell functions regulate insulin release in response to glucose and other secretagogues , including glucose sensing and metabolism through the enzyme glucokinase , and use of ATP-dependent potassium channels ( KATP ) and voltage-gated calcium channels to induce insulin exocytosis ( reviewed in Suckale and Solimena , 2010 ) . Deficiency or malfunctioning of β-cells produces impaired glucose regulation and diabetes mellitus , a disease with autoimmune ( type 1 , T1DM ) and pandemic forms ( type 2; Ashcroft and Rorsman , 2012 ) . Thus , replacement or regeneration of functional human β-cells is an intensely-sought goal . Human islet transplantation can be used to replace β-cell function in T1DM ( reviewed in Vardanyan et al . , 2010 ) , but a shortage of donors currently precludes broad use of human pancreatic islets for β-cell replacement . Because of their expandability and multipotency , human embryonic stem cells ( hESCs ) and induced pluripotent stem cells ( iPSCs ) have been explored as sources of replacement insulin-producing cells ( reviewed in Hebrok , 2012 ) . However , directing the differentiation of these developmentally ‘primitive’ cells through multiple sequential fates into β-cell-like progeny that synthesize , process , store , and secrete insulin while lacking tumorigenic potential has challenged investigators worldwide ( Fujikawa et al . , 2005; McKnight et al . , 2010; Cheng et al . , 2012 ) . Moreover , different hESC and iPSC cell lines exhibit significant variability during development into insulin-producing cells ( Nostro and Keller , 2012 ) . Recent work demonstrated that differentiated cell types in adult organs , including the mouse pancreas , can be experimentally ‘reprogrammed’ into progeny resembling islet cells , suggesting a new strategy for β-cell replacement ( Vierbuchen and Wernig , 2011 ) . For example , adult mouse pancreatic acinar cells can be converted into insulin-producing cells in vitro and in vivo ( Minami et al . , 2005; Zhou et al . , 2008 ) . However , little progress has been made in reprogramming primary human epithelial cells into different cell types , including conversion of pancreatic non-β-cells toward a human β-cell fate ( Vierbuchen and Wernig , 2011 ) . Thus , systems permitting expansion and genetic modulation of human pancreatic cells could powerfully influence studies of β-cell biology and replacement . Pancreatic ducts constitute 30–40% of human pancreas and have been proposed as a potential source of replacement β-cells ( Bouwens and Pipeleers , 1998; Bonner-Weir et al . , 2004 ) . During pancreas development , fetal endocrine cells derive from primitive ductal epithelium ( reviewed by Pan and Wright , 2011; Pictet and Rutter , 1972 ) . In addition , some studies have suggested that in adult mice , β-cells may be produced from pancreatic ductal epithelium ( Inada et al . , 2008; Xu et al . , 2008; Rovira et al . , 2010 ) . However , recent lineage tracing evidences have not supported this view ( Solar et al . , 2009; Furuyama et al . , 2011; Kopp et al . , 2011 ) . In humans , prior studies have suggested that adult human primary ductal cells in heterogeneous cell mixtures may harbor the potential to generate endocrine-like progeny ( Bonner-Weir et al . , 2000; Heremans et al . , 2002; Swales et al . , 2012 ) , but interpretation in these studies was limited by the probability of islet cell contamination . Therefore , the potential for conversion of pancreatic ductal cells toward an endocrine fate remains unclear . Moreover , prior studies have revealed only limited proliferative capacity of primary human pancreatic ductal cells in culture ( Rescan et al . , 2005 ) . Thus , despite their relative abundance , multiple practical issues have prevented development of human pancreatic ductal cells as a source of replacement β-cells . Here we report that normal human adult pancreatic duct cells can be sorted , clonally expanded , and genetically converted into endocrine cells . Human insulin-producing cells ( IPCs ) produced from sorted duct cells exhibited hallmark features of functional neonatal β-cells including high-level preproinsulin ( INS ) expression , proinsulin processing and dense-core granule formation . Moreover , secretion of insulin and insulin C-peptide from IPCs is stimulated by glucose and KATP channel stimulants in a calcium-dependent manner . Together these studies reveal a new system for investigating human pancreatic duct cell biology , genetics , and β-cell regeneration . To identify human pancreatic epithelial cells that can be grown and maintained in culture , we systematically screened cell isolation methods and culture conditions with dispersed adult human pancreatic cells obtained from cadaveric donors without known pancreatic cancer , diabetes mellitus , or other pancreatic diseases ( Table 1 ) . With primary cells plated at low density , we observed formation of multicellular epithelial spheres , when cultured in Matrigel with a serum-free culture medium without feeder cells ( ‘Materials and methods’ , Figure 1—figure supplement 1A ) . The multicellular sphere formation suggested primary cell expansion , so based on this assay we fractionated cells by fluorescence-activated cell sorting ( FACS ) to isolate and characterize sphere-forming pancreatic cells . A survey of cell surface markers used for fetal mouse pancreatic cell isolation ( Sugiyama et al . , 2007 ) revealed that antibodies recognizing CD133 enriched sphere-forming cells by four fold , whereas sphere-forming cells were depleted in the CD133neg fraction ( Figure 1A , B ) . Immunohistochemical analysis of the human adult pancreas revealed CD133 expression at the apical portion of duct epithelial cells that co-expressed keratin 19 ( KRT19 ) , whereas CD133 was undetectable in islet endocrine cells or acinar cells ( Figure 1C , Figure 1—figure supplement 1B ) , consistent with prior reports ( Lardon et al . , 2008 ) . We have achieved sphere formation from over 35 consecutive adult donors ( Table 1 ) ; thus , the sphere formation of primary adult human pancreatic CD133+ cells was highly reproducible . 10 . 7554/eLife . 00940 . 003Table 1 . Phenotypes of pancreas donorsDOI: http://dx . doi . org/10 . 7554/eLife . 00940 . 003Anonymous IDAge ( year ) GenderBody mass index131Male28 . 1252Male31 . 6352MaleNot provided616Female20 . 4934Male35 . 41050Female231132Female36 . 21235Male45 . 71323Female26 . 61451Female23 . 31548Male36 . 71625Male21 . 81763Female30 . 91844Male24 . 71939Male27 . 362044Male23 . 52150Female312240Female262353Male312419Female20 . 832534Male22 . 82655Male37 . 72717Female31 . 12833Male18 . 82948Male36 . 63040Female28 . 43143Female35 . 33247Female213348Female23 . 33728Male24 . 24034Male32 . 84122Male19 . 64253Female22 . 44416Male33 . 94554Male29 . 64618Male21 . 84824Male25 . 510 . 7554/eLife . 00940 . 004Figure 1 . The ductal cell surface marker CD133 enriches sphere-forming cells from dissociated human adult pancreas . ( A ) Left panel , FACS plot of the dissociated human adult pancreas stained with ( gray ) or without ( blue ) antibodies specific for CD133 . Right panel , A schematic of the sphere culture system and a representative sphere after culture . ( B ) Quantification of spheres generated from CD133+ , CD133neg , and unsorted cells . Data are presented as mean ± SEM ( n = 4 ) . ( C ) Immunostaining of CD133 ( green ) with a ductal marker KRT19 ( red ) and C-peptide ( red ) in adult human pancreas . ( D ) The gene expression profiles of FACS-sorted human adult pancreatic cells and isolated islets ( islet values normalized to 1 ) . Data are presented as mean ± SEM ( n = 3 ) . ( E ) Representative immunostaining pictures of sorted cells with KRT19 ( green ) or C-peptide ( green ) . ( F ) Quantification of cell immunostaining after FACS . ≥7200 cells were counted per staining condition . n . d . = not detected . Scale bars , 50 µm . See also Supplementary file 1D . DOI: http://dx . doi . org/10 . 7554/eLife . 00940 . 00410 . 7554/eLife . 00940 . 005Figure 1—figure supplement 1 . Sorted CD133+ cells originate from pancreatic ducts . ( A ) Schematic diagram of the experimental procedure . Dissociated pancreatic cells were embedded and cultured as previously described ( Lawson et al . , 2007 ) . Scale bar , 200 µm . ( B ) Confocal images of CD133 ( green ) and CPA1 ( red ) co-staining in adult human pancreas tissue . Scale bar , 20 µm . ( C ) CEL expression profiles of FACS-sorted human adult pancreatic cells and isolated islets ( islet values normalized to 1 ) . Data are presented as mean ± SEM ( n=3 ) . ( D ) Representative immunostaining pictures of sorted cells . Scale bar , 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00940 . 005 To assess the properties of FACS-purified adult pancreatic CD133+ cells , we performed quantitative reverse transcription PCR ( qRT-PCR ) . This revealed that CD133+ cells expressed high levels of mRNA encoding ductal markers ( KRT19 and CAR2 ) , while mRNAs expressed in acinar ( CPA1 and CEL ) and endocrine ( CHGA , INS , and GCG ) cells were exclusively enriched in the CD133neg fraction ( Figure 1D , Figure 1—figure supplement 1C ) . Immunostaining confirmed that >98% of sorted CD133+ cells produced KRT19 , whereas CD133+ cells produced no detectable islet hormone ( Figure 1E , F , Figure 1—figure supplement 1D ) . Thus , FACS efficiently eliminated islet endocrine and acinar cells , and enriched for a population of primary adult pancreatic duct cells that expanded as epithelial spheres in feeder- and serum-free culture . After commencing in vitro cultures , the epithelial spheres from CD133+ ductal cells attained diameters ranging from 40 to 520 µm in 2 weeks ( Figure 1A , Figure 1—figure supplement 1A and Figure 2—figure supplement 1A ) . Spheres 350–500 µm in diameter were composed of 1470 ± 310 cells ( n = 5 ) ; thus , based on evidence of clonal expansion ( see below ) , we calculated that spheres resulted from a minimum of 10 cell divisions in 2 weeks . Sphere epithelium maintained KRT19 protein expression and a polarized monolayer as indicated by apical localization of CD133 ( Figure 2A , Figure 2—figure supplement 1A , D ) . Neither acinar ( CPA1 ) nor islet endocrine ( CHGA and insulin C-peptide ) markers were detectable ( Figure 3C and data not shown ) , suggesting epithelial cells in cultured spheres maintained ductal characteristics . 10 . 7554/eLife . 00940 . 006Figure 2 . Clonal expansion and passaging of ductal spheres . ( A ) Confocal images of 2-week-old spheres immunostained with KRT19 , CD133 , Ki-67 , and Phospho-Histone H3 ( all green ) . Note the apical localization of CD133 . Scale bars , 50 µm . ( B ) Representative time-lapse images of sphere formation from single cell ( arrowhead ) . Images taken every 12 hr for 9 days are shown . Arrows point a non-sphere forming cell used as a landmark . ( C ) Representative pictures of spheres after each passage . Scale bars , 100 µm . ( D ) Quantification of cell number in spheres after each indicated passage . Y axis represents fold increase of total cell numbers relative to the one measured in the first ‘generation’ of spheres ( G1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00940 . 00610 . 7554/eLife . 00940 . 007Figure 2—figure supplement 1 . Quantification of sphere growth and passaging . ( A ) A representative image of human spheres grown for 2 weeks in culture . Note variable sphere sizes . Scale bar , 100 µm . ( B ) Quantification of Ki-67-expressing cells as a percentage of total cell number shown in Figure 2A . More than 200 cells per slide , from three or more slides per sample were counted . Data are presented as means ± S . D . ( C ) Quantification of the total cell number in each passage of the individual samples shown in Figure 2D . Y axis represents fold increase of total cell numbers relative to the one measured in the first ‘generation’ of spheres ( G1 ) . ( D ) Representative confocal images of G1 and G7 spheres co-immunostained with KRT19 ( red ) and CD133 ( green ) . Scale bars , 50 µm . ( E ) Quantification of the total cell number of CD133+ cells in G1 and G7 spheres . Data are presented as means ± S . D . DOI: http://dx . doi . org/10 . 7554/eLife . 00940 . 00710 . 7554/eLife . 00940 . 008Figure 3 . Neurog3 is sufficient to convert pancreatic ductal spheres into hormone-expressing endocrine-like cells . ( A ) Schematic of growth and reprogramming strategies . See ‘Materials and methods’ for details . ( B ) Schematics of adenoviral constructs used . ( C ) Relative mRNA level of Neurog3 targets ( NEUROD1 , INSM1 , and RFX6 ) , endocrine cell-specific genes ( PAX4 , NKX2 . 2 , and CHGA ) , and pancreatic hormones ( SST and GHRL ) . Data are presented as mean ± SEM ( n ≥ 3 ) . ( D ) Representative confocal images of Ad-RFP-Neurog3 infected spheres after immunostaining with antibodies specific to mouse Neurog3 , NEUROD1 , NKX2 . 2 , SST , and GHRL . Note that all hormone-positive cells are CHGA-positive . Right: co-staining of SST and GHRL . Scale bar , 20 µm . ( E ) Quantification of the staining results shown in ( D ) . Pie graph represents the percentage of the hormone+ cells . ( F ) A representative FACS plot of dissociated ductal spheres infected with Ad-RFP-Neurog3 adenovirus ( red ) or uninfected control ( gray ) . Fractions P1 through P5 were sorted based on RFP fluorescence intensity . ( G ) qRT-PCR analysis of fractions P1 through P5 from ( F ) . ‘U’ indicates unsorted cells . Analytical duplicates are shown . Data are presented as mean ± SD . DOI: http://dx . doi . org/10 . 7554/eLife . 00940 . 00810 . 7554/eLife . 00940 . 009Figure 3—figure supplement 1 . Representative confocal images of spheres infected with control virus ( Ad-RFP ) . ( A ) Representative confocal images of control virus ( Ad-RFP ) infected spheres after immunostaining with antibodies specific to mouse Neurog3 , NEUROD1 , NKX2 . 2 , SST , and GHRL . Right: immunostaining to detect cells co-expressing SST and GHRL . DOI: http://dx . doi . org/10 . 7554/eLife . 00940 . 009 To assess whether sphere growth was achieved by cell proliferation or by other mechanisms like cell migration and aggregation , we analyzed spheres by immunostaining and time-lapse imaging . Immunohistochemistry revealed the proliferation marker Ki-67 in more than 25% of cells comprising 2-week-old spheres ( Figure 2A , Figure 2—figure supplement 1B; labeling index 26 . 5 ± 5 . 1% ) , data further supported by detection of a second proliferation marker , phospho-histone H3 ( Figure 2A ) . Time-lapse imaging revealed that spheres arose from single cells ( Figure 2B ) , providing strong evidence that sphere formation resulted from CD133+ ductal cell proliferation , rather than through cell migration and aggregation . Enzymatic dispersion of 2-week-old G1 spheres and subsequent culture revealed that the spheres can be passaged up to seven generations ( G7 , 3 months ) and that the total number of cells increased with each generation ( Figure 2C , D , Figure 2—figure supplement 1C ) . After G7 , ductal cell expansion was not achieved , and the spheres were not formed ( Figure 2—figure supplement 1C and data not shown ) , supporting the view that ductal epithelial cells are not immortalized , and consistent with the origin of pancreatic cells from donors without neoplasia . The endocrine potential of human or mouse pancreatic ductal cells remains controversial . To investigate the potential of purified human pancreatic ductal cells to achieve an endocrine fate , we used an adenovirus-mediated transgenic system . Neurog3 is a transcription factor necessary and sufficient for pancreatic endocrine cell differentiation in vivo ( Gradwohl et al . , 2000; Gu et al . , 2002 ) and , combined with other factors , can induce pancreatic acinar-to-islet cell conversion in mice ( Zhou et al . , 2008 ) . To test if Neurog3 expression could respecify human duct cells toward an endocrine fate , we infected cultured spheres as well as primary CD133+ cells with recombinant adenovirus co-expressing red fluorescent protein and Neurog3 ( Ad-RFP-Neurog3 ) , and assessed changes in gene expression by qRT-PCR ( Figure 3A–C and 4C ) . Neurog3 induced the expression of NEUROD1 , INSM1 , and RFX6 ( Figure 3C ) , genes whose mouse homologs are known direct targets of Neurog3 in pancreas development ( Mellitzer et al . , 2006; Smith et al . , 2010 ) . Ad-RFP-Neurog3 infection induced expression of the pan-endocrine markers chromogranin A ( CHGA ) and synaptophysin in both primary CD133+ duct cells and cultured spheres ( Figures 3C and 4C , and data not shown ) . Ad-RFP-Neurog3 infection also induced expression of mRNA encoding PAX4 and NKX2 . 2 , transcriptional regulators of pancreatic endocrine cell fate ( Sosa-Pineda et al . , 1997; Sussel et al . , 1998 ) , and mRNA encoding crucial β-cell factors such as the prohormone processing enzymes PCSK1 ( PC1/3 ) and PCSK2 ( PC2 ) , KATP channel components KCNJ11 ( KIR6 . 2 ) and ABCC8 ( SUR1 ) , and glucokinase ( GCK ) ( Figure 4D ) . Moreover , Ad-RFP-Neurog3 significantly induced mRNA encoding the pancreatic hormones ghrelin and somatostatin , but not mRNAs encoding insulin , glucagon , PPY or the intestinal hormones cholecystokinin and gastrin ( Figures 3C and 4D , Figure 4—figure supplement 1A , and data not shown ) . These findings support the conclusion that human adult pancreatic ductal cells harbor pancreatic endocrine potential upon induction of Neurog3 . 10 . 7554/eLife . 00940 . 010Figure 4 . Induction of four transcription factors ( Neurog3 , MafA , Pdx1 , and Pax6 ) produces Insulin+ endocrine cells in pancreatic ductal spheres in vitro . ( A ) Schematics of adenoviruses used . ( B ) INS qRT-PCR analysis of human spheres infected with control ( R = RFP ) or a combination of MafA ( M ) , Neurog3 ( N ) , and Pdx1 ( P ) ( MNP ) n = 4 . ( C ) qRT-PCR analysis of INS , SST , and CHGA with freshly sorted CD133+ ductal cells infected with adenoviruses encoding Neurog3 or all four genes ( 4V ) ( n = 2 ) . ( D ) qRT-PCR analysis of the spheres infected with a combination of adenoviruses . Pax6 abbreviated as ‘6’ , ( n ≥ 3 ) . ( E ) qRT-PCR analysis of the spheres infected with 4V minus each indicated factor n = 2 . All bar graph data are presented as mean ± SEM with mRNA levels from purified adult human islets normalized to 1 . ( F ) Confocal images of infected spheres after staining with antibodies recognizing C-peptide . Note that adenoviruses encoding Neurog3 ( N ) and Pdx1 ( P ) also express RFP . Scale bar , 20 µm . ( G ) Quantification of the CHGA- , SST- , and C-peptide-immunoreactive cells in the spheres infected with the indicated combination of adenoviruses . Note that the number of C-peptide-positive cells increased in 4V than MNP by 18–20-fold . DOI: http://dx . doi . org/10 . 7554/eLife . 00940 . 01010 . 7554/eLife . 00940 . 011Figure 4—figure supplement 1 . GCG , PPY , and PAX6 mRNA levels after sphere infection with adenovirus combinations . ( A ) qRT-PCR analysis of the spheres infected with a combination of adenoviruses . PAX6 abbreviated as ‘6’ , ( n ≥ 3 ) . Note that PAX6 qRT-PCR probe recognizes both endogenous and exogenous PAX6 mRNA . DOI: http://dx . doi . org/10 . 7554/eLife . 00940 . 011 Immunostaining confirmed these qRT-PCR findings and demonstrated that only RFP+ cells produced by Ad-RFP-Neurog3 infection were immunostained with antibodies recognizing NEUROD1 , NKX2 . 2 , CHGA , SST or GHRL ( Figure 3B , D , Figure 3—figure supplement 1 ) . We also confirmed that no insulin- , glucagon- or PPY-positive cells were observed by immunostaining ( data not shown ) . While only a subset of cells infected with Ad-RFP-Neurog3 ( RFP+ ) expressed CHGA , we noted all GHRL+ or SST+ cells co-expressed CHGA ( Figure 3D ) . Quantification of CHGA+ and hormone+ cells revealed that 30% of infected cells ( RFP+ ) expressed CHGA . At least 45% of CHGA+ cells produced SST or GHRL , and less than 2% of CHGA+ cells expressed both hormones ( Figure 3D , E ) . Thus , Neurog3 expression efficiently converted primary human ductal cells and cultured ductal epithelial spheres into hormone-expressing cells with cardinal features of endocrine pancreas . In mice , Neurog3 gene dosage can determine commitment between exocrine and endocrine lineages in pancreas development ( Wang et al . , 2010 ) . Therefore , we next assessed the possibility that the 70% of RFP+ cells infected by Ad-RFP-Neurog3 failing to express CHGA may have achieved inadequate levels of Neurog3 expression . We fractionated cells produced by Ad-RFP-Neurog3 infection by RFP intensity and measured mRNA expression of Neurog3 , CHGA , SST and GHRL by qRT-PCR ( Figure 3F , G ) . We found that cell fractions with the highest levels of RFP expression ( ‘P4 and P5’ , Figure 3F ) had the highest levels of mouse Neurog3 mRNA , and only these cell fractions produced mRNA encoding CHGA , SST or GHRL ( Figure 3G ) . These data suggest that relatively high threshold levels of Neurog3 may be necessary and sufficient for directing endocrine differentiation of human pancreatic cells . The transcription factors MafA , Neurog3 , and Pdx1 ( a combination hereafter summarized as ‘MNP’ ) were sufficient to convert adult mouse acinar cells into insulin-producing cells ( IPCs: Zhou et al . , 2008 ) . We constructed three adenoviruses expressing MafA , Neurog3 , or Pdx1 ( see ‘Materials and methods’; Figure 4A ) , and infected cultured spheres with this MNP combination . Within 5 days after infection , we reproducibly detected INS mRNA induction but at extremely low levels relative to adult human islet controls ( 0 . 0035 ± 0 . 0012% of islet levels; Figure 4B ) . Thus , we sought additional factors and discovered that mRNA encoding PAX6 , an important regulator of mouse pancreatic endocrine cell development ( Sander et al . , 1997 ) , was induced by MNP to only 0 . 03% of levels in control islets ( Figure 4—figure supplement 1A ) . When combined with MafA , Neurog3 , and Pdx1 ( encoded in four viruses , ‘4V’ ) , Pax6 induced INS expression in primary CD133+ ductal cells or spheres by over 30-fold relative to MNP ( Figure 4A , C , D ) . We observed ductal conversion to IPCs with four consecutive , independent donors ( INS , Figure 4D ) . We also detected substantially increased expression of other islet endocrine markers , including SST , GCK , PCSK1 , KCNJ11 , and ABCC8 ( Figure 4D ) . Immunohistochemical analyses demonstrated that the number of Insulin+ cells was increased by 18 to 20-fold in spheres transduced by the four factor combination ( 4V ) compared to the MNP combination ( Figure 4F , G ) . ELISA studies quantified and confirmed this increase of proinsulin levels , showing that the spheres derived from 4V exposure contained proinsulin levels that averaged 0 . 7% of those in human islets ( Figure 5E ) . Systematic removal of individual factors from this four virus combination revealed that omission of Neurog3 prevented expression of INS , CHGA or SST ( Figure 4E ) . Omission of virus expressing MafA or Pax6 from this combination significantly reduced INS expression ( Figure 4E–G ) , whereas omission of virus expressing Pdx1 did not significantly decrease INS expression . Thus , Neurog3-mediated endocrine cell conversion is required for the production of IPCs as well as other hormone-producing cells from ductal spheres . 10 . 7554/eLife . 00940 . 012Figure 5 . Induced insulin-secreting cells resemble functional β-cells . ( A ) Schematic of adenoviral constructs used . See ‘Materials and methods’ and Figure 5—figure supplement 5 for details . ( B ) A schematic diagram of growth , conversion , and maturation procedures . ( C ) qRT-PCR analysis of spheres infected with Ad-eGFP ( black ) or Ad-4TF ( gray ) followed by extended culture . Data are normalized to adult human islet samples ( red dotted line ) . ( D ) Representative confocal images of 4TFM spheres immunostained with indicated antibodies . Scale bar , 20 µm . ( E ) Quantification of total proinsulin and C-peptide content in GFP , 4TF , 4TFM spheres and human adult islets ( Top ) . Total protein level ( pmol ) was normalized by total genomic DNA content ( µg ) . Ratio of proinsulin and C-peptide content is presented as % ( Bottom ) . Sph . = Spheres . ( F ) Representative electron microscopic images of 4TFM spheres . Dotted white line demarks cell boundary between converted , granulated ( left ) and non-converted ( right ) cells . Dense core vesicles with different morphology in converted cells are shown in the right panels . Scale bar , 1 µm . ( G ) Human C-peptide secretion assay of 4TFM spheres stimulated by the indicated secretagogues and drugs . Gluc = Glucose , Tol = Tolbutamide , Diaz . = Diazoxide . Data are presented as means ± SEM ( n = 2 for Diaz . ; n ≥ 3 for all other conditions ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00940 . 01210 . 7554/eLife . 00940 . 013Figure 5—figure supplement 1 . Phenotypes of induced Insulin-secreting cells . ( A ) qRT-PCR analysis of INS and IAPP with GFP , 4TF or 4TFM spheres . Note that spheres from extended culture ( 4TFM ) had significantly increased mRNA levels of INS and IAPP ( n = 6 ) . ( B ) Quantification of CHGA- and C-peptide-immunoreactive cells in the 4TFM spheres . ( C ) Representative confocal images of 4TFM spheres with NKX6 . 1 and SST . Note the non-overlapping staining . Blue = DAPI . ( D and E ) Representative images of electron microscopy . ( D ) Dense core vesicles found adjacent to the plasma membrane . ( E ) Rare cells contain vesicles with irregular shape , reminiscent of δ-cells . ( F ) Four different culture media used for this study . See ‘Experimental procedures’ for details . ( G ) C-peptide secretion was shown as a percentage of total C-peptide content ( n = 4 ) . ( H ) Human C-peptide secretion assay with step increase of glucose concentration ( left ) and with KCl in the presence or absence of extracellular calcium ( right ) ( n ≥ 3 ) . All bar graphs are presented as means ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 00940 . 01310 . 7554/eLife . 00940 . 014Figure 5—figure supplement 2 . Grafted IPCs survive long term and secrete insulin C-peptide upon glucose stimulation . ( A ) Representative confocal images of kidney-transplanted IPCs immunostained with indicated antibodies ( HuNu = human nuclei-specific antibody ) . ( B ) Human insulin levels in serum of an IPC-grafted mouse ( ID51 ) before ( fasting ) or 30 min after glucose challenge ( glucose injection ) . Data are presented as means ± S . D . ( C ) Representative confocal images of liver-transplanted IPCs immunostained with indicated antibodies . ( D ) Representative confocal images of human islets transplanted in the indicated sites ( Kidney or EFP ) and immunostained with indicated antibodies . ( E ) Human insulin level in serum of human islet-grafted mice before ( gray ) or 30 min after glucose challenge ( black ) . Data are presented as means ± SD . Scale bars , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00940 . 01410 . 7554/eLife . 00940 . 015Figure 5—figure supplement 3 . Sustained expression of exogenous factors after maturation period . ( A ) Schematic of qRT-PCR probes designed against adenoviral constructs used . ( B ) qRT-PCR of spheres infected with Ad-eGFP ( eGFP ) or Ad-Neurog3-IRES-eGFP and Ad-eGFP-M6P ( 4TF ) with or without extended culture . Note that insulin expression is markedly elevated independent of transgene expression . DOI: http://dx . doi . org/10 . 7554/eLife . 00940 . 01510 . 7554/eLife . 00940 . 016Figure 5—figure supplement 4 . Conversion of human dermal fibroblasts . qRT-PCR of human dermal fibroblasts or spheres infected with Ad-eGFP ( GFP ) or Ad-Neurog3-IRES-eGFP and Ad-eGFP-M6P ( 4TFM ) and cultured . qRT-PCR probes for INS ( A ) and CHGA ( B ) were used to assess cell conversion . See ‘Materials and methods’ for detail . DOI: http://dx . doi . org/10 . 7554/eLife . 00940 . 01610 . 7554/eLife . 00940 . 017Figure 5—figure supplement 5 . Protein expression of viral transgenes . ( A ) Schematic of adenoviral constructs used . To differentiate viral transgenes from endogenously encoded proteins , MAFA , PAX6 , and PDX1 were epotpe-tagged with Myc ( N-terminus ) , HA ( C-terminus ) , and Flag ( C-terminus ) , respectively . ( B ) Representative confocal images of 4TFM spheres with antibodies against mouse Neurog3 , Myc , HA , and Flag . Scale bar , 20 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 00940 . 017 Although ELISA studies readily detected proinsulin production by IPCs in our 4V spheres , we failed to detect processed C-peptide by ELISA ( Figure 5E ) or by immunostaining with antibodies recognizing cleaved C-peptide ( data not shown ) . Thus , we sought methods to enhance proinsulin processing in IPCs produced by genetic conversion . For this , we used Ad-Neurog3-IRES-eGFP and a second adenovirus constructed to express simultaneously the three transcription factors MAFA , PAX6 , and PDX1 ( Ad-eGFP-M6P ) in cultured G1 spheres ( Figure 5A , referred to as ‘4TF’ combination ) . Compared to our standard 5 day post-infection culture ( 4TF ) , we found that two additional weeks of culture ( referred to as ‘4TFM’ ) resulted in a 10-fold increase of INS mRNA expression in spheres ( Figure 5B , C , Figure 5—figure supplement 1A ) . We observed conversion to IPCs with five consecutive , independent donors ( INS , Figure 5C ) , demonstrating the robustness of our conversion method . The total number of converted IPCs appeared unchanged after this extended culture compared to 4V cultures ( Figure 4G , Figure 5—figure supplement 1B ) , suggesting that INS mRNA levels per cell were increased in the 4TFM ( 4 transcription factors in two viruses plus maturation period ) condition . In addition , mRNA encoding islet amyloid pancreatic polypeptide ( IAPP ) , a β-cell dense core granule component not detectable in standard 4TF conditions , was readily detected in 4TFM spheres ( Figure 5—figure supplement 1A ) . Likewise , multiple mRNAs encoding β-cell factors were expressed at levels comparable to those in purified human islets ( Figure 5C ) , including the transcription factors NKX2 . 2 and NKX6 . 1 , GCK , glucose transporters SLC2A1 ( GLUT1 ) and SLC2A2 ( GLUT2 ) , PCSK1 , PCSK2 , Zinc transporter SLC30A8 , KCNJ11 , ABCC8 , the voltage-gated calcium channel component CACNA1C , regulators of Ca++-induced insulin exocytosis like RAB3A , SYT3 , and VAMP2 , and the postulated maturation marker Urocortin 3 ( UCN3 ) ( Suckale and Solimena , 2010; Blum et al . , 2012 ) . Immunohistochemical analyses corroborated our qRT-PCR analysis , and showed that converted Insulin+ IPCs did not express other islet hormones ( Figure 5D ) . Although we were unable to assess endogenous MAFA and PDX1 in cells with virally-expressed exogenous MAFA and PDX1 protein , we readily detected other known β-cell specific markers including NKX6 . 1 , IAPP , and PC1/3 ( Figure 5D ) . Moreover , Insulin+ cells , but not other hormone+ cells , expressed NKX6 . 1 , a transcription factor with expression normally restricted in islets to β-cells ( Figure 5—figure supplement 1C and data not shown ) . To assess enhanced IPC maturation after extended culture ( 4TFM ) , we measured proinsulin and insulin C-peptide by ELISA . Total insulin ( proinsulin + C-peptide ) levels ranged from 3 . 4 to 15 . 2 pmol/µg DNA ( Figure 5E ) , equal to approximately 9 . 6% of the total insulin protein level found in human adult islets ( Figure 5E ) . Moreover , the percentage of insulin C-peptide processing in IPCs was comparable to that found in adult human islets ( IPCs 77–92%; human islets 96–97% ) , indicating that maturation of IPCs during extended culture permitted proinsulin processing ( Figure 5E ) . Ultrastructural studies by electron microscopy demonstrated round dense-core vesicles ( Figure 5F ) resembling those in adult human β-cells , including subsets of immature ( light core ) and mature ( dense or crystallized core ) vesicles , and vesicles adjacent to the plasma membrane ( Figure 5F , Figure 5—figure supplement 1D ) . Consistent with the detection of SST mRNA ( Figure 5C , D ) , we also observed rare cells with irregular electron-dense granules characteristic of islet δ-cells ( Figure 5—figure supplement 1E; Klimstra et al . , 2007 ) . Native islet β-cells depolarize and secrete insulin and C-peptide in response to glucose and other physiological or pharmacological stimuli , but reconstructing these hallmark functions in progeny of purified primary human non-β-cells has not been previously achieved during in vitro culture . Compared to baseline secretion in media with 0 . 1 mM glucose , IPCs increased insulin C-peptide secretion by 2 . 4-fold upon exposure to 2 mM glucose ( Figure 5G ) . Similar to insulin release by human islet β-cells ( Lupi et al . , 1999 ) , glucose stimulated the secretion of approximately 4% of total insulin C-peptide in IPCs ( Figure 5—figure supplement 1G ) . This effect was blocked when the cells were incubated with glucose and Diazoxide , a drug that opens KATP channels and prevents glucose-stimulated insulin secretion ( Figure 5G ) . However , unlike adult human islet β-cells , the release of insulin by IPCs was not further increased by 11 mM glucose . Islets from fetal or neonatal stages do not show elevated insulin secretion by high level glucose challenge ( Rozzo et al . , 2009 ) , suggesting that IPCs are similar to immature islet β-cells and that further maturation is possible ( Figure 5G ) . Calcium and voltage-dependent calcium channels are important regulators of normal insulin secretion after KATP channel-mediated membrane depolarization in β-cells ( Henquin , 2005 ) . When calcium was omitted in secretion buffer , C-peptide secretion stimulated by glucose was abolished , but restored upon calcium addition ( Figure 5G ) . Insulin C-peptide release by cultured IPCs was also induced by the depolarizing agent potassium chloride ( 30 mM KCl ) , an effect reversed by a subsequent wash in media with 4 . 8 mM potassium ion ( Figure 5G , figure 5—figure supplement 1H ) . Treatment with tolbutamide , a KATP channel blocker causing membrane depolarization , also stimulated insulin secretion by IPCs , an effect prevented by omission of calcium ( Figure 5G ) . Together with data showing expression of key regulators of stimulus-secretion coupling , these findings provide strong evidence that IPCs produced by conversion and extended culture in our system develop regulated insulin secretion . We examined the stability of the conversion of human ducts into IPCs by long-term transplantation of the converted spheres into specific transplantation sites of NOD scid gamma ( NSG ) mice ( Figure 5—figure supplement 2; Supplementary file 1A ) . Human C-peptide was readily detected in kidney grafts harvested at specific times by immunostaining ( 8/12 cases , Figure 5—figure supplement 2A; Supplementary file 1A ) and by ELISA ( 9/10 cases , Supplementary file 1A ) without detectable tumor formation . This also included C-peptide+ IPCs left in the transplant location beyond 5 months ( Figure 5—figure supplement 2A , d151 ) , suggesting converted IPCs were stable . However , we observed that the total number of grafted C-peptide+ cells was drastically reduced within 2 weeks after transplantation , likely due to the apoptotic cell death . In three independent IPC transplants , however , we were able to detect circulating human insulin in the serum of host mice , and its level increased following intraperitoneal glucose challenge ( Figure 5—figure supplement 2B; Supplementary file 1B ) . Therefore , these data suggest that despite extensive cell death in early stages of transplantation , IPCs can further mature in vivo and release increased levels of insulin in response to acute glucose challenge . Methods to regenerate lost or injured cells in diseases like diabetes mellitus are the focus of intensive investigations ( McKnight et al . , 2010; Benitez et al . , 2012 ) . Generation of insulin-producing cells from human stem cell lines like human ES cells ( D’Amour et al . , 2006; Kroon et al . , 2008 ) is an important , and oft-cited ‘benchmark’ , in efforts to achieve β-cell replacement . However , in these prior reports , progeny of human ES cells developed largely as poly-hormonal cells , most frequently expressing both glucagon and insulin . Moreover such hESC progeny failed to secrete insulin in response to glucose or other secretagogues unless transplanted as progenitors for >2 months in mice ( Nostro and Keller , 2012 ) . This transplant-based maturation strategy was complicated by tumor formation ( Fujikawa et al . , 2005 ) . Thus , it has remained unknown whether human cells can develop solely in vitro to generate glucose-responsive insulin-secreting progeny without tumorigenicity . Our data indicate that in principal this can be achieved , using a small number of genes in sorted human pancreatic ductal cells that convert them toward an islet fate , including progeny that produce , store , and secrete insulin in response to glucose . Conversion of mouse acinar cells into insulin-producing cells using adenoviral delivery of Neurog3 , Pdx1 , and MafA was previously reported ( Zhou et al . , 2008 ) . However , it has remained unknown whether human pancreatic cells can be converted using transgenic methods toward a β-cell fate . We were unable to culture and expand primary human pancreatic acinar cells ( Figure 1B and data not shown ) ; moreover , we found that the combination of these three genes ( MNP ) was insufficient to reprogram primary or expanded human pancreatic ductal cells toward a β-cell fate , suggesting transgenic conversion may be restricted by species and cell type . Thus , we postulated that additional transcriptional regulators might be needed to promote human ductal conversion toward a β-cell fate . Like Neurog3 , MafA , and Pdx1 , the transcription factor Pax6 is expressed in both fetal and adult pancreas , and required to achieve appropriately high levels of Ins and Gcg expression in mouse islet cell development ( Sander et al . , 1997; Wang et al . , 2009 , 2010; Pan and Wright , 2011 ) . Together with the other factors , we found that Pax6 significantly enhanced expression of β-cell markers during ductal reprogramming into β-cells , and was shown as an essential factor for this process . By systematic addition or omission of each transcription factor , we found PDX1 is not required for IPC formation . Thus , unlike mouse acinar cells ( Zhou et al . , 2008 ) and human hepatocytes ( Sapir et al . , 2005 ) , human ductal cells do not require exogenous Pdx1 expression for conversion toward an endocrine fate , for reasons that remain unclear . Our findings are also consistent with recent reports that transgenic adult mouse ductal cells can generate endocrine cells in vivo ( Al-Hasani et al . , 2013 ) . We initially attempted to induce spontaneous differentiation of pancreatic ductal cells using systematic variations of culture conditions , but these efforts proved unsuccessful ( J Lee , unpublished results ) . During pancreas development , Neurog3 level surges in a subset of pancreatic progenitors located in primitive ducts , inducing development of endocrine cell fates ( Zhou et al . , 2007; Miyatsuka et al . , 2009 ) . Therefore , based on this model , we attempted to mimic induction of Neurog3 in human ductal cells using adenoviral overexpression of Neurog3 . We found that Neurog3 was necessary and sufficient for reprogramming human ductal cells , and that the level of ectopic Neurog3 mRNA expressed in ductal cells correlated well with the extent of endocrine reprogramming , including expression of islet hormones ( Figure 3G ) . These findings are reminiscent of studies by Gu et al . showing that reduced Neurog3 gene dosage in mice leads to respecification of pancreatic endocrine progenitors into ductal and acinar cells ( Wang et al . , 2010 ) . Thus , Neurog3 functions may be evolutionarily conserved in allocating cells toward an exocrine or endocrine fate ( whether in development or experimental cell conversion ) in a dosage-dependent manner . Consistent with prior work revealing that Neurog3 attenuates islet cell proliferation ( Miyatsuka et al . , 2011 ) , we did not observe multiple rounds of cell division , an important prerequisite for some de-differentiation events ( Hanna et al . , 2009 ) , during Neurog3-dependent cell conversion . Also , we observed Neurog3 induction alone can rapidly upregulate endocrine molecular signatures in cultured human ductal cells . Thus endocrine cell conversion described here may involve direct conversion of human ductal cells into endocrine cells , rather than de-differentiation , but additional studies are required to assess this possibility . Our findings , albeit with enforced transcription factor expression in adult cells , indicate that Neurog3 expression is sufficient to induce latent endocrine programs in human adult ductal cells , a capacity not yet clearly demonstrated , to our knowledge . We demonstrated robust expansion of purified human ductal cells in 3-dimensional culture . The cells were clonally expanded and serially passaged up to seven generations over 3 months , achieving an increase in cell number calculated to be up to 3 , 200-fold . By contrast , in prior studies , the maximum duration of sustained culture achieved with primary human pancreatic ductal cells was 5 weeks ( Trautmann et al . , 1993; Bonner-Weir et al . , 2000; Rescan et al . , 2005; Hao et al . , 2006; Yatoh et al . , 2007; Hoesli et al . , 2012 ) . Moreover , cultured cells in spheres maintained cardinal features of primary pancreatic ducts such as apical-basal polarity and KRT19 expression up to seven generations ( Figure 2—figure supplement 1D , E ) . Thus , features of our culture system may be useful for studying the genetics and biology of human ductal cells . Prior studies have reported that duct-containing fractions from human adult pancreas can form insulin-producing cells in vitro ( Bonner-Weir et al . , 2000; Hao et al . , 2006; Heremans et al . , 2002; Noguchi et al . , 2006; Koblas et al . , 2008; Swales et al . , 2012 ) or after xeno-transplantion in mice ( Yatoh et al . , 2007 ) . However , the possibility of endocrine cell contamination in the initial ductal fraction or feeder/stromal cells used for co-culture was raised by the detection of mRNAs encoding islet cell hormones and other endocrine markers in these and other studies ( Heremans et al . , 2002; Gao et al . , 2005 ) . Therefore , it remained elusive whether human pancreatic ducts retained the potential to produce islet endocrine cells in adult . In this report , we used FACS to fractionate CD133+ ductal cells and used molecular and immunocytological studies to demonstrate complete elimination of cells expressing markers of differentiated endocrine cells ( including islet hormones ) . Therefore , subsequent conversion of these cells into functional endocrine cells provided unequivocal evidence that endocrine cell-free human adult CD133+ ductal cell fraction can be converted into islet endocrine cells . Centroacinar cells are located at the junction of acini and tip of intercalated ducts ( Cleveland et al . , 2012 ) and their properties remain poorly understood . These cells express CD133 ( Immervoll et al . , 2008 ) , raising the possibility that our fractionated CD133+ cells also include centroacinar cells . Based on their relative paucity in the pancreas , it is unlikely that centroacinar cells are the exclusive source of spheres within this CD133+ fraction , as more than 11% of CD133+ cells were capable of generating spheres ( Figure 1B ) . However , because of difficulties performing lineage-tracing experiments with human samples , we cannot exclude the possibility that centroacinar cells may also contribute to the conversion into endocrine cell lineages . While expression of Pax6 along with Neurog3 , Pdx1 and MafA significantly enhanced expression of INS and other β-cell marker genes in converted ductal cells , this transcription factor combination alone was not sufficient to generate mature IPCs . We found that extending the culture period for 2 weeks after viral infection led to maturation of several hallmark β-cell functions , including expression of key β-cell factors , significant increases of INS mRNA and protein levels , proinsulin processing , dense-core granule formation , and Insulin secretion in response to glucose or other depolarizing stimuli . We tested four distinct culture media with or without serum for this extended culture , and all media permitted maturation of these β-cell properties in IPCs ( Figure 5—figure supplement 1F and see ‘Materials and methods’ ) , indicating that the duration of culture is a key variable for promoting β-cell maturation in vitro . After maturation , the spheres contained an average of 7% total insulin compared to human islet controls , and 7–11% of cells comprising these spheres produced insulin C-peptide . Thus , we calculate that each reprogrammed Insulin+ cell produced between 49 and 77% of insulin levels observed in native β-cell controls , a comparable level to the IPCs derived from human ES cells ( D’Amour et al . , 2006 ) . Is the capacity of human ductal cells to be converted toward endocrine islet fates unique ? A prior study by Sapir et al . ( 2005 ) suggests that human hepatocytes may be induced to express insulin . However , the conversion toward an insulin-producing fate was comparatively poor; resulting cells produced about 10 , 000-fold lower insulin mRNA level than that of human islets , about 3–4 orders of magnitude lower than from conversion of pancreatic duct spheres . In addition , characteristic dense core vesicles in converted hepatocytes were not observed , indicating insufficient conversion towards β-cells . Here , we also assessed the endocrine potential of primary human dermal fibroblasts , cells successfully ‘reprogrammed’ toward many non-fibroblast fates , including induced pluripotent stem cells ( Takahashi et al . , 2007 ) , but detected no clear evidence of conversion toward an endocrine or β-cell fate ( Figure 5—figure supplement 4 , see ‘Materials and methods’ for details ) . Thus , conversion of human adult duct spheres into cells that produce and secrete insulin is singularly robust . Moreover , unlike prior studies of human ES cells that have high variability among ES cell lines used ( D’Amour et al . , 2006; Kroon et al . , 2008 ) , we demonstrated conversion toward insulin+ fates by ductal cells from multiple unrelated donors , another feature of the robustness of our methods . Expression of factors produced from viral transgenes persisted in Insulin+ cells for at least 5 months , evidenced by the GFP expression in transplanted insulin-producing cells ( Figure 5—figure Supplement 2A and 3 ) . The transgenes delivered by adenovirus do not generally persist in dividing cells ( Zhou et al . , 2008 ) . We speculate that cell cycle arrest in Insulin+ cells may be induced by Neurog3 ( Miyatsuka et al . , 2011 ) , thereby preventing dilution of viral transgene-encoded factors . Thus , further studies are needed to investigate how persistent expression of conversion factors like Neurog3 affects maintenance and maturation of endocrine phenotypes in converted cells . Survival of transplanted insulin-secreting cells produced from ductal cells was poor , and reduced yields following transplantation of ductal cells precluded physiological studies in mouse models of diabetes . Promoting survival of transplanted insulin-secreting cells is a general problem for transplant-based islet replacement approaches . Thus , studies of factors that enhance survival of Insulin+ ductal cell progeny are an important current focus . In conclusion , our study provides unique evidence that primary human cells can generate progeny that produce , store and secrete insulin in response to glucose or depolarizing agents , the hallmark features of pancreatic β-cells . We also show that human pancreatic exocrine cells , like in mice ( Zhou et al . , 2008 ) , can be converted by transgenes toward an endocrine islet-like cell fate . We speculate that gene-based strategies like those described here may be combined with other methods , including culture modulation by growth factors and small molecules ( Warren et al . , 2010 ) , to optimize endocrine differentiation or conversion of diverse cellular sources to advance cell replacement for diabetes . We speculate that our cell culture system may also serve as the foundation to investigate the genetics and pathogenesis of diverse human diseases rooted in pancreatic ductal cells , including pancreatitis , cystic fibrosis , and adenocarcinoma . Institutional review board approval for research use of human tissue was obtained from the Stanford University School of Medicine . Human islet-depleted cell fractions were obtained with appropriate consent from healthy , non-diabetic organ donors deceased due to acute traumatic or anoxic death by overnight shipping from the following facilities: Division of Transplantation ( Massachusetts General Hospital , MA ) , UAB Islet Resource Facility ( University of Alabama at Birmingham , AL ) , UCSF Diabetes Center ( University of California , San Francisco , CA ) , Kidney/pancreas transplantation center ( University of Pennsylvania , PA ) , Islet Core of the University of Pittsburgh ( Pittsburgh , PA ) , and Human Islet Isolation Program ( The Hospital of the University of Virginia , VA ) . Donor samples with the age range 16–63 years ( mean 38 . 24 years ) used for this study are listed in Table 1 . On receipt , the cell fractions were washed with PBS and cultured with CMRL media ( Mediatech , Inc , Manassas , VA ) supplemented with 10% heat inactivated fetal bovine serum ( FBS , HyClone , Logan , UT ) , 2 mM GlutaMax ( Life Technologies , Grand Island , NY ) , 2 mM nicotinamide ( prepared in PBS , Sigma , St . Louis , MO ) , and 100 U Penicillin and 100 µg Streptomycin ( Pen/Strep , Life Technologies ) in a non-coated culture dish at 25 . 5°C in 5% CO2 until use . For dissociation , the cell pellet was washed with PBS , trypsinized with 0 . 05% Trypsin-EDTA solution ( Life Technologies ) for 5 min , and quenched with 5 vol of FACS buffer ( 10 mM EGTA , 2% FBS in PBS ) . Cells were collected by centrifugation and further digested in 1 U/ml dispase solution ( Life Technologies ) containing 0 . 1 mg/ml DNaseI in PBS on a nutating mixer at 37°C for 30 min . PBS washing was performed after each enzymatic digestion step . After centrifugation , the cell pellet was resuspended in FACS buffer and passed through a 40-µm-cell strainer . Cell viability and number were assessed using a Vi-Cell analyzer ( Beckman Coulter , Fullerton , CA ) and the samples exceeding 70% cell viability were used for subsequent antibody staining for FACS . Dissociated cells were stained with biotin-conjugated CD133 antibodies ( clone AC133 and 293C3 , Miltenyi Biotec , Auburn , CA ) and then Allophycocyanin-conjugated Streptavidin ( eBioscience , San Diego , CA ) for 15 min , each at room temperature . Cell pellets were collected by centrifugation and washed with PBS after each staining steps . Propidium Iodide ( Life Technologies ) staining was used to exclude dead cells . The cells were sorted using a FACSAria II ( BD Biosciences , Bedford , MA ) and collected in 100% FBS , washed with PBS twice , and resuspended in ice-cold Advanced DMEM/F-12 media ( Life Technologies ) at a density of 8000 cells/µl . The average percentage of CD133+ fraction was 32 . 73% ( n = 32 ) . 50 µl of growth factor-reduced Matrigel ( BD Biosciences ) was then added to 30 µl cell suspension and the mixture was placed around the bottom rim of each well . After solidification at 37°C for 60 min , each well was overlaid with 500 µl of modified crypt culture media ( Sato et al . , 2009 ) comprised of Advanced DMEM/F-12 media supplemented with recombinant human ( rh ) EGF ( 50 ng/ml , Sigma ) , rhR-spondin I ( 500 ng/ml , R&D systems , Minneapolis , MN ) , rhFGF10 ( 50 ng/ml , R&D systems ) , recombinant mouse Noggin ( 100 ng/ml , R&D systems ) , 10 mM Nicotinamide in PBS , and Pen/Strep . The media was changed twice weekly . The spheres were harvested after 2 to 3 weeks for passaging or viral infection . Static and time-lapse images of sphere growth were collected using Zeiss Axiovert 200 inverted microscope and Zeiss Observer . Z1 equipped with a temperature- and CO2-controlled chamber using Axiovision ( Carl Zeiss , Germany ) and MetaMorph ( Molecular Devices , Sunnyvale , CA ) softwares , respectively . For harvesting spheres , 500 µl of 2 U/ml dispase ( Life Technologies ) solution containing 0 . 1 mg/ml DNaseI in PBS was added in each well and the Matrigel was mechanically disrupted by pipetting and incubated at 37°C for 45 min . The released spheres were collected , washed twice with PBS and used for subsequent applications . For passaging spheres , the harvested spheres were trypsinized at 37°C for 5 min followed by quenching with FBS . The dispersed cells were then used for cell counting with a hemocytometer or were plated as described above . Ad-eGFP and Ad-RFP control adenoviruses were purchased from Vector Biolabs ( Philadelphia , PA ) . Ad-MafA and Ad-Neurog3-IRES-eGFP were described previously ( Tashiro et al . , 1999 ) . To construct Ad-RFP-Neurog3 and Ad-RFP-Pdx1 adenoviruses , mouse cDNAs for Neurog3 ( BC104326 ) and Pdx1 ( BC103581 ) were purchased from Open Biosystems ( Lafayette , CO ) and the inserts were obtained by restriction enzyme digestion with EcoR V/BamH I and EcoR V/Msc I , respectively . The inserts were then subcloned into multiple cloning sites of Dual-RFP-CCM shuttle vector ( Vector Biolabs ) and adenoviruses were constructed by Vector Biolabs . For Ad-eGFP-M6P , human MAFA cDNA ( gift from M German ) , PDX1 ( NM_000209; GeneCopoeia , Rockville , MD ) , and PAX6 ( BC011953; Open Biosystems ) were used for PCR amplification with the primers shown in Supplementary file 1C to add T2A , P2A , restriction enzyme sites , and/or tagging proteins ( Figure 5—figure supplement 5 ) . A fused construct of MAFA-T2A-PAX6 was generated by PCR with MAFA and PAX6 PCR amplicons as templates . Similarly , PCR products for PAX6 and PDX1 were used to construct PAX6-P2A-PDX1 . Next , MAFA-T2A-PAX6 , PAX6-P2A-PDX1 , and pDual-GFP-CCM vector ( Vector Biolabs ) were cut with BglII/PstI , PstI/EcoRI , and BglII/EcoRI , respectively , and ligated with NEB quick ligation kit ( New England Biolabs , Ipswich , MA ) followed by transformation of TOP10 chemically competent cells ( Invitrogen , Carlsbad , CA ) . The construct was then used for generating adenoviruses by Vector Biolabs . Spheres were infected at 37°C in suspension overnight at a multiplicity of infection ( MOI ) 100 for Ad-MafA and Ad-eGFP-M6P , or MOI 500 for the rest of viruses used . The spheres were then washed twice with culture medium and embedded in Matrigel as described above . The infected spheres were overlayed with sphere growth media without R-spondin I and with 0 . 33 µM all-trans retinoic acid ( Sigma ) , and cultured for 5 days . For extended culture , the media was replaced with either ( 1 ) DMEM with high glucose ( Life Technologies ) supplemented with 10% FBS ( Hyclone ) and Pen/Strep ( Life Technologies ) for 2 weeks ( referred as ‘DF’ in Figure 3—figure supplement 1F ) , ( 2 ) DF plus 20 mM KCl and 10 µM R0-28-1675 ( glucokinase activator; Axon Ligands ) for 2 weeks ( referred as ‘DFK’ ) , ( 3 ) DF for one week and then DMEM/F-12 media ( Life Technologies ) supplemented with 0 . 5 × N2 supplement ( Life technologies ) , 0 . 5 × B27 ( Life technologies ) , 0 . 2% BSA ( Sigma ) , 1% ITS supplement ( Life Technologies ) , 10 mM nicotinamide , 10 ng/ml recombinant human basic FGF ( R&D systems ) , 50 ng/ml Exendin-4 ( R&D systems ) , recombinant human BMP-4 ( R&D systems ) for additional 1 week ( referred as ‘Z’; Zhang et al . , 2009 ) , or ( 4 ) DMEM high glucose supplemented with 1 × B27 , 55 nM GLP-1 , 50 ng FGF10 ( R&D Systems ) , and Pen/Strep for 3 days followed by 5 days with DMEM high glucose supplemented with 1 × B27 , 55 nM GLP-1 ( Sigma ) , 10 µM DAPT ( Sigma ) , and Pen/Strep , then for 6 days with CMRL1066 media ( Mediatech ) supplemented with 1 × B27 , 55 nM GLP-1 , 50 ng HGF ( R&D Systems ) , 50 ng IGF-1 ( R&D Systems ) , and Pen/Strep ( referred as ‘T’; Thatava et al . , 2011 ) . The media was replaced every other day unless otherwise noted . Total RNA was prepared from sorted cells or cultured spheres with QIAGEN RNeasy micro kit ( QIAGEN Sciences , MD ) , and used for cDNA synthesis using QIAGEN Omniscript RT kit ( QIAGEN ) , according to the manufacturer’s protocol . Relative mRNA level was measured by qRT-PCR of each cDNA in duplicate with gene-specific probe sets ( Applied Biosystems , Foster City , CA ) with TaqMan Universal PCR Master Mix ( Applied Biosystems ) and the ABI Prism 7500 detection system ( Applied Biosystems ) . Normalizations across samples were performed using β-actin primers . Information of the primer and probe sets is available upon request . For immunohistochemical analyses , cultured spheres were harvested , washed with PBS , mixed with 20 µl of Collagen Gel Kit ( Nitta Gelatin , Osaka , Japan ) , solidified at 37°C for 1 hr , fixed with 4% paraformaldehyde for 2 hr at 4°C , cryoprotected in 30% sucrose solution in PBS overnight , embedded in OCT on dry ice , and sectioned in 8 µm thickness . For sorted cells , the cell suspension was washed once and resuspended with 20 µl of PBS , placed on a Polysine slide ( Thermo scientific , Waltham , MA ) , and waited for 30 min at room temperature ( RT ) to let the cells sit on the slide glass by gravity . Then the solution was removed carefully and 40 µl of 4% paraformaldehyde was added . After 10 min of incubation at RT , the fixative was removed and the slides were washed with PBS three times for 5 min each . After removal of PBS , the slides were dried at RT for 1 hr and stored at −20°C . For immunostaining transplanted IPCs , grafted organs ( kidney , EFP , or liver ) were harvested , fixed with 4% paraformaldehyde overnight at 4°C , cryoprotected in 30% sucrose solution in PBS overnight , embedded in OCT on dry ice , and sectioned in 8 µm ( kidney and liver ) or 40 µm ( EFP ) thickness . The primary antibodies used were rabbit anti-Amylase ( 1:1000; Sigma ) , goat anti-Amylase ( sc-12821; 1:200; Santa Cruz Biotechnology , Dallas , TX ) , CD133 ( 1:100 each; clone AC133 and 293C3; Miltenyi Biotec , Auburn , CA ) , rabbit anti-ChromograninA ( 20085; 1:100; Immunostar , Hudson , WI ) , mouse anti-ChromograninA ( LK2H10; 1:200; Cell Marque , Rocklin , CA ) , mouse anti-CK19 ( KRT19 ) ( M0888; 1:200; DAKO , Carpinteria , CA ) , rabbit anti-CK19 ( 319R-15; 1:200; Cell Marque ) , rabbit anti-CPA1 ( 1810-0006; 1:100; AbD Serotec , UK ) , rabbit anti-C-peptide ( #4593B; 1:200; Cell Signaling Technology , Danvers , MA ) , mouse anti-C-peptide ( capt ) ( 1:100; Mercodia , Sweden ) , mouse anti-Flag ( F1804; 1:1000; Sigma ) , goat anti-GHRL ( sc-10368; 1:200; Santa Cruz Biotechnology ) , guinea pig anti-Glucagon ( 4031-01; 1:200; Linco , Billerica , MA ) , mouse anti-HA ( MMS-101P-1000; 1:1000; Covance ) , mouse anti-HuNu ( MAB1281; 1:200; Millipore , Billerica , MA ) , mouse anti-IAPP ( MCA1126T; 1:200; AbD serotec ) , rabbit anti-Ki-67 ( NCL-Ki67p; 1:100 , Leica Microsystems , Germany ) , rabbit anti-Myc ( sc-789; 1:1000; Santa Cruz Biotechnology ) , mouse anti-NeuroD ( sc-46684; 1:10; Santa Cruz Biotechnology ) , mouse anti-Neurog3 ( F25A1B3; 1:4000; DSHB , Iowa City , IA ) , mouse anti-Nkx2 . 2 ( 74 . 5A5; 1:10; DSHB ) , mouse anti-Nkx6 . 1 ( F55A10; 1:200; DSHB ) , rabbit anti-PC1/3 ( PCSK1 , AB10553; 1:200; Millipore ) , rabbit anti-phospho-H3 ( 06-570; 1:500; Millipore ) , goat anti-PPY ( NB100-1793; 1:200; Novus Biologicals , Littleton , CO ) , rabbit anti-Somatostatin ( 1:200 , DAKO ) , goat anti-Somatostatin ( sc-7819; 1:200; Santa Cruz Biotechnology ) , goat anti-SUR-1 ( sc-5789; 1:50; Santa Cruz Biotechnology ) . Tyramide signal amplification ( Perkin Elmer , Waltham , MA ) was used for antibodies against Neurog3 , NeuroD , Nkx2 . 2 , Nkx6 . 1 , and PC1/3 . Antigen unmasking ( H-3300; Antigen Unmasking Solution , Citric Acid Based , Vector Laboratories , Burlingame , CA ) was performed for anti-Flag antibody staining . The Neurog3 , Nkx2 . 2 , and Nkx6 . 1 antibodies developed by Dr OD Madsen were obtained from the Developmental Studies Hybridoma Bank ( DSHB ) developed under the auspices of the NICHD and maintained by The University of Iowa , Department of Biological Sciences , Iowa City , IA 52242 . Secondary antibodies used were from Jackson ImmunoResearch ( West Grove , PA ) or Molecular Probes ( Eugene , OR ) . Stained sections were mounted with VECTASHIELD Mounting Medium with DAPI ( Vector Laboratories ) . Fluorescence images were taken using Zeiss Axio Imager . M1 or Leica SP2 inverted confocal laser scanning microscope . The samples were fixed in Karnovsky’s fixative: 2% Glutaraldehyde ( EMS Cat# 16000 ) and 4% Paraformaldehyde ( EMS; Electron Microscopy Sciences , Hatfield , PA ) in 0 . 1 M Sodium Cacodylate ( EMS ) pH 7 . 4 for 1 hr at RT then cut , post fixed in 1% Osmium tetroxide ( EMS ) for 1 hr at RT , washed three times with ultrafiltered water , then en bloc stained for 2 hr at RT or moved to 4°C overnight . The samples were then dehydrated in a series of ethanol washes for 15 min each at 4°C beginning at 50% , 70% , 95% , where the samples are then allowed to rise to RT , changed to 100% two times , followed by Acetonitrile for 15 min . The samples are infiltrated with EMbed-812 resin ( EMS ) mixed 1:1 with Acetonitrile for 2 hr followed by two parts EMbed-812 to 1 part Acetonitrile for 2 hr . The samples were then placed into EMbed-812 for 2 hr and then placed into molds , and resin filled gelatin capsules with labels were orientated over the cells of interest and placed into 65°C oven overnight . Sections were taken between 75 and 90 nm on a Leica Ultracut S ( Leica , Wetzlar , Germany ) , picked up on formvar/Carbon coated slot grids ( EMS Cat#FCF2010-Cu ) or 100 mesh Cu grids ( EMS ) . Grids were contrast stained for 15 min in 1:1 saturated UrAcetate ( ∼7 . 7% ) to 100% ethanol followed by staining in 0 . 2% lead citrate for 3 to 4 min . JEOL JEM-1400 TEM was used to observe at 120 kV and photos were taken using a Gatan Orius digital camera . C-peptide secretion assay and content measurement were performed as described previously with minor modification ( Chen et al . , 2001 ) . Briefly for secretion assay , media was replaced a day before assay was performed . On the day , each well with matrigel-embedded spheres was incubated with fresh media for 2 hr , washed twice with plain Krebs-Ringer bicarbonate buffer ( KRBB ) , and incubated twice with plain KRBB for 1 hr each for thorough washing . Next , the spheres were incubated consecutively with 400 µl KRBB containing indicated concentrations of glucose ( Sigma ) with or without 0 . 5 mM Diazoxide ( Sigma ) , KCl ( 30 mM , Sigma ) , or Tolbutamide ( 0 . 2 mM , Sigma ) for 2 hr each . KRBB without Calcium ( No Ca++ ) was prepared by omission of CaCl2 and addition of 1 mM EGTA ( Sigma ) . Secreted C-peptide level was measured with Human Ultrasensitive C-peptide ELISA kit ( Mercodia ) . For C-peptide content measurement , the spheres were harvested in 1 . 5 ml microfuge tube , washed with PBS , resuspended with 300 µl of ice-cold TE/BSA buffer ( 10 mM Tris-HCl , 1 mM EDTA , 0 . 1% wt/vol BSA , pH 7 . 0 ) , and sonicated with Bioruptor Sonicator ( Diagenode , Denville , NJ ) . Half of the lysate was used for genomic DNA isolation and quantification with Quant-iT PicoGreen dsDNA Assay Kit ( Invitrogen ) . Same volume of acid alcohol ( 75% vol/vol ethanol , 2% vol/vol concentrated HCl , 23% vol/vol H2O ) was added to the rest of lysate to extract C-peptide by rocking overnight at 4°C . The extract was then neutralized with 10 vol of PBS and used for C-peptide ELISA . Transplantation in kidney capsule , epididymal fat pad ( EFP ) , or in the liver by portal vein injection was performed as previously described ( Kroon et al . , 2008; Alipio et al . , 2010; Wang et al . , 2011 ) . For transplantation in kidney or EFP , converted spheres with or without extended culture were harvested and mixed with or without mouse embryonic fibroblasts ( Supplementary file 1A ) . The spheres were then mixed with matrigel to make a final volume of 10 µl for kidney transplantation or overlayed on pre-wet gelfoam for EFP transplantation . For liver transplantation , single cells produced by trypsinization of harvested spheres were resuspended in 100 µl PBS and injected into the portal vein with a 27 G needle . All animal experiments and methods were approved by the Institutional Animal Care and Use Committee ( IACUC ) of Stanford University . Secretion of human Insulin or C-peptide by glucose injection was measured as previously described ( Kroon et al . , 2008 ) . Briefly , transplanted mice were fasted overnight ( 14–16 hr ) and 120 µl of blood was collected from tail into Microvette CB300LH ( Sarstedt , Germany ) to prepare 50 µl of serum . 3 g/kg glucose was then injected and blood was collected again 30 min after glucose administration . Secreted C-peptide or insulin level was measured with Human Ultrasensitive C-peptide or Insulin ELISA kits ( Mercodia ) . Human adult dermal fibroblasts ( Coriell Institute for Medical Research , Camden , New Jersey , USA ) were cultured and maintained as described previously ( Yoo et al . , 2011 ) . The cells were either trypsinized for suspension infection ( as was described above for ductal spheres ) or infected as adherent cells in six-well plates by direct addition of virus into the culture medium , with Ad-eGFP ( GFP ) or Ad-eGFP-M6P and Ad-Neurog3-IRES-eGFP ( 4TFM ) . The same MOIs used for ductal sphere infection were also used . The suspension-infected cells were harvested the following day and embedded in Matrigel as described above for infected ductal spheres . The culture was maintained for additional 18 days to match the duration of infected ductal sphere maturation . The infected adherent cells were cultured with virus for 48 hr and the media was replaced . The culture was maintained for additional 10 days , passaged in 1:3 ratio due to confluency , re-plated , and cultured additional 7 days to match the duration of infected ductal sphere maturation . In both cases , media was replaced every other day . Three independent experiments were performed for both conditions and each experiment at least in duplicates . RNA isolation , cDNA preparation , and qRT-PCR were performed with primers specific to human INS , CHGA , and β-actin as described above .	Diabetes mellitus is a disease that can lead to dangerously high blood sugar levels , causing numerous complications such as heart disease , glaucoma , skin disorders , kidney disease , and nerve damage . In healthy individuals , beta cells in the pancreas produce a hormone called insulin , which stimulates cells in the liver , muscles and fat to take up glucose from the blood . However , this process is disrupted in people with diabetes , who either have too few pancreatic beta cells ( type 1 diabetes ) or do not respond appropriately to insulin ( type 2 diabetes ) . All patients with type 1 diabetes , and some with type 2 , must inject themselves regularly with insulin , but this does not always fully control the disease . Some type 1 patients have been successfully treated with beta cells transplanted from deceased donors , but there are not enough donor organs available for this to become routine . Thus , intensive efforts worldwide are focused on generating insulin-producing cells in the lab from human stem cells . However , the cells produced in this way can give rise to tumors . Now , Lee et al . have shown that duct cells , which make up about 30% of the human pancreas , can be converted into cells capable of producing and secreting insulin . Ductal cells obtained from donor pancreases were first separated from the remaining tissue and grown in cell culture . Viruses were then used to introduce genes that reprogrammed the ductal cells so that they acquired the ability to make , process and store insulin , and to release it in response to glucose—hallmark features of functional beta cells . As well as providing a potential source of cells for use in transplant or cell conversion therapies for diabetes , the ability to grow and maintain human pancreatic ductal cells in culture may make it easier to study other diseases that affect the pancreas , including pancreatitis , cystic fibrosis , and adenocarcinoma .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2013	Expansion and conversion of human pancreatic ductal cells into insulin-secreting endocrine cells
Primary sensory neurons are generally considered the only source of dorsal horn calcitonin gene-related peptide ( CGRP ) , a neuropeptide critical to the transmission of pain messages . Using a tamoxifen-inducible CalcaCreER transgenic mouse , here we identified a distinct population of CGRP-expressing excitatory interneurons in lamina III of the spinal cord dorsal horn and trigeminal nucleus caudalis . These interneurons have spine-laden , dorsally directed , dendrites , and ventrally directed axons . As under resting conditions , CGRP interneurons are under tonic inhibitory control , neither innocuous nor noxious stimulation provoked significant Fos expression in these neurons . However , synchronous , electrical non-nociceptive Aβ primary afferent stimulation of dorsal roots depolarized the CGRP interneurons , consistent with their receipt of a VGLUT1 innervation . On the other hand , chemogenetic activation of the neurons produced a mechanical hypersensitivity in response to von Frey stimulation , whereas their caspase-mediated ablation led to mechanical hyposensitivity . Finally , after partial peripheral nerve injury , innocuous stimulation ( brush ) induced significant Fos expression in the CGRP interneurons . These findings suggest that CGRP interneurons become hyperexcitable and contribute either to ascending circuits originating in deep dorsal horn or to the reflex circuits in baseline conditions , but not in the setting of nerve injury . Calcitonin gene-related peptide ( CGRP ) is the most prominent molecular marker of the peptidergic subpopulation of primary afferent nociceptors ( Basbaum et al . , 2009 ) . When released from peripheral terminals of sensory neurons , CGRP acts on endothelial cells that line blood vessels , producing pronounced vasodilation ( Brain et al . , 1985 ) . Recent efforts to develop novel therapeutics in the management of migraine led to the successful development of antibodies that scavenge CGRP , reducing the vasodilation that triggers migraine ( Ho et al . , 2010 ) . When released into the superficial dorsal horn from the central branches of sensory neurons , CGRP , along with its co-occurring neuropeptide , substance P , potentiates the glutamatergic excitation of postsynaptic neurons , contributing to injury-provoked central sensitization ( Ryu et al . , 1988; Woolf and Wiesenfeld-Hallin , 1986 ) . The latter process , in turn , contributes to the ongoing pain and profound hypersensitivity characteristic of both inflammatory and neuropathic pains . Interestingly , a recent study showed that pharmacological inhibition of CGRP receptor signaling in the periphery alleviates incision-induced mechanical and heat hypersensitivity , but not neuropathic pain , suggesting that primary sensory neuron-derived CGRP differentially influences injury-induced persistent pain ( Cowie et al . , 2018 ) . Despite much earlier reports , which used colchicine to enhance somatic CGRP levels ( Kruger et al . , 1988; Tie-Jun et al . , 2001 ) and a more recent report ( McCoy et al . , 2012 ) of small CGRP-positive cells in the dorsal horn of a reporter mouse , the prevailing view is that dorsal horn CGRP derives exclusively from afferents . Here , we took advantage of a tamoxifen-inducible CalcaCreER mouse line , which when crossed with a tdTomato reporter mouse , reveals a discrete population of CGRP-expressing interneurons that are concentrated in lamina III and inner lamina II of the spinal cord dorsal horn and trigeminal nucleus caudalis . Unlike dorsal horn vertical cells , which have ventrally directed dendrites and a dorsally directed axon , the CGRP interneurons have mainly dorsally directed dendrites and ventrally directed axons . A comprehensive functional analysis showed that these interneurons are minimally responsive to a host of acute , innocuous or noxious mechanical and chemical stimuli , despite the fact that electrical stimulation of Aβ afferents readily activates the cells . On the other hand , an innocuous mechanical stimulus evoked significant Fos expression in the setting of peripheral nerve injury and chemogenetic activation of the interneurons produced clear mechanical hypersensitivity . Conversely , caspase-mediated ablation of the neurons increased mechanical thresholds . We conclude that these CGRP-expressing interneurons engage deep dorsal horn nociresponsive circuits that contribute either to ascending circuits originating in deep dorsal horn or to the reflex circuits in baseline conditions , but not in the setting of nerve injury . We next asked whether these CGRP-expressing neurons include both projection and interneurons . First , we injected the retrograde tracer Fluorogold ( 1% ) into several brain areas that receive projections from the spinal cord dorsal horn . Despite an extensive analysis , which included injections into the ventrobasal and nucleus submedius ( Yoshida et al . , 1991 ) of the thalamus , lateral parabrachial nucleus ( see Figure 5—figure supplement 2b–c for injection site in the parabrachial nucleus ) , and dorsal column nuclei , which are targeted by postsynaptic dorsal column neurons located in the region of lamina IV of the dorsal horn , we found no evidence of CGRP-expressing projection neurons . This finding was confirmed with an anterograde-tracing approach in which we injected an AAV1-flex-GCaMP6s virus unilaterally into the nucleus caudalis of CalcaCreER/tdTomato mice ( Figure 5—figure supplement 2a ) . After 4 weeks , we examined the brainstem , thalamus , and hypothalamus for GFP-labeled fibers , but found no evidence of long-distance axonal projections deriving from the lamina III CGRP cells . By immunolabeling the CGRP-tdTomato neurons , we next determined that these cells are excitatory and define a unique subset of interneurons . First , the CGRP-tdTomato cells co-express Lmx1b ( 98%; 92/94 tdTomato cells ) , but not Pax2 ( Figure 2—figure supplement 1 ) , which are excitatory and inhibitory markers , respectively . Some of the CGRP-tdTomato cells populate inner lamina II , and here approximately 16% co-expressed PKCγ ( 31/187 tdTomato cells ) , a marker of a large population of excitatory interneurons ( Malmberg et al . , 1997 ) . Sixty-three ( 97/158 tdTomato cells ) and 9% ( 9/97 tdTomato cells ) of the CGRP interneurons co-expressed calbindin and calretinin , respectively , calcium binding proteins that mark subpopulations of excitatory dorsal horn interneurons ( Figure 2 ) . The incomplete immunohistochemical overlap with major neurochemical classes of dorsal horn interneurons indicates that the CGRP interneurons are heterogeneous consistent with previously described populations of dorsal horn neurons . However , as there is a limited number of quality antibodies that can be used for comprehensive neurochemical profiling we turned to in situ hybridization ( Figure 3 ) . Consistent with the concentration of tdTomato-CGRP interneurons in lamina III , particularly notable is that 56% of the Calca mRNA-expressing ( CGRP ) cells double-labeled for Rora message ( 639/1134 Calca mRNA-expressing cells ) , a marker of excitatory interneurons in lamina III ( Bourane et al . , 2015b ) . Interestingly , however , only 4% co-expressed Cck ( 27/595 Calca mRNA-expressing cells ) , which marks a significant subset of the RORα population ( Liu et al . , 2018 ) . As for other populations of excitatory interneurons , we found minimal overlap with the population that transiently expresses VGLUT3 ( examined at P7 ) ( Peirs et al . , 2015 ) or others that express Nptx2 , BDNF or the NK1 receptor , a marker of many projection neurons . Similar results were found in the dorsal horn of the spinal cord and in the trigeminal nucleus caudalis ( viz . , dorsal horn of the medulla ) . We conclude that a substantial portion of the CGRP interneuron population overlaps with a subset of the Cck-negative RORα population of lamina III interneurons . Despite the very intense tdTomato labeling of the cell bodies of the dorsal horn neurons , it was difficult to distinguish axonal processes from the dense primary sensory neuron-derived CGRP innervation . This was particularly the case when an antibody to tdTomato was used to detect the dorsal horn CGRP neurons . And unfortunately , although the cell body of the intracellularly recorded cells was readily filled with biotin dextran in electrophysiological slice preparations ( see below ) , we never successfully filled dendrites or axons . Therefore , in a separate set of experiments , we first reduced the complement of primary afferent-derived CGRP-derived by making an intrathecal injection of capsaicin , 7 days prior to perfusing the mice ( Cavanaugh et al . , 2009 ) . In addition , tdTomato-immunoreactivity was revealed with immunoperoxidase staining so that sections could be analyzed by either light or electron microscopy ( EM ) . The results from this approach were both striking and especially informative . Figure 4 illustrates that the CGRP interneurons have many dorsally-directed , spine-laden dendrites . These dendritic arbors often penetrated lamina II , and some labeled processes appeared to reach lamina I . Nevertheless , despite the capsaicin treatment , the latter were rare and difficult to distinguish from residual primary afferent-derived CGRP . Based on their remarkably uniform dendritic morphology , the dorsal horn CGRP neurons appear to represent a subpopulation of excitatory , so-called radial interneurons ( Grudt and Perl , 2002 ) ; however , the morphology of the CGRP-expressing radial interneurons differ considerably from those previously described in lamina II . First , the majority of lamina II radial cells have dendrites that arborize ventrally and axons that , if anything , project and collateralize dorsally , occasionally targeting presumptive projection neurons in lamina I . In contrast , not only do the CGRP interneurons have dorsally-directed dendrites , but almost all of their axons project ventrally and/or ventrocaudally . In some instances , we could trace the axons well into the neck of the dorsal horn , including lamina V ( Figure 5 and Figure 5—figure supplement 1 ) . Furthermore , EM analysis of these interneurons ( Figure 5 ) illustrates that there is significant synaptic input to the soma , dendrites , and spines of the CGRP interneurons . Finally , given the concentration of the CGRP interneurons in lamina III , we assumed that they receive primary afferent input from large myelinated afferents . Indeed when we double-immunostained for tdTomato and VGLUT1 , a glutamate transporter that is highly expressed in large myelinated afferents ( Oliveira et al . , 2003 ) , we observed many close appositions of VGLUT1-immunoreactive axon terminals onto the cell bodies and dendrites of the CGRP interneurons ( Figure 4f–j ) . To confirm that the VGLUT1 appositions indeed mark a monosynaptic input from Aβ afferents to the CGRP-tdTomato interneurons , we prepared transverse lumbar and caudal medullary slices ( 350–400 µm ) from 3-week-old mice for whole-cell patch-clamp recordings . The slices contained large numbers of fluorescent tdTomato-labeled CGRP neurons ( Figure 6a–c ) . We first characterized the intrinsic properties of the CGRP-tdTomato neurons by inducing depolarizing current steps . The CGRP-tdTomato neurons in the dorsal horn and nucleus caudalis showed mostly delayed firing patterns , consistent with their excitatory and radial phenotype ( delayed 19 , tonic 1 , reluctant 2 , single 2 , no response 3 , Figure 6—source data 1 table ) . In some preparations we stimulated an attached dorsal root . At near threshold stimulation intensities ( 10 Hz ) , we recorded a very short latency component , which likely corresponds to a monosynaptic Aβ-fiber input . Of five cells recorded in three mice , all received monosynaptic Aβ input . In two additional mice , we recorded from four cells that responded to dorsal root stimulation , but we could not unequivocally establish whether they received a monosynaptic input ( Figure 6d–e ) . Overall , the intrinsic properties of neurons recorded from lumbar dorsal horn ( 22 cells , eight mice ) and nucleus caudalis ( 5 cells , two mice ) were comparable ( see Figure 6—source data 1 table ) . Taken together , we conclude that the predominant ( monosynaptic ) input to the CGRP interneurons derives from low threshold ( Aβ ) mechanoreceptors . In a separate set of experiments , we specifically sought evidence that the neurons , under baseline conditions , are under inhibitory control . To this end , cells were patched and then rheobase determined , before and after application of a combination of bicuculline and strychnine . Figure 7a–d illustrate that concurrent blocking of the GABA and glycine receptors significantly reduced rheobase , from 46 . 0 ± 7 . 4 pA before antagonist treatment to 31 . 0 ± 4 . 9 pA after antagonist treatment ( two-tailed , paired T-test; p=0 . 0005 , n = 25 ) . Of the 25 neurons studied , rheobase decreased in 21 , increased in one and did not change in 3 . Figure 7c shows that application of bicuculline and strychnine to neurons in which current was maintained 10 pA below rheobase also generated action potentials . Figure 7e shows that resting membrane potential also showed a significant depolarization after application of the GABA and glycine receptor antagonists , from −53 . 8 ± 1 . 8 mV before antagonist treatment to −49 . 1 ± 1 . 7 mV after antagonist . Taken together , these results demonstrate that the CGRP interneurons , under resting conditions , are under tonic inhibitory control . To provide a global activity measure of the stimuli that engage the CGRP interneurons , we first monitored Fos expression using a battery of noxious and innocuous stimuli . As expected , a unilateral injection of dilute formalin into the cheek ( 10 µl of 2% formalin , Figure 8—figure supplement 1c ) or a unilateral hindpaw injection of capsaicin ( Figure 8—figure supplement 2a–b ) , produced considerable Fos immunolabeling of dorsal horn neurons , but not of the CGRP-tdTomato interneurons ( Figure 8—figure supplements 1c and 2a–b ) . Unexpectedly , however , selectively engaging non-nociceptive afferents by having the animal walk for 90 min on a rotarod , which provokes considerable Fos in laminae III and IV ( Neumann et al . , 2008 ) , did not induce Fos expression in the CGRP interneurons ( Figure 8—figure supplement 1a ) . The same was true for brushing of the cheek , another innocuous stimulus that activates Aβ afferents ( Figure 8 ) . Finally , although CGRP is strongly implicated in the generation of migraine , largely but not exclusively via its peripheral vasodilatory action ( Brain et al . , 1985 ) , systemic injection of nitroglycerin , which triggers migraine in humans and profound mechanical hypersensitivity in animals ( Bates et al . , 2010 ) , did not induce Fos in the CGRP interneurons ( Figure 8—figure supplement 1b ) . We conclude that despite our electrophysiological evidence that Aβ afferents engage the CGRP interneurons , there does not appear to be sufficient input to activate these cells under natural innocuous mechanical stimulus conditions in uninjured mice ( 5 . 3% , 5/88 tdTomato cells; Figure 8a ) . We , therefore , next asked whether an injury state would render the CGRP interneurons more responsive to an innocuous stimulus . In fact six days after inducing the spared nerve injury ( SNI ) model of neuropathic pain , we found that brushing the ipsilateral paw evoked Fos expression in 50% ( 53/110 tdTomato cells ) of the dorsal horn CGRP interneurons ( Figure 8c and d ) . Importantly , although we recorded significant dorsal horn Fos expression in nerve-injured mice without brushing ( Figure 8b ) , no Fos expression occurred in the CGRP interneurons ( 3%; 6/205 tdTomato cells ) . We conclude that activation of the CGRP interneurons only occurs when the innocuous input , which could include contact of the plantar surface of the paw with the ground ( Liu et al . , 2018 ) , engages the interneurons in the setting of nerve injury . As electrical stimulation of the dorsal root at Aβ intensity readily excites the CGRP interneurons , the inability of brush stimulation to activate the neurons in the absence of injury was surprising . The discrepancy may reflect the fact that dorsal root stimulation involves a synchronous activation of many primary sensory neurons . In contrast , natural stimuli ( e . g . brushing or walking on a rotarod ) trigger an asynchronous afferent drive . However , as brushing was effective in the nerve injury setting , we hypothesized that a central sensitization rendered the CGRP neurons hyperexcitable . To test this hypothesis , we asked whether a different mode of activation , namely chemogenetic ( direct ) activation of the CGRP interneurons , could generate behaviors indicative of mechanical allodynia , comparable to what is observed in response to innocuous mechanical stimuli in the setting of nerve injury . In these studies , we used an intersectional approach to target expression of a Designer Receptor Exclusively Activated by Designer Drugs ( DREADD ) selectively in the CGRP interneurons . To this end , we crossed the CalcacreER mice to a FLPo mouse line , driven by the Lbx1 gene . The latter gene is only expressed in neurons of dorsal spinal cord and hindbrain , but not in sensory neurons of the DRG ( Bourane et al . , 2015b ) . We then made a unilateral microinjection of an adenoassociated virus ( AAV ) expressing a Cre and FLPo-dependent DREADD ( hM3Dq ) into the dorsal horn of the CalcaCreER/FLPo mice . Four weeks later , we evaluated the behavioral effects of a systemic injection of CNO , which activates the DREADD . We first established that there was no constitutive effect of virus infection . Thus , CNO injection , compared to saline , did not alter the latency to fall from an accelerating rotarod ( Figure 9c ) . Furthermore , baseline von Frey mechanical thresholds of the DREADD-expressing mice , measured prior to injection of CNO , did not differ from mice injected with the AAV-GFP virus . In distinct contrast , Figure 9 shows that CNO injection in the experimental group produced a significant reduction of von Frey threshold of the ipsilateral hindpaw , compared to baseline or to saline-injected mice ( Figure 9b ) . Mechanical thresholds did not change from baseline in the AAV-GFP control animals , whether they received saline or CNO ( Repeated Measures Two-way ANOVA , F ( 1 , 20 ) =6 . 964 , p=0 . 012 , interaction effect between DREADD group and CNO treatment ) . The groups contained the same numbers of males and females ( DREADD animals: 8 of each; GFP controls: 3 of each ) , but there was no significant interaction between sex and treatment ( CNO versus saline ) . Nor did factoring in sex reduce the error ( R2 ) in the full Repeated Measures Two-way ANOVA . Consistent with a contribution of CGRP-expressing interneurons to mechanical sensitivity , mice in which we ablated selectively the CGRP-expressing spinal cord interneurons with a virally derived caspase ( Figure 10a and Figure 10—figure supplement 1 ) exhibited significantly higher mechanical thresholds than did control mice ( Figure 10b ) . On the other hand , and somewhat unexpectedly , the mechanical hypersensitivity produced in the SNI model of neuropathic pain was not altered by the ablation . Lastly , we evaluated heat and cold responsiveness after CNO injection . Neither latency to withdraw the hindpaw to noxious heat in the Hargreaves test ( n = 16; Figure 9d ) nor time spent paw lifting after exposure of the plantar surface of the hindpaw to a cold ( acetone ) stimulus ( n = 11; Figure 9e ) , differed when comparing CNO and control saline injection ( p>0 . 05 , Students T-test and Wilcoxon Signed Ranks Test , respectively ) . Similarly , responses to noxious heat ( Hargreaves; Figure 10c ) or cold ( acetone; Figure 10d ) stimuli were unchanged after caspase-mediated ablation of the CGRP-expressing interneurons . We conclude that direct and likely synchronous activation of the CGRP interneurons produces a selective mechanical hypersensitivity , mimicking the mechanical allodynia observed in response to low threshold ( Aβ ) mechanical stimulation ( brush ) in the setting of nerve injury . Based on their single cell transcriptome analysis , Häring and colleagues ( Häring et al . , 2018 ) concluded that several populations of dorsal horn excitatory neurons that express Calca mRNA co-express gastrin-releasing peptide ( GRP ) , a peptide linked to dorsal horn circuits that drive itch-provoked scratching ( Albisetti et al . , 2019; Sun and Chen , 2007 ) . To confirm this , we performed double in situ hybridization for Calca and Grp . Although the Grp interneurons predominated in a band just dorsal to the Calca interneurons , consistent with our previous report ( Solorzano et al . , 2015 ) , we did find several instances of co-localization of Calca mRNA and Grp mRNA . Interestingly , however , when using immunohistochemistry , we found almost no overlap of GRP and CGRP in a double transgenic GRP-GFP/CGRP-tdTomato mouse line ( Figure 11a–d ) . This difference is likely related to the fact that neurons labeled in the reporter mouse constitute less than half of the Grp mRNA-positive population ( Solorzano et al . , 2015; Dickie et al . , 2019 ) . Despite these discordant findings , we also examined the pattern of Fos expression provoked by injection of chloroquine ( CQ ) , a strong pruritogen , into the cheek or hindpaw . To prevent scratching-induced Fos , the CQ injections were performed in anesthetized mice . As Figure 11e–f illustrates , despite considerable chloroquine-induced Fos expression , we found only an occasional double-labeled neuron . Furthermore , the number of scratching bouts induced by a subcutaneous calf injection of 100 μg chloroquine was comparable between control and CGRP-ablated mice ( Figure 10e ) . We conclude that the CGRP interneurons , despite some overlap with GRP , likely do not transmit chemical itch , a finding consistent with the effects of deleting RORα ( Bourane et al . , 2015b ) . Whether the CGRP interneurons are engaged in conditions in which mechanical stimulation can trigger itch ( alloknesis ) remains to be determined . Despite overwhelming evidence that primary sensory neurons are the predominant source of dorsal horn CGRP , here we describe a morphologically uniform population of dorsal horn CGRP-expressing interneurons . Many of these interneurons correspond to the Cck-negative subset of the RORα population in lamina III of the dorsal horn and trigeminal nucleus caudalis , are excitatory and are activated by electrical stimulation of non-nociceptive , Aβ primary afferents . In contrast to the Cck-expressing subset of RORα neurons , and despite their location in the so-called , low threshold mechanoreceptive recipient zone of the dorsal horn ( Abraira et al . , 2017 ) , the CGRP interneurons do not express Fos in response to natural Aβ-mediated , innocuous mechanical stimulation ( brushing or walking on a rotarod ) . We hypothesize that this reflects competition with the ongoing inhibition of these neurons ( see below ) . As for the RORα population , the CGRP interneurons do not respond to noxious chemical stimulation . Even peripheral nerve injury , without superimposed stimulation , did not activate these neurons . On the other hand , brush stimulation in the nerve injury setting did activate the CGRP interneurons . Furthermore , and consistent with a limited contribution of these neurons in the setting of nerve injury , ablation of CGRP interneurons did not influence the magnitude of mechanical allodynia that develop following peripheral nerve injury . Interestingly , however , brush stimulation in the nerve injury setting did induce Fos in the CGRP interneurons . This distinction suggests that unless these neurons are rendered hyperexcitable , as occurs after nerve injury , only synchronous afferent input or direct neuronal sensitization ( e . g . by DREADD activation ) is sufficient to engage the circuits in which the CGRP interneurons participate . Consistent with this conclusion , chemically provoked ( chemogenetic ) synchronous activation of these neurons produced a significant mechanical hypersensitivity and conversely their ablation increased mechanical thresholds . Based on the predominant ventrally directed axonal arbors of these interneurons we suggest that the dorsal horn CGRP interneurons contribute either to ascending circuits originating in deep dorsal horn or to the reflex circuits in baseline conditions , but not in the setting of nerve injury . The fact that nerve injury-induced mechanical hypersensitivity persisted after ablation of the CGRP interneurons undoubtedly reflects the major contribution of other mechanosensitive afferents and dorsal horn interneurons . Indeed , we previously reported that the MrgprD subpopulation of sensory neurons is an important driver of the nerve-injury induced mechanical sensitivity ( Cavanaugh et al . , 2009 ) and these afferents target interneurons located dorsal to the predominant band of CGRP interneurons . RNA-Seq analyses have now defined at least 15 subsets of excitatory interneurons and 15 subsets of inhibitory neurons in the dorsal horn of the spinal cord ( Häring et al . , 2018; Sathyamurthy et al . , 2018 ) . Ablation , optogenetic and chemogenetic studies further characterized those classes based on functional properties . Of note , an increasing number of dorsal horn interneurons that ‘gate’ mechanical pain have been identified . These include neurochemically distinct excitatory interneuron populations: transient VGLUT3 , somatostatin , RORα , calretinin , and Tac1 ( Bourane et al . , 2015a; Cheng et al . , 2017; Duan et al . , 2014; Huang et al . , 2019; Peirs et al . , 2015; Petitjean et al . , 2019 ) and distinct inhibitory interneuron populations: dynorphin , calretinin , parvalbumin , and enkephalin ( Boyle et al . , 2019; Duan et al . , 2014; François et al . , 2017; Petitjean et al . , 2019; Petitjean et al . , 2015 ) . The CGRP-expressing interneurons define yet another population of dorsal horn interneurons that contributes to spinal cord processing of mechanical inputs . Interestingly , there is a striking laminar organization of these molecularly distinct populations of interneurons . For example , the transiently expressing VGLUT3 population is located ventral to the CGRP interneurons , receives low-threshold mechanoreceptive input and their chemogenetic activation also enhances mechanical sensitivity ( Cheng et al . , 2017; Peirs et al . , 2015 ) . Dorsal to the CGRP interneuron are PKCγ and calretinin excitatory interneurons that contribute to nerve injury induced mechanical allodynia ( Malmberg et al . , 1997; Neumann et al . , 2008; Peirs et al . , 2015; Petitjean et al . , 2019; Smith et al . , 2019 ) . To what extent these mechanically driven neuronal populations are interconnected or whether they represent parallel , independent circuits activated under different mechanical pain conditions ( e . g . naive vs injury vs inflammation ) remains to be determined . Here , the unique morphology of the CGRP interneurons is instructive . In contrast to many of the interneuron populations whose axons arborize longitudinally ( e . g . PKCγ cells ) or dorsally ( e . g . calretinin cells ) , the CGRP interneurons have ventrally-directed axons . In some respects , the CGRP interneurons resemble the lamina II radial cells described by Grudt and Perl , 2002 in the mouse , many of which are nociceptive , and the lamina III interneurons demonstrated in Golgi preparations in the cat and primate ( Beal and Cooper , 1978; Maxwell , 1985 ) . The fact that the CGRP interneurons show delayed firing patterns is also consistent with the properties of excitatory lamina II radial cells ( Dickie et al . , 2019; Grudt and Perl , 2002; Punnakkal et al . , 2014; Yasaka et al . , 2010 ) . Surprisingly , despite their dorsal dendrites , which extend into lamina II , where many nociceptive afferents terminate , we found no evidence that the CGRP interneurons are activated by acute noxious inputs ( capsaicin or formalin ) . On the other hand , we did detect an occasional polysynaptic input following synchronous electrical stimulation of primary afferent C fibers . Most importantly , compared to the lamina II radial cells , we recorded much more extensive ventral axon trajectories of the CGRP interneurons , which suggests that these interneurons engage very different circuits in the dorsal and potentially ventral horn . In this regard , the CGRP interneurons are distinct from the calretinin interneurons that target lamina I projection neurons ( Petitjean et al . , 2019 ) . An RNA sequencing study of dorsal horn interneurons demonstrated expression of Calca , the gene that encodes CGRP , in different clusters of neurons ( Häring et al . , 2018 ) , including several that express Rora , the gene that encodes RORα . Consistent with those results , our in situ hybridization studies found extensive co-expression of Calca and Rora . In fact , almost 55% of the CGRP interneurons co-express RORα message and there are significant similarities in their anatomical and functional properties ( Bourane et al . , 2015b ) . Specifically , the majority of RORα interneurons are excitatory and approximately 1/3 has a radial morphology , with ventrally arborizing axons . Furthermore , both the CGRP and RORα interneurons receive a monosynaptic Aβ afferent input and interestingly , despite the lack of response to capsaicin , some neurons in both populations receive a polysynaptic A delta and C input . Consistent with the report that deletion of the RORα population did not influence itch ( Bourane et al . , 2015b ) , and despite some overlap of the CGRP and GRP subsets of interneurons , we found that pruritogens did not activate ( induce Fos ) in the CGRP interneurons and ablation of CGRP interneurons did not influence scratching in response to exogenous pruritogens . There are , however , some striking differences between the RORα and CGRP interneurons . For example , although a majority of the RORα interneurons co-express Cck , the CGRP interneurons rarely do . Furthermore , whereas RORα interneurons are activated by innocuous mechanical stimuli ( e . g . brushing ) in both naive and injured conditions , the CGRP interneurons respond to innocuous stimuli only in the setting of nerve injury . To our knowledge , the CGRP interneurons represent the first class of excitatory interneurons in lamina III that are unresponsive to innocuous mechanical stimulation under basal conditions despite receiving a monosynaptic Aβ input . One possibility is that the CGRP interneurons are tonically inhibited under normal conditions , which is consistent with our electrophysiological recordings showing bicuculline-mediated facilitation of these interneurons . Reduction of these inhibitory inputs in the setting of injury ( Torsney and MacDermott , 2006 ) would render the neurons responsive to an innocuous stimulus ( e . g . brush ) . In turn , the ventrally directed axons of these interneurons could drive reflex withdrawal circuits , which is consistent with increased mechanical thresholds in the CGRP-interneuron ablated mice , and/or engage ascending nociceptive pathways located in deep dorsal horn . The fact that DREADD-mediated direct activation of many CGRP interneurons lowered mechanical withdrawal thresholds is consistent with that hypothesis . In other words , we suggest that sensitization of these neurons is critical to mechanisms that underlie Aβ-mediated mechanical allodynia in the setting of nerve injury . Interestingly , Lu et al . , 2013 provided evidence for convergence of a primary afferent-derived Aβ and a tonic glycinergic inhibitory input to PKCγ interneurons , some of which we found express CGRP . Loss of this glycinergic inhibition allowed Aβ input to access lamina I nociceptive circuits . Other studies demonstrated a comparable outcome , in this case by a presynaptic glycinergic inhibition of non-nociceptive inputs to superficial dorsal horn neurons ( Sherman and Loomis , 1996 ) . Furthermore , Imlach et al . , 2016 proposed that decreased glycinergic inhibition is selective for radial cells in lamina II and likely contributes to neuropathic pain . We suggest that a comparable circuit involving the CGRP radial cells could uncover low threshold inputs to ventrally located nociceptive circuits , which in recent years have been largely ignored ( Wercberger and Basbaum , 2019 ) . Mice were housed in cages on a standard 12:12 hr light/dark cycle with food and water ad libitum . Permission for all animal experiments was obtained and overseen by the Institutional Animal Care and Use Committee ( IACUC ) at the University of California San Francisco . All experiments were carried out in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and the recommendations of the International Association for the Study of Pain . The CalcaCreER mouse strain was kindly provided by Dr . Pao-Tien Chuang ( UC San Francisco ) ( Song et al . , 2012 ) . CalcaCreER mice were then bred with C57BL/6J -Ai14 mice ( Jackson Laboratory , Stock No: 007914 ) or with mice that selectively express green fluorescent protein ( GFP ) in gastrin-releasing peptide ( GRP ) -expressing cells ( GrpGFP mouse Solorzano et al . , 2015 ) . Lbx1FlpO mice , in which FLPo is driven from the Lbx1 promoter , were a kind gift from Dr . Martin Goulding at the Salk Institute , La Jolla CA . We dissolved tamoxifen ( T5648 , Sigma-Aldrich ) in corn oil and injected it ( 150 mg/kg , i . p . ) into the CGRP-tdTomato mice on two consecutive days . For immunohistochemistry , electrophysiology and tracing experiments we injected the tamoxifen into P21-22 mice . We waited 5 and 7–10 days before recording and perfusion for immunostaining , respectively . For Fluorogold ( 1% ) tracing experiments , we injected the tracer into 6- to 8-week-old mice . For intraspinal surgeries intended for DREADD receptor expression studies , we injected tamoxifen into P11-12 mice and subsequently , between P14 and P16 , made an intraspinal injection of hM3Dq without laminectomy . Mice of either sex were transcardially perfused with 10 mL phosphate-buffered saline ( PBS ) followed by 30 mL cold 4% formaldehyde in PBS . After dissection , dorsal root ganglia ( DRG ) , trigeminal ganglia ( TG ) , spinal cord , and caudal medullary tissue were post fixed for ~3 hr at room temperature and subsequently cryoprotected in 30% sucrose in PBS overnight at 4°C . The spinal cord and caudal medulla were sectioned in a cryostat at 25 μm; DRG and TG at 16 μm . After mounting and drying on slides , the sections were incubated for 1 . 5 hr in 10% normal goat serum with 0 . 3% Triton X-100 ( NGST ) to block non-specific antibody binding , and then for 24 hr in primary antibodies diluted in 10% NGST . The sections were then washed three times for 10 min in PBS and then incubated for 2 hr with a secondary antibody diluted in 1% NGST . After washing with PBS three times for 10 min , the sections were dried and coverslipped with Fluoromount G . The following primary antibodies were used: rabbit anti-CGRP ( 1:1000 , Peninsula ) , rabbit anti-calbindin ( 1:2000 , Swant ) , mouse anti-calretinin ( 1:5000 , Swant ) , guinea pig anti-PKCγ ( 1:7000 , Strategic Bio ) , chicken anti-GFP ( 1:2500 , Abcam ) , rabbit anti-Fos ( 1:5000 , Calbiochem; 1:2000 , Cell Signaling ) , guinea pig anti-Fluorogold ( 1:1000 , Protos Biotech ) , guinea pig anti-Lmx1b ( 1:10000 , kind gift from T . Müller and C . Birchmeier , Max Delbrück Center for Molecular Medicine , Berlin , Germany ) , rabbit anti-Pax2 ( 1:4000 , Abcam ) , or rabbit anti HA ( 1:800 , Cell Signaling ) . Secondary antibodies were conjugated to Alexa-488 or Alexa-647 ( 1:1000 , Thermo Fisher Scientific ) . To ablate the central terminals of CGRP-expressing DRG neurons , CGRP-tdTomato mice were anesthetized with 2% isoflurane and injected intrathecally with 5 . 0 μl of a solution containing 10 µg of capsaicin , dissolved in 10% ethanol , 10% Tween-80% and 80% saline . Five days later , the mice received 5 i . p . injections of 150 mg/kg tamoxifen ( one injection per day , on 5 consecutive days ) . Seven days later , the mice were processed for immunohistochemistry . Mice were perfused with phosphate-buffered 4% formaldehyde ( n = 3 ) or 4% formaldehyde plus 0 . 3% glutaraldehyde ( n = 5 ) . Transverse or parasagittal Vibratome sections ( 50 μm ) were processed for detection of tdTomato for either light ( LM ) or electron microscopic ( EM ) ( Llewellyn-Smith et al . , 2018 ) examination . For EM analysis , the sections were washed for 2 hr in 50% ethanol , incubated for 30 min in 10% normal horse serum diluted with Tris-PBS ( TPBS ) , then in 1:25 , 000 or 1:100 , 000 rabbit anti DSRed ( Takara Bio USA ) in 10% NHS-TPBS . The sections were subsequently exposed to 1:500 biotinylated donkey anti-rabbit IgG ( Jackson ImmunoResearch ) in 1% NHS-TPBS and then to 1:1500 ExtrAvidin-horseradish peroxidase ( Sigma-Aldrich ) in TPBS . Incubations in immunoreagents were for 3 days at room temperature on a shaker; sections were washed 3 × 30 min between incubations . To visualize CGRP-tdTomato-immunoreactivity in the dorsal horn , we used a nickel-intensified diaminobenzidine ( DAB ) reaction and hydrogen peroxide generated by glucose oxidase ( Llewellyn-Smith et al . , 2005 ) . After the peroxidase reaction , sections containing tdTomato-immunoreactive neurons were osmicated , stained en bloc with aqueous uranyl acetate , dehydrated with acetone and propylene oxide , and infiltrated with Durcupan resin ( Sigma-Aldrich ) . Finally , sections were embedded on glass slides under Aclar coverslips ( Electron Microscopy Sciences ) and polymerized at 60°C for at least 48 hr . Dorsal horn regions containing CGRP-tdTomato neurons were re-embedded in resin on blank blocks under glass coverslips and repolymerized . Ultrathin sections were collected on copper mesh grids , stained with aqueous uranyl acetate , and examined with a JEOL 100CXII transmission electron microscope . Transverse or parasagittal Vibratome sections of tissue from mice perfused with phosphate-buffered 4% formaldehyde ( n = 3 ) or 4% formaldehyde , 0 . 3% glutaraldehyde ( n = 3 ) were either single stained to show tdTomato-immunoreactivity or double stained to demonstrate the relationships between VGLUT1-immunoreactive axons and CGRP-tdTomato neurons . All sections were washed 3 × 20 min in TPBS containing 0 . 3% Triton X-100 and exposed to 10% NHS in TPBS-Triton for 30 min . Single labeling involved exposure of sections to 1:25 , 000 or 1:100 , 000 anti-DSRed ( Takara ) , 1:500 anti-rabbit IgG , 1:1500 ExtrAvidin-HRP and a nickel-intensified DAB reaction . For double labeling , VGLUT1-immunoeractivity was first detected with 1:50 , 000 or 1:100 , 000 rabbit anti-VGLUT1 ( Synaptic Systems ) , biotinylated donkey anti-rabbit IgG , ExtrAvidin-horseradish peroxidase and a cobalt +nickel intensified DAB reaction ( Llewellyn-Smith et al . , 2005 ) . Then , after another blocking step in 10% NHS , DSRed-immunoreactivity was detected as for single labeling except that the peroxidase reaction was intensified with imidazole ( Llewellyn-Smith et al . , 2005 ) rather than nickel . For LM labeling , primary antibodies were diluted with 10% NHS in TPBS-Triton; secondary antibodies , in 1% NHS-TPBS-Triton; and avidin-HRP complex , in TPBS-Triton . For LM , all incubations in immunoreagents were done on a shaker at room temperature for at least 24 hr and washes between incubations were 3 × 20 min in TPBS . Stained sections were mounted on subbed slides , dehydrated and coverslipped with Permaslip Mounting Medium ( Alban Scientific ) . In situ hybridization was performed using fresh spinal cord or caudal medullary tissue from adult mice ( 8–10 week-old ) , except for transient VGLUT3 assessment ( Peirs et al . , 2015 ) , where the mice were 7 days old . We followed the protocol outlined by Advanced Cell Diagnostics ( Newark , CA ) . The tissue was dissected out , instantaneously frozen on dry ice , and kept at –80°C until use . Cryostat sections of DRG ( 12 µm ) were fixed at 4°C in 4% formaldehyde for 15 min , washed twice in PBS , and dehydrated through successive 5 min ethanol steps ( 50% , 70% , and 100% ) and then dried at room temperature . After a 30 min incubation with protease IV , sections were washed twice in PBS and incubated at 40°C with RNAscope-labeled mouse probes: calcitonin gene-related peptide ( Calca ) , RAR-related orphan receptor alpha ( RORα ) , cholecystokinin ( Cck ) , vesicular glutamate transporter 3 ( Slc17a8 ) , neurokinin receptor 1 ( Tacr1 ) , gastrin releasing peptide ( Grp ) for 2 hr in a humidified chamber . Sections were then washed twice in washing buffer and incubated with four 15–30 min ‘signal amplifying’ solutions at 40°C . After two washes , the sections were dried and covered with mounting media containing 4′ , 6-diamidino-2-phenylindole ( DAPI ) . Images of fluorescent immunostained sections were acquired on an LSM 700 confocal microscope using ZEN Software ( Carl Zeiss ) . The microscope was equipped with 405 , 488 , 555 , and 639 nm diode lasers . For co-localization studies we used a 20x Plan-Apochromat ( 20×/0 . 8 ) objective ( Zeiss ) and image dimensions of 1024 × 1024 pixels with an image depth of 12 bits . Two times averaging was applied during image acquisition . Laser power and gain were adjusted to avoid saturation of single pixels and kept constant for each experiment . Image acquisition was performed with fixed exposure times for each channel and a 10% overlap of neighboring images where tiling was used . Stitching was done in ZEN using the ‘stitching/fuse tiles’ function . Adjustment of brightness/contrast and maximum projections of Z-stack images were done in Fiji/Image J . All images of the same experiment were processed in an identical manner . Images of peroxidase immunostained sections were acquired on an Olympus BH2 brightfield microscope equipped with SPlanApo lenses and a SPOT Insight CMOS Color Mosaic 5MP camera running SPOT 5 . 3 Advanced software . For assessment of VGLUT1 appositions on DSRed-immunoreactive CGRP neurons , an x100 oil immersion lens was used . A VGLUT1-positive terminal was classified as forming a close apposition when ( 1 ) there was no space between the terminal and the DSRed-positive neuron for terminals lying side-by-side with a cell body or dendrite or when ( 2 ) the terminal and the DSRed-positive neuron were in the same focal plane for terminals overlying cell bodies or dendrites . To analyze overlap by immunohistochemistry or in situ hybridization , we counted cells from four to five sections in at least three animals per experiment . By immunohistochemistry , we first counted the number of neurons in the DRG and TG that were tdTomato-positive ( total 1266 cells , three mice ) or CGRP-positive ( total 1050 cells , three mice ) and then determined the percentage of tdTomato-positive neurons that were CGRP double-labeled and vice versa . The number of dorsal horn tdTomato-positive cells that double-labeled for different markers ( e . g . PKCγ , Lmx1b , Fos , calretinin , calbindin ) are indicated in the Results . To conclude that cells were double-labeled by in situ hybridization we set a threshold of at least five fluorescent ‘dots’ for each probe in conjunction with a DAPI-positive nucleus . Quantification of caspase-mediated ablation of CGRP-positive spinal cord neurons was performed in 5 Caspase-injected CGRP-tdTomato mice and four saline-injected , CGRP-tdTomato control mice . We counted neurons positive for tdTomato or PKCγ ipsilateral and contralateral to the injection side , in 5–10 sections per mouse , and then determined the percentage of tdTomato- or PKCγ-positive neurons in the ipsilateral side relative to the contralateral side . For DREADD experiments we used a Cre and FlpO-dependent hM3D ( Gq ) adeno-associated virus: AAV1--hEF1alpha/hTLV1-Fon/Con[dFRT-HA_hM3D ( Gq ) -dlox-hM3D ( Gq ) -I-dlox-I-HA_hM3D ( Gq ) ( rev ) -dFRT]-WPRE-hGHp custom made by the University of Zurich Viral Vector Facility of the Neuroscience Center . For control injections , we used an AAV1 . hSyn . eGFP . WPRE . SV40 from Addgene . For GCaMP-tracing experiments , we used an AV1 . Syn . Flex . GCaMP6s . WPRE . SV40 from the Penn Vector Core , University of Pennsylvania . Note that we evaluated several Cre-dependent viral vectors for the tracing studies and only used those where specificity of expression was confirmed by lack of expression after injection into wild type mice . We waited at least 4 weeks to achieve stable viral expression before beginning the behavioral or neuroanatomical experiments . For Caspase-mediated ablation experiments , we used a Cre-dependent adeno-associated virus-expressing Caspase ( AAV1‐flex‐taCasp3‐TEVp , titer: 1 . 5–2 . 8 × 1012 viral particles/ml; Gene Therapy Vector Core at the University of North Carolina at Chapel Hill ) . To study potential projection targets of the dorsal horn CGRP interneurons , we injected Fluorogold ( 1% ) into several supraspinal sites known to receive projections from the spinal and medullary dorsal horns . We studied two mice for each location and allowed 5–9 days for tracer transport after which the mice were perfused with formaldehyde for subsequent histological analysis . We injected tracer into the following locations: ventrolateral thalamus ( X:ML = 1 . 5 , Y:AP = −1 . 82 , Z:DV = 3 . 5; 500 or 800 nl ) ; parabrachial nucleus ( X = 1 . 25 , Y = −4 . 95 , Z = −3 . 6; 600 nl , see Figure 5—figure supplement 2 ) ; nucleus submedius of the thalamus ( X = 0 . 5 , Y = −1 . 43 , Z = 4 . 25; 250 or 450 nl ) : dorsal column nuclei ( 400 nl ) . For all surgeries , the mice were administered carprofen ( 0 . 1 mg/kg , i . p . ) just prior to surgery and lidocaine ( 0 . 5% ) was applied to the incision site . For the DREADD experiments , under 2% isoflurane anesthesia , we injected P14-16 CalcaCreER-LbxFLPo mice and littermates with an AAV-GFP . We removed muscles that overlay the left side of the T13 and L1 vertebra to expose the lumbar enlargement . Without laminectomy , we then slowly inserted a glass micropipette ( 50 µm tip ) through the dura and made two 400 nl rostrocaudally separated injections of viral solution . The micropipette was left in place for ~2 min after which overlying muscle and skin were closed . After recovering from the anesthesia , the mice were returned to their home cages . For the GCaMP6-tracing studies , we made injections ( 300–800 nl ) into the medullary dorsal horn in 8-week-old mice anesthetized with i . p . ketamine ( 100 mg/kg ) and xylazine ( 10 mg/kg ) or isofluorane ( 2% ) . For injections into the nucleus caudalis , we incised the dura overlying the cisterna magna , exposing the caudal medulla and made a unilateral injection of viral solution with a glass micropipette . After recovering from anesthesia , the mice were returned to their home cage . For Caspase-mediated ablation studies , each mouse received a total of 2 . 0 μl of viral stock solution of the AAV1‐flex‐taCasp3‐TEVp ( ablated group ) into the lumbar spinal cord . To measure post‐virus mechanical and thermal thresholds , we tested the mice 3 weeks after virus injection , before and 7 and 14 days after spared nerve injury . At the end of all behavioral testing , control and ablated mice were euthanized , perfused and tissues harvested for quantification of the caspase-mediated ablation . We took several measures to blind the behavioral experiments . ( 1 ) DREADD-injected and control ( GFP-injected ) mice were housed together . ( 2 ) A different experimenter performed the injections of CNO ( 5 . 0 mg/kg in saline ) or saline before behavioral testing . ( 3 ) The behavioral tester recorded each mouse’s eartag number after the test and was blind to the treatment ( saline or CNO ) that the mouse received or to which group the mouse belonged ( AAV-GFP-injected or DREADD-injected ) . ( 4 ) Identification was made using records of eartag numbers after all testing was finalized . For these experiments , we determined hindpaw mechanical thresholds with von Frey filaments , and quantified results using the updown method ( Chaplan et al . , 1994 ) . The animals were habituated on a wire mesh for 2 hr on 2 consecutive days . On the next 2 days we recorded baseline thresholds , after a 1 . 5 hr of acclimatization on the wire mesh . After baseline determinations , the mice were injected with CNO or saline and then tested 30 min later . For all behavioral tests , either CNO or saline was injected every other day in a randomized fashion . Mice were habituated for 30 min on a mesh in plexiglass cylinders . Next , we used a syringe to squirt 50 µl acetone onto the plantar surface of the paw . The responses of the mice directly after application of acetone were recorded on video for 30 s . Each paw was tested five times and we measured time ( in seconds ) spent lifting , licking or flinching the paw . Results are displayed as the average time across the five trials . Testing began 1 hr post injection of CNO or saline , with test days 48 hr apart . For thermal threshold testing ( heat ) , we first acclimatized the mice for 30 min in Plexiglass cylinders . The mice were then placed on the glass of a Hargreaves apparatus and the latency to withdraw the paw from the heat source was recorded . Each paw was tested five times and we averaged latencies over the five trials . Hargreaves tests were done 1 hr after the tests of static dynamic mechanical allodynia . We made a subcutaneous injection of 100 μl chloroquine ( 100 μg diluted in saline; Sigma‐Aldrich ) into the left calf . Mice were immediately placed into cylinders and video recorded for 30 min . Behavior was scored as number of scratching/biting bouts of the injection area over the 30 min . Mice were acclimatized to the testing room and trained by placing them on an accelerating rotarod for a maximum of 60 s at low speed , three times with training taking place on two consecutive days . On testing days ( 48 hr apart ) , mice were injected with CNO or saline 30 min before being placed on the rotarod . Latency to fall was measured for up to 300 s . The procedure was repeated three times and latencies averaged across trials . To induce mechanical hypersensitivity in a model of neuropathic pain we used the approach described by Shields et al . , 2003 . Under isofluorane anesthesia ( 2% ) , two of the three branches of the sciatic nerve were ligated and transected distally , sparing the tibial nerve . To study the effects of a chemical algogen , we injected 10 µl of 2% formalin in saline into the cheek ( n = 3 ) . In a separate group of anesthetized animals ( n = 3 ) , we made a unilateral injection of 20 µl capsaicin ( 1 . 0 µg/µl ) into the hindpaw or the cheek . We perfused all mice ~ 1 . 5 hr after injection and immunostained sections of the lumbar cord ( paw injections ) or caudal medulla ( cheek injections ) for Fos . To study the effects of a pruritogen , under isofluorane anesthesia , mice ( n = 3 ) received unilateral injections of chloroquine ( 200 µg ) into either the hindpaw ( 20 μl ) or cheek ( 50 μl ) . The mice were perfused ~1 . 5 hr after injection and sections of the lumbar cord ( paw injections ) or caudal medulla ( cheek injections ) were immunostained for Fos . We injected mice ( n = 3 ) with nitroglycerin ( 10 mg/kg , i . p . ) , which in humans can trigger a migraine and in rodents provokes behavioral signs of widespread thermal hyperalgesia and mechanical hypersensitivity ( Bates et al . , 2010 ) , beginning 30–60 min after injection and subsiding within 4 hr . Based on this time course , the mice were perfused 2 hr after nitroglycerin injection and sections of caudal medulla were immunostained for Fos . To assess Fos expression in uninjured animals ( n = 3 ) , we first acclimatized the mice to brushing of the cheek , ( Utrecht 225 , pure red sable brush 6 , Germany ) while lightly restraining the mouse in a towel with its head exposed . We brushed the left cheek along the direction of the hairs for 45 min , with a one minute break every 10 min . To monitor Fos expression in the injured animals , we performed unilateral partial sciatic nerve injury ( SNI , see above ) . One week after SNI , we used a paintbrush ( 5/0 , Princeton Art and Brush Co . ) to lightly stroke the injured hind paw , from heel to toe ( velocity:~2 cm/s ) . Ninety minutes to 2 hr after brushing , the mice were anesthetized , perfused and spinal cord sections were immunostained for Fos . In a separate experiment , we also assessed Fos expression 1 week after SNI without applying a stimulus . Three mice were trained on a rotating rod for 60 min at a constant speed of 10 rpm . One week later the mice walked on the rotarod at 10 rpm for 1 . 5 hr ( Neumann et al . , 2008 ) , after which they were anesthetized , perfused and lumbar spinal cord sections immunostained for Fos . Following our previous protocol ( Etlin et al . , 2016 ) , we collected transverse lumbar and caudal medullary Vibratome ( Leica ) slices ( 350–400 μm ) from 3 to 10 weeks old CGRP-tdTomato mice 5–7 days after tamoxifen injection . The sections were incubated in recording solution at 37°C for 1 hr and then transferred to a recording chamber ( Automate Scientific ) under an upright fluorescence microscope ( E600FN; Nikon ) . The sections were superfused with recording solution at a rate of 1 . 0 ml/min and viewed with a CCD digital camera ( Hamamatsu or DAGE-MTI ) . The transparent appearance of lamina II of the superficial dorsal horn and tdTomato-positive CGRP cells were obvious under near-infrared ( IR ) illumination . The patch pipettes were pulled to yield an impedance of 6–8 MΩ on a horizontal pipette puller ( Sutter Instrument ) from thin-walled , fire-polished , borosilicate glass filaments . The pipette solution composition was ( in mM ) : K-methane sulfonate 140 , NaCl 10 , CaCl2 1 . 0 , EGTA 1 . 0 , HEPES 10 , Mg-ATP 5 . 0 , and NaGTP 0 . 5 and included 5 . 0 mg/ml of Biocytin ( Sigma-Aldrich ) for intracellular filling of the recorded cells . Neurons were approached with a micromanipulator ( Sutter Instrument ) while monitoring the resistance in voltage-clamp mode using the ‘Membrane Test’ module of pClamp10 software ( Molecular Devices ) . To prevent clogging of the tip , we applied positive pressure to the pipette via a 1 . 0 ml syringe . After a seal was established with a cell , we ruptured its membrane by gently applying negative pressure to the pipette to secure a whole-cell configuration . Current and voltage signals were amplified using a DC amplifier ( MultiClamp 700 ) and digitized using Digidata 1440a system ( Molecular Devices ) at 10 kHz and then stored for subsequent offline analysis . In some experiments , we placed an attached dorsal root in a suction electrode to be stimulated electrically while simultaneously measuring evoked responses of the tdTomato-expressing neurons . To determine the fiber types providing input to the recorded neurons , and to assess the monosynaptic/polysynaptic nature of the Aβ , Aδ , and C fiber inputs , the dorsal roots were stimulated 20 times at the following frequencies and intensities ( 25 μA , 20 Hz for Aβ fibers; 100 μA , 2 Hz and occasionally 10 Hz for Aδ fibers; 500 μA , 1 Hz for C fibers ) . For current clamp recordings , after whole cell configuration was achieved , action potentials were induced by current steps , from −10 to 150 pA , with an increment of 5 . 0 or 10 pA ( pulse duration 300 ms ) . The rheobase was determined using a 5 . 0 pA increment current step ( pulse duration 300 ms ) . Only neurons with a resting membrane potential of at least −40 mV and stable baseline were used for further experiments and analysis . The recording was abandoned with loss of spike overshoot . To determine the effect of the inhibitory inputs on excitability of the CGRP interneurons , slices were continuously perfused with 20 µM bicuculline and 4 . 0 µM strychnine and the rheobase measured before and after application of the GABA and glycine antagonists . After establishing a stable baseline recording , we maintained the neurons 10 pA below rheobase ( pulse duration 300 ms , sweep intervals of 30 s ) , a current at which action potentials were never evoked . Next the bicuculline/strychnine solution was applied to the recording chamber . The appearance of an action potential signaled that the antagonists had removed a tonic inhibition of the CGRP interneurons . Statistical analyses were performed using SPSS ( IBM-SPSS version 24 ) . Similarity of normality and variance were assessed before applying parametric or non-parametric tests . For analysis of the effect of CNO on mechanical hypersensitivity , we assessed interaction between treatment ( CNO , saline or baseline ) with group ( DREADD-virus injected animals or GFP-virus injected animals ) by repeated measures two-way ANOVA , including all conditions and groups . Statistics were calculated based on a type III sum of squares model and significant interaction effects were assessed using deviation from the mean of the control groups . The N was estimated based on variance for von Frey experiments using an a priori power calculation . Hargreaves and rotarod results were analyzed using Student’s t-tests . For acetone sensitivity we used the Wilcoxon signed rank test . Parametric and non-parametric tests are reported as mean ± SEM or by medians and inter-quartiles , respectively . Electrophysiological recordings of intrinsic membrane and action potential properties were calculated using custom-written Matlab scripts ( MathWorks , Illinois ) as previously described ( Etlin et al . , 2016 ) . p Values were considered significant if p<0 . 05 .	The ability to sense pain is critical to our survival . Normally , pain is provoked by intense heat or cold temperatures , strong force or a chemical stimulus , for example , capsaicin , the pain-provoking substance in chili peppers . However , if nerve fibers in the arms or legs are damaged , pain can occur in response to touch or pressure stimuli that are normally painless . This hypersensitivity is called mechanical allodynia . A protein called calcitonin gene-related peptide , or CGRP , has been implicated in mechanical allodynia and other chronic pain conditions , such as migraine . CGRP is found in , and released from , the neurons that receive and transmit pain messages from tissues , such as skin and muscles , to the spinal cord . However , only a few distinct groups of CGRP-expressing neurons have been identified and it is unclear if these nerve cells also contribute to mechanical allodynia . To investigate this , Löken et al . genetically engineered mice so that all nerve cells containing CGRP produced red fluorescent light when illuminated with a laser . This included a previously unexplored group of CGRP-expressing neurons found in a part of the spinal cord that is known to receive information about non-painful stimuli . Using neuroanatomical methods , Löken et al . monitored the activity of these neurons in response to various stimuli , before and after a partial nerve injury . This partial injury was induced via a surgery that cut off a few , but not all , branches of a key leg nerve . The experiments showed that in their normal state , the CGRP-expressing neurons hardly responded to mechanical stimulation . In fact , it was difficult to establish what they normally respond to . However , after a nerve injury , brushing the mice’s skin evoked significant activity in these cells . Moreover , when these CGRP cells were artificially stimulated , the stimulation induced hypersensitivity to mechanical stimuli , even when the mice had no nerve damage . These results suggest that this group of neurons , which are normally suppressed , can become hyperexcitable and contribute to the development of mechanical allodynia . In summary , Löken et al . have identified a group of nerve cells in the spinal cord that process mechanical information and contribute to touch-evoked pain . Future studies will identify the nerve circuits that are targeted by CGRP released from these nerve cells . These circuits represent a new therapeutic target for managing chronic pain conditions related to nerve damage , specifically mechanical allodynia , which is the most common complaint of patients with chronic pain .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2021	Contribution of dorsal horn CGRP-expressing interneurons to mechanical sensitivity
Humans often evaluate sensory signals according to their reliability for optimal decision-making . However , how do we evaluate percepts generated in the absence of direct input that are , therefore , completely unreliable ? Here , we utilize the phenomenon of filling-in occurring at the physiological blind-spots to compare partially inferred and veridical percepts . Subjects chose between stimuli that elicit filling-in , and perceptually equivalent ones presented outside the blind-spots , looking for a Gabor stimulus without a small orthogonal inset . In ambiguous conditions , when the stimuli were physically identical and the inset was absent in both , subjects behaved opposite to optimal , preferring the blind-spot stimulus as the better example of a collinear stimulus , even though no relevant veridical information was available . Thus , a percept that is partially inferred is paradoxically considered more reliable than a percept based on external input . In other words: Humans treat filled-in inferred percepts as more real than veridical ones . In order to make optimal and adaptive decisions , animals integrate multiple sources of sensory information across time and space . One of the prime examples of this is observed when animals are confronted with coherently-moving stimuli during random-dot motion experiments . In such experiments , performance and the corresponding neural activity vary proportionally to signal strength in a way that is consistent with the progressive integration of evidence over time ( Shadlen et al . , 1996; Shadlen and Newsome , 2001 ) . Besides temporal accumulation , sensory integration is also possible by combining the information from multiple sensory sources ( Quigley et al . , 2008; Schall et al . , 2009; Hollensteiner et al . , 2015; Wahn and König , 2015a , 2015b , 2016 ) . In the case of multisensory perception , several experiments have shown that integration often occurs in a statistically optimal way . This has been best demonstrated in cue-integration tasks in which humans perform as if they were weighting the different sources of information according to their respective reliabilities ( Ernst and Banks , 2002; Alais and Burr , 2004; Körding and Wolpert , 2004; Tickle et al . , 2016 ) . This form of statistical inference has also been demonstrated for cortical neurons of the monkey brain , with patterns of activity at the population level that are consistent with the implementation of a probabilistic population code ( Gu et al . , 2008; Fetsch et al . , 2011 ) . In most of these sensory integration experiments , the perceptual reliability of different inputs is probed through quantitative manipulations of the inputs’ signal-to-noise ratios ( Heekeren et al . , 2004; Tassinari et al . , 2006; Bankó et al . , 2011 ) . However , some percepts are unreliable not because they are corrupted by noise but because they are inferred only from the context and thus intrinsically uncertain . This occurs naturally in monocular vision at the physiological blind spot , where content is ‘filled-in’ based on information from the surroundings . In this case , no veridical percept is possible at the blind spot location . Even though changes in reliability due to noise directly result in behavioral consequences , the effects of the qualitative difference between veridical and inferred percepts , that are otherwise apparently identical , are unknown . We recently reported differences in the processing of veridical and inferred information at the level of EEG responses ( Ehinger et al . , 2015 ) . We demonstrated that a qualitative assessment of differences in reliability exists at the neural level in the form of low- and high-level trans-saccadic predictions of visual content . Notably , active predictions of visual content differed between inferred and veridical visual information presented inside or outside the blind spot . Although no difference was found between low-level error signals , high-level error signals differed markedly between predictions based on inferred or veridical information . We concluded that the inferred content is processed as if it were veridical for the visual system , but knowledge of its reduced precision is nevertheless preserved for later processing stages . In the present experiment , we address whether such an assessment of a dichotomous , qualitative difference in reliability is available for perceptual decision-making . Using 3D shutter glasses , we presented one stimulus partially in the participant’s blind spot to elicit filling-in and a second stimulus at the same eccentricity in the nasal field of view outside of the blind spot . The subject’s task was to indicate which of the two stimuli was continuously striped and did not present a small orthogonal inset ( see Figure 1a ) . Crucially , stimuli within the blind spot are filled-in and thus perceived as continuous , even when they present an inset . In the diagnostic trials , both stimuli were physically identical and continuous , and subjects were confronted with an ambiguous decision between veridical and partially inferred stimuli . 10 . 7554/eLife . 21761 . 003Figure 1 . Stimuli and stimulation . ( a ) Striped stimuli used in the study . The inset was set to ~50% of the average blind spot size . The global orientation of both stimuli was the same , but in different trials it could be either vertical ( as shown here ) or horizontal ( not shown ) . ( b ) Each stimulus was displayed individually either ( partially ) inside or ( completely ) outside the blind spot . This example presents an inset stimulus inside the subject’s left blind spot . However , due to filling-in , it is perceived as continuous ( right column ) . The task required subjects to select the continuous stimulus , and it was designed to differentiate between two mutually exclusive predictions: First , subjects cannot differentiate between the two different types of stimuli and thus answer randomly . Alternatively , subjects have implicit or explicit knowledge about the difference between inferred ( filled-in ) and veridical contents and consequently select the stimulus outside the blind spot in ambiguous trials . ( c ) Two stimuli were displayed using shutter glasses . Each stimulus was presented to one eye only , and it is possible that both are presented to the same eye ( as in the example depicted here ) . That is , the left stimulus could be shown either in the temporal field of view ( nasal retina ) of the left eye ( as in the plot ) or in the nasal field of view ( temporal retina ) of the right eye ( not shown ) . In this case , the trial was unambiguous: The stimulus with an inset was presented outside the blind spot and could be veridically observed , therefore , the correct answer was to select the left stimulus . ( d ) The locations of stimulus presentation in the five experiments . All stimuli were presented relative to the blind spot location of each subject . All five experiments included the blind spot location ( green ) . In the second and fifth experiment , effects at the blind spot were contrasted with a location above it ( purple ) . In the third experiment , the contrasts were in positions located to the left or the right of the blind spot . Note that both stimuli were always presented at symmetrical positions in a given trial , the position of the stimuli differed only across trials . DOI: http://dx . doi . org/10 . 7554/eLife . 21761 . 00310 . 7554/eLife . 21761 . 004Figure 1—figure supplement 1 . Trial balancing of all experiments . Each row is one condition in one experiment ( depicted in the left most column ) . The graph is split in a physical stimulation ( what is shown , left ) and a perception column ( what do subjects perceive due to fill-in in the blind spot , right ) . The dark-blue fields depict trials where an inset stimulus ( dark-blue ) was shown but partially inside the blind spot . On the right side ( perception ) we added these trials to the respective continuous ( blue ) cases . We mark with red the columns that indicate trials where an inset was shown in the temporal field . Note that perceptually these trials only exist in the locations above/inward/outward the blind spot , but are impossible inside the blind spot ( due to fill-in ) . Because the resulting statistical distribution might influence decisions by the subjects ( see results of experiment 5 , probability matching ) , experiment 5 was a repetition of experiment 2 without these trials . DOI: http://dx . doi . org/10 . 7554/eLife . 21761 . 004 We evaluated two mutually exclusive hypotheses on how perceptual decision-making could proceed when confronted with an ambiguous decision between veridical and inferred percepts . In the first case , we hypothesized that subjects are unable to make perceptual decisions based on an assessment of differences in reliability between veridical and inferred stimuli . Therefore , subjects would have an equal chance of selecting stimuli presented inside or outside the blind spot . Alternatively , it might be possible to use the information about the reduced reliability of filled-in information . In this case , we expect subjects to follow an optimal strategy and trust a stimulus presented outside the blind spot , where the complete stimulus is seen , more often than when the stimulus is presented inside the blind spot , where it is impossible to know the actual content within the filled-in part . In the first experiment , 24 subjects performed a 2-AFC task in which they had to indicate which of two stimuli was continuously striped instead of presenting a small orthogonal central inset ( Figure 1a , b ) . The stimuli were presented simultaneously in the periphery at the locations of the blind spots or at equivalent eccentricity on the opposite side ( Figure 1c , d ) . We used a 3D monitor and shutter glasses that allowed for the controlled monocular display of the stimuli . That means each stimulus was visible to a single eye only . There were always two stimuli , therefore , in a given trial either one or both eyes were stimulated ( Figure 1b ) . Importantly , subjects always perceived the two stimuli at the same locations , to the left and the right of the fixation cross . In this experiment there were perceptually ambiguous trials , where two continuous stimuli were perceived , and unambiguous trials where one stimulus contained a visible inset . In the unambiguous trials , an orthogonal inset was present in one of the stimuli . Importantly , in these trials , the stimulus with the inset was outside the blind spot and therefore clearly visible . As expected , subjects performed with near-perfect accuracy ( Figure 2 , unambiguous trials , blue data ) , choosing the continuous stimulus in an average of 98 . 8% of trials ( 95%-quantile over subjects [96 . 4–100%] ) . 10 . 7554/eLife . 21761 . 005Figure 2 . First experiment . ( a ) The left column shows schematics of the actual stimulation and the associated percepts for the corresponding data presented in the right panel . A dark-lined circle , where present , indicates that the stimulus was presented in the blind spot and , consequently , an inset stimulus within was perceived as a continuous stimulus due to filling-in . The plot to the right shows each subject’s ( n = 24 ) average response and the group average ( 95% bootstrapped confidence intervals , used only for visualization ) . The results from unambiguous trials ( blue ) show that subjects were almost perfect in their selection of the continuous stimulus when an inset was visible . For the first type of ambiguous control trials ( red ) , both stimuli were presented either outside or inside the blind spot . Here , only a global bias toward the left stimulus can be observed ( solid line , the mean across all observed conditions in red ) . Note that the performance when presenting an inset in the blind spot was identical to the one of presenting a continuous stimulus in the blind spot . The ambiguous diagnostic conditions ( green ) show the , unexpected , bias toward the blind spot ( for either side ) . ( b ) Statistical differences were evaluated by fitting a Bayesian generalized mixed linear model . In the model , the left and right ambiguous diagnostic conditions were combined in a single estimate of the bias for nasal or temporal stimuli ( outside or inside the blind spot respectively ) . The plot shows the average effect of each subject ( small yellow dots ) , the bootstrapped summary statistics of the data ( yellow errorbar ) , and the posterior 95% credibility interval model estimate ( black errorbar ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21761 . 005 There were two types of ambiguous trials . In the first type ( Figure 2 , ambiguous control , red data ) , one of the following applied: both stimuli were continuous and appeared outside the blind spots in the nasal visual fields ( Figure 2 , row 3 ) ; both were continuous and appeared inside the blind spots ( Figure 2 , row 4 ) ; or one was continuous , the other had an inset , and both appeared inside the blind spots with the inset either in the left or the right blind spot ( Figure 2 , rows 5 and 6 ) . In the case when a stimulus with an inset was present , this central part was perfectly centered inside the blind spot ( Figure 1a ) , and in consequence was perceived as continuous due to filling-in . Thus , in all four versions , subjects perceived two identical stimuli , and there was no single correct answer . In this type of ambiguous trial , subjects showed a small global leftward bias and chose the left stimulus in 53 . 6% of trials ( Figure 2 , continuous vertical line ) . In addition , no difference can be seen between the perception of pairs of filled-in stimuli and pairs of veridical continuous stimuli ( Figure 2 , rows 3 vs . 4–6 ) . This type of ambiguous control trial confirms that filling-in was accurately controlled in our experiment . In the second type of ambiguous trials one stimulus was presented inside and the other outside the blind spot ( Figure 2 , ambiguous diagnostic , data in green ) . This allowed us to test directly between two rival predictions: whether subjects will show a bias against the stimulus that is partially inferred ( inset area inside the blind spot ) and in favor of the veridical stimulus ( in the opposite visual field ) , or no bias . Selecting the filled-in stimulus is a suboptimal decision because the stimulus presented partially in the blind spot is the only one which could possibly contain the inset . This is explicit in the cases where an inset is shown in the blind spot but rendered invisible by filling-in ( Figure 2a , ambiguous trials with an inset stimulus ) . To analyse the data , we modeled the probability increase of choosing the right stimulus if the right stimulus was presented in either the temporal visual field of the right eye ( blind spot ) or the nasal visual field of the left eye ( non-blind spot ) . A similar factor was used for the left stimulus . Subsequently , the two one-sided model estimates were collapsed to a single measure of preference for stimulus presented at the nasal or temporal visual field ( outside or inside the blind spot respectively ) . As a model for inference , we used a Bayesian generalized mixed linear model . There were three additional factors in the model ( handedness , dominant eye , and precedent answer ) that are not the focus of the experiment and are thus reported in the methods section ( see ‘Effects not reported in the Results section’ ) . Figure 2a ( ambiguous diagnostic , data in green ) and 2b show that subjects indeed presented a bias . However , in contrast to our expectations , subjects were more likely to choose the filled-in percept with a 15 . 01% preference for stimuli presented in the temporal visual field ( CDI958 . 49–21 . 08% ) . In other words , when subjects had to decide which of the two stimuli ( both perceived as being continuous , and in most cases actually physically identical ) was less likely to contain an inset , they showed a bias for the one in which the critical information was not sensed directly but inferred from the surrounding spatial context . Remarkably , this result is at odds with both of our experimental predictions that postulated either no bias or a bias in favor of the veridical stimulus . The second experiment was designed to replicate the unexpected result of the first experiment and evaluate whether the blind spot bias observed was due to systematic differences between nasal and temporal retinae . In experiment 1 , we presented stimuli at mirror eccentricities inside and outside the blind spot , that is , temporal and nasal respectively ( see Figure 1c ) . In experiment 2 , we tested whether the bias in experiment 1 was specific to the blind spot location or related to known differences between the temporal and nasal retina ( Fahle and Schmid , 1988 ) . There is higher photoreceptor density ( Curcio et al . , 1990 ) , spatial resolution ( Rovamo et al . , 1982 ) , luminance discrimination ( Pöppel et al . , 1973 ) and orientation discrimination ( Paradiso and Carney , 1988 ) at locations that project to the nasal retina ( the temporal visual field where the blind spots are located ) . Thus , we repeated our experiment with a new group of subjects ( n = 27 ) and an additional experimental condition . In this new condition , the two stimuli were displayed at symmetrical locations above the blind spot ( 25° above the horizontal meridian; see Figure 1d , purple location ) . The results of this second experiment replicate the observations of experiment 1 ( Figure 3a ) : subjects showed a bias for selecting the stimulus presented inside the blind spot ( 12 . 5% , CDI957 . 35–17 . 49% ) . However , subjects also presented a bias in the control condition , toward the stimuli presented in the temporal visual field above the blind spot ( 6 . 63% , CDI950 . 77–12 . 3% ) . The bias was nevertheless stronger inside the blind spot ( paired difference: 6 . 11% , CDI951 . 16–10 . 78% ) . In summary , additionally to the bias inside of the blind spot area , we observed that subjects also showed a smaller bias for stimuli presented to the nasal retina ( temporal visual field ) . 10 . 7554/eLife . 21761 . 006Figure 3 . Location control experiments . Two control experiments were designed to test whether the observed bias for the blind spot could be explained by a general bias for stimuli presented in the temporal visual field . ( a ) Results of experiment 2 . In a given trial , stimuli were presented either at the locations corresponding to the blind spot or at locations above it . Results are presented as in Figure 2b , with the addition of within-subject differences between blind spot and control locations ( in purple ) . ( b ) Results of experiment 3 . In a given trial , stimuli were presented at the locations corresponding to the blind spot or at locations to inward ( toward the fixation cross ) or outward ( away from the fixation cross ) to it . Note that the blind spot effect is replicated in both experiments . In addition , both blind spot effects are larger than in any control location . DOI: http://dx . doi . org/10 . 7554/eLife . 21761 . 006 To better delineate the distribution of bias across the temporal visual field and to clarify if the blind spot location is , in fact , special , we performed a third experiment on a new group of subjects ( n = 24 ) . Here , we compared biases in the blind spot to two other control conditions flanking the blind spot region from either the left or the right ( Figure 3b ) . The blind spot location again revealed the strongest effect of a bias for the temporal visual field ( 13 . 18% CDI95 6 . 47–19 . 64% ) , while the locations inwards and outwards resulted in a 2 . 85% and 4 . 8% bias , respectively ( CDI95 −1 . 1–6 . 65%; CDI95 0 . 58–8 . 89% ) . The bias of both control locations was different from the bias of the blind spot location ( BS vs . inward: 10 . 51% , CDI95 3 . 55–17 . 29%; BS vs . outward: 8 . 61% , CDI950 . 98–16 . 04% ) . In this experiment , as in experiments 1 and 2 , we observed a bias that is specific to the blind spot region . The results of the three previous experiments suggest that subjects considered the filled-in percept a better exemplar of a continuous non-inset stimulus , in disregard of the physical possibility of the presence on an inset inside the blind spot . To confirm this , we performed a fourth experiment with a new group of subjects ( n = 25 ) . This experiment was identical to the first experiment , except that in this case , the subjects’ task was to choose the stimulus with an inset , instead of the continuous one . In this case , if a filled-in stimulus is indeed considered a more reliable exemplar of a continuous stimulus , the non blind spot stimulus should be preferred in the diagnostic trials . This was the case; subjects showed a bias for selecting the stimulus presented outside the blind spot ( 7 . 74% , CDI951 . 56–13 . 68% , Figure 4a ) , thus resulting in the expected reversal of the bias pattern observed in the first three experiments . This pattern is again suboptimal , since this time the filled-in stimulus is the one that could conceal the target . The result of this experiment indicates that the observed biases do not correspond to an unspecific response bias for the blind spot , and instead are a consequence of considering the inferred percepts as more reliable exemplars of a continuous stimulus . 10 . 7554/eLife . 21761 . 007Figure 4 . Task instruction control and probability matching control . ( a ) Results of experiment 4 . This control was the same as experiment 1 , except that subjects have to choose the stimulus with an inset ( instead of the continuous one ) . ( b ) Results of experiment 5 . This control was similar to experiment 2 , except that no inset stimulus was ever experienced in the control location above in the temporal visual field . DOI: http://dx . doi . org/10 . 7554/eLife . 21761 . 00710 . 7554/eLife . 21761 . 008Figure 4—figure supplement 1 . Correlation between experiment 4 and 5 . Global-bias subtracted blind spot effect for each subject in experiment 4 and 5 . Subjects showed a negative bias towards the nasal stimulus outside the blind spot in experiment 4 ( the task-switch experiment ) and a positive bias towards the temporal stimulus inside the blind spot in experiment 5 . The reference line has a slope of 1 . The red line is the first principle component ( representing total least squares ) . The Pearson correlation coefficient is 0 . 61 ( p=0 . 0016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21761 . 008 We performed a final control to evaluate whether the observed bias for a filled-in stimulus was not a result of subjects using a probability matching heuristic . It is possible that , in order to solve the ambiguous task , subjects used their knowledge of the rate of appearance of continuous and inset stimuli at different locations as learned during unambiguous trials . As it is impossible to experience an inset in the blind spot , the base rate of continuous stimuli at that location is 1 . 0 . Therefore , when confronted with two stimuli that are apparently identical , one inside and one outside the blind spot , subjects might just apply the base rate they have learned instead of relying on a perceptual estimate . If this is the case , subjects should show a bias for the location where they experienced exclusively continuous stimuli during unambiguous trials , which could result in a bias pattern similar to the one observed in experiments 1–3 . To evaluate this alternative explanation , we performed a further experiment with the same group of subjects that participated in experiment 4 . Experiment 5 was similar to experiment 2 , with control trials presenting stimuli above the blind spot . However , in contrast to experiment 2 , subjects never experienced an inset in the temporal field in the above positions during unambiguous trials ( see Figure 1—figure supplement 1 for a detailed overview of trial randomization ) . This results in an identical base rate of occurrence of a continuous stimulus in the temporal field for both the above and blind spot locations . Consequently , if the behavior observed in the previous experiments was a result of probability matching , in this experiment we should observe the same bias at both the blind spot and the temporal field above locations . Subjects showed a bias for selecting the stimulus presented inside the blind spot ( 14 . 53% , CDI957 . 56–21 . 09% , Figure 4b ) , replicating again the results of experiment 1–3 . At odds with the probability matching hypothesis , the bias for the temporal field in the above location was only 5 . 84% , not different from 0 ( CDI95−1 . 33–13 . 01% ) and similar to what was observed in experiment 2 . This bias was different from the bias observed in the blind spot ( paired-diff: 8 . 95% , CDI953 . 91–13 . 85% ) . The same group of subjects participated in experiment 4 and 5 , allowing us to make a within subjects comparison between the two tasks . Subjects’ performance in these two tasks was negatively correlated ( r = −0 . 61 , p=0 . 002 , see Figure 4—figure supplement 1 ) . Taking the task reversal of experiment 4 into account , this result indicates that subjects were consistently biased to consider the inferred filled-in stimulus a better exemplar of a continuous stimulus . The result of experiment 5 thus gives evidence that the bias for the filled-in stimulus was not a consequence of subjects matching the base rate of the occurrence of different stimuli during unambiguous trials . A bias for the temporal visual field , especially the blind spot , could also be reflected in the distribution of reaction times . We compared the reaction times of trials where subjects selected a stimulus in the temporal visual field against trials where the stimulus in the nasal visual field was selected . The reaction time analysis was not a planned comparisons , thus , in contrast to the other analyses presented here , it is explorative . In the first experiment , we observed an average reaction time of 637 ms ( minimum subject average: 394 ms , maximum 964 ms; Figure 5 ) . We used a linear mixed model to estimate the reaction time difference for selecting a stimulus presented inside the blind spot ( temporally ) against one outside the blind spot ( nasally ) . In the first experiment , after excluding three outliers , we observed this effect with a median posterior effect size of 13 ms faster reaction times when selecting the blind spot region ( CDI95% 2–42 ms ) . The three outliers ( on the right of the vertical dashed line in Figure 5 ) were identified visually and removed because they were distinctively different from the rest of the population . The mean of the outliers was 5 . 2 SD away from the remaining subjects . The outliers were nevertheless in the same direction of the reaction time effect and did not change its significance ( with outliers , 63 ms , CDI957–124 ms ) . However , faster reaction times while selecting the blind spot stimulus were not present individually in the other four experiments . The nominal differences were in the same direction as experiment 1 but not significant ( Exp . 2: 4 ms , CDI95−14–23 ms; Exp . 3: 22 ms . CDI95 −3–39 ms; Exp . 4: −1 ms CDI95 −20–21 ms; Exp . 5: 4 ms CDI95 −15–23 ms ) . Non-significant results were obtained for the other locations tested ( above Exp . 2: 8 ms , CDI95 −38–53 ms; above Exp . 4: 8 ms CDI95 −17–32 ms; outward: 2 ms CDI95 −13–16 ms; inward: 4 ms , CDI95 −29–37 ms ) . After combining all data ( without experiment 4 as the task was reversed ) , we observed a reduced reaction time for decisions for the blind spot stimulus with 10 ms ( CDI952–17 ms ) but not in any other location . We do not find this small bias in any experiment individually ( except Exp . 1 ) but only after pooling over experiments and therefore , we should interpret it cautiously . In conclusion , subjects selected stimuli in the blind spot slightly faster than stimuli outside the blind spot . The same effect does not appear for the other temporal control locations . 10 . 7554/eLife . 21761 . 009Figure 5 . Reaction times . Reaction times of trials where the nasal stimulus was chosen minus the reaction times of trials where the temporal stimulus was chosen . Single subject estimates and 95% CI posterior effect estimates are shown . The black ( combined ) estimate results from a model fit of all data combined , the individual confidence intervals represent the experiment-wise model fits . We observe a reaction time effect only inside the blind spot . DOI: http://dx . doi . org/10 . 7554/eLife . 21761 . 009 For an overview of all experiments and the results of a Bayesian logistic mixed effects model that combines all experiments , see Figure 6 , Figure 6—figure supplement 1 and Supplementary file 1 . In the combined model , we did not find any differences between the temporal field effects at locations other than the blind spots ( Figure 6 , fourth last to second last row ) . That is , the temporal field effects of the locations inward , outward and above were not different from each other . For the sake of clarity , we combined these location levels . Keeping everything else constant , we observed that if we present one stimulus in the blind spot against the equidistant nasal location , subjects are 13 . 82% ( CDI9510 . 84–16 . 78% , t-test , t = 8 . 7 , df = 98 , p<0 . 001 ) more likely to choose the stimulus in the blind spot . This bias is stronger than the effect observed elsewhere in the temporal field by 9 . 35% ( CDI95 6 . 25%–12 . 47%; paired t-test , t = 4 . 8 , df = 74 , p<0 . 001 ) . In summary , subjects showed a robust bias for the blind spot locations that could not be explained by a non-specific bias for the temporal visual field . In conclusion , when confronted with an ambiguous choice between veridical and inferred sensory information , human subjects showed a suboptimal bias for inferred information . 10 . 7554/eLife . 21761 . 010Figure 6 . Summary and overview of blind spot effects . Posterior GLMM-effect estimates of all data combined ( black ) except experiment 4 ( inversed task ) . We also show for each experiment the 95% CI of bootstrapped means summary statistics of the data ( yellow ) . Next , we show difference values between the blind spot and all other control locations ( model dark , raw data pink ) . As discussed in the text , the control locations outward , inward and above do not differ ( fourth last to second last row ) , and thus we compare the blind spot effect to all locations combined ( last row ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21761 . 01010 . 7554/eLife . 21761 . 011Figure 6—figure supplement 1 . Normalized data with control locations for all experiments . Fraction of choosing the right stimulus dependent on location ( indicated by icon ) and experiment ( Exp . 1: n = 24 , Exp . 2: n = 27 , Exp . 3: n = 24 , Exp . 4: n = 25 , Exp . 5: n = 24 ) . For plotting purposes , we preprocessed the data of each subject by subtracting their respective global bias . Each gray dot depicts one subject . The error bars depict mean , and 95% bootstrapped CI . A bias for the blind spot was visible in the form of ‘left’ responses when the left stimulus was presented in the temporal visual field of the left eye ( green , nasal/blind spot retina of the left eye ) and of more ‘right’ responses when the right stimulus was presented in the temporal visual field of the right eye ( green , nasal/blind spot of the right eye ) in all experiments . A bias was visible in the other tested locations , but it was much smaller . Control conditions show that there was no bias if the stimuli were shown either both inside the temporal fields ( dark blue ) or both inside the nasal fields ( light blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21761 . 01110 . 7554/eLife . 21761 . 012Figure 6—figure supplement 2 . Posterior predictive checks . ( a ) First level posterior predictive check: We simulated 100 new datasets from the posterior with the subject-wise posterior effect estimates . The observed values ( black dots ) are adequately captured for all subjects depicted here . ( b ) Second level posterior predictive check: Here we estimated datasets with new subjects randomly sampled from the multi-variate mixed-effects population distribution . We calculated the subject-wise blind spot effect , collapsed over both blind spots . The posterior predictive estimates are shown in red . They match the empirical distribution of blind spot effects ( black line , histogram and scatter plot below ) closely . We conclude that our model is adequate to capture the patterns of interest in our study . DOI: http://dx . doi . org/10 . 7554/eLife . 21761 . 012 When confronted with identical physical stimulation , subjects showed a consistent bias for blind spot inferred percepts which was stronger than the bias at any other location in the temporal visual field . Why do subjects choose the blind spot location when it is objectively the least reliable ? Our interpretation takes the results at face value: subjects must possess at least implicit information about whether a percept originates from the blind spot in order to show a bias for or against it . At the same time , the veridical information from the other stimulus is also available . This indicates that perceptual decision-making can rely more on inferred than veridical information , even when there is some knowledge about the reduced reliability of the inferred input available in the brain ( Ehinger et al . , 2015 ) . This is also supported by the results of the reaction time analyses that indicated a faster evidence accumulation for the inferred percepts . In other words , the implicit knowledge that a filled-in stimulus is objectively less reliable does not seem to be used for perceptual decision-making . This suboptimal decision between qualitatively different veridical and inferred inputs is in contrast to properties of standard sensory integration . There , reduced reliability derived from noisy but veridical signals results in a corresponding weighting of inputs and consequently in optimal decisions ( Körding et al . , 2007 ) . In the following , we discuss two potential explanations of this discrepancy of processing filled-in information and standard sensory integration . The first explanation focuses on physiological properties of neuronal and small circuits’ response properties at and around the blind spot region . The second explanation addresses the conceptual level and uses the general notion of predictive coding . First , although the filled-in percept is by definition independent of the stimulus within the blind spot , it is nevertheless based on the information sensed by the region around the blind spot in the nasal retina . We might assume that an area , e . g . in the nasal retina around the blind spot region , that has a lower contrast threshold also shows stronger neuronal signals for super-threshold stimuli . This could in principle lead to a filled-in stimulus with increased salience as compared to the veridical stimulus . Effectively , this explanation proposes that differences in physiological properties of nasal and temporal retinae are transferred to the filling-in process making it the ‘better’ candidate stimulus in an ambiguous condition . Above we already introduced some evidence for psychophysical differences between the nasal and temporal visual field ( Fahle and Schmid , 1988 ) . There is also some evidence for the superiority of the blind spot in a Vernier task ( Crossland and Bex , 2009 ) . The areas around the blind spot showed greater performance compared to areas at similar eccentric locations in the nasal visual field . It is still unclear whether this goes over and beyond the aforementioned temporal/nasal bias . Unfortunately , this explanation runs into the problem that the sensitivity in the region corresponding to the blind spot in the other eye is also enhanced compared to regions at similar eccentricities ( Wolf and Morandi , 1962; Midgley , 1998 ) . This suggests that differences between the eyes in the area around the blind spot should be the smallest within the contrast between temporal and nasal retina . Moreover , we explicitly controlled for temporal-nasal differences in experiments 2 and 3 , and found that it is not enough to explain the effect specific to the blind spot . Thus , an explanation of the observed effects based on known differences in retinal properties is currently tentative at best . An alternative explanation is based on the framework of predictive coding ( Friston et al . , 2006 , 2012; Summerfield and de Lange , 2014 ) . Specifically , context information of static stimuli would be used to predict local stimulus properties leading to the phenomenon of filling-in . The predicted sensory input would then be compared to the incoming sensory input , and an error signal representing the mismatch would be returned . In the absence of veridical information , no deviation and thus no error signal would occur . Effectively , the filled-in signal might have less noise . Reduced noise , in turn , results in a smaller prediction error and higher credibility at later stages . A faster reaction time to the filled-in stimulus compared to the veridical stimulus could suggest that the integration process is indeed biased with less noise . In summary , although the results reported here seem compatible with the predictive coding framework , this explanation presently remains vague and speculative . In conclusion , we find a new behavioral effect where subjects prefer a partially inferred stimulus to a veridical one . Though both appear to be continuous , the filled-in one could hide an inset and is , therefore , less reliable . In this perceptual decision-making task , subjects do not make use of high-level assessments about the reliability of the filled-in stimulus . Even more so , they prefer the unreliable percept . Overall , 175 subjects took part in the experiments . Of the subjects , 32% ( n = 56 ) were removed due to the screening experiments described below . An additional 3% ( n = 6 ) were removed due to low performance ( n = 2 , <75% in at least two conditions with a visible unique inset ) or because they responded to the stimuli with the inset stimulus instead of the continuous stimulus ( n = 4 ) . The experimental data were not recorded in 7% ( n = 13 ) due to eye tracking calibration problems ( n = 4 ) and other issues during data collection ( n = 9 ) . The remaining 100 subjects were recorded and analyzed in the following experiments . For the first experiment , 24 subjects entered the analysis ( average age 21 . 9 years , age range 18–28 years , 12 female , 20 right-handed , 16 right-eye dominant ) . Fifteen of these subjects participated in the EEG study reported by Ehinger et al . , 2015 . In the second experiment , 27 subjects entered the analysis ( average age 22 . 4 years , age range 19–33 years , 15 female , 25 right-handed , 19 right-eye dominant ) . In the third , 24 subjects entered the analysis ( average age 21 . 9 years , range 19–27 years , 19 female , 23 right-handed , 16 right-eye dominant ) . In the fourth experiment , we report the results of 25 subjects ( average age 22 . 1 , range 18–35 , 20 female , 24 right-handed , 14 right-eye dominant ) . In the last experiment , the same set of subjects participated as in experiment four with the exception of a single subject , who did not finish the both parts of the combined session with experiment 4 and 5 . All subjects gave written informed consent , and the experiment was approved by the ethics committee of the Osnabrück University . For the initial experiment , we set out to analyze 25 subjects . For the second experiment , we calculated a sample size of 18 subjects based on the results of experiment one in order to have a power of 90% ( calculated with gPower , ( Faul et al . , 2009 ) , matched pair means cohen's-d = 0 . 72 , planned-power 90% ) . We disclose that the results of the initial analysis with this group were not conclusive about differences between the location inside and the location above the blind spot . Although the sample size was large enough to replicate the blind spot main effect , it was not adequate to find the difference between locations . Therefore , we decided to increase the number of subjects by 50% ( n = 9 ) . For the third , fourth and fifth experiments , we used an empirical power analysis based on MLE of a linear mixed model in order to achieve 90% power for the smallest effect observed outside the blind spot . This resulted in a sample of 24 subjects . As described above , many subjects failed a simple screening test . In this pre-experiment , we showed a single stimulus in the periphery either inside or outside the blind spot in the left or right visual field . In two blocks of 48 trials , subjects indicated which stimulus ( no inset vs . inset ) had been perceived . We thought of this simple experiment to evaluate our blind spot calibration method , as an inset stimulus inside the blind spot should have been reported as no inset . The first block was used as a training block . In the second block , we evaluated the performance in a conservative way . No feedback was given to the subjects . If the performance was below 95% ( three errors or more ) , we aborted the session because the participant was deemed to be too unreliable to proceed further with our experiment . We analyzed the data of those that failed the screening experiment , in four categories of failures that demonstrate the heterogeneity of subjects: Subjects reported inset when an inset was shown in the left blind spot ( 44% ) , or in the right blind spot ( 78% ) . Subjects did not report the inset of a stimulus presented outside the blind spot ( 37% ) , and subjects reported an inset , even though a continuous stimulus was shown ( 80% ) . The percentage represents how many subjects had at least one trial where a classification-criterion was fulfilled and thus do not add to 100% . The rates for subjects that did not fail the criterion were 16% , 21% , 13% and 22% respectively . The high percentage in the last category of removed subjects , in which they report an inset even though no inset was visible , strongly suggests that subjects failed the task not due to blind spot related issues , but due to inattention or perceptual problems . Even though we observe more wrong reports in the right than the left blind spot position , there was nevertheless no correlation with calibration position or size . Overall , 57% ( n = 100 ) of the recruited subjects passed this test and were admitted to subsequent experiments . A remote , infrared eye-tracking device ( Eyelink 1000 , SR Research ) with a 500 Hz sampling rate was used . The average calibration error was kept below 0 . 5° with a maximal calibration error of 1 . 0° . Trials with a fixation deviation of 2 . 6° from the fixation point were aborted . We used a 24-inch , 120 Hz monitor ( XL2420t , BenQ ) with a resolution of 1920 × 1080 pixels in combination with consumer-grade shutter glasses for monocular stimulus presentation ( 3D Vision , Nvidia , wired version ) . The shutter glasses were evaluated for appropriate crosstalk/ghosting using a custom-manufactured luminance sensor sampling at 20 kHz . The measured crosstalk at full luminance was 3 . 94% . The subject screen distance was 60 cm in experiment 1 , 2 , 4 , and 5 and 50 cm in the third experiment . Modified Gabor patches with a frequency of 0 . 89 cycles/° and a diameter of 9 . 6° were generated . Two kinds of patterns were used ( Figure 1a ) : one completely continuous and one with a small perpendicular inset of 2 . 4° . For comparison , the blind spot typically has a diameter of 4°–5° . The Gabor had constant contrast in a radius of 6 . 3° around the center . This ensured the same perception of the continuous stimulus outside the blind spot in comparison to a filled-in stimulus , where the inner part is inside the blind spot . To account for possible adaptation effects , horizontal and vertical stimuli were used in a balanced and randomized way across the trials . Stimuli were displayed using the Psychophysics Toolbox ( Brainard , 1997 , RRID: SCR:002881 ) and Eyelink Toolbox ( Cornelissen et al . , 2002 ) . The stimuli were displayed centered at the individually calibrated blind spot location . The stimulus at the location above the blind spot in experiment two was at the same distance as the blind spot but was rotated by 25° to the horizon around the fixation cross . For the inward and outward condition of experiment 3 , stimuli were moved nasally or temporally by 8 . 6° . Thus the stimuli had an overlap of only 1° . Less overlap is not possible without either cutting the border of the screen or overlapping with the fixation cross . After a fixation period of 500 ms , we presented two stimuli simultaneously to the left and right of the fixation cross . Subjects were instructed to indicate via button press ( left or right ) which stimulus was continuous . Each stimulus was presented either in the temporal or nasal field of view . In some trials , the required response was unambiguous , when one of the stimuli showed an inset and the other did not ( and the inset stimulus was presented outside the blind spot ) . In many trials ( 80% of all experiments and locations , 46% when the stimulus was shown above the blind spot in experiment 2 ) , both stimuli were continuous and no uniquely correct answer existed ( see Figure 1—figure supplement 1 for a detailed overview of the balancing ) . All trials were presented in a randomized order . If the subject had not given an answer after 10 s , the trial was discarded , and the next trial started . All in all , subjects answered 720 trials over six blocks; in experiment one the trials were split up into two sessions . After each block , the eye tracker and the blind spot were re-calibrated . After cleaning trials for fixation deviation and blinks , an average of 662 trials ( 90%-quantile: 585 , 710 ) remained . For two subjects , only 360 trials could be recorded . In several figures , we present data with summary statistics . To construct the confidence intervals we used bias-corrected , accelerated 95% bootstrapped confidence intervals of the mean with 10 , 000 resamples . Note that the summary statistics do not need to conform to the posterior summary estimates because they are marginals . Only the posterior model values reflect the estimated effect . In order to calibrate the blind spot locations , subjects were instructed to use the keyboard to move a circular monocular probe on the monitor and to adjust its size and location to fill the blind spot with the maximal size . They were explicitly instructed to calibrate it as small as necessary to preclude any residual flickering . The circular probe flickered from dark gray to light gray to be more salient than a probe with constant color ( Awater et al . , 2005 ) . All stimuli were presented centered at the respective calibrated blind spot location . In total , each subject calibrated the blind spot six times . For the following comparisons of blind spot characteristics , we evaluated one-sample tests with the percentile bootstrap method ( 10 , 000 resamples ) of trimmed means ( 20% ) with alpha = 0 . 05 ( Wilcox , 2012 ) . For paired two-sample data , we used the same procedure on the difference scores and bias-corrected , accelerated 95% bootstrapped confidence intervals of the trimmed mean ( 20% ) . We report all data combined over all experiments . In line with previous studies ( Wolf and Morandi , 1962; Ehinger et al . , 2015 ) , the left and right blind spots were located horizontally at −15 . 52° ( SD = 0 . 57° CI:[−15 . 69° , −15 . 36°] ) and 15 . 88° ( SD = 0 . 61° CI:[15 . 70° , 16 . 07°] ) from the fixation cross . The mean calibrated diameter was 4 . 82° ( SD = 0 . 45° CI:[4 . 69° , 4 . 95°] ) for the left and 4 . 93° ( SD = 0 . 46° CI:[4 . 79° , 5 . 07°] ) for the right blind spot . Left and right blind spots did significantly differ in size ( p=0 . 009 , CI:[−0 . 17° , −0 . 03°] and in absolute horizontal position ( in relation to the fixation cross; p<0 . 001 , CI: [0 . 27° , 0 . 45°] ) . On average , the right blind spot was 0 . 36° further outside of the fixation cross . No significant difference was found in the vertical direction ( p=0 . 37 ) , but this is likely due to the oval shape of the blind spot in this dimension and the usage of a circle to probe the blind spot . These effects seem small , did not affect the purpose of the experiments and will not be discussed further . We fitted a Bayesian logistic mixed effects model predicting the probability of responding ‘right’ with multiple factors that represent the temporal over nasal bias and several other covariates described below . Because we were interested in the bias between the nasal fields and the temporal fields of view , we combined both predictors for the left and right temporal ( and nasal , respectively ) locations and reported the combined value . Data were analyzed using a hierarchical logistic mixed effects models fitted by the No-U-Turn Sampler ( NUTS , STAN Development Team ) . The model specification was based on an implementation by Sorensen , Hohenstein and Vasisth ( Sorensen et al . , 2016 ) . In the results section , we report estimates of linear models with the appropriate parameters fitted on data of each experiment independently . We also analyzed all data in one combined model: there were no substantial differences between the results from the combined model and the respective submodels ( Supplementary file 1 ) . The models are defined as follows using the Wilkinson notation:answerRight ∼1+TemporalLeft∗Location + TemporalRight∗Location+AnswerRight ( t−1 ) +HandednessRight+DominantEyeRight+ ( 1+TemporalLeft∗Location + TemporalRight∗Location+AnswerRight ( t−1 ) \|Subject ) AnsweriRight∼Bernoulli ( θi ) θi=logit−1 ( Xwithinβwithin+Xbetweenβbetween+N ( 0 , τXwithin ) +N ( 0 , ei ) ) Two factors were between subjects: handedness and dominant eye . In total , we have four within-subject factors , resulting in eight parameters: There are two main factors representing whether the left , and respectively the right , stimulus was inside or outside the temporal field . Depending on the experiment , the main factor location had up to three levels: the stimuli were presented outward ( Exp . 3 ) , inward ( Exp . 3 ) , above ( Exp 2 , 5 ) or on ( all experiments ) the blind spot . In addition , we modeled the interactions between location and whether the left stimulus ( and the right stimulus , respectively ) was shown temporally . In order to assure independence of observation , an additional within-subject main factor answer ( t-1 ) was introduced , which models the current answer based on the previous one . In frequentist linear modeling terms , all within-subject effects were modeled using random slopes clustered by subject and a random intercept for the subjects . We used treatment coding for all factors and interpreted the coefficients accordingly . In the model , we estimated the left and right temporal field effects separately . For the statistical analysis , we combined these estimates by inverting the left temporal effect and averaging with the right temporal effect . We did this for all samples of the mcmc-chain and took the median value . We then transformed these values to the probability domain using the inverse-logit function , subtracting the values from 0 . 5 and multiplying by 100 . All results were still in the linear range of the logit function . We calculated 95% credible intervals the same way and reported them as parameter estimates ( CDI95 lower-upper ) in the text . These transformed values represent the additive probability ( in % ) of choosing a left ( right ) stimulus that is shown in the left ( right ) temporal field of view compared to presenting the left ( right ) stimulus in the nasal field of view , keeping all other factors constant . Initially , we did not plan to analyze the reaction time data . These analyses are purely explorative . The response setup consisted of a consumer keyboard . Thus delays and jitters are to be expected . However , with an average of 494 ambiguous trials per subject , we did not expect a spurious bias between conditions due to a potential jitter . Reaction time data was analyzed with a simple Bayesian mixed linear model:RT∼1+Temporalselected∗Location+ ( 1+Temporalselected∗Location\|Subject ) Only trials without a visible inset stimulus were used . Temporal selected consists of all trials where a temporal stimulus was selected . Because of the bias described in the results , there was no imbalance between the number of trials in the two conditions ( difference of 10 trials bootstrapped-CI [−2 , 23] ) . We did not make use of prior information in the analysis of our data . We placed implicit , improper , uniform priors from negative to positive infinity on the mean and 0 to infinity for the standard deviations of our parameters , the default priors of STAN . An uninformative lkj-prior ( ν=2 ) was used for the correlation matrix , slightly emphasizing the diagonal over the off-diagonal of the correlation matrix ( Sorensen et al . , 2016; Carpenter et al . , 2017 ) . We used six mcmc-chains using 2000 iterations each , with 50% used for the warm-up period . We visually confirmed convergence through autocorrelation functions and trace plots , then calculated the scale reduction factors ( Gelman et al . , 2014 ) , which indicated convergence ( Rhat < 1 . 1 ) . Posterior predictive model checks were evaluated to test for model adequacy ( Gelman et al . , 2013 ) . Posterior predictive checks work on the rationale that newly generated data based on the model fit should be indistinguishable from the data that the model was fitted by originally . Due to our hierarchical mixed model , we perform posterior predictive checks on two levels: trial , and subject . In the first case , we generate new datasets ( 100 samples ) based on the posterior estimates of each subject’s effect . We compare the distribution of this predicted data with the actual observed values for each ( Figure 6—figure supplement 2a ) . At the subject level , we draw completely new data sets , based on the multivariate normal distribution given by the random effects structure . We then compare the collapsed blind spot effect once for the newly drawn subjects with the observed data ( Figure 6—figure supplement 2b ) . Taken together , these posterior predictive model checks show that we adequately capture the very diverse behavior of our subjects but also correctly model the blind spot effect on a population basis . Here we report the result of the covariate factor based on the combined model ( all experiments modeled together ) . Note that the interpretation of such effects naturally occurs on logit-transformed values . Summation of different parameter-levels ( as necessary for treatment coding ) on logit-scale can be very different to summations of raw-percentage values . It can also be similar , close to the linear scale of the logit-transform , that is , close to 50% ( which we made use of for the blind spot effects reported at other points of the manuscript ) . We did not find evidence for a different global bias ( main effect location ) in any of the four stimulation positions tested here . The dominant eye factor had an 11 . 51% effect ( CDI952 . 78–19 . 59% ) on the global bias . Thus subjects with a dominant right eye also showed a preference to the right stimulus over the left one , irrespective of whether the stimulus was visible through the left or the right eye . We find a global bias ( the intercept , −26 . 75% CDI95−38 . 18% to −9 . 29% , with treatment coding ) toward choosing the left stimulus; this might reflect that in the first two experiments we instructed subjects to use the right hand , thus they used their index and middle fingers . In the third experiment we instructed subjects to use both index fingers , resulting in a decreased bias to the left , with a shift more to the right , and thus more to balanced answers , of 12 . 24% ( CDI95−1 . 98–24 . 16%] ) . We did not find evidence for a bias due to handedness ( 7 . 71% , CDI95−8 . 96–22 . 75% ) . There was an influence of the previous answer on the current answer . We observe a global effect of 7 . 86% ( CDI95 0 . 53%–14 . 95% ) , which suggests that subjects are more likely to choose e . g . the right stimulus again when they have just chosen ‘right’ in the previous trial . For this effect it is more important to look at random effect variance , which is quite high with a standard deviation of −31 . 4 ( CDI9528 . 27–34 . 69% ) , suggesting that there is a large variation between subjects . Indeed , a closer look at the random slopes of the effect reveals three different strategies: Some subjects tend to stick the same answer , some subjects are balanced in their answers without any trend , and some subjects tend to regularly alternate their answers in each trial . Note that this behavior does not seem to influence any of the other effects: We do not see any correlation between the random effects , except for the correlation between the n-1 effect and the intercept ( −0 . 55 , CI: −0 . 72 , –0 . 34 ) . This correlation means that subjects who tend to alternate their key presses will not have a strong bias in the intercept , or the other way around , subjects who press the same key all the time also have a bias towards this key . Other extended models we considered showed no effect when both stimuli were in the temporal field , nor any three-way interaction . Following standard procedures to avoid spurious effects of unnecessary degrees of freedom , we removed these variables from the final model .	To make sense of the world around us , we must combine information from multiple sources while taking into account how reliable they are . When crossing the street , for example , we usually rely more on input from our eyes than our ears . However , we can reassess the reliability of the information: on a foggy day with poor visibility , we might prioritize listening for traffic instead . But how do we assess the reliability of information generated within the brain itself ? We are able to see because the brain constructs an image based on the patterns of activity of light-sensitive proteins in a part of the eye called the retina . However , there is a point on the retina where the presence of the optic nerve leaves no space for light-sensitive receptors . This means there is a corresponding point in our visual field where the brain receives no visual input from the outside world . To prevent us from perceiving this gap , known as the visual blind spot , the brain fills in the blank space based on the contents of the surrounding areas . While this is usually accurate enough , it means that our perception in the blind spot is objectively unreliable . To find out whether we are aware of the unreliable nature of stimuli in the blind spot , Ehinger et al . presented volunteers with two striped stimuli , one on each side of the screen . The center of some of the stimuli were covered by a patch that broke up the stripes . The volunteers’ task was to select the stimulus with uninterrupted stripes . The key to the experiment is that if the central patch appears in the blind spot , the brain will fill in the stripes so that they appear to be continuous . This means that the volunteers will have to choose between two stimuli that both appear to have continuous stripes . If they have no awareness of their blind spot , we might expect them to simply guess . Alternatively , if they are subconsciously aware that the stimulus in the blind spot is unreliable , they should choose the other one . In reality , exactly the opposite happened: the volunteers chose the blind spot stimulus more often than not . This suggests that information generated by the brain itself is sometimes treated as more reliable than sensory information from the outside world . Future experiments should examine whether the tendency to favor information generated within the brain over external sensory inputs is unique to the visual blind spot , or whether it also occurs elsewhere .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Humans treat unreliable filled-in percepts as more real than veridical ones
The progenitor cells of the developing liver can differentiate toward both hepatocyte and biliary cell fates . In addition to the established roles of TGFβ and Notch signaling in this fate specification process , there is increasing evidence that liver progenitors are sensitive to mechanical cues . Here , we utilized microarrayed patterns to provide a controlled biochemical and biomechanical microenvironment for mouse liver progenitor cell differentiation . In these defined circular geometries , we observed biliary differentiation at the periphery and hepatocytic differentiation in the center . Parallel measurements obtained by traction force microscopy showed substantial stresses at the periphery , coincident with maximal biliary differentiation . We investigated the impact of downstream signaling , showing that peripheral biliary differentiation is dependent not only on Notch and TGFβ but also E-cadherin , myosin-mediated cell contractility , and ERK . We have therefore identified distinct combinations of microenvironmental cues which guide fate specification of mouse liver progenitors toward both hepatocyte and biliary fates . The cells which populate the hepatic diverticulum during development and later serve as the source of liver parenchyma are termed bipotential progenitor cells , or hepatoblasts , as they are capable of differentiating toward both hepatocytic and biliary epithelial cell fates . While differentiation of liver progenitors toward a hepatocytic fate is guided chiefly by signaling through Wnt , HGF , and FGF ( Micsenyi et al . , 2004; Berg et al . , 2007; Schmidt et al . , 1995 ) , biliary fate is regulated by Notch and TGFβ signaling ( Kodama et al . , 2004; Clotman et al . , 2005; Zong et al . , 2009 ) . Specifically , a gradient of TGFβ activity caused in part by expression of TGFβR2 and TGFβR3 in the periportal region leads to differentiation of progenitors toward a biliary epithelial fate ( Clotman et al . , 2005 ) . In patients with Alagille syndrome , mutations in the ligand JAG1 or receptor NOTCH2 are associated with bile duct paucity and cholestasis ( Li et al . , 1997; Oda et al . , 1997; McDaniell et al . , 2006 ) . Zong et al . further underlined the importance of Notch in particular for both biliary cell fate and morphogenesis by showing that deletion of the Notch effector Rbpj results in reduction of both biliary fate and abnormal tubulogenesis ( Zong et al . , 2009 ) . Thus , the progenitor cells of the developing liver integrate a diverse set of biochemical cues during fate specification . Several recent lines of evidence suggest , however , that liver progenitor cells are influenced not only by biochemical cues but also biophysical parameters in their microenvironment . Using combinatorial extracellular matrix ( ECM ) protein arrays , we showed that TGFβ-induced biliary differentiation of liver progenitor cells is coordinated by both substrate stiffness and matrix context and is further correlated with cell contractility ( Kourouklis et al . , 2016 ) . Several groups have established mechanosensing via the transcriptional co-activator YAP and further elaborated a novel role for this protein in the developing cells of the liver ( Camargo et al . , 2007; Dupont et al . , 2011; Yimlamai et al . , 2014; Lee et al . , 2016 ) . This is particularly interesting in the context of liver progenitor fate specification because YAP has been shown to regulate both Notch signaling and TGFβ in liver cells ( Yimlamai et al . , 2014; Lee et al . , 2016 ) . However , the potential link between mechanical sensing and the fate specification of liver progenitor cells has yet to be fully defined . Here , we utilize microarrayed patterns of ECM co-printed with Notch ligands to provide a controlled biochemical and biomechanical environment for liver progenitor cell differentiation . We characterize spatially-localized , segregated differentiation of these progenitor cells toward biliary fates at the periphery of patterns and hepatocytic fates near the center of patterns . We employ traction force microscopy ( TFM ) to measure cell-generated forces , observing high stresses coincident with peripheral biliary differentiation . Further , we explore the dependence of peripheral biliary differentiation of progenitors on mechanotransduction pathway activity and expression of the Notch ligands Jag1 and Dll1 . Collectively , our findings provide support for a model of liver progenitor differentiation which includes mechanical signaling as a key regulator of spatially-segregated progenitor differentiation and downstream biliary morphogenesis . We have previously observed peripheral expression of the biliary marker osteopontin ( OPN ) in liver progenitors on arrayed patterns containing both ECM proteins and Notch ligands ( Kaylan et al . , 2016 ) . In order to better characterize the expression profile of cells in the periphery vs . center , we fabricated arrays of circular patterns ( ∼600 μm diameter ) containing the ECM protein collagen I paired with either control IgG or Fc-recombinant Notch ligands ( DLL1 , DLL4 , and JAG1 ) . These ligands were pre-conjugated to Protein A/G so as to increase ligand functionality by clustering and orientation . Bipotential mouse embryonic liver ( BMEL ) progenitor cells , which are capable of assuming a hepatocytic or biliary fate ( Strick-Marchand and Weiss , 2002 ) , were seeded on these Notch ligand arrays and cultured under differentiation conditions for t=72h , at which point we immunolabeled for OPN and the hepatocytic marker albumin ( ALB ) . Within these defined multicellular geometries , we observed OPN+ cells at the periphery of patterns while ALB+ cells were located centrally ( Figure 1A ) . Counts of cells that were OPN+ peaked at the periphery and increased with the presentation of Notch ligands , particularly DLL4 ( Figure 1B ) . However , counts of cells that were ALB+ cells indicated central localization and only moderate induction by ligand in the center of patterns ( Figure 1C ) . Multiple regression analysis of these data generated coefficient estimates ( β ) for each presented ligand , corresponding to the mean change in cell counts from control IgG ( Figure 1E and Figure 1F ) . These coefficient estimates confirmed increases in both peripheral OPN+ ( β=37 . 5 , P<0 . 001 ) and central ALB+ ( β=5 . 64 , P<0 . 001 ) cell counts upon presentation with DLL4 . Evaluation of the expression of the biliary transcription factor SOX9 and hepatocytic transcription factor HNF4A revealed segregation similar to that of OPN and ALB ( Figure 1D ) . Specifically , SOX9-expressing cells were at the periphery while HNF4A-expressing cells were central . We also evaluated expression of the biliary marker cytokeratin 19 ( CK19 ) ( Figure 1—figure supplement 1 ) and observed 1 . 6 times greater intensity in cells at the periphery compared to those in the center ( P<0 . 001 ) ( Figure 1—figure supplement 1 ) . We observed peripheral expression of both OPN and CK19 at t=24h , suggesting that segregation starts earlier than t=72h and is less likely to be dependent on cell motility mechanisms ( Figure 1—figure supplement 2A ) . Measurements of cell density across the island at t=72h indicated uniform density with radius , ruling out cell condensation as a mechanism of differentiation ( Figure 1—figure supplement 2B ) . In preliminary experiments , we determined that patterns of approximately 600 µm diameter would lead to consistent patterned differentiation . Accordingly , for most our studies here , 600 µm diameter patterns were utilized . However , to examine potential effects of pattern diameter , we generated complementary array sets that resulted in cellular island diameters of 300 µm and 1000 µm , in addition to 600 µm ( Figure 1—figure supplement 3A ) . Quantification of peak OPN+ cell counts on these pattern sizes indicated that biliary differentiation remained confined to the periphery independent of pattern size ( Figure 1—figure supplement 3B ) . Together , these data establish spatially-segregated liver progenitor fates in arrayed patterns with central hepatocytic differentiation and peripheral biliary differentiation . We next asked if Notch signaling is necessary for peripheral biliary differentiation in arrayed patterns . We treated cultures with an inhibitor of Notch signaling ( γ-secretase inhibitor X , GSI ) and observed reduction in OPN+ cell counts at the periphery ( Figure 2A and B ) . Prompted by previous experiments which showed that liver progenitor differentiation is sensitive to substrate stiffness ( Kourouklis et al . , 2016 ) , we also evaluated progenitor differentiation on soft ( 4 kPa ) rather than stiff ( 30 kPa ) substrates , observing decreased counts of peripheral OPN+ cells and similar responsiveness to GSI ( Figure 2A and B , Figure 2—figure supplement 1 ) . Multiple regression analysis of these data confirmed reduction in peripheral OPN+ with both GSI treatment ( β=-9 . 99 , P<0 . 001 ) and culture on 4 kPA substrates ( β=-3 . 10 , P=0 . 00292 ) ( Figure 2—figure supplement 2 ) . ALB+ cell counts increased with GSI treatment ( β=5 . 41 , P<0 . 001 ) , suggesting that hepatocytic differentiation is inhibited by active Notch signaling ( Figure 2—figure supplement 3 and Figure 2—figure supplement 2 ) . We also evaluated expression of SOX9 and HNF4A , observing reduction in peripheral SOX9 expression and an increase in central HNF4A expression on soft substrates compared to stiff ( Figure 2C ) . Quantification of immunolabel intensity for SOX9 and HNF4A on both soft and stiff substrates confirmed our qualitative observations ( Figure 2D ) , indicating a 74 . 7% increase in overall SOX9 intensity on 30 kPa substrates ( relative to 4 kPa substrates , P<0 . 001 ) and 40 . 6% increase in overall HNF4A intensity on 4 kPa substrates ( relative to 30 kPa substrates , P<0 . 001 ) . Using in situ hybridization of mRNA , we characterized the expression of Notch family members in arrayed patterns of liver progenitors . To do so , we validated several probes against Jag1 , Dll1 , and Notch2 ( data not shown ) . When used to detect mRNA in arrayed patterns fabricated on stiff ( 30 kPa ) substrates , we observed peripheral localization of Jag1 , Dll1 , and Notch2 ( Figure 2E ) . Presentation of the ligand DLL4 induced rearrangement of this expression pattern , specifically causing an increase in centrally-located cells expressing mRNA for each gene . On soft ( 4 kPa ) substrates , we observed similar mRNA expression for cells presented with IgG but no longer observed ligand-induced central expression for Jag1 and Notch2 . This loss of ligand-induced central expression on soft substrates suggests that the responsiveness of liver progenitors to Notch ligand is enhanced by stiffer substrates . Collectively , these data show that segregation of liver progenitor fates is dependent on both Notch signaling and substrate stiffness . Previous studies have delineated a role for TGFβ in liver progenitor differentiation ( Clotman et al . , 2005 ) , and we have described interactions between TGFβ and Notch signaling in this context ( Kaylan et al . , 2016 ) . To determine if TGFβ is involved in the generation of biliary cells at the periphery of the arrayed patterns , we treated cells with an inhibitor of TGFβ type I receptor kinase signaling ( SB-431542 ) or stimulated with exogenous TGFβ1 ( Figure 3A ) . Treatment with SB-431542 reduced the peripheral count of OPN+ cells while increasing central expression of HNF4A ( Figure 3A , Figure 3—figure supplement 1 ) . In contrast , treatment with TGFβ1 increased counts of OPN+ cells uniformly across the patterns irrespective of ligand presented ( Figure 3A and B ) , in agreement with previous efforts showing uniform induction of OPN on patterns of smaller diameter ( 150 µm ) ( Kourouklis et al . , 2016 ) . Similarly , in situ hybridization for Jag1 , Dll1 , and Notch2 mRNA showed uniform induction across the patterns with TGFβ1 treatment ( Figure 3—figure supplement 2 ) . Interestingly , we observed loss of cell–cell junctional interactions in cells treated with TGFβ1 , which is thought to be a consequence of inhibition of E-cadherin expression by the Snail family of transcription factors ( Cano et al . , 2000; Vincent et al . , 2009 ) . To ascertain the impact of E-cadherin function without activation of the other regulatory programs of TGFβ , we treated cells with a functional antibody against E-cadherin ( DECMA ) ( Figure 3A ) . In contrast with our observations following treatment with TGFβ1 , we observed differential responsiveness to control IgG and DLL4 presentation ( Figure 3B ) . Specifically , presentation of DLL4 to cells treated with DECMA resulted in uniform induction of OPN+ cells across the patterns . We confirmed this observation using the Kolmogorov–Smirnov test , which showed that the difference between the IgG and DLL4 probability density distributions ( measured by D , the supremum distance ) was greater for DECMA ( D=0 . 437 , P<0 . 001 ) compared to both DMSO ( D=0 . 0655 , P<0 . 001 ) and TGFβ1 ( D=0 . 0848 , P=0 . 0350 ) . Last , although inhibition of TGFβ by treatment with SB-431542 reduced OPN+ cell counts , mRNA in situ hybridization of cultures treated with SB-431542 indicated that both Jag1 and Notch2 remain expressed at the periphery ( Figure 3C ) . However , SB-4315412 treatment reduced expression of both Jag1 and Notch2 in centrally-located cells presented with DLL4 ( Figure 3C ) , which we had previously observed in untreated cultures ( Figure 2E ) . These data therefore demonstrate that TGFβ only partially regulates fate segregation and that these effects are additionally mediated by cell–cell junctional interactions through E-cadherin . Others have established a role for mechanical stresses in multicellular pattern formation and stem cell differentiation , specifically observing collection of mechanical stresses at the corners and edges of geometric shapes ( Nelson et al . , 2005; Ruiz and Chen , 2008; Kilian et al . , 2010; Ma et al . , 2015 ) . Having previously demonstrated a combinatorial role for biochemical and biomechanical stimuli in liver progenitor cell fate ( Kourouklis et al . , 2016 ) , we hypothesized that mechanical stress gradients are involved in the segregation of liver progenitor fates arrayed patterns . To obtain theoretical predictions of mechanical stress , we used finite element modeling ( FEM ) of an active layer ( i . e . , the cell monolayer ) of 600 µm diameter bound to a passive substrate with fixed lower boundary ( Figure 4A ) . We observed peak stresses of 150 Pa at the periphery of the active layer ( Figure 4B ) , in agreement with previous simulations ( Nelson et al . , 2005 ) . Next , we used TFM to obtain experimental measurements in liver progenitor cells , observing that traction stresses are collected at the periphery of patterns on both 30 kPa and 4 kPa substrates ( Figure 4C ) . The peak magnitude and distribution of stresses across the patterns did not vary with ligand presentation ( Figure 4D ) . However , we did observe that central cells ( R<0 . 75 ) on 30 kPa substrates exerted stresses averaging to 32 . 9 Pa , which was statistically greater than the 16 . 2 Pa of stress exerted by cells on 4 kPa substrates ( P<0 . 001 ) . TFM measurements of cells treated with GSI showed that Notch signaling was not upstream of traction stress generation at the periphery ( Figure 4—figure supplement 1 ) . In contrast , inhibition of TGFβ by treatment with SB-431542 resulted in more uniform traction stress distributions in cells presented with both IgG and DLL4 ( Figure 4—figure supplement 2 ) . Intriguingly , treatment with functional antibody against E-cadherin ( DECMA ) resulted in more uniform traction stress distribution in cells presented with IgG but not DLL4 , indicating that ligand presentation in the context of reduced cell–cell interactions induces cell-generated traction stresses . In sum , these data show that mechanical stresses are collected at the periphery , coincident with peripheral biliary fate , and are further dependent on TGFβ signaling and E-cadherin interactions between cells . Having established the presence of gradients of mechanical stress in patterns , we next hypothesized that these gradients are involved in the segregated differentiation of liver progenitors . In order to first determine whether regulation by this gradient of mechanical stress is consistent with known modes of Notch signaling , we used a lattice-based computational model described by the groups of Elowitz and Sprinzak ( Sprinzak et al . , 2010; Formosa-Jordan and Sprinzak , 2014 ) . We adapted their computational model to include: ( 1 ) fixed boundary conditions to better represent the physical boundary of our arrayed patterns; and ( 2 ) an additional term representing the effect of the stress gradient on expression of both Notch receptor and ligand , as observed in our mRNA in situ hybridization experiments . A model of trans-activation ( Kt=10 and Kc=0 ) with increasing stress gradient strength ( b=0 , 0 . 5 , 5 ) produced segregation of fates qualitatively similar to our experimental results ( Figure 5A ) . Concentration profiles of Notch receptor and repressor , a measure of Notch signaling activity , in models including trans-activation and steeper stress gradients were also qualitatively consistent with our experimental data ( Figure 5B ) . Notably , simulations without stress suggested a biphasic distribution of receptor , which we did not observe experimentally . As experimental validation , we treated cells with blebbistatin ( Figure 5C ) , an inhibitor of myosin II ATPases , and observed reduced peripheral OPN+ cell percentages ( Figure 5D ) . These observations are in agreement with our previous experiments ( Kourouklis et al . , 2016 ) , indicating that myosin-mediated contractility is necessary for peripheral biliary differentiation . TFM measurements obtained in parallel indicated loss of peripheral traction stresses in cells treated with blebbistatin ( Figure 5D ) , in agreement with the known action of this inhibitor . These simulations demonstrate that a simple model of Notch trans-activation coupled with an external stress gradient is consistent with our experimental findings . In order to ascertain which specific mechanotransduction pathways are involved in this process , we treated cells with inhibitors for ERK ( FR180204 ) and ROCK ( Y-27632 ) ( Figure 6A ) . We observed that FR180204 reduced OPN+ cell percentages at the outer edge of the patterned domains ( Figure 6B ) , which is in accordance with our previous studies suggesting involvement of ERK in biliary differentiation ( Kourouklis et al . , 2016 ) . In contrast , inhibition of ROCK resulted in increased peripheral OPN+ cell percentages extending centrally ( Figure 6A and B ) . Consistent with the respective functions of the proteins targeted by these inhibitors , TFM measurements indicated loss of peripheral traction stresses in cells treated with Y-27632 but not FR180204 ( Figure 6C ) . Analysis of Jag1 and Notch2 mRNA expression in cells treated with FR180204 indicated that inhibition of ERK signaling results in direct reduction in expression of both ligand and receptor ( Figure 6D ) . Furthermore , we observed that the Hippo pathway effector YAP exhibited increased expression at the periphery of arrayed patterns on both 30 kPa and 4 kPa substrates ( Figure 6E ) , though the expression of YAP was not altered by the presence of Notch ligands in the arrayed domains ( data not shown ) . This observation of peripheral YAP expression is especially interesting in light of recent findings regarding the demonstrated role of YAP in biliary fate ( Yimlamai et al . , 2014 ) and suggests a potential role for the Hippo pathway in progenitor fate segregation . Collectively , these data show that peripheral biliary differentiation is dependent on myosin-mediated cell contractility and ERK signaling and is decoupled from mechanical stress when ROCK is inhibited . Studies of Alagille syndrome , a genetic disorder which results in bile duct paucity , have shown that the Notch ligand JAG1 is necessary for bile duct formation ( Li et al . , 1997; Oda et al . , 1997 ) . Our previous work has also shown that the Notch ligand Dll1 can modulate differentiation toward both biliary and hepatocytic fates ( Kaylan et al . , 2016 ) . We therefore hypothesized that the Notch ligands Jag1 and Dll1 are involved in the segregation of liver progenitor fates in arrayed patterns . Using lentiviral shRNA vectors , we knocked down Jag1 ( shJag1 ) and Dll1 ( shDll1 ) in liver progenitors and cultured them on arrayed patterns ( Figure 7A and B ) . We observed that shJag1 cells exhibited reduced OPN+ cell counts at the periphery ( β=-16 . 3 , P<0 . 001 ) while , in contrast , counts of peripheral shDll1 cells that were OPN+ increased ( β=20 . 1 , P<0 . 001 ) , observations confirmed by quantification ( Figure 7C ) and regression analysis ( Figure 7—figure supplement 1 ) . In agreement with the data for OPN , only shJag1 cells exhibited loss of peripheral SOX9 expression ( Figure 7D ) . Interestingly , knockdown of both Jag1 and Dll1 resulted in decreased central HNF4A expression ( Figure 7B and D ) . TFM of shJag1 and shDll1 cells showed no reduction in cell-generated traction stresses by ligand knockdown compared to control cells ( data not shown ) . These data establish contrasting roles for Jag1 and Dll1 in biliary differentiation in which Dll1 has the unanticipated function of antagonizing biliary fate and , further , suggest that the ligands are involved in hepatocytic differentiation of progenitor cells . Here , we utilized microarrayed patterns of ECM co-presented with Notch ligands to provide a biochemically- and biophysically-defined microenvironment for liver progenitor differentiation . In these patterns , we observed spatially-localized , segregated differentiation of progenitors toward biliary fates peripherally and hepatocytic fates centrally . Other groups have made similar observations using both 2D and 3D engineered systems as part of studies investigating the differentiation of mesenchymal and induced pluripotent stem cells ( Ruiz and Chen , 2008; Kilian et al . , 2010; Ma et al . , 2015; Lee et al . , 2015 ) . In these other cell types , pathways related to cell contractility ( e . g . , RhoA , ROCK , RAC1 ) and cell–cell adaptor proteins ( e . g . , E-cadherin ) were both implicated . We show in this work that cell contractility is a key inducer of biliary fate in liver progenitors and elaborate roles for cell–cell interactions and mechanotransduction pathway activity in addition to established regulation by Notch and TGFβ signaling . We have previously examined the role of substrate stiffness in the context of TGFβ-induced biliary differentiation , finding that progenitor cells cultured on fibronectin are sensitive to stiffness whereas cells cultured on collagen IV differentiated independent of stiffness ( Kourouklis et al . , 2016 ) . On collagen I patterns , we observed that high substrate stiffness ( E∼30kPa ) increases peripheral biliary differentiation , Notch family member expression , and responsiveness to cell-extrinsic ligand presentation ( Figure 2 ) . In contrast , low substrate stiffness ( E∼4kPa ) was more supportive of hepatocytic fate , particularly in the pattern center ( Figure 1 and Figure 2 ) . These findings are consistent with other recent efforts toward delineating the impact of substrate stiffness on hepatocyte function , which have identified potential mechanisms of transcriptional and epigenetic repression of HNF4A in hepatocytes experiencing increased cytoskeletal tension ( Desai et al . , 2016; Cozzolino et al . , 2016 ) . By integrating TFM with the array platform , we were able to localize traction stresses and associated cell contractility to the pattern periphery , coincident with biliary differentiation ( Figure 4 ) . Paradoxically , treatment with inhibitors of actomyosin contractility ( blebbistatin and Y-27632 ) resulted in divergent fate trajectories . Blebbistatin , a direct inhibitor of myosin ATPase , reduced both peripheral traction stress and downstream biliary differentiation as expected , whereas Y-27632 , an inhibitor of myosin light chain phosphorylation by ROCK , increased peripheral differentiation and extension of differentiation centrally ( Figure 5 and Figure 6 ) . It is possible this divergence is due to the antagonism of ROCK against RAC1-induced adherens-junction formation ( Wildenberg et al . , 2006 ) , suggesting that increased cell–cell interactions in the context of reduced cytoskeletal tension is supportive of biliary fate . Further evidence for this hypothesis is our observation of uniform induction of biliary fate by DLL4 presentation in cells with adherens junctions inhibited by DECMA ( Figure 3 ) , results which raise the additional possibility that Notch ligand–receptor binding is dependent on adherens junction formation . Lowell et al . provide evidence of such a mechanism in human keratinocytes , observing mutual exclusion of E-cadherin and Delta ligand and further noting that ligand expression promotes cell–cell interactions independent of adherens junction formation ( Lowell et al . , 2000 ) . Interestingly , treatment with an inhibitor of TGFβ ( SB-431542 ) reduced biliary differentiation and increased hepatocytic differentiation but failed to abolish peripheral expression of Jag1 and Notch2 ( Figure 3 ) . It is therefore not likely that TGFβ signaling is the single factor responsible for peripheral biliary fate and associated gradient formation , though it may act through autocrine or paracrine regulation to enable differentiation by other mechanisms . For instance , Zavadil et al . showed that TGFβ serves as a leading signal in the biphasic activation of HEY1 via interactions with SMAD3 and SMAD4 transcriptional regulators , whereas the lagging signal consisted of sustained HEY1-mediated activation of JAG1 signaling dependent on ERK ( Zavadil et al . , 2004 ) . In the context of liver progenitor fate , this model would require only moderate amounts of autocrine TGFβ to activate the Notch transcriptional machinery leading to ligand expression and associated biliary differentiation . In support of this model , inhibition of ERK signaling with FR180204 reduced both biliary differentiation and peripheral expression of Jag1 and Notch2 ( Figure 6 ) . Last , our observation of peripherally-expressed cytoplasmic YAP ( Figure 6 ) is intriguing in light of recent literature regarding the role of YAP as a mechanosensor ( Dupont et al . , 2011 ) and regulator of liver cell fate ( Yimlamai et al . , 2014; Lee et al . , 2016 ) and might serve as a mechanistic effector downstream of peripherally-induced cytoskeletal tension in progenitor cells . Our observations of peripherally-expressed Jag1 , Dll1 , and Notch2 ( Figure 2 ) are especially striking in light of the TFM data showing colocalization with peak traction stresses . Although we have demonstrated dependence of peripheral expression of ligand and receptor on substrate stiffness and ERK signaling , the exact mechanism linking traction stress to Notch ligand and receptor expression remains unidentified . Answering this question is crucial in order to define the role of cytoskeletal stress relative to Notch and TGFβ in biliary differentiation of liver progenitors . TFM of progenitor cells treated with GSI places generation of traction stresses prior to Notch-mediated biliary differentiation ( Figure 4—figure supplement 1 ) . In contrast , TFM of cells treated with SB-431542 provides evidence that TGFβ is upstream of traction stress ( Figure 4—figure supplement 2 ) , in accordance with the biphasic model described above in which TGFβ serves as an initial stimulus to Notch activity as well as potential feed-forward induction of cell contractility by TGFβ under conditions of mechanical stress ( Tomasek et al . , 2002 ) . Recent descriptions of new modes of non-canonical Notch signaling provide other potential mechanisms linking cytoskeletal stress and Notch through ligand–intermediate filament interactions ( Antfolk et al . , 2017 ) or Notch transmembrane domain-mediated activation of RAC1 signaling ( Polacheck et al . , 2017 ) . To gain insight into how cell mechanical stress may influence the Notch pathway , we explored the utility of incorporating mechanical stress into a multicellular model of Notch pathway dynamics ( Figure 5 ) . The results of this integrated model demonstrate that the introduction of mechanical stress as a positive regulator of Notch receptor and Notch ligand expression is sufficient to generate a patterning response with enhanced peripheral Notch activation . Notably from the in situ hybridization experiments , the presence of the Notch ligand DLL4 in the arrayed domains appeared to enhance central expression of Jag1 and Notch2 mRNA on 30 kPa but not 4 kPa substrates ( Figure 2 ) . This observation would suggest that DLL4 is acting to enhance Notch signaling centrally on 30 kPa . However , cells presented with DLL4 on 30 kPa substrates exhibited preferential biliary differentiation at the periphery with minimal biliary differentiation centrally , indicating that central expression of Notch pathway components may not be sufficient for biliary differentiation . Taken together , these findings suggest that the spatial distribution of mechanical stress signals may impact cell differentiation not only by influencing the expression of Notch pathway members but also through interactions with downstream Notch-mediated transcription or through cooperation with TGFβ and ERK , which is required for differentiation . Furthermore , future experiments incorporating additional quantitative measurements of spatial mRNA expression will be useful in identifying subtler patterns of Notch ligand and receptor expression . Knockdown of cell-intrinsic Jag1 and Dll1 further revealed distinct roles in both biliary and hepatocytic differentiation of progenitor cells ( Figure 7 ) . The reduction of central HNF4A with knockdown of either ligand is particularly interesting and suggests a role for cell–cell interactions with ligand-presenting cells in hepatocytic differentiation . The loss of biliary differentiation with Jag1 knockdown is consistent with the known role of Jag1 expressed in the mesenchyme of the portal vein ( Hofmann et al . , 2010 ) . The unanticipated increase in OPN+ cells as a consequence of Dll1 knockdown , however , has fewer precedents and suggests an cell-intrinsic inhibitory role in contrast with that of Jag1 . Although we used multiple Notch ligands ( DLL1 , DLL4 , JAG1 ) in arrays , we have largely focused on presentation of DLL4 to progenitor cells due to its consistent activation of progenitor cells . The differential cell-extrinsic activity of the ligands might be explained in part by the known preferential affinity of ligands for specific receptors ( Yamamoto et al . , 2012; Andrawes et al . , 2013 ) as well as recent evidence showing that DLL4 binds Notch receptors with greater affinity and requires less mechanical tension to activate signaling ( Luca et al . , 2017 ) . It may also be a consequence of ligand presentation in the array format , which is known to be a function of molecular weight and charge ( Flaim et al . , 2005; Reticker-Flynn et al . , 2012 ) . Despite the previously established role of substrate stiffness in hepatocellular differentiation , one of the unexpected observations of these studies was the significant cooperative effect that substrate stiffness exhibited with multicellular geometry . Although substrate stiffness did not substantially influence mechanical stress profiles as measured by TFM ( Figure 4 ) , substrate stiffness altered the baseline levels of hepatocyte and biliary markers , with stiffer substrates promoting biliary differentiation and reducing hepatocyte differentiation ( Figure 2 ) . Overall , this observation highlights the importance of considering tissue stiffness as a potential variable within current and future studies examining other regulatory signals , such as Notch . In addition , future studies could examine a broader range of geometries , including non-circular . Our analysis of different pattern sizes suggested that peripheral differentiation was independent of diameter ( Figure 1—figure supplement 3 ) . As a result , with decreasing diameter , a greater fraction of the cells are at the periphery and exhibit biliary differentiation . Subsequent studies could be aimed at further reducing pattern size or even patterning single cells to determine if there is a size that balances mechanical stress and other intercellular signals for achieving optimal biliary differentiation . Although the array patterns we used represent a relatively simple 2D geometry , we anticipate that the mechanisms regulating progenitor differentiation investigated here will serve as a foundation for future efforts employing 3D culture models while also helping to identify candidates for future manipulation in vivo . Interestingly , during liver development , biliary differentiation is initiated as a ductal plate consisting of a layer of differentiating progenitor cells that encircle the portal vein ( Ober and Lemaigre , 2018 ) . Based on our findings related to the spatial patterning of progenitor differentiation , it is reasonable to hypothesize that the structure of the portal vein may play a role in defining the geometric and mechanical cues presented to the nascent biliary cells . In these studies , we utilized BMEL cells , which are untransformed and have been demonstrated to exhibit bipotential differentiation both in vitro and in vivo ( Strick-Marchand and Weiss , 2002; Strick-Marchand et al . , 2004 ) . Accordingly , they represent a robust model cell type for controlled in vitro studies investigating microenvironmental regulation of progenitor fate specification . Building on our findings presented here , the cellular microarray approach could be adapted for investigating the differentiation of immortalized human bipotential cell lines and , ultimately , primary or stem cell-derived human liver progenitors . Finally , the mechanoresponsiveness of liver progenitors has crucial implications not only for development but also disease . Cholangiocytic cells derived from transitional progenitors have been implicated in the pathogenesis of cholangiopathies , cholangiocarcinomas , and related disorders through compensatory ductular reactions ( Gouw et al . , 2011 ) and are further thought to play a role in regenerating the liver by transdifferentiation ( Boulter et al . , 2012; He et al . , 2014; Raven et al . , 2017 ) . The mechanisms we describe here may contribute to early sensing of and differentiation responses to the stiff , fibrotic microenvironments in both ductular reactions and regeneration , contributing to the biliary fates observed in these contexts . We utilized BMEL 9A1 cells between passages 30 and 36 . These cells were cultured as previously described ( Strick-Marchand and Weiss , 2002 ) . Briefly , cells were seeded on tissue culture plastic coated with collagen I ( 0 . 5 mg/ml ) and subsequently cultured under controlled environmental conditions ( 37°C and 5% CO2 ) . Treatment with trypsin-EDTA ( 0 . 25% v/v ) for ≤10 min was used to detach cells for subculturing . Basal media for expansion consisted of RPMI 1640 with fetal bovine serum ( 10% v/v , FBS ) , penicillin/streptomycin ( 1% v/v , P/S ) , L-glutamine ( 1% v/v ) , human recombinant insulin ( 10 µg/ml , Life Technologies , 12585–014 ) , IGF-2 ( 30 ng/ml , PeproTech , 100–12 ) , and EGF ( 50 ng/ml , PeproTech , AF-100–15 ) . Differentiation media consisted of Advanced RPMI 1640 ( Life Technologies , 12633–012 ) with FBS ( 2% v/v ) , P/S ( 0 . 5% v/v ) , L-glutamine ( 1% v/v ) , and minimum non-essential amino acids ( 1% v/v , Life Technologies , 11140–050 ) . BMEL cells tested negative for Mycoplasma spp . using the MycoProbe Mycoplasma Detection Kit ( R&D Systems , #CUL001B ) . We confirmed expression of liver-specific genes and proteins in bulk cultures using PCR , immunocytochemistry , and western blot as previously described ( Strick-Marchand and Weiss , 2002; Kaylan et al . , 2016; Kourouklis et al . , 2016 ) . Additionally , bipotential differentiation capacity of BMEL cells was confirmed using bulk cultures within standard tissue culture plates with or without treatment with TGFβ1 ( Kaylan et al . , 2016; Kourouklis et al . , 2016 ) . During microarray-based differentiation experiments , cells were seeded on arrays at 1E6 cells/slide ( immunocytochemistry ) and 500E3 cells/dish ( TFM ) . Cells were allowed to adhere to arrays for 2 hr before addition before 2× washes with differentiation media and subsequent addition of experiment-specific treatments . All growth factors and drugs used in these experiments were prepared and reconstituted according to the instructions of the manufacturers; see Table 1 . The control , shJag1 , and shDll1 cells were generated by lentiviral transduction with shRNA constructs targeting a non-mammalian sequence , Jag1 , and Dll1 , respectively , the details and validation of which we have described elsewhere ( Kaylan et al . , 2016 ) . Polyacrylamide ( PA ) hydrogels were prepared following previous protocols ( Aratyn-Schaus et al . , 2010; Tse and Engler , 2010; Wen et al . , 2014 ) . Briefly , 25×75 mm glass microscope slides were washed with 0 . 25% v/v Triton X-100 in dH2O and placed on an orbital shaker for 30 min . After rinsing with dH2O , slides were immersed in acetone and placed on the shaker for 30 min . The acetone wash was followed by immersion in methanol and another 30 min on the shaker . The slides were then washed with 0 . 2 N NaOH for 1 hr , rinsed with dH2O , air-dried , and placed on a hot plate at 110°C until dry . For silanization , the cleaned slides were immersed in 2% v/v 3- ( trimethoxysilyl ) propyl methacrylate in ethanol and placed on the shaker for 30 min . The silanized slides were washed with ethanol on the shaker for 5 min , air-dried , and again placed on the hot plate at 110°C until dry . For fabrication of hydrogels with specific elastic moduli , two prepolymer solutions with different acrylamide/bis-acrylamide percentage ( w/v ) ratios were prepared to achieve elastic moduli of 4 kPa ( 4% acrylamide , 0 . 4% bis-acrylamide ) and 30 kPa ( 8% acrylamide , 0 . 55% bis-acrylamide ) with similar porosity ( Wen et al . , 2014 ) . Each of these prepolymer solutions were mixed with Irgacure 2959 ( BASF , Corp . ) solution ( 20% w/v in methanol ) at a final volumetric ratio of 9:1 ( prepolymer:Irgacure ) . This working solution was then deposited onto slides ( 100 µl/slide ) and covered with 22×60 mm cover glasses . The sandwiched working solution was transferred to a UV oven and exposed to 365 nm UV A for 10 min ( 240E3 µJ ) . After removing the cover glasses , the slides were immersed in dH2O at room temperature for 3 d in order to remove excess reagents from the hydrogel substrates . Before microarray fabrication , hydrogel substrates were thoroughly dehydrated on a hot plate for ≥15 min at 50°C . Microarrays were fabricated as described previously ( Flaim et al . , 2005; Brafman et al . , 2012; Kaylan et al . , 2016 ) . Biomolecules for arraying were diluted in 2× growth factor buffer ( 38% v/v glycerol in 1× phosphate-buffered saline [PBS] , 10 . 55 mg/ml sodium acetate , 3 . 72 mg/ml EDTA , 10 mg/ml CHAPS ) and loaded in a 384-well V-bottom microplate . Collagen I ( rat tail , EMD Millipore , 08–115 ) was prepared at a final concent µg/ml . Fc-recombinant Notch ligand solutions were prepared at a final concentration of 104 µg/ml and included: Fc-JAG1 ( R&D Systems , 599-JG-100 ) , Fc-DLL1 ( R&D Systems , 5026-DL-050 ) , and Fc-DLL4 ( Adipogen , AG-40A-0145-C050 ) . All Notch ligand conditions were pre-conjugated with Protein A/G ( Life Technologies , 21186 ) at a minimum 1:6 molar ratio ( A/G:ligand ) before arraying . Human IgG ( 104 µg/ml final , R&D Systems , 1–001-A ) was arrayed as a control in experiments involving Notch ligands . A robotic benchtop microarrayer ( OmniGrid Micro , Digilab ) loaded with SMPC Stealth microarray pins ( ArrayIt ) was used to transfer biomolecules from source plate to polyacrylamide hydrogel substrate , producing ∼600 µm diameter arrayed domains . For other pattern sizes , we used Xtend pins ( LabNEXT ) at 200 µm and 700 µm diameter . Fabricated arrays were stored at room temperature and 65% RH overnight and left to dry under ambient conditions in the dark . Prior to cell culture , the arrays were sterilized with 30 min UVC while immersed in 1× PBS supplemented with 1% ( v/v ) P/S , after which cells were seeded on arrays as described above . Images of entire arrays were converted to individual 8-bit TIFF files per channel ( i . e . , red , green , blue , and gray ) by Fiji ( ImageJ version 1 . 51n ) ( Schneider et al . , 2012; Schindelin et al . , 2012 ) . Image size was reduced to ∼50 megapixels/channel by binning to reduce memory requirements during computational analysis . The IdentifyPrimaryObjects and IdentifySecondaryObjects modules of CellProfiler ( version 2 . 2 . 0 ) ( Kamentsky et al . , 2011 ) were used to identify nuclei for cell counts and regions marked by fluorescence . The MeasureObjectIntensity module was used to quantify single-cell intensity . The location of arrayed conditions within each image was automatically determined relative to manually-located dextran-rhodamine markers . The centroid of each island was calculated and used to assign a radial distance to each cell for analyses of spatial localization within arrayed patterns . Samples were fixed in paraformaldehyde ( 4% w/v in 1× PBS ) for 15 min . Samples intended for labeling of secreted proteins ( namely ALB and OPN ) were treated with brefeldin A ( 10 µg/ml , R&D Systems , 1231/5 ) for 2 hr prior to fixation . Fixed samples were permeabilized with Triton X-100 ( 0 . 25% v/v in 1× PBS ) for 10 min and incubated in blocking buffer ( 5% v/v donkey serum and 0 . 1% v/v Triton X-100 in 1× PBS ) for 1 hr at room temperature . We incubated samples for 1 hr at room temperature or overnight at 4°C with one or two of the primary antibodies listed in Table 2 diluted in blocking buffer . The next day , we incubated samples for 1 hr at room temperature with one or two of the following secondary antibodies diluted in blocking buffer: DyLight 488-conjugated donkey anti-rabbit IgG ( 1/50 from stock , Abcam , ab96919 ) , DyLight 550-conjugated donkey anti-mouse IgG ( 1/50 from stock , Abcam , ab98767 ) , and DyLight 488-conjugated donkey anti-goat IgG ( 1/50 from stock , Abcam , ab96935 ) . Samples were mounted in Fluoromount G with DAPI ( Southern Biotech , 0100–20 ) and imaged no earlier than the day after mounting using an Axiovert 200M microscope ( Carl Zeiss , Inc . ) and associated Zen Pro software . In order to capture entire arrays as one image for later analyses , we utilized the tiling feature of Zen Pro . We performed in situ hybridization as previously-described ( Biehl and Raetzman , 2015; Aujla et al . , 2015 ) . Samples were fixed in paraformaldehyde ( 4% w/v in 1× PBS ) for 10 min , permeabilized with 0 . 3% Triton X-100 in 1× PBS for 15 min , and digested with Proteinase K ( 0 . 1 µg/ml ) for 15 min at 37°C . Afterwards , samples were acetylated , pre-hybridized , and incubated in hybridization solution with linearized , digoxigenin-labeled probes for Jag1 , Dll1 , or Notch2 at 55°C . Prior to initiation of hybridization , probes were denatured for 3 min at 95°C . After overnight incubation , samples were washed in 50% 0 . 5× formamide solution and 0 . 5× sodium citrate and subsequently blocked ( 10% heat-inactivated sheep serum , 2% bovine serum albumin and 0 . 1% Triton X-100 in tris-buffered saline ) . Following blocking , slides were incubated with anti-digoxigenin antibody ( see Table 2 ) diluted in blocking buffer for 1 hr . Next , samples were washed with tris-buffered saline of increasing alkalinity ( pH = 7 . 5 , 9 . 5 ) and incubated overnight in NBT/BCIP developing solution ( Roche , 11 681 451 001 ) . Samples were subsequently fixed with paraformaldehyde ( 4% w/v in 1× PBS for 10 min ) , mounted in Fluoromount G with DAPI ( Southern Biotech , 0100–20 ) , and imaged similarly to the immunofluorescently-labeled samples described above . For TFM experiments , we adjusted our protocol in order to fabricate the PA hydrogels in glass-bottom 35 mm Petri dishes ( Cell E&G , GBD00002-200 ) rather than on 25×75 mm microscope slides . This enabled us to perform TFM on live cells at 37°C and 5% CO2 . To measure the cell-generated forces , we added 1 µm far-red fluorescent beads ( 0 . 2% v/v , Life Technologies , F-8816 ) to the working solution ( Wang and Lin , 2007; Wang et al . , 2002 ) and fabricated hydrogels with embedded beads by exposure to 365 nm UV A for 10 min . We subsequently completed the hydrogel and array fabrication protocols as described above and seeded cells on the arrays . After completion of experiment-specific treatments , the arrays were transferred to an incubated ( 37°C and 5% CO2 ) Axiovert 200M microscope ( Carl Zeiss , Inc . ) . The microscope was used to capture phase contrast and far-red fluorescent micrographs to record cellular position and morphology along with bead displacement before and after cell dissociation with sodium dodecyl sulfate ( 1% v/v in 1× PBS ) . For analysis , we calculated the traction fields from the displacements using standard methods ( Butler et al . , 2002; Wang et al . , 2002 ) which we have adapted for analysis of cell microarrays elsewhere ( Kaylan et al . , 2017 ) . We next analyzed the captured images in MATLAB software ( MathWorks , Inc ) using Bio-Formats ( Linkert et al . , 2010 ) in conjunction with a set of custom scripts ( see Source code 1–7 ) . Specifically , the border of each island was identified , allowing calculation of a best fit ellipse and centroid . A previously-described digital image correlation program was used to calculate the displacement field between the contracted and relaxed state ( Bar-Kochba et al . , 2015 ) . Cell island contraction was simulated using COMSOL Multiphysics software ( COMSOL Inc . , Burlington , MA ) as already described ( Nelson et al . , 2005 ) using previously-determined parameter values ( Sato et al . , 1990; Folkman and Moscona , 1978 ) . Briefly , the model was comprised of an active layer bound to a passive substrate with fixed lower boundary . The cell island ( 20 µm height , 600 µm diameter ) was modeled as an isotropic linearly-elastic material with Young’s modulus of 1 . 5 kPa , Poisson’s ratio of 0 . 48 , thermal conductivity of 10 Wm-1K-1 , and coefficient of expansion of 0 . 05 K-1 . The substrate was modeled as an isotropic linearly-elastic material with Young’s modulus of 30 kPa and Poisson’s ratio of 0 . 48 . Contraction was induced in the model by reducing the temperature by 5 K ( see Source code 8 ) . Our computational model for Notch signaling is based on that of the groups of Elowitz and Sprinzak ( Sprinzak et al . , 2010; Formosa-Jordan and Sprinzak , 2014 ) , extending their approach to include the effect of an external gradient of a morphogen which regulates expression of Notch ligand and receptor . The model outputs a hexagonal lattice with fixed ( rather than periodic ) boundaries containing individual cells with their respective Notch , Delta , and repressor concentrations as determined by the following equations: ( 1 ) dNidτ=αn−KtNi⟨Di⟩−KcNiDi−γnNi+σ ( b ) ( 2 ) d⁢Did⁢τ=αd1+ ( Riθr ) h-Kt⁢Di⁢⟨Ni⟩-Kc⁢Ni⁢Di-γd⁢Di+σ⁢ ( b ) ( 3 ) d⁢Rid⁢τ=αr⁢ ( Kt⁢Ni⁢⟨Di⟩γ𝑛𝑑 ) mθ𝑛𝑑m+ ( Kt⁢Ni⁢⟨Di⟩γ𝑛𝑑 ) m-γR⁢Ri Where N is Notch receptor concentration , D is Delta ligand concentration , R is repressor concentration , σ is the stress gradient function , Kc is the constant representing strength of cis-interactions , Kt is the constant representing strength of trans-interactions , b is the base constant for steepness of the stress gradient , α is the maximal production rate , γ is the maximal degradation rate , h is the cooperativity of Delta inhibition , m is the cooperativity of repressor activation , and θ is the Hill coefficient . Subscript i indicates index within the hexagonal lattice while angle brackets denote ensemble value of neighbers of cell i . These equations were evaluated with and without the stress function ( σ ) under various strengths of cis- and trans-interactions . The model defines σ to be a linear function of the radius of the island , thereby increasing expression of Notch ligand and receptor with radius in accordance with our mRNA in situ hybridization data ( see Source code 9–11 ) . Array experiments consisted of at least three biological replicates with 18 total islands per combination of arrayed condition , treatment , cell type , and readout . Counts of cells positive for immunolabels are plotted as mean values representative of an individual island . Line plots of both percentages of positive cells and mechanical stress were calculated using local polynomial regression fitting and are shown with 95% CI ribbons in gray to allow for direct statistical comparisons , that is P<0 . 05 if the 95% CI ribbons for two conditions do not overlap . The percentage of cells positive for an immunolabel ( namely ALB and OPN ) was calculated relative to cell counts in each of 30 radial bins across every island . Multiple regression analyses were performed in R using the base lm ( ) function ( R Core Team , 2017 , R Foundation for Statistical Computing ) and are presented as coefficient estimates ( β ) and associated 95% CI . All β coefficients in regressions represent mean changes in cell counts , for which positive β represents increased cell counts and negative β represents decreased cell counts . For each regression model , we confirmed homoscedasticity , normal distribution of residuals , and the absence of leveraged outliers using residual-fit , Q-Q , and scale-location plots . For select comparisons in the text , Welch’s two-sample t-test was performed in R using the base t . test function . For all hypothesis testing , P<0 . 05 was considered significant and P-values below P=0 . 001 are denotated as P<0 . 001 .	Children are said to be a product of both nature and nurture – of their genes and the environment in which they are raised . The cells of the growing liver are not so different in this sense . As the liver of a fetus develops , immature cells called liver progenitors mature to become one of two types of adult cells: the hepatocytes that form the bulk of the liver , or the biliary cells that make up the bile duct . The traditional view is that genetic factors mainly control which cell type the progenitor cells become . However , recent research suggests that the environment around the cells matters more in this process than once thought . Cells can respond to the physical properties of their environment , such as the structure and stiffness of the surrounding tissue . These properties change as the liver develops , and can also be altered by disease . For example , damaged liver cells can spit out proteins that harden and form stiff scars . This raises a question: do changes in stiffness affect how progenitor cells behave ? To answer this question , Kaylan et al . printed collagen in circular patterns and grew liver progenitor cells on them . The cells at the edges of the circular patterns matured into bile duct cells , while those in the center became hepatocytes . The stiffness felt by the cells was then determined by measuring the level of mechanical stress that they experienced . This revealed that the cells at the edge of the collagen pattern – the cells that became bile duct cells – were under most stress . In addition , more bile duct cells formed when progenitor cells were grown on a stiffer collagen pattern . Overall , the results reported by Kaylan et al . suggest that the stiffness of the environment , and the resulting stresses on a progenitor cell , can influence how it matures . As well as helping us to understand how the liver develops , this knowledge could also help us to treat a group of diseases called cholangiopathies , in which the bile ducts become inflamed . These diseases are thought to be caused by certain cells ( which are similar to liver progenitor cells ) maturing to become incorrect cell types . Future studies could determine if preventing changes in stiffness in the environment of these cells , or slowing their response to such changes , would help patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2018	Spatial patterning of liver progenitor cell differentiation mediated by cellular contractility and Notch signaling
Candida albicans is both a member of the healthy human microbiome and a major pathogen in immunocompromised individuals . Infections are typically treated with azole inhibitors of ergosterol biosynthesis often leading to drug resistance . Studies in clinical isolates have implicated multiple mechanisms in resistance , but have focused on large-scale aberrations or candidate genes , and do not comprehensively chart the genetic basis of adaptation . Here , we leveraged next-generation sequencing to analyze 43 isolates from 11 oral candidiasis patients . We detected newly selected mutations , including single-nucleotide polymorphisms ( SNPs ) , copy-number variations and loss-of-heterozygosity ( LOH ) events . LOH events were commonly associated with acquired resistance , and SNPs in 240 genes may be related to host adaptation . Conversely , most aneuploidies were transient and did not correlate with drug resistance . Our analysis also shows that isolates also varied in adherence , filamentation , and virulence . Our work reveals new molecular mechanisms underlying the evolution of drug resistance and host adaptation . Virtually all humans are colonized with Candida albicans , but in some individuals this benign commensal organism becomes a serious , life-threatening pathogen . C . albicans possesses an arsenal of traits that promote its pathogenicity , including phenotypic switching ( Alby and Bennett , 2009 ) , yeast–hyphae transition ( Kumamoto and Vinces , 2005 ) and the secretion of molecules that promote adhesion to abiotic surfaces ( Chandra et al . , 2001 ) . As a commensal , an intricate balance is maintained between the ability of C . albicans to invade host tissues and the host's defense mechanisms ( Kim and Sudbery , 2011; Kumamoto and Pierce , 2011 ) . Alteration of this delicate host–fungus balance can result in high levels of patient mortality ( Pittet et al . , 1994; Charles et al . , 2003 ) : systemic C . albicans infections are fatal in 42% of cases ( Wisplinghoff et al . , 2003 ) , despite the use of antifungal therapies , and C . albicans is the fourth most common infection in hospitals ( Gudlaugsson et al . , 2003; Pappas et al . , 2003 ) . While compromised immune function contributes to pathogenesis ( Gow and Hube , 2012 ) , it is less clear how C . albicans evolves to better exploit the host environment during the course of infection . Two classes of antifungals in clinical use target ergosterol , a major component of the fungal cell membrane: polyenes and azoles . Polyenes ( e . g . , Amphotericin B ) are used sparingly due to toxicity ( Rex et al . , 1994 ) , whereas azoles ( e . g . , fluconazole ) are used widely because they can be administered orally and have few side effects ( Rex et al . , 2003 ) . However , resistance to the azoles arises within the commensal population of the treated individual , primarily because azoles are fungistatic ( inhibit growth but do not kill ) ( Cowen et al . , 2002 ) . Epidemiological data suggest that the intensity of fluconazole use is driving the appearance of resistant isolates ( Pfaller et al . , 1998 ) . Studies of clinical isolates of C . albicans suggest that drug resistance can increase during an infection through the acquisition of aneuploidies ( Selmecki et al . , 2009 ) due to genomic plasticity and rapid evolutionary selection during infection . Previous studies have identified two molecular mechanisms of azole resistance in C . albicans . First , increased activity or level of the enzymes of the ergosterol pathway ( e . g . , ERG11 ) reduces direct impact of the drug on its target ( Asai et al . , 1999; Oliver et al . , 2007 ) . Second , increased efflux of the drug from cells by ABC transporters ( encoded by CDR1 and CDR2 ) ( Coste et al . , 2006 ) or by the major facilitator superfamily efflux pump ( encoded by MDR1 ) ( Dunkel et al . , 2008 ) reduces the effective intracellular drug concentration . In both cases , such alterations can result from point mutations in genes encoding these proteins ( Marichal et al . , 1999 ) , in transcription factors that regulate mRNA expression levels ( MacPherson et al . , 2005; Coste et al . , 2006; Dunkel et al . , 2008 ) , or from increased copy number of the relevant genes , via genome rearrangements such as whole chromosome and segmental aneuploidies ( Selmecki et al . , 2006; 2008; 2009 ) . Indeed , the genomes of drug-resistant strains isolated following clinical treatment often exhibit large-scale changes , such as loss of heterozygosity ( LOH ) ( Coste et al . , 2006; Dunkel and Morschhauser , 2011 ) , copy-number variation ( CNV ) , including short segmental CNV , and whole chromosome aneuploidy ( Selmecki et al . , 2010 ) accompanied by point mutations . While we understand some aspects of the molecular basis of resistance , we understand less about the mechanisms that drive the evolution of drug resistance and overall pathogenicity in C . albicans . It is challenging to use forward genetic approaches in C . albicans due to its diploid genome and lack of a complete sexual cycle . Although C . albicans has conserved the genomic elements needed for mating , mating occurs instead through rare mating-competent haploids ( Hickman et al . , 2013 ) or via a parasexual cycle consisting of mating of diploid strains to form tetraploids followed by chromosome loss to regenerate diploids ( Bennett and Johnson , 2005 ) . An alternative approach is to use isolates sampled consecutively from the same patient to study the changes in the frequency of variants in natural populations under selection for drug resistance . Studies in evolved isolates have implicated multiple mechanisms in drug resistance , but have focused on large-scale aberrations such as aneuploidies and LOH ( Selmecki et al . , 2008; 2009 ) or candidate genes ( Perea et al . , 2001; White et al . , 2002 ) , and do not comprehensively chart the genetic basis of adaptation . Here , we used genome sequencing of isolates sampled consecutively from patients that were clinically treated with fluconazole to systematically analyze the genetic dynamics that accompany the appearance of drug resistance during oral candidiasis in human HIV patients . Most isolates from each individual patient were highly related , suggesting a clonal population structure and facilitating the identification of variation . Because each clinical sample was purified from a single colony , we cannot assess the population structure at any single time point . Instead , we have measured the occurrence of single-nucleotide polymorphisms ( SNPs ) , CNV , and LOH events in each isolate and then compared them between isolates from the same patient and across patients' series . Consistent with previous studies , we found that LOH events were recurrent across patients' series and were associated with increased drug resistance . To identify SNPs with likely functional impact in the context of substantial genetic diversity , we focused on those events that were both persistent across isolates within a patient and were recurrent in the same gene across multiple patient series . We found 240 genes that recurrently contain persistent SNPs , many of which may be related not only to antifungal exposure but also to the complex process of adaptation to the host and antifungal exposure . In contrast , aneuploidies were prevalent in the isolates , yet they were more likely to be transient , and aneuploidy , per se , did not correlate with changes in drug resistance . Our work uses comparative analysis of a fungal pathogen to reveal new molecular mechanisms underlying drug resistance and host adaptation and provides a general model for such studies in other eukaryotic pathogens . To study the in vivo evolution of azole resistance in C . albicans , we analyzed 43 longitudinal isolates from 11 HIV-infected patients with oropharyngeal candidiasis ( White , 1997a; Perea et al . , 2001 ) ( Table 1 ) . The isolates were previously collected during incidences of infection and form a time series from each patient ( 2–16 isolates per series; Figure 1 , Figure 2A ) . Each isolate was derived from a single colony , and thus , represents a single diploid genotype sampled from the within-host C . albicans population at the respective time point . In each series , the first isolate ( ‘progenitor’ ) was collected prior to any treatment with azole antifungals and the remaining isolates were collected at later , typically consecutive , time points , culminating in the final ‘endpoint’ isolate ( Table 1 ) . 10 . 7554/eLife . 00662 . 003Table 1 . Isolate history and sequencing summaryDOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 003Publication namePTStrainEntry dateDrug treatmentDose ( mg/day ) E-test MIC ( ug/mL ) Depth of coverageReadsPercent alignedWhite , T . C . 119/10/90Fluconazole1000 . 25111 . 969 , 896 , 46887 . 17%212/14/90Fluconazole100169 . 2012 , 797 , 32887 . 43%312/21/90Fluconazole100492 . 0416 , 987 , 81486 . 87%412/31/90Fluconazole100380 . 6914 , 858 , 71087 . 81%52/8/91Fluconazole1004110 . 8020 , 484 , 58486 . 75%62/22/91Fluconazole1004101 . 9418 , 837 , 95486 . 63%73/25/91Fluconazole100481 . 6515 , 123 , 02086 . 66%84/8/91Fluconazole1004112 . 5320 , 778 , 56286 . 64%96/4/91Fluconazole1004113 . 1822 , 223 , 22883 . 20%117/15/91Fluconazole100453 . 289 , 896 , 46887 . 17%1211/26/91Fluconazole200496 . 1018 , 282 , 47285 . 54%1312/13/91Fluconazole40032123 . 6722 , 070 , 51889 . 13%141/28/92Fluconazole4002498 . 6618 , 114 , 91687 . 41%152/21/92Clotriminazole5024120 . 9022 , 401 , 37486 . 57%164/1/92Fluconazole4009687 . 4416 , 061 , 56087 . 17%178/25/92Fluconazole8009697 . 8318 , 317 , 11885 . 91%Perea , S . et al . 74122/15/95Fluconazole00 . 2593 . 1517 , 417 , 58886 . 69%230711/22/95Fluconazole4000 . 7595 . 7918 , 014 , 24285 . 25%Perea , S . et al . 910024/20/95Fluconazole1000 . 125188 . 4934 , 834 , 97086 . 74%28234/6/96Fluconazole800282 . 6252 , 839 , 28886 . 30%37952/26/97Fluconazole80012877 . 6313 , 901 , 06288 . 78%Perea , S . et al . 145803/13/95Fluconazole01 . 577 . 0814 , 711 , 80485 . 00%24401/3/96Fluconazole8001 . 582 . 9315 , 446 , 88285 . 69%25011/4/96Fluconazole8009688 . 5917 , 480 , 27481 . 98%Perea , S . et al . 159454/14/95Fluconazole3004108 . 5920 , 591 , 04485 . 19%16197/11/95Fluconazole5006493 . 1417 , 565 , 08084 . 69%Perea , S . et al . 1631076/5/96Fluconazole800497 . 0118 , 361 , 26684 . 84%31196/5/96Fluconazole8009687 . 9216 , 615 , 46284 . 67%31206/5/96Fluconazole80096105 . 9519 , 442 , 01686 . 79%31847/1/96Fluconazole800101 . 8918 , 487 , 46287 . 50%32817/16/96Fluconazole80076 . 4414 , 327 , 37685 . 69%Perea , S . et al . 3051061/7/98Fluconazole8000 . 587 . 2116 , 466 , 52484 . 67%51081/7/98Fluconazole8000 . 7582 . 3217 , 480 , 27481 . 98%Perea , S . et al . 4216918/3/95Fluconazole100122 . 6022 , 072 , 56288 . 38%373112/27/96Fluconazole400256119 . 9021 , 436 , 03488 . 72%373312/27/96Fluconazole40025695 . 5117 , 295 , 88888 . 00%Perea , S . et al . 4316497/19/95Fluconazole00 . 125102 . 1019 , 545 , 53084 . 08%30345/15/96Fluconazole4000 . 7592 . 9717 , 300 , 04085 . 64%Perea , S . et al . 5939172/19/97Fluconazole8002113 . 2721 , 549 , 70483 . 86%46178/28/97Fluconazole4006475 . 3715 , 242 , 90481 . 42%46399/2/97Fluconazole400128115 . 3225 , 468 , 19075 . 69%Perea , S . et al . 6440184/2/97Fluconazole200110 . 1620 , 118 , 73686 . 78%43807/14/97Fluconazole20018 . 0320 , 970 , 9469 . 26%Strains and coverage . ( Red ) Not clonally derived from progenitor . isolated on same day from same patient as previously published strain , 2500 . 10 . 7554/eLife . 00662 . 004Figure 1 . Overview of study design . ( A ) Background , persistent , transient , recurrent , and driver mutations in patient time courses . Shown is a schematic illustration of the genomes of isolates ( gray bars ) from two patient time courses ( Patient A and B , left and right panels , respectively ) , ordered from the first isolate ( progenitor , top ) to the last ( evolved , bottom ) . Background mutations ( purple ) exist in the all isolates; persistent mutations ( yellow ) are not in the progenitor , but found in all subsequent isolates after their first occurrence; transient mutations ( pink ) are not in the progenitor and only in some later isolates; recurrently polymorphic genes contain persistent mutations that occur in the same gene in more than one patient ( black box ) . LOH events were also evaluated for persistence ( light teal bar ) . Driver mutations , where a new persistent homozygous allele appears ( e . g . , G/T > A/A ) , are annotated in association with persistent LOH events ( dark teal ) and independent of these events ( not shown ) . Each of these can be associated with a change in phenotype , such as drug resistance ( boxes , right ) . ( B ) Sampling in the context of de novo mutation and selection bottlenecks . Each strain is a single clone ( circle ) isolated from an evolving population ( represented by a phylogenetic tree ) . The population evolves and undergoes selective sweeps ( dashed lines ) , with phenotypic changes occurring during the course of infection and treatment ( i . e . , drug resistance , black: high , white: low; gray scale at bottom ) . Persistent mutations ( yellow lightning bolt ) have likely swept through the population , whereas transient mutations ( pink lightning bolt ) have not . ( C ) Sampling in the context of selection on existing variation . Selection acts to vary the frequency of different pre-existing genotypes in the population . Persistent mutations ( yellow lightning bolt ) have risen in the population to a frequency that they are repeatedly sampled ( large circles ) whereas transient mutations ( pink lightning bolt ) have not ( small circle ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 00410 . 7554/eLife . 00662 . 005Figure 1—figure supplement 1 . Analysis of discordant sites . ( A ) Degree of concordance ( Y axis ) with Sequenom iPlex genotyping for 1973 SNP X strain combinations overall ( leftmost red bar; 93 . 9% ) and in each tested strain ( X axis ) . ( B ) Shown are the classes of discordant sites by genotype as defined by Illumina ( orange ) or Sequenome ( teal ) ( X axis ) and the prevalence ( Y axis ) of that genotype call in Sequenom ( blue ) and Illumina ( orange ) based discordant calls . The most common discrepancies arose when Sequenom typing classified a site as homozygous , but Illumina sequencing identified it as heterozygous . ( C–G ) Comparison on distributions of quality features between concordant ( blue bars ) and discordant ( green bars ) sites: ( C ) depth of coverage , ( D ) RMS Mapping Quality ( MQ ) score , ( E ) PHRED scaled quality score for each base call , shown as log-normalized ‘QUAL’ scores , ( F ) quality by depth ( QD ) score for each variant site , and ( G ) the allele balance ratio ( AB Score ) for each variant site . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 00510 . 7554/eLife . 00662 . 006Figure 2 . Most isolates from the same patient are clonal . ( A ) Two possible models of infection may underlie serial isolates . In the ‘clonal model’ ( top ) each subsequent sample ( circle ) is related to the other isolates . In the non-clonal model ( bottom ) isolates in a series are un-related . ( B ) The phylogenetic relationship of the isolates ( black ) from 11 patients ( blue ) was inferred based on 201 , 793 informative SNP positions using maximum parsimony in PAUP* . Isolates from the same patient separated by a branch distance greater than 20 , 000 were considered non-clonal ( 3281 , 2823 , 3184 , 1691 , red ) . Most nodes were supported by 100% of 1000 bootstrap replicates ( indicated by * ) , expect as indicated ( in gray ) . Clade identifiers were included as appropriate . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 00610 . 7554/eLife . 00662 . 008Figure 2—source data 1 . ( A ) SNP category summary and all patient-series SNPs SNP category summary . Listed for each series ( PT series SNP summary ) are the number of filtered ( ‘Materials and methods’ ) coding and noncoding SNPs . Coding SNPs are further classified as synonymous or nonsynonymous . Noncoding SNPs are classified as intronic , promoter region ( <800 bps from the start of an ORF ) , or general noncoding . Patient1–Patient 59: Listed is each base that is mutated in at least one isolate in the respective series . For this base , listed are the chromosomal position , the base in the SC5314 reference genome , the base in each isolate in the series ( hyphen ( ‘-’ ) : homozygous , same as reference; upper case: homozygous mutation; lower case: heterozygous mutation ) , whether the mutation is a background mutation , transient ( trans ) or persistent ( pers ) , if it is upstream , downstream or within an ORF , and in the latter case , the effect on the amino acid sequence of the encoded protein . ( B ) Frequency of nonsynonymous SNP occurrence between serial isolates using different filters . All SNP arising aft prev: For each clinical series ( PT1-PT59 ) listed are the number of ORFs in each chromosome ( columns ) containing for each isolate ( rows ) all the ‘newly arising’ SNPs , defined as those not present in the immediately preceding isolate ( rows ) . All NS in ORF aft prev: the same as above , but only for NS SNPs . All SNPs are only outside of LOH regions . All instances of Pers NS SNPs: the same as above , but only for those NS SNPs that persist once they arose . All Rec SNP aft Prev: the same as above but restricted to those ORFs that contain persistent mutations in three or more clinical series . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 00810 . 7554/eLife . 00662 . 007Figure 2—figure supplement 1 . SNP heterozygosity profiles for each strain . The heterozygosity profiles shows , in chromosomal order ( top ) , each variant locus that exists in at least one strain in the series ( white is a heterozygous SNP , blue is homozygous for the SC5314 allele , red is a homozygous SNP relative to SC5314 ) . ( A ) Patient 1; ( B ) Patient 7; ( C ) Patient 9; ( D ) Patient 14; ( E ) Patient 15; ( F ) Patient 16; ( G ) Patient 30; ( H ) Patient 42; ( I ) Patient 43; ( J ) Patient 59 . Only Patient 9 ( C ) , Patient 16 ( F ) , Patient 42 ( H ) , and Patient 64 ( not shown ) contain un-related isolates . Patient 64 contained an isolate ( 4380 ) whose genome aligned poorly to the C . albicans reference , but aligned well to C . dubliniensis . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 007 The progenitor isolates were more sensitive to fluconazole than subsequent isolates , as defined by the minimum inhibitory concentration ( MIC ) ( Table 1 , ‘Materials and methods’ ) . Previous studies with some of these patient isolates identified several genomic alterations that may contribute to azole resistance , including segmental aneuploidy ( Selmecki et al . , 2006 ) , and LOH across large chromosomal segments ( Coste et al . , 2006; Dunkel et al . , 2008 ) , as well as targeted alterations including increased expression of drug efflux genes ( Coste et al . , 2006 ) , mutations in ergosterol biosynthetic genes ( Asai et al . , 1999; Oliver et al . , 2007 ) , and buffering by the chaperone heat shock protein 90 ( Hsp90 ) ( Cowen and Lindquist , 2005 ) . We sequenced the genomic DNA of the isolates as well as the C . albicans lab strain , SC5314 , using Illumina sequencing ( 53-283X coverage , 103X on average , ‘Materials and methods’ , Table 1 ) and identified in each series point mutations , LOH events and aneuploidies that were not present in the first strain in that series . By convention , all mutations were defined relative to SC5314 , the C . albicans genome reference strain . We validated our pipeline for detection of point mutations using Sequenom iPlex genotyping ( Storm et al . , 2003 ) ( ‘Materials and methods’ ) . We interrogated 1973 SNPs in 27 isolates from nine clinical series and found that the iPlex base calls matched 1853 ( 93 . 9% , Figure 1—figure supplement 1A , Table 2 ) of the calls from our computational analysis of the sequencing data . Evaluation of the discordant sites showed somewhat lower quality scores by certain metrics but did not identify any metrics that could be used to systematically revise filtering in our computational pipeline without a radical reduction in sensitivity ( Figure 1—figure supplement 1B–G ) . 10 . 7554/eLife . 00662 . 009Table 2 . Sequenom iPLEX genotyping assay validationDOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 009PatientIsolateTotal discordantTotal concordantTotal Assayed% ConcordantPatient_1TWTC12313393 . 94%Patient_1TWTC21323396 . 97%Patient_1TWTC31323396 . 97%Patient_1TWTC121323396 . 97%Patient_1TWTC131323396 . 97%Patient_1TWTC151323396 . 97%Patient_1TWTC161313296 . 88%Patient_1TWTC171323396 . 97%Patient_74123606395 . 24%Patient_723074596393 . 65%Patient_91002169611285 . 71%Patient_93795910311291 . 96%Patient_145803495294 . 23%Patient_1424402272993 . 10%Patient_1425013333691 . 67%Patient_15945812112993 . 80%Patient_1516191012013092 . 31%Patient_1631072515396 . 23%Patient_1631193505394 . 34%Patient_1631202505296 . 15%Patient_305106321521898 . 62%Patient_3051081920422391 . 48%Patient_4316497899692 . 71%Patient_4330348889691 . 67%Patient_5939173626595 . 38%Patient_5946172636596 . 92%Patient_5946394596393 . 65%TOTAL1201853197393 . 92% We designated as background those polymorphisms those that are common to all isolates in a series , including the first ( ‘progenitor’ ) isolate ( Figure 1A , purple ) and use them to determine that isolates within most series were clonally related , suggesting a single ( primary ) infection source ( Figure 1B , C , Figure 2 , Figure 2—figure supplement 1 , ‘Materials and methods’ ) . To distinguish between a single primary ( clonal ) infection ( Figure 2A , top ) and repeated , independent infections ( Figure 2A , bottom ) , we determined the distance between every two isolates based on their SNP profile and used as a heuristic a neighbor-joining algorithm to construct a phylogenetic tree from this distance metric ( ‘Materials and methods’ , Figure 2B ) . Patient 64 contained one C . albicans isolate ( 4018 ) and one C . dubliniensis isolate ( 4380 ) ; therefore , we have excluded this series from further analysis . Additionally , we detected at least one non-clonal C . albicans isolate in three of the remaining ten patient series ( PT 9 , 16 , 42; Figure 2B , red ) , indicating that at least ∼36% of the 11 patients sampled carried more than one unrelated Candida strain . We removed the four non-clonal samples ( Figure 2B , red ) from further consideration , and all subsequent analyses focused on samples from the 10 patients with at least two clonal isolates . Despite these clonal relationships , the distance between isolates indicated significant genetic diversity within each patient series ( Figure 2B ) , typically with each isolate differing by several thousand SNPs from its ‘progenitor’ isolate ( Figure 2—source data 1 ) . These data are consistent with two different evolutionary scenarios: accumulation of de novo mutations followed by selection ( Figure 1B ) , or selection acting on pre-existing variation to vary the frequency of different genotypes in the population ( Figure 1C ) . The large number of SNPs detected suggests that isolates from later time points in a series are not simply direct descendants of the earlier isolate; however , since mutation and mitotic recombination rates can be elevated under stressful conditions ( e . g . , drug treatment Galhardo et al . , 2007; Forche et al . , 2011 ) , we cannot rule out the possibility that some of the variation may be due to de novo events occurring between time points . Formally distinguishing between these two models is not possible with the samples and data at hand . However , the role of pre-existing diversity is supported by the observation that different isolates collected on the same day from the same patient ( patient 14 [2440 and 2501] and patient 16 [3107 and 3119] ) differed by 9668 and 18 , 291 SNPs , respectively ( Figure 2—source data 1 ) and had very different fluconazole MIC levels ( Table 1 ) and different fitness phenotypes ( see below ) , although in each case the strains were clearly genetically related ( Figure 2B ) . Thus , we conclude that a population of related but divergent genotypes of the same lineage exists within a given patient . We next sought to identify potentially adaptive genetic changes by focusing on large-scale events ( LOH and aneuploidies ) as well as single-nucleotide polymorphisms . Given the high number of SNPs , LOH events and aneuploidies , we next devised a strategy to identify those changes that are more likely to play an adaptive role in drug resistance and host adaptation . We previously filtered all background polymorphisms , defined as any SNP relative to the reference present in all isolates from a series . Next , we defined alterations as persistent if present within the same patient at all subsequent time points after the ‘non-progenitor’ isolate in which they are first identified . We reasoned that such persistent changes will include those variants that were driven to sufficiently high frequency by selection to ensure repeated sampling ( Figure 1B , C , yellow lightning bolt ) , whereas non-persistent ( transient ) ones do not ( Figure 1B , C , pink lightning bolt ) . We consider the special case of a genetic change detected only in the endpoint isolate as ‘persistent’ as well , since several of the time courses consist of only two or three isolates . We apply the persistence filter to better identify potentially adaptive aneuploidies , LOH events , and SNPs . Next , we further focused on non-synonymous polymorphisms in coding regions and employed two different strategies to identify potentially adaptive changes . In the first strategy , to identify potential drivers of adaptation , we focused on non-synonymous SNPs that were homozygous for a genotype not found in the progenitor strain that persisted in the subsequent isolates ( e . g . , G/T > A/A ) consistent with positive selection . In the second strategy , we analyzed genes that were recurrently polymorphic across patients , such that persistent , non-synonymous polymorphisms appeared within the same open reading frame ( ORF ) in different patient series ( Figure 1A and Figure 5—source data 1A ) . For recurrence , we considered only those that were not included in LOH regions , as these regions artificially inflate the estimates of persistence and recurrence . Recurrence allows us to better handle polymorphisms from the endpoint isolate in a series for which ‘persistence’ does not provide a meaningful filter . Thus , we further considered polymorphisms occurring only in the terminal isolate in one patient if polymorphisms also recurred in the same ORF in a series from two other patients . For example , filtering for both persistence and recurrence across at least three series reduced the number of polymorphisms for patient 1 from 13 , 562 polymorphisms in 5022 genes to 23 recurrent genes ( Figure 2—source data 1 , Figure 5—source data 1A ) . LOH events were detected in all of the series and were often persistent , recurrent , and associated with increased drug resistance ( Figures 3 and 4 , Figure 3—source data 1 ) . For example , three of four LOH events in Patient 1 were persistent and associated with an increase in MIC and both of these events were recurrent , such that LOH events in these genomic regions coincided with increases in MIC in other patients . Highly recurrent LOH events occurred on the right arm of chromosome 3 ( in Patients 1 , 9 , 14 , 16 , 42 , and 59; Figure 3A , Figure 4A , B , D , F , H , Figure 3—source data 1 ) and on the left arm of chromosome 5 ( in Patients 1 , 14 , 15 , and 43; Figure 3A , Figure 4B , C , G , Figure 3—source data 1 ) . These regions include key genes implicated in drug resistance: on Chromosome 3 , genes encoding the Cdr1 and Cdr2 efflux pumps and the Mrr1 transcription factor that regulates the Mdr1 major facilitator superfamily efflux pump ( Schubert et al . , 2011 ) , and on Chromosome 5 , genes encoding the drug target Erg11 , and Tac1 , a transcription factor that positively regulates expression of CDR1 and CDR2 ( Coste et al . , 2006 ) . The extent of persistence and recurrence of these two LOH events is statistically significant under a naïve binary model ( p < 5 × 10−4 for the Chr3R LOH; p < 0 . 01 for the Chr5L LOH ) . The recurrence of LOH events that coincide with changes in MIC suggests that they have been positively selected to rise in frequency relative to the progenitor strain . Notably , some of the recurrent LOH events may have been difficult to detect previously on SNP arrays ( Forche et al . , 2008; Forche et al . , 2004; Forche et al . , 2005 ) due to the relative paucity of SNPs in those regions in the reference strain , SC5413 , itself a clinical isolate . 10 . 7554/eLife . 00662 . 010Figure 3 . LOH events were often persistent while aneuploidies were often transient . For each time series shown are the genomes of all isolates ( rows ) from a patient , ordered from the first isolate ( progenitor , top ) to the last ( evolved , bottom ) . Boxes on right indicate the MIC of the respective strain ( black: high , white: low; gray scale at bottom ) . Persistent LOHs: blue , transient LOHs: pink; trisomies ( all transient ) : green . The sequence coverage along each chromosome is indicated by black tickmarks . ( A ) Patient 1 has four LOH events , each coinciding with an increase in MIC ( gray scale boxes , right ) . One LOH is transient ( isolate 2 , chromosome R , pink ) and three are persistent ( isolate 3 , chromosome 3; isolate 13 , chromosome 5; and isolate 16 , chromosome 5 , blue ) . The ploidy changes ( isolates 6 , 8 , 13 ) are all transient . ( B ) Patient 7 has one LOH event ( isolate 2307 , chromosome 3 , blue ) which coincides with an increase in MIC . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 01010 . 7554/eLife . 00662 . 011Figure 3—source data 1 . Persistent LOH regions LOH map . For each isolate ( strain column ) in each series ( patient column ) , listed are the coordinates of any persistent LOH in that isolate . Coordinates in blue are persistent LOH events , coordinates in red are transient LOH events . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 01110 . 7554/eLife . 00662 . 012Figure 4 . Persistent and transient LOH and aneuploidies . For each time series shown are the genomes of all isolates ( rows ) , ordered from the first isolate ( progenitor , top ) to the last ( evolved , bottom ) . Boxes on right indicate the MIC of the respective strain ( black: high , white: low , gray scale at bottom ) . Persistent LOHs: light blue , transient LOHs: pink; trisomies ( all transient ) : green . The coverage along each chromosome is indicated by black tickmarks . ( A ) Patient 9; ( B ) Patient 14; ( C ) Patient 15; ( D ) Patient 16; ( E ) Patient 30; ( F ) Patient 42; ( G ) Patient 43; ( H ) Patient 59 . Several LOHs are recurrent ( right arm of chromosome 3 , left arm of chromosome 5 , and chromosome 1 ) . Please note: data in Figure 3—source data 1 also applies to this figure . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 01210 . 7554/eLife . 00662 . 013Figure 4—source data 1 . Persistent LOH regions LOH map . For each isolate ( strain column ) in each series ( patient column ) , listed are the coordinates of any persistent LOH in that isolate . Coordinates in blue are persistent LOH events , coordinates in red are transient LOH events . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 013 Putative driver mutations ( defined above as non-synonymous SNPs that were homozygous for a genotype not found in the progenitor strain that persisted in the subsequent isolates; e . g . , G/T > A/A ) in these regions are suggestive of a point mutation followed by an LOH of the mutant allele that confers an advantage . There were 131 such mutations in 86 ORFs from 18 LOH regions from the 10 clonal patient series ( Table 1 and Figure 5; Figure 5—source data 1B , C ) . Some of the SNPs were in genes that encode proteins with key known roles in drug resistance and were associated with large LOH events . For example , a nonsynonymous homozygous change in the fluconazole drug target ERG11 was associated with the formation of the persistent LOH on the left arm of chromosome 5 in Patient 1 ( Figure 3A ) , consistent with previous reports ( White , 1997b ) , as was TAC1 in Patient 42 . In another example , the persistent and recurrent LOH on the right arm of chromosome 3 in Patients 9 and 16 ( Figure 4A , D ) was associated with the presence of a homozygous mutation in MRR1 ( Schubert et al . , 2011 ) , a regulator of MDR1 expression . Other mutations were in genes not previously related to fluconazole resistance , including cell adhesion ( ALS3 , 5 and 7 and HYR3; [Hoyer et al . , 1998; Sheppard et al . , 2004; Hoyer et al . , 2008] ) , filamentous growth ( FGR14 , FGR28 , and EFH , [Uhl et al . , 2003; Connolly et al . , 2013] ) , and biofilm formation ( BCR1 and YAK1; [Nobile and Mitchell , 2005; Goyard et al . , 2008; Noble et al . , 2010; Nobile et al . , 2012] ) . Thus , the detection of known genes involved in drug resistance confirms the approach works and that detection of genes involved in processes implicated in virulence , suggests that these process are co-evolving . 10 . 7554/eLife . 00662 . 014Figure 5 . Co-occurrence of nonsynonymous substitutions across isolates reveals functional clusters . ( A ) For each of the recurrently mutated 240 genes ( genes in which nonsynonymous persistent SNPs appear in more than three patients and are not within an LOH region ) , we constructed a patient-by-gene binary vector . We clustered the resulting patient-by-gene matrix using NMF clustering to reveal five coherent clusters ( correlation matrix of the clusters left; red: positive correlation; blue: negative correlation; white: no correlation ) . ( B ) Co-occurrence clusters . For the genes in each cluster ( rows ) , shown are their mutated occurrences in each patient ( columns ) ; green: gene is persistently mutated in patient , white: no persistent mutation , yellow circle: driver mutation . Functional enrichment of clusters was revealed using gene ontology , and genes matching the enriched cluster function are bolded . We have overlaid recurrent driver mutations ( e . g . , G/T > A/A ) ( n = 17 ) occurring outside of LOH regions ( yellow circle , green box ) and inside LOH regions ( yellow circle , white box ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 01410 . 7554/eLife . 00662 . 016Figure 5—source data 1 . ( A ) Recurrence lists and clusters . 1 All Pers NS Genes . Listed are all the ORFs with a persistent nonsynonymous SNPs , the series in which they occur as such ( 1 in relevant Patient 1-Patient 59 column ) , and the total number of series in which they recur ( SUM column ) . 2 All Pers NS in LOH: Listed are all the ORFs with a persistent nonsynonymous SNPs within an LOH region . 3 All Pers NS not in LOH: Listed are all the ORFs with a persistent nonsynonymous SNPs NOT within an LOH region . 4 Cluster Rec . genes not in LOH: NMF Clustering of the occurrence matrix from ‘All Pers NS not in LOH’ . 5 Cluster GO Enrichment: The GO enrichments for each of the clusters identified in ‘4 Cluster Rec . genes not in LOH’ . ( B ) Driver mutations . Patient 1—59 . Shown are all the positions where a nonsynonymous SNP changed from one homozygous genotype to another . Each column represents the base-call in that isolate of a given patient series . The formatting is consistent with Figure 2—source data 1 . Drivers Recurrence in genome: for each of the driver candidates identified in the previous tabs , shown are the occurrence of a driver mutation in that ORF across each of the patient series . ( C ) Driver mutations in LOH regions . As above ( Figure 5—source data 1B ) , but restricted to only driver mutations occurring within LOH regions . ( D ) . Recurrence lists and clusters for MIC associated mutations . As above ( Figure 5—source data 1A ) , but restricted to only recurrent mutations that occur in parallel with changes in MIC . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 01610 . 7554/eLife . 00662 . 015Figure 5—figure supplement 1 . Co-occurrence of nonsynonymous SNPs occurring in conjunction with a shift in MIC . ( A ) For each of the 166 recurrently mutated genes associated with a change in MIC , we constructed a patient-by-gene binary vector . We clustered the resulting patient by gene matrix using NMF clustering to reveal 5 coherent clusters ( correlation matrix of the clusters left; red: positive correlation; blue: negative correlation; white: no correlation ) . ( B ) Co-occurrence clusters . For the genes in each cluster ( rows ) , shown are their mutated occurrences in each patient ( columns ) ; green: gene is persistently mutated in patient , white: no persistent mutation . Functional enrichment of clusters was revealed using gene ontology , and genes matching the enriched cluster function are bolded . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 015 Aneuploidies , either whole chromosomal or segmental , were evident in at least one isolate from 80% ( 8/10 ) of the clonal patient series , with the most prevalent aneuploidies involving Chromosome 5 ( 6 of 8 patients with at least one aneuploid isolate; Figure 3A and Figure 4B–G , green ) . In contrast to the persistence of most LOH events , persistent aneuploidies were rarer and were not consistently associated with adaptive increases in MIC levels ( Figures 3 and 4 ) . This is consistent with the irreversibility of LOH events in the absence of mating highlights the reversible nature ( instability ) of aneuploidy chromosomes . While we cannot definitively infer an ordering of events from our singly sampled isolates , we hypothesize that aneuploidy could contribute to the evolution of LOH by increasing the likelihood of its occurrence . For example , in 4 of the 6 patients with a Chromosome 5 LOH , the isolate with an LOH event also harbors a Chromosome 5 trisomy or is preceded by an isolate with a Chromosome 5 trisomy . Thus , the additional copy may increase the likelihood of an LOH event on that chromosome . In three of these cases , ERG11 , located on the region of Chromosome 5 with LOH , was mutated . Additionally , isolates in 2 of the 7 patients with a Chromosome 3 LOH were trisomic for this chromosome . We identified persistent nonsynonymous coding SNPs within 1470 genes outside LOH tracts , 167 of them harboring 336 driver-like polymorphisms ( Figure 5—source data 1B ) . These again include ERG11 in patients 9 , 14 , 30 , and 59 and TAC1 in patients 1 , 7 , 14 , 15 , 30 and 43 ( Figure 5 ) . Applying the recurrence filter ( i . e . , persistent nonsynonymous SNPs that appeared in the same ORFs in three or more patient series ) , we identified 240 polymorphic genes that are more likely to have contributed to adaptation ( Figure 5—source data 1A ) . This number of genes is higher than expected by chance ( empirical p < 10−4 based on a Poisson model of background mutation , ‘Materials and methods’ ) . Though the coding sequence for these 240 recurrent genes is longer than average ( 2 . 21 ± 1 . 53 kb vs 1 . 83 ± 1 . 29 kb for non-recurrent persistent genes , p < 3 . 68 × 10−5 , t-test ) , and thus a larger target for mutation , our simulation accounts for gene length . Notably , 17 persistent recurrently polymorphic genes also had driver-like polymorphisms , eight of which were also homozygosed in an LOH tract in at least one patient series ( Figure 5 , Figure 5—source data 1A , B , C ) . Finally , polymorphisms in 166 of the 240 genes appeared together with an increase in MIC and are thus stronger candidates for making a significant functional contribution to resistance ( Figure 5—figure supplement 1 , Figure 5—source data 1D , empirical p < 10−5 based a binomial model , ‘Materials and methods’ ) . The set of 240 recurrently mutated genes was enriched for fungal-type cell wall ( 18 genes , p < 0 . 0012 ) and cell surface genes ( 24 genes , p < 0 . 00012 ) , including several members in each of three cell wall gene families important for biofilm formation and virulence ( Hoyer et al . , 2008 ) : the Hyr/Iff proteins ( HYR1 and 3 , IFF8 and 6 ) , the ALS adhesins ( ALS1-4 , 7 , 9 ) , and the PGA-30-like proteins ( seven genes ) ( Figure 5—source data 1A ) . All three families are specifically expanded in the genomes of pathogenic Candida species ( Butler et al . , 2009 ) . In addition , seven members of the FGR genes ( Uhl et al . , 2003 ) , involved in filamentous growth and specifically expanded in C . albicans ( Butler et al . , 2009 ) , are also among the 240 genes ( Figure 5—source data 1A ) . The most recurrently mutated gene outside of an LOH region was AXL1 that encodes a putative endoprotease , whose transcript is upregulated in an RHE model of oral candidiasis and in clinical isolates from HIV+ patients with oral candidiasis ( Zakikhany et al . , 2007 ) . The gene is persistently mutated in eight series , ( three of which were driver mutations ) , followed by ten genes mutated in seven series ( Figure 5—source data 1A ) . ERG11 , which encodes the drug target of fluconazole , was affected in 70% ( 7/10 ) of the patient series with persistent SNPs in four series ( Patients 9 , 14 , 30 , and 59 ) and mutations in the LOH events in three series ( Patients 1 , 15 , and 43 ) ( Figure 5—source data 1A ) . Likewise HYR3 , a known virulence gene , was persistently mutated in nine of the patients , three of which occurred in LOH tracts , including one in which a new allele was homozygosed ( Figure 5 , Figure 5—source data 1A , B , C ) . More generally , 171 of the 240 genes were also mutated in an LOH tract in at least one additional patient ( 15/171 in three or more additional patients and 34/171 in two additional patients ) . Next , we partitioned the 240 recurrently mutated genes into 5 ‘co-occurrence clusters’ based on the correlation in their mutation occurrence patterns ( Figure 5 , Figure 5—source data 1A ) . These correlations are significantly higher than expected in a null model ( p < 5 . 2 × 10−182 , permutation test , ‘Materials and methods’ ) . The characterized genes in most of the clusters have coherent functions . Cluster 1 is enriched for cell wall and cell surface genes , Cluster 2 for cell cycle and stress genes , Cluster 3 for genes involved in drug response , and Cluster 5 for carbohydrate binding ( Figure 5 , Figure 5—source data 1A ) . Most of the genes in these clusters are not well characterized and represent new candidates involved drug resistance and adaptation to the host environment . The full list of genes and descriptions is given in Figure 5—source data 1A . To explore the possibility that some of the mutations reflect adaptation to other factors besides drug , we next measured phenotypes associated to virulence and interaction with the host ( ‘Materials and methods’ ) . Adhesion , filamentation , and virulence in a C . elegans model of infection ( Jain et al . , 2013 ) were measured for a large panel of isolates ( Figure 6 , Figure 6—figure supplement 1 , Figure 6—source data 1 ) . Additionally , we measured competitive fitness in standard tissue culture medium ( RPMI ) with and without drug in vitro ( Figure 7 ) . 10 . 7554/eLife . 00662 . 017Figure 6 . Filamentation , adhesion and virulence increase concurrently with fitness . For each pair of consecutive isolates ( green preceding blue ) , shown are the fitness , adhesion , filamentation , and virulence in a worm model of infection ( each described in ‘Materials and methods’ ) . A subset of fitness values are duplicated from Figure 7A , with selection coefficient ( s ) shown on the Y-axis . A subset of adhesion values are plotted from Figure 6—source data 1 , with Abs590 nm on the Y-axis . A subset of images showing filamentation on spider media are shown , with the full set found in Figure 6—figure supplement 1 . For virulence , shown are Kaplan–Meier plots of survival rates from C . elegans infection with the specified C . albicans isolates ( ‘Materials and methods’ ) . For each isolate pair , significant changes in virulence were observed between the two isolates ( in all cases , p < 0 . 001 , log-rank test ) , with three of the four evolved isolates being more virulent than their corresponding progenitor . ( A ) Patient 30 isolates 5106 and 5108; ( B ) Patient 43 isolates 1649 and 3034; ( C ) Patient 1 isolates 12 and 13; ( D ) Patient 59 isolates 3917 and 4617 . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 01710 . 7554/eLife . 00662 . 019Figure 6—source data 1 . Adhesion values for the majority of isolates . Adhesion was defined as described in ‘Materials and methods’ and measured eight times to determine the average adherence as measured by Abs590 . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 01910 . 7554/eLife . 00662 . 018Figure 6—figure supplement 1 . Filamentation increases in many patient series . For several patient series , shown are the filamentation assay results after 7 days of growth on Spider Media ( ‘Materials and Methods’ ) . These data , a subset of which is shown in Figure 6 , demonstrate the heterogeneity seen between strains , as well as the general trend for filamentation to increase over time . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 01810 . 7554/eLife . 00662 . 020Figure 7 . Emergence of increased drug resistance often coincides with reduction in fitness in the absence of drug , but an increase in the presence of drug . ( A ) For each patient ( panel ) shown is the fitness ( ‘Materials and methods’ ) of each strain ( Y axis , mean ± STDV ) , ordered from the progenitor to evolved isolates ( left to right , X axis ) . Fitness is calculated relative to an ENO1::YFP SC5314 reference isolate . The MIC of each strain is shown in the gray boxes on top ( white: low; black: high , color bar at bottom ) . Green: isolates with aneuploidies; Blue: euploid isolates . ( B ) Shown is the mean difference between fitness in the absence and presence of drug ( Y axis , error bars are ± STDV; n > 3 ) for isolates ( X axis ) that showed a decrease in fitness ( Figure 7A ) in the absence of drug concomitant with an increase in MIC ( asterisks ) , and flanking isolates in Patient 1 and 59 ( ordered from the progenitor to evolved isolates , left to right , X axis ) . The difference in fitness is calculated as the difference in selection coefficient ( s , Y axis ) between matching competition experiments in RPMI and those in RPMI with one half the MIC for fluconazole ( Table 1 ) for each isolate tested ( X axis ) . Negative values indicate that the strain had higher fitness in the presence of fluconazole vs assays without fluconazole . For each assay , the fluconazole-resistant isolate 4639 ENO1::YFP was used as the reference strain . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 020 We found substantial variation in many of these phenotypes between isolates in the same series ( Figure 6 and Figure 7 ) , supporting the notion that the isolates are samples from a broad range of genetic variants within clonal ( single infection ) populations . In general , increased fitness in vitro ( in the absence of drug ) correlated with an increase in traits associated with virulence ( adhesion , filamentation , and virulence in nematode ) . For example , the later isolates in the series from patients 30 and 43 had increased fitness and higher virulence by all three measures ( Figure 6A , B , Figure 7 ) ; whereas , a decrease in fitness in isolate 13 of patient 1 was accompanied by a decrease in virulence ( Figures 6C and Figure 7 ) . A notable exception was patient 59 , where fitness in vitro decreased while virulence phenotypes increased in a later isolate ( Figures 6D and Figure 7 ) . This is consistent with the observations of Noble and co-workers that in vitro fitness is not always a reflection of virulence ( Noble et al . , 2010 ) . Initially in a series , drug resistance ( MIC ) and in vitro fitness ( in the absence of drug ) were inversely related , suggesting that these are competing selective pressures . When MIC increases first appeared , they were usually accompanied by a decrease in fitness in the absence of drug ( Patient 1 , isolates 2 , 13 , and 16 , Patients 9 , 14 , 15 16 , and 59 , Figure 7A ) . Consistent with the elevated MIC , these isolates exhibited increased relative fitness in the presence of the drug ( Figure 7B ) . This is also consistent with a recent study ( Sasse et al . , 2012 ) showing that resistance conferred in C . albicans by gain-of-function mutations in the transcription factors Mrr1 , Tac1 , and Upc2 is associated with reduced fitness under non-selective conditions in vitro as well as in vivo during colonization of a mammalian host . Consistent with subsequent selection of strains with compensatory variations , isolates from later time points were often more fit than those from earlier time points ( measured in vitro , in the absence of drug ) without further changes in MIC ( e . g . , patient 1 , isolates 5-7 , isolate 14 , Figure 7A ) , with notable exceptions ( e . g . , isolates 8 and 11 ) . This general trend is consistent with previous studies in bacteria ( Bjorkman and Andersson , 2000 ) ; ( Gagneux et al . , 2006 ) and in a single documented case in C . glabrata ( Singh-Babak et al . , 2012 ) , suggesting that compensatory mutations may subsequently arise to offset the major fitness cost of mutations conferring drug resistance . Nevertheless , substantial additional sampling will be required per time point to fully interpret such patterns . In this context , it appears that aneuploidies ( Figure 7A , green ) , while largely transient , may be an important intermediate giving rise to more stable adaptive genotypes in some cases , as was recently demonstrated in budding yeast adapting to a stressful environment in vitro ( Yona et al . , 2012 ) . For example , in Patient 1 isolate 13 , an increase in MIC and a trisomy of 5 of 8 chromosomes accompanies a large decrease in fitness ( in the absence of drug ) relative to the preceding isolate 12 ( Figure 7A ) but has increased fitness in the presence of drug ( Figure 7B ) . Isolate 14 has a similar MIC phenotype to isolate 13 but is euploid ( Figure 3 ) and is much more fit ( Figure 7A ) . Consistent with the general negative effect of aneuploidy on fitness ( Tang and Amon , 2013 ) , the absence of the extra chromosomes resulted in improved overall fitness . The analysis of clinical isolates identified a range of new candidate genes that may affect drug resistance , fitness , and/or virulence . To test the contribution of some of the recurrently identified genes to specific C . albicans phenotypes , we profiled all 23 recurrently mutated loci for which a homozygous deletion mutant was available from a deletion strain collection ( Noble et al . , 2010 ) . We measured the MIC of fluconazole and the in vitro fitness in the absence of drug for each of these 23 mutants . Deletion of one gene ( orf19 . 4658 ) caused a twofold decrease in MIC , whereas the other 22 mutants tested did have no significant effect on MIC ( data not shown ) . Deletion mutants are loss-of-function mutations , whereas the previously identified mechanisms of fluconazole resistance are ‘gain of function’ , resulting in the increase in the amount or activity level of Erg11 ( Asai et al . , 1999; Oliver et al . , 2007 ) or the efflux of drug transporters ( Coste et al . , 2006; Dunkel et al . , 2008 ) . Therefore , it is possible that the recurrent non-synonymous coding SNPs in the new loci , we identified in the clinical isolates confer resistance . Alternatively , these loci may not be involved in fluconazole resistance per se and may have a more general role in adaptation to the complex host environment . Consistent with a role in host adaptation , 5 of the 22 deletion mutants reduced in vitro fitness in a culture medium thought to approximate in vivo conditions ( Figure 8 , ‘Materials and methods’ ) . Three were significantly more fit than the WT parental strain ( SN250 , red , Figure 8 ) , including CCN1 , that encodes a G1 cyclin required for hyphal growth maintenance ( Loeb et al . , 1999 ) and orf19 . 4471 , an ortholog of Saccharomyces cerevisiae VPS64 , which is required for cytoplasm-to-vacuole targeting of proteins ( Bonangelino et al . , 2002 ) , is involved in recycling pheromone receptors ( Kemp and Sprague , 2003 ) , and is identified as an ‘aneuploidy-tolerating mutant’ ( Torres et al . , 2010 ) . Among the least fit were cell wall protein genes ( HYR1 , HYR3 , and PIR1; De Groot et al . , 2003 ) . 10 . 7554/eLife . 00662 . 021Figure 8 . Deletion mutants of recurrently mutated genes reveal changes in relative fitness . Shown is the fitness ( ‘Materials and methods’ ) for each deletion mutant strain and the corresponding wild-type strain ( Y axis , mean ± STDV ) . The wild-type parental strain ( SN250 ) is on the far left ( red bar and dashed line ) . Fitness is calculated relative to an ENO1::YFP SC5314 reference isolate . Locus names are given for the mutant isolates ( X axis ) . Asterisks denote statistical significance ( * < 0 . 05 , < 0 . 01 , * < 0 . 001 , **** < 0 . 0001 ) by one-way ANOVA with Holm–Sidak correction for multiple comparisons . DOI: http://dx . doi . org/10 . 7554/eLife . 00662 . 021 We sequenced the genomes of serial clinical isolates of C . albicans and analyzed them by comparing consecutive isolates from one patient to reach novel insights into drug resistance within the human host . This approach allowed us to distinguish ( and remove from further analysis ) isolates that were non-clonal and to estimate that at least ∼30% of the patients ( 3/10 ) carried at least one non-clonal strain of C . albicans . We used the clonal isolates to identify persistent SNPs , and the different series to identify those persistent SNPs that recurred within the same ORFs , thereby focusing the analysis on a small number of loci where the identified variants are more likely to be adaptive and excluding the substantial background of likely neutral variation that hitchhike along with selective beneficial mutations . Our study identified substantial genetic diversity in each series , in contrast to the report of only 26 SNPs detected in a single clinical series of Candida glabrata isolates that spanned a 10-month period ( Singh-Babak et al . , 2012 ) . Several reasons may account for this difference . First , fluconazole , the antifungal drug used to treat the patients in our series , is fungistatic , such that many cells exposed to the drug arrest their growth but do not die . Thus , the range of diversity in the initial population is not entirely lost . In contrast , the C . glabrata study involved exposure to caspofungin , an echinocandin fungicidal drug . Therefore , most cells likely died upon drug exposure and only the rare survivors went on to seed the remaining population . Accordingly , the C . glabrata isolates may have been subjected to selection that would have removed much of the initial diversity in the population , whereas in the C . albicans series diversity persisted and selection acted mostly to change the relative proportions of different genotypes . Second , C . albicans is a highly heterozygous diploid whereas C . glabrata is haploid . Mutations can be more readily assimilated in a diploid than in a haploid organism , since deleterious mutations are potentially buffered by a functional version ( Thompson et al . , 2006 ) . Furthermore , because C . albicans genomes are initially much more diverse ( with tens of thousands of heterozygous SNPs in a given isolate ) , LOH is a high frequency mechanism available to reveal mutations more readily . Third , C . albicans lab isolates likely undergo a stress-induced elevation of mutation and mitotic recombination rates ( Ponder et al . , 2005; Forche et al . , 2011; Rosenberg , 2011 ) , and exposure to a mammalian host results in elevated frequencies of LOH and aneuploidy ( Forche et al . 2009 ) . Thus , it is possible that C . albicans isolates within the human host also undergo elevated levels of LOH and of point mutations to generate a wider range of diversity . Thus , C . albicans like S . cerevisiae ( Gresham et al . , 2008; Pavelka et al . , 2010; Yona et al . , 2012; Chang et al . , 2013 ) generates large scale genetic variation as a means of adaptation . This adds another level of variation to the genome and protein diversity ( Santos et al . , 2004; Selmecki et al . , 2010 ) that C . albicans is able to tolerate . LOH in several genes important for fluconazole resistance has been reported previously for ERG11 ( Oliver et al . , 2007 ) , TAC1 ( Coste et al . , 2006 ) , and MRR1 ( Schubert et al . , 2011 ) , but the degree to which LOH is important in clinical infections was not known . In the ten patients studied here , LOH was commonly observed and was associated with changes in MIC . As we detected LOH of mutations in ERG11 in three patients , it would be of interest to know if the LOH in these known genes was sufficient to increase MIC , or if other genes within the homozygous region make important contributions . Aneuploidies appeared frequently within the drug-resistant isolates , consistent with previous reports ( Selmecki et al . , 2006 ) . Unlike LOH events , aneuploidies were often transient and not consistently correlated to increases in drug resistance . Perhaps these aneuploidies provide a mechanism akin to genetic assimilation ( ‘phenotype precedes genotype’ ) , in which cells are provided with a phenotypic mechanism that facilitates survival until a more stable and/or less costly mechanism is attained . In this case , the ‘phenotypic’ mechanism would be genetic but unstable—the acquisition of one or more extra chromosomes . Nevertheless , aneuploidies may cause increased frequencies of LOH events through whole chromosome loss , as well as by increasing the likelihood of recombination events . A transient role for aneuploidy is consistent with recent findings from in vitro evolution studies in S . cerevisiae ( Yona et al . , 2012 ) in which a transient aneuploidy was responsible for fitness at elevated temperature , but was eventually replaced by a more stable mutation . In addition , a substantial number of persistent and recurrent SNPs , and clusters of co-occurring SNPs , implicate a broad range of pathways and functions that likely provide some growth advantage in the presence of the complex selective pressures found in the host . In particular , there was strong enrichment for cell wall gene families thought to be critical determinants of the transition from commensalism to pathogenesis ( Gow and Hube , 2012 ) . The genes in several of these families ( e . g . , ALS1-4 , 7 , 9 and HYR/IFF genes ) frequently contain intragenic tandem repeats . Variation in intragenic repeat number modulates phenotypic diversity in adhesion and biofilm formation ( Verstrepen et al . , 2005 ) . This functional diversity of cell surface antigens has been proposed to allow rapid adaptation to the environment as well as evasion of the host immune system in fungi and other pathogens ( Gemayel et al . , 2010 ) . Notably , the cell wall deletion mutants ( HYR1 , HYR3 , and PIR1 ) were among the least fit in vitro ( Figure 8 ) . Indeed , many of the isolates evolved additional phenotypes , including changes in in vitro fitness , filamentation , adhesion , and in vivo virulence , and the data presented here points to candidate genes that underlie some of these evolved traits . For example , the evolved isolate 4617 in patient 59 had a dramatic increase in filamentation relative to the progenitor , which was concomitant with the appearance of persistent SNPs in genes associated with filamentous growth: CHO1 , MNN2 , and 7 different FGR ( filamentous growth regulator ) genes ( Uhl et al . , 2003 ) . The evolution of drug resistance in C . albicans has many parallels with the somatic evolution of cancer cells undergoing chemotherapy or treated with specific inhibitors . These include variation on a background of clonal descent , lack of sexual recombination , acquisition of drug resistance , tolerance of aneuploidy and genome plasticity , and increased mutation and mitotic recombination rates under stress . Indeed , several recent studies have shown a similar spectrum of genetic alterations to those observed here during the somatic evolution of cancers in patients undergoing chemotherapy ( Podlaha et al . , 2012; Landau et al . , 2013 ) or treated with specific inhibitors ( Ding et al . , 2012 ) to those observed here . Finally , our data and analyses provide a rich and novel resource for Candida researchers and a host of candidate genes for further functional studies . While our analysis focused on recurrent SNPs in ORFs , we nonetheless cataloged the many genetic alterations found in intergenic regions ( Figure 2—source data 1 ) , some of which could affect gene regulation . It will be especially interesting to analyze the similarities and differences in additional C . albicans genome sequences that are likely to become available in the near future . Our results suggest there may be complex population dynamics during the transition from commensal to pathogen and across the course of treatment . As sequencing capacity continues to grow , it will be especially interesting to more fully sample this population-level diversity during longitudinal collection to better understand these dynamics . In particular , it will be interesting to determine the degree to which specific mutations recur in different isolates prior to and after the acquisition of drug resistance . Isolates were obtained from HIV-infected patients with oropharyngeal candidiasis , as previously described ( White , 1997a; Perea et al . , 2001 ) . The patients were not on azole antifungal treatment at time of enrollment; subsequent samples were collected during recurrence of infection . Isolates were colony purified at collection and represent a single clone . The isolates are detailed in Table 1 . Minimal inhibitory concentrations ( MIC ) were determined for each strain ( clinical and mutant ) using fluconazole E-test strips ( 0 . 016–256 μg/ml , bioMérieux , Durham , NC ) on RPMI 1640-agar plates ( Remel , Lenexa , KS ) . Overnight YPD cultures were diluted in sterile 0 . 85% NaCl to an OD600 of 0 . 01 and 250 μl was plated using beads . After a 30-min pre-incubation , 2–3 E-test strips were applied and plates were incubated at 35°C for 48 hr . The susceptibility endpoint was read at the first growth-inhibition ellipse , and the median value is reported here . Genomic DNA was prepared from different clinical time courses via a Qiagen Maxiprep kit ( Qiagen , Valencia , CA ) and sequenced using 101 base paired-end Illumina sequencing ( Mardis , 2008 ) . Library preparation included an eight base barcode ( Grabherr et al . , 2011 ) ; 43 samples from 11 patients were sequenced . All reads were mapped to the SC5314 reference genome ( Candida Genome Database Assembly 21 , gff downloaded on 4 January 2010 ) using the BWA alignment tool ( version 0 . 5 . 9 ) ( Li and Durbin , 2009 ) . To minimize false positive SNP calls near insertion/deletion ( indel ) events , poorly aligning regions were identified and realigned using the GATK RealignerTargetCreator and IndelRealigner ( GATK version 1 . 4-14 , [version 1 . 4-14] ) ( McKenna et al . , 2010 ) . Coverage for each strain is reported in Table 1 . Coverage was defined as the total number of bases with BWA mapping quality greater than 10 divided by the total number of sites in the nuclear genome . Isolate 4380 aligned poorly to the SC5314 genome; however , this sequence aligned at high identity to the C . dubliniensis genome and was therefore removed from further analysis . These data can be accessed from a genomics portal hosted by the Broad Institute at: http://www . broadinstitute . org/pubs/candidadrugresistance/ . Reads are deposited for access to the NCBI SRA under project accession number PRJNA257929 . SNPs were identified using Unified Genotyper ( GATK version 1 . 4 . 14 ) ( McKenna et al . , 2010 ) , using read alignments to the SC5314 reference sequence . Unreliable SNPs were identified using the GATK Variant Filtration module , with the version 3 best practice recommended annotation filters ( QD < 2 . 0 , MQ < 40 . 0 , FS > 60 . 0 , HaplotypeScore >13 . 0 , MQRankSum < −12 . 5 , ReadPosRankSum < −8 . 0 ) except that the HaplotypeScore was also filtered if greater than two standard deviations above the mean of all HaplotypeScore values . The combined list of SNP positions across all strains was used to evaluate those matching the reference allele; by emitting all sites using Unified Genotyper , high quality reference matches were identified as positions with quality of 30 or greater , with positions with extremes of read depth ( top or bottom 0 . 5% quantile ) eliminated . A matrix of all strains by all positions was created from the SNP calls , with reference calls added where identified . Non-clonal isolates ( see below ) were removed from further analysis . We chose 1973 genetic locus X strain combination ( 523 unique sites across nine patients ) for iPLEX genotyping as either ( 1 ) persistent within their time course ( 605 sites ) , ( 2 ) background mutations ( 1263 sites ) , or ( 3 ) transient mutations ( 105 sites ) . All selected loci were at least 150 bp away from any other SNP in either direction to avoid ambigious iPlex calls . Sites producing multiple iPlex results were eliminated from further consideration . 1 , 853 predictions were confirmed as correct by Sequenom genotyping and 120 were discordant ( Table 2 ) , to a calling accuracy of 93 . 9% . We investigated the phylogenetic relationship of all strains using SNP calls to determine relatedness between strains; positions with missing data in 10% or more of strains were eliminated , resulting in a total of 201 , 793 parsimony informative positions . A distance based tree was estimated using maximum parsimony with PAUP* ( 4 . 0 ) ( Swofford , 2002 ) ; a step matrix was used to estimate the distance between homozygous and heterozygous positions , where each of the homozygotes is two steps apart from each other and one step from the heterozygote . SNP positions were resampled using 1000 bootstrap replicates , and the phylogeny re-estimated to test the branch support . We define isolates with a branch distance of greater than 20 , 000 as non-clonal . For each strain , we calculated a per-locus depth-of-coverage using GATK ( McKenna et al . , 2010 ) , with a minimal mapping quality of 10 . The number of reads aligning to each 5 kb window across the nuclear genome was calculated and then normalized to the genome median . Each bin was then multiplied to the ploidy for the majority of the genome as determined by a FACS assay ( below ) . We then applied a sliding window across each bin , defining a potential CNV if 70% of 10 consecutive bins had a normalized count >2 . 5× . Regional amplifications are identified if >15% of the chromosome is identified as having a CNV . Boundaries were confirmed by visual inspection in the Integrative Genome Viewer ( Robinson et al . , 2011 ) . C . albicans cultures were grown to log phase . 200 μl of culture was centrifuged in a round bottom microtiter plate , and pellets were resuspended in 20 μl of 50 mM Tris pH8/50 mM EDTA ( 50/50 TE ) . 180 μl of 95% ethanol was added and suspensions were stored overnight at 4°C . Cells were centrifuged and pellets washed twice with 200 μl of 50/50 TE , then resuspended in 50 μl of RNAse A at 1 mg/ml in 50/50 TE and incubated 1 hr at 37°C . Cells were centrifuged and pellets resuspended in 50 μl of Proteinase K at 5 mg/ml in 50/50 TE for 30 min at 37°C . Cells were washed in 50/50 TE and pellets resuspended in 50 μl of a 1:85 dilution SYBR Green I ( Invitrogen , Carlsbad , CA ) in 50/50 TE and incubated overnight in the dark at 4°C . Cells were centrifuged and pellets were resuspended in 700 μl 50/50 TE and read on a FACS caliber flow cytometer ( BD Biosciences , San Jose , CA ) . Flow data were fitted with a multi-Gaussian cell cycle model to produce estimates for whole genome ploidy . For each time course , we assembled the high quality SNPs ( post-filtering , above ) from multi-sample calling into the columns of a matrix , ordered by genome position , with the isolates in rows , ordered temporally . The genetic state of each locus in each sample was coded to distinguish loci homozygous for the haploid reference ( −1 ) , heterozygous SNPs ( 0 ) , and homozygous SNPs for the non-reference state ( 1 ) . We then applied a sliding window method across each chromosome , only looking at sites in which a SNP call was made in at least one isolate . An LOH event was defined as occurring if ( 1 ) at least one isolate had a heterozygosity content >40% , and ( 2 ) at least one other isolate had a heterozygosity content <5% . Window sizes were of length 500 . Boundaries were trimmed such that if a window terminated in a heterozygous site in the isolate for which the LOH occurred , it was trimmed back until it was homozygous . If two 500+ windows were within 7 KB of each , the region was assessed to determine if the event was actually one event and merged if the heterozygous sites in the inter-window space had homozygosed . If two isolates had LOHs that overlapped but did not have precisely identical boundaries , the LOH regions were combined such that the LOH interval for both isolates was the same . All LOH regions were confirmed by visual inspection and are listed in Figure 3 and Figure 4—source data 1 . For each time course , each SNP was classified for its position in the genome ( Figure 2—source data 1 ) . If the SNP fell within an ORF , the reference and altered amino acids were reported . If the SNP fell outside of an ORF , the distance to the closest flanking ORF ( s ) was reported , as well as the SNP's orientation with respect to these ORFs . SNP genotypes that are common to all isolates ( including the ‘progenitor’ ) were classified as background mutations . Genotypes not present in the progenitor or evolved strain , but that occur in one or more intermediate strain , are classified as transient . Finally , genotypes that occur after the progenitor , and persist through the terminally evolved time point , are classified as persistent . To determine if the number of persistent non-synonymous SNPs ( nsSNPs ) occurring in conjunction with changes in MIC was greater than expected , we developed a simple model to simulate the occurrence of nsSNPs outside of LOH regions at each time point . For each time point ( i ) , a random variable Xi ∼ Pois ( λi ) was assigned , where λi represents the Poisson parameter for each time point:λi = m/T × ti;m = 1471 , the number of ORFs with persistent nsSNP;T = 23 , the number of time points;ti = the length of time ( days ) for time point ( i ) divided by the mean length of time . The number of persistent nsSNP-containing ORFs for each of the 14 time points associated with a change in MIC was summed , and this was repeated 100 , 000 times to build a probability distribution , where p ( observing x mutated ORFs ) was determined by dividing the number of successes for each bin by the number of trials . To determine the probability of observing x recurrent , persistently mutated ORFs outside of LOH regions , we developed an additional stochastic simulation model . For each patient series ( i ) , at each non-LOH ORF ( j ) , a random variable , Xij ∼ B ( n , pij ) was assigned , where n represents the number of trials , 1 , and p represents the probability of a SNP occurring in that ORF: p = mi/Mi × hj; mi = number of ORFs with persistent nsSNPs found outside of LOH events for patient series ( i ) ; Mi = number of ORFs outside of LOH events for patient series ( i ) ; hj = log normalized ORF length divided by mean lognormalized ORF length for ORF ( j ) . This was repeated 10 , 000 times to build a distribution , and p ( observing x recurrent nsSNPs ) was determined by dividing the number of successes in each bin by the cumulative number of trials . For co-occurrence analysis we focused only SNPs that ( 1 ) had persistent nonsynonymous coding SNPs that did not occur in LOH regions and ( 2 ) recurred in three or more time courses . We generated for each such gene a binary patient vector , and we used NMF clustering ( Brunet et al . , 2004 ) to identify the optimal number of clusters , based on local maximas . This was accomplished using the ‘NMFConsensus’ module ( version 5 ) in GenePattern ( Reich et al . , 2006 ) . To determine the most appropriate number of clusters , k was selected such that it was the smallest value for which the cophenetic correlation begins decreasing . We then tested each of the co-occurrence gene clusters for functional enrichment ( below ) . To determine if the degree of co-occurrence would have arisen by chance , we ran 1000 iterations of 1 million edge-pair swaps from the original binary matrix , calculating a Pearson correlation matrix for each of the 1000 iterations . We compared the distribution of Pearson correlations on the real and permuted vectors using a two-sample Kolmogorov–Smirnov ( KS ) test and Wilcoxon Rank Sum test . We calculated the overlap of each co-occurring cluster with Gene Ontology gene sets using the Gene Ontology toolset from the Candida Genome Database ( Arnaud et al . , 2009; Inglis et al . , 2012 , 2013 ) . Bonferroni adjusted p-values as well as the False Discovery Rate are reported ( Figure 5—source data 1A ) . We measured the relative fitness of the progenitor and evolved lines in RPMI Cell Culture medium ( Gibco , Grand Island , NY ) , competing them against a reference strain ( SC5314 ) , expressing ENO1::YFP . Isolates stored at −80°C were revived on rich media petri plates and then grown overnight in 3-ml cultures of minimal media in a roller drum at 35°C . An aliquot of cells in each culture was removed , sonicated in a Branson 450 sonifier , and the concentration of cells was determined using a Cellometer M10 ( Nexcelom , Lawrence , MA ) . The reference strain and experimental competitors were added to fresh RPMI medium in a 1:1 ratio and a final cell concentration of 1 × 107 cells/ml . The cultures were grown for 24 hr in a roller drum at 35°C . Cells were then counted as above , and 3 × 106 cells were transferred to fresh RPMI medium grown for 24 hr in a roller drum at 35°C ( transfer cycle 1 ) . This procedure was repeated ( transfer cycle 2 ) . This protocol represents 5–10 generations of growth , depending on the strain genotype . The ratio of the two competitors was quantified at the initial and final time points by flow cytometry ( Accuri , San Jose , CA ) . 3 to 6 independent replicates for each fitness measurement were performed . The selective advantage , s , or disadvantage of the evolved population was calculated as previously described ( Thompson et al . , 2006 ) , where E and R are the numbers of evolved and reference cells in the final ( f ) and initial ( i ) populations , and T is the number of generations that reference cells have proliferated during the competition . Fitness assays were performed as described above except that the reference strain used was a derivative of the drug-resistant isolate 4639 from patient 59 ( Table 1 ) expressing ENO1::YFP and competition experiments were performed in RPMI and in RPMI with one half the MIC for fluconazole ( Table 1 ) initiated from replicate 1:1 mixtures of the same population of cells for each isolate tested . In order to quantify fitness in the presence of fluconazole , we constructed a derivative of the fluconazole-resistant ( 128 μg/ml ) isolate 3795 from patient 9 ( Table 1 ) that expresses ENO1::YFP to use as the reference for competitive fitness assays in the presence and absence of fluconazole . This strain was chosen since it was a euploid strain with the highest MIC . There were two strains from patient 42 ( 3731 and 3733 ) with a higher MIC but these strains are aneuploid and thus the potential loss of additional chromosomes during the course of the competition could alter fitness and confound our results . The addition of the YFP marker reduced the fitness of the strain relative to the unaltered one and slightly reduced the fluconazole MIC as measured by the E-strip test . Fitness assays were performed as described above competition experiments except they were performed in RPMI and in RPMI with one half the MIC for fluconazole ( Table 1 ) initiated from replicate 1:1 mixtures of the same population of cells for each isolate tested . A C . elegans survival assay was performed as previously described ( Jain et al . , 2009 ) . Briefly , Escherichia coli OP50 and the different C . albicans clinical isolates were grown overnight respectively in LB at 37°C and YPD at 30°C . E . coli was then centrifuged and resuspended to a final concentration of 200 mg/ml , while C . albicans isolates were diluted with sterile water to OD600 = 3 . Small petri dishes ( 3 . 5 cm ) containing NGM agar were spotted with a mixture of 10 μl streptomycin ( stock solution 50 mg/ml ) , 2 . 5 μl of E . coli , 0 . 5 μl of C . albicans , and 7 μl of sterile water . The plates were incubated overnight at 25°C and 20 young synchronized N2 C . elegans adults were transferred on the spotted plates . Synchronous populations of adult worms were obtained by plating eggs on NGM plates seeded with E . coli OP50 at 20°C for 2–3 days . In this time frame , the eggs hatch and the larvae reach young adulthood . The survival assay was carried at 20°C , and worms were scored daily by gentle prodding with a platinum wire; dead worms were discarded while live ones were transferred to seeded plates grown overnight at 25°C . Worms accidentally killed while transferring or found dead on the edges of the plates were censored . Statistical analysis was performed using SPSS software; survival curves were obtained using the Kaplan–Meier method and p-values by using the log-rank test . Overnight cultures grown in YPD at 30°C were normalized to OD600 = 1 with sterile water and spotted on Spider agar media ( 1% mannitol , 1% Difco nutrient broth , 0 . 2% K2HPO4 ) . Plates were incubated at 37°C and colonies were photographed 3 , 7 , and 10 days post spotting . As a negative control for filamentation cph1/cph1 efg1/efg1 ( Lo et al . , 1997 ) double mutant strain was used . The in vitro adhesion assay was performed as previously described for S . cerevisiae ( Reynolds and Fink , 2001 ) . Briefly , cultures were grown in Synthetic Complete ( SC ) media + 0 . 15% glucose at 30°C overnight . Cells were then centrifuged at maximal speed and resuspended to OD600 = 0 . 5 in fresh media . 200 ml of each culture were dispensed into 8 wells of a flat bottom 96-well plate and incubated at 37°C for 4 hr . The content of the plate was then decanted and 50 ml of crystal violet added to each well . After 45 min of incubation at room temperature , the content of the plate was decanted and the plate was rinsed ten times in DI water by alternate submerging and decanting . 200 ml of 75% methanol was added to each well and absorbance was measured after 30 min at OD590 . An edt1/edt1 knockout mutant ( Zakikhany et al . , 2007 ) was used as a negative control for adhesion .	Nearly all humans are infected with the fungus Candida albicans . In most people , the infection does not produce any symptoms because their immune system is able to counteract the fungus' attempts to spread around the body . However , if the balance between fungal attack and body defence fails , the fungus is able to spread , which can lead to serious disease that is fatal in 42% of cases . How does C . albicans outcompete the body's defences to cause disease ? This is a pertinent question because the most effective antifungal medicines—including the drug fluconazole—do not kill the fungus; they only stop it from growing . This gives the fungus time to develop resistance to the drug by becoming able to quickly replace the fungal proteins the drug destroys , or to efficiently remove the drug from its cells . In this study , Ford et al . studied the changes that occur in the DNA of C . albicans over time in patients who are being treated with fluconazole . Ford et al . took 43 samples of C . albicans from 11 patients with weakened immune systems . The experiments show that the fungus samples collected early on were more sensitive to the drug than the samples collected later . In most cases , the genetic data suggest that the infections begin with a single fungal cell; the cells in the later samples are its offspring . Despite this , there is a lot of genetic variation between samples from the same patient , which indicates that the fungus is under pressure to become more resistant to the drug . There were 240 genes—including those that can alter the surface on the fungus cells to make it better at evading the host immune system—in which small changes occurred over time in three or more patients . Laboratory tests revealed that many of these genes are likely important for the fungus to survive in an animal host in the presence of the drug . C . albicans cells usually have two genetically distinct copies of every gene . Ford et al . found that for some genes—including some that make surface components or are involved in expelling drugs from cells—the loss of genetic information from one copy , so that both copies become identical , is linked to resistance to fluconazole . However , the gain of whole or partial chromosomes—which contain large numbers of genes—is not linked to resistance , but may provide additional genetic material for generating diversity in the yeast population that may help the cells to evolve resistance in the future . These experiments have identified many new candidate genes that are important for drug resistance and evading the host immune system , and which could be used to guide the development of new therapeutics to treat these life-threatening infections .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "microbiology", "and", "infectious", "disease" ]	2015	The evolution of drug resistance in clinical isolates of Candida albicans
Which virological factors mediate overdispersion in the transmissibility of emerging viruses remains a long-standing question in infectious disease epidemiology . Here , we use systematic review to develop a comprehensive dataset of respiratory viral loads ( rVLs ) of SARS-CoV-2 , SARS-CoV-1 and influenza A ( H1N1 ) pdm09 . We then comparatively meta-analyze the data and model individual infectiousness by shedding viable virus via respiratory droplets and aerosols . The analyses indicate heterogeneity in rVL as an intrinsic virological factor facilitating greater overdispersion for SARS-CoV-2 in the COVID-19 pandemic than A ( H1N1 ) pdm09 in the 2009 influenza pandemic . For COVID-19 , case heterogeneity remains broad throughout the infectious period , including for pediatric and asymptomatic infections . Hence , many COVID-19 cases inherently present minimal transmission risk , whereas highly infectious individuals shed tens to thousands of SARS-CoV-2 virions/min via droplets and aerosols while breathing , talking and singing . Coughing increases the contagiousness , especially in close contact , of symptomatic cases relative to asymptomatic ones . Infectiousness tends to be elevated between 1 and 5 days post-symptom onset . Intrinsic case variation in rVL facilitates overdispersion in the transmissibility of emerging respiratory viruses . Our findings present considerations for disease control in the COVID-19 pandemic as well as future outbreaks of novel viruses . Natural Sciences and Engineering Research Council of Canada ( NSERC ) Discovery Grant program , NSERC Senior Industrial Research Chair program and the Toronto COVID-19 Action Fund . Severe acute respiratory syndrome coronavirus 2 ( SARS-CoV-2 ) has spread globally , causing the coronavirus disease 2019 ( COVID-19 ) pandemic with more than 129 . 2 million infections and 2 . 8 million deaths ( as of 1 April 2021 ) ( Dong et al . , 2020 ) . For respiratory virus transmission , airway epithelial cells shed virions to the extracellular fluid before atomization ( from breathing , talking , singing , coughing and aerosol-generating procedures ) partitions them into a polydisperse mixture of particles that are expelled to the ambient environment . Aerosols ( ≤100 μm ) can be inhaled nasally , whereas droplets ( >100 μm ) tend to be excluded ( Prather et al . , 2020; Roy and Milton , 2004 ) . For direct transmission , droplets must be sprayed ballistically onto susceptible tissue ( Liu et al . , 2017a ) . Hence , droplets predominantly deposit on nearby surfaces , potentiating indirect transmission . Aerosols can be further categorized based on typical travel characteristics: short-range aerosols ( 50–100 μm ) tend to settle within 2 m; long-range ones ( 10–50 μm ) often travel beyond 2 m based on emission force; and buoyant aerosols ( ≤10 μm ) remain suspended and travel based on airflow profiles for minutes to many hours ( Liu et al . , 2017a; Wei and Li , 2015 ) . Although proximity has been associated with infection risk for COVID-19 ( Chu et al . , 2020 ) , studies have also suggested that long-range airborne transmission occurs conditionally ( Hamner et al . , 2020; Lu et al . , 2020a; Park et al . , 2020 ) . While the basic reproductive number has been estimated to be 2 . 0–3 . 6 ( Hao et al . , 2020; Li et al . , 2020a ) , transmissibility of SARS-CoV-2 is highly overdispersed ( dispersion parameter k , 0 . 10–0 . 58 ) , with numerous instances of superspreading ( Hamner et al . , 2020; Lu et al . , 2020a; Park et al . , 2020 ) and few cases ( 10–20% ) causing many secondary infections ( 80% ) ( Bi et al . , 2020; Endo et al . , 2020; Laxminarayan et al . , 2020 ) . Similarly , few cases drive the transmission of SARS-CoV-1 ( k , 0 . 16–0 . 17 ) ( Lloyd-Smith et al . , 2005 ) , whereas influenza A ( H1N1 ) pdm09 transmits more homogeneously ( k , 7 . 4–14 . 4 ) ( Brugger and Althaus , 2020; Roberts and Nishiura , 2011 ) , despite both viruses spreading by contact , droplets and aerosols ( Cowling et al . , 2013; Yu et al . , 2004 ) . Although understanding the determinants of viral overdispersion is crucial towards characterizing transmissibility and developing effective strategies to limit infection ( Lee et al . , 2020 ) , mechanistic associations for k remain unclear . As an empirical estimate , k depends on myriad extrinsic ( behavioral , environmental and invention ) and host factors . Nonetheless , k remains similar across distinct outbreaks for a virus ( Lloyd-Smith et al . , 2005 ) , suggesting that intrinsic virological factors mediate virus overdispersion . Here , we investigated how intrinsic case variation in respiratory viral loads ( rVLs ) facilitates overdispersion in SARS-CoV-2 transmissibility . By systematic review , we developed a comprehensive dataset of rVLs from cases of COVID-19 , SARS and A ( H1N1 ) pdm09 . Using comparative meta-analyses , we found that heterogeneity in rVL was associated with overdispersion among these emerging infections . To assess potential sources of case heterogeneity , we analyzed SARS-CoV-2 rVLs across age and symptomatology subgroups as well as disease course . To interpret the influence of heterogeneity in rVL on individual infectiousness , we modeled likelihoods of shedding viable virus via respiratory droplets and aerosols . We conducted a systematic review on virus quantitation in respiratory specimens taken during the infectious periods of SARS-CoV-2 ( −3 to 10 days from symptom onset [DFSO] ) ( Arons et al . , 2020; He et al . , 2020; Wölfel et al . , 2020 ) , SARS-CoV-1 ( 0–20 DFSO ) ( Pitzer et al . , 2007 ) and A ( H1N1 ) pdm09 ( −2 to 9 DFSO ) ( Ip et al . , 2017 ) ( Materials and methods ) . The systematic search ( Figure 1—source data 1 , Figure 1—source data 2 , Figure 1—source data 3 , Figure 1—source data 4 , Figure 1—source data 5 ) identified 4274 results . After screening and full-text review , 64 studies met the inclusion criteria and contributed to the systematic dataset ( Figure 1 ) ( N = 9631 total specimens ) , which included adult ( N = 5033 ) and pediatric ( N = 1608 ) cases from 15 countries and specimen measurements for asymptomatic ( N = 2387 ) , presymptomatic ( N = 28 ) and symptomatic ( N = 7161 ) infections . According to a hybrid Joanna Briggs Institute critical appraisal checklist , risk of bias was low for most contributing studies ( Appendix 1—table 1 ) . We hypothesized that individual case variation in rVL facilitates the distinctions in k among COVID-19 , SARS and A ( H1N1 ) pdm09 . For each study in the systematic dataset , we used specimen measurements ( viral RNA concentration in a respiratory specimen ) to estimate rVLs ( viral RNA concentration in the respiratory tract ) ( Materials and methods ) . To investigate the relationship between k and heterogeneity in rVL , we performed a meta-regression using each contributing study ( Figure 2—figure supplement 1 ) , which showed a weak , negative association between the two variables ( meta-regression slope t-test: p=0 . 038 , Pearson’s r = −0 . 26 ) . Using contributing studies with low risk of bias , meta-regression ( Figure 2 ) showed a strong , negative association between k and heterogeneity in rVL for these three viruses ( meta-regression slope t-test: p<0 . 001 , Pearson’s r = −0 . 73 ) . In this case , each unit increase ( one log10 copies/ml ) in the standard deviation ( SD ) of rVL decreased log ( k ) by a factor of −1 . 41 ( 95% confidence interval [CI]: −1 . 78 to −1 . 03 ) , suggesting that broader heterogeneity in rVL facilitates greater overdispersion in the transmissibility of SARS-CoV-2 than of A ( H1N1 ) pmd09 . To investigate mechanistic aspects of this association , we conducted a series of analyses on rVL and then modeled the influence of heterogeneity in rVL on individual infectiousness . We first compared rVLs among the emerging infections . We performed a random-effects meta-analysis ( Figure 2—figure supplement 2 ) , which approximated the expected rVL when encountering a COVID-19 , SARS or A ( H1N1 ) pdm09 case during the infectious period . This showed that the expected rVL of SARS-CoV-2 was comparable to that of SARS-CoV-1 ( one-sided Welch’s t-test: p=0 . 111 ) but lesser than that of A ( H1N1 ) pdm09 ( p=0 . 040 ) . We also performed random-effects subgroup analyses for COVID-19 ( Figure 3 ) , which showed that expected SARS-CoV-2 rVLs were similar between pediatric and adult cases ( p=0 . 476 ) and comparable between symptomatic/presymptomatic and asymptomatic infections ( p=0 . 090 ) . Since these meta-analyses had significant between-study heterogeneity among the mean estimates ( Cochran’s Q test: p<0 . 001 for each meta-analysis ) , we conducted risk-of-bias sensitivity analyses; meta-analyses of low-risk-of-bias studies continued to show significant heterogeneity ( Figure 3—figure supplements 1–5 ) . We next analyzed rVL distributions . For all three viruses , rVLs best conformed to Weibull distributions ( Figure 4—figure supplement 1 ) , and we fitted the entirety of individual sample data for each virus in the systematic dataset ( Figure 4A , Figure 4—figure supplement 1N ) . While COVID-19 and SARS cases tended to shed lesser virus than those with A ( H1N1 ) pdm09 ( Figure 2—figure supplement 2 ) , broad heterogeneity in SARS-CoV-2 and SARS-CoV-1 rVLs inverted this relationship for highly infectious individuals ( Figure 4A , Figure 4—figure supplement 2A-C ) . At the 90th case percentile ( cp ) throughout the infectious period , the estimated rVL was 8 . 91 ( 95% CI: 8 . 83–9 . 00 ) log10 copies/ml for SARS-CoV-2 , whereas it was 8 . 62 ( 8 . 47–8 . 76 ) log10 copies/ml for A ( H1N1 ) pdm09 ( Figure 4—figure supplement 3 ) . The SD of the overall rVL distribution for SARS-CoV-2 was 2 . 04 log10 copies/ml , while it was 1 . 45 log10 copies/ml for A ( H1N1 ) pdm09 , showing that heterogeneity in rVL was indeed broader for SARS-CoV-2 . To assess potential sources of heterogeneity in SARS-CoV-2 rVL , we compared rVL distributions among COVID-19 subgroups . In addition to comparable mean estimates ( Figure 3 ) , adult , pediatric , symptomatic/presymptomatic and asymptomatic COVID-19 cases showed similar rVL distributions ( Figure 4B , C ) , with SDs of 2 . 03 , 2 . 06 , 2 . 00 and 2 . 01 log10 copies/ml , respectively . Thus , age and symptomatology minimally influenced case variation in SARS-CoV-2 rVL during the infectious period . To analyze the influences of disease course , we delineated individual SARS-CoV-2 rVLs by DFSO and fitted the mean estimates to a mechanistic model for respiratory virus kinetics ( Figure 4D and Materials and methods ) . The outputs indicated that , on average , each productively infected cell in the airway epithelium shed SARS-CoV-2 at 1 . 33 ( 95% CI: 0 . 74–1 . 93 ) copies/ml day−1 and infected up to 9 . 25 susceptible cells ( Figure 4—figure supplement 4 ) . The turnover rate for infected epithelial cells was 0 . 71 ( 0 . 26–1 . 15 ) days−1 , while the half-life of SARS-CoV-2 RNA before clearance from the respiratory tract was 0 . 21 ( 0 . 11–2 . 75 ) days . By extrapolating the model to an initial rVL of 0 log10 copies/ml , the estimated incubation period was 5 . 38 days , which agrees with epidemiological findings ( Li et al . , 2020a ) . Conversely , the expected duration of shedding was 25 . 1 DFSO . Thus , SARS-CoV-2 rVL increased exponentially after infection , peaked around 1 DFSO along with the proportion of infected epithelial cells ( Figure 4—figure supplement 5 ) and then diminished exponentially . To evaluate case heterogeneity across the infectious period , we fitted distributions for each DFSO ( Figure 4E ) , which showed that high SARS-CoV-2 rVLs also increased from the presymptomatic period , peaked at 1 DFSO and then decreased towards the end of the first week of illness . For the 90th cp at 1 DFSO , the rVL was 9 . 84 ( 95% CI: 9 . 17–10 . 56 ) log10 copies/ml , an order of magnitude greater than the overall 90th cp estimate . High rVLs between 1 and 5 DFSO were elevated above the expected values from the overall rVL distribution ( Figure 4—figure supplement 3 ) . At −1 DFSO , the 90th cp rVL was 8 . 30 ( 6 . 88–10 . 02 ) log10 copies/ml , while it was 7 . 93 ( 7 . 35–8 . 56 ) log10 copies/ml at 10 DFSO . Moreover , heterogeneity in rVL remained broad across the infectious period , with SDs of 1 . 83–2 . 44 log10 copies/ml between −1 to 10 DFSO ( Figure 4—figure supplement 2H-S ) . Towards analyzing the influence of heterogeneity in rVL on individual infectiousness , we first modeled the likelihood of respiratory particles containing viable SARS-CoV-2 . Since rVL is an intensive quantity , the volume fraction of virions is low and viral partitioning coincides with atomization , we used Poisson statistics to model likelihood profiles . To calculate an unbiased estimator of partitioning ( the expected number of viable copies per particle ) , our method multiplied rVL estimates with particle volumes during atomization and an assumed viability proportion of 0 . 1% in equilibrated particles ( Materials and methods ) . When expelled by the mean COVID-19 case during the infectious period , respiratory particles showed low likelihoods of carrying viable SARS-CoV-2 ( Figure 5—figure supplement 1 ) . Aerosols ( equilibrium aerodynamic diameter [da] ≤ 100 µm ) were ≤3 . 16% ( 95% CI: 2 . 61–3 . 71% ) likely to contain a virion . Droplets also had low likelihoods: at da = 200 µm , they were 22 . 3% ( 21 . 4–23 . 2% ) , 3 . 36% ( 3 . 03–3 . 69% ) and 0 . 34% ( 0 . 29–0 . 39% ) likely to contain one , two or three virions , respectively . COVID-19 cases with high rVLs , however , expelled particles with considerably greater likelihoods of carrying viable copies ( Figure 5A , B , Figure 5—figure supplement 1D , E ) . For the 80th cp during the infectious period , aerosols ( da ≤ 100 µm ) were ≤87 . 9% ( 95% CI: 87 . 2–88 . 5% ) likely to carry at least one SARS-CoV-2 virion . For the 90th cp , larger aerosols tended to contain multiple virions ( Figure 5—figure supplement 1E ) . At 1 DFSO , these estimates were greatest , and ≤98 . 8% ( 98 . 1–99 . 4% ) of buoyant aerosols ( da ≤ 10 µm ) contained at least one viable copy of SARS-CoV-2 for the 98th cp . When expelled by high cps , droplets ( da > 100 µm ) tended to contain tens to thousands of SARS-CoV-2 virions ( Figure 5B , Figure 5—figure supplement 1E ) . Using the partitioning estimates in conjunction with published profiles of the particles expelled by respiratory activities ( Figure 5—figure supplement 2 ) , we next modeled the rates at which talking , singing , breathing and coughing shed viable SARS-CoV-2 across da ( Figure 5C-F ) . Singing shed virions more rapidly than talking based on the increased emission of aerosols . Voice amplitude , however , had a significant effect on aerosol production , and talking loudly emitted aerosols at similar rates to singing ( Figure 5—figure supplement 2E ) . Based on the generation of larger aerosols and droplets , talking and singing shed virions significantly more rapidly than breathing ( Figure 5C-E ) . Each cough shed similar quantities of virions as in a minute of talking ( Figure 5C , F ) . Each of these respiratory activities expelled aerosols at greater rates than droplets , but particle size correlated with the likelihood of containing virions according to our model . Talking , singing and coughing expelled virions at comparable proportions via droplets ( 55 . 6–59 . 4% ) and aerosols ( 40 . 6–44 . 4% ) , whereas breathing did so predominantly within aerosols ( Figure 5G ) . Moreover , short-range aerosols mediated most of the virions ( 79 . 2–81 . 9% ) shed via aerosols while talking normally and coughing . In comparison , while singing , or talking loudly , buoyant ( 14 . 5% ) and long-range ( 17 . 5% ) aerosols carried a larger proportion of the virions shed via aerosols ( Figure 5G ) . To interpret how heterogeneity in rVL influences individual infectiousness , we modeled total SARS-CoV-2 shedding rates ( over all particle sizes ) for each respiratory activity ( Figure 5H , Figure 5—figure supplement 3 ) . Between the 1st and the 99th cps , the estimates for a respiratory activity spanned ≥8 . 48 orders of magnitude on each DFSO; cumulatively from −1 to 10 DFSO , they spanned 11 . 0 orders of magnitude . Hence , many COVID-19 cases inherently presented minimal transmission risk , whereas highly infectious individuals shed considerable quantities of SARS-CoV-2 . For the 98th cp at 1 DFSO , singing expelled 313 ( 95% CI: 37 . 5–3158 ) virions/min to the ambient environment , talking emitted 293 ( 35 . 1–2664 ) virions/min , breathing exhaled 1 . 54 ( 0 . 18–15 . 5 ) virions/min and coughing discharged 249 ( 29 . 8–25111 ) virions/cough; these estimates were approximately two orders of magnitude greater than those for the 85th cp . For the 98th cp at −1 DFSO , singing shed 14 . 5 ( 0 . 15–4515 ) virions/min and breathing exhaled 7 . 13 × 10−2 ( 7 . 20 × 10−4–220 . 2 ) virions/min . The estimates at 9–10 DFSO were similar to these presymptomatic ones ( Figure 5H , Figure 5—figure supplement 3B ) . As indicated by comparable mean rVLs ( Figure 3 ) and heterogeneities in rVL ( Figure 4B , C ) , adult , pediatric , symptomatic/presymptomatic and asymptomatic COVID-19 subgroups presented similar distributions for shedding virions through these activities . We also compared the influence of case variation on individual infectiousness between A ( H1N1 ) pdm09 and COVID-19 . Aerosol spread accounted for approximately half of A ( H1N1 ) pdm09 transmission events ( Cowling et al . , 2013 ) , and the 50% human infectious dose for aerosolized influenza A virus is approximately 1–3 virions in the absence of neutralizing antibodies ( Fabian et al . , 2008 ) . Based on the model , 62 . 9% of A ( H1N1 ) pdm09 cases were infectious ( shed ≥1 virion ) via aerosols within 24 hr of talking loudly or singing ( Figure 5—figure supplement 4A ) , and the estimate was 58 . 6% within 24 hr of talking normally and 22 . 3% within 24 hr of breathing . In comparison , 48 . 0% of COVID-19 cases shed ≥1 virion via aerosols in 24 hr of talking loudly or singing ( Figure 5—figure supplement 4C ) . Notably , only 61 . 4% of COVID-19 cases shed ≥1 virion via either droplets or aerosols in 24 hr of talking loudly or singing ( Figure 5—figure supplement 4D ) . While the human infectious dose of SARS-CoV-2 by any exposure route remains unelucidated , it must be at least one viable copy . Thus , at least 38 . 6% of COVID-19 cases were expected to present negligible risk to spread SARS-CoV-2 through either droplets or aerosols in 24 hr . The proportion of potentially infectious cases further decreased as the threshold increased: 55 . 8 , 42 . 5 and 25 . 0% of COVID-19 cases were expected to shed ≥2 , ≥10 and ≥100 virions , respectively , in 24 hr of talking loudly or singing during the infectious period . While these analyses indicated that a greater proportion of A ( H1N1 ) pdm09 cases were inherently infectious , 18 . 8% of COVID-19 cases shed virions more rapidly than those infected with A ( H1N1 ) pdm09 ( Figure 4A ) . At the 98th cp for A ( H1N1 ) pdm09 , singing expelled 4 . 38 ( 2 . 85–6 . 78 ) virions/min and breathing exhaled 2 . 15 × 10−2 ( 1 . 40 × 10−2–30 . 34×10−2 ) virions/min . Highly infectious COVID-19 cases expelled virions at rates that were up to 1–2 orders of magnitude greater than their A ( H1N1 ) pdm09 counterparts ( Figure 5H , Figure 5—figure supplement 5 ) . This study provided systematic analyses of several factors characterizing SARS-CoV-2 transmissibility . First , our results indicate that broader heterogeneity in rVL facilitates greater overdispersion for SARS-CoV-2 than A ( H1N1 ) pdm09 . They suggest that many COVID-19 cases infect no one ( Bi et al . , 2020; Endo et al . , 2020; Laxminarayan et al . , 2020 ) because they inherently present minimal transmission risk via respiratory droplets or aerosols , although behavioral and environmental factors may further influence risk . Meanwhile , highly infectious cases can shed tens to thousands of SARS-CoV-2 virions/min , especially between 1 and 5 DFSO , potentiating superspreading events . The model estimates , when corrected to copies rather than virions , align with recent clinical findings for exhalation rates of SARS-CoV-2 ( Ma et al . , 2020 ) . In comparison , a greater proportion of A ( H1N1 ) pdm09 cases are infectious but shed virions at low rates , which concurs with more uniform transmission and few superspreading events observed during the 2009 H1N1 pandemic ( Brugger and Althaus , 2020; Roberts and Nishiura , 2011 ) . Moreover , our analyses suggest that heterogeneity in rVL may be generally associated with overdispersion for viral respiratory infections . In this case , rVL distribution can serve as an early correlate for transmission patterns , including superspreading , during outbreaks of novel respiratory viruses . When considered jointly with contact-tracing studies , this provides epidemiological triangulation on k: heterogeneity in rVL indirectly estimates k via an association , whereas contact tracing empirically characterizes transmission chains to estimate k but is limited by incomplete or incorrect recall of contact events by cases . When transmission is highly overdispersed , targeted interventions may disproportionately mitigate infection ( Lee et al . , 2020 ) , with models showing that focused control efforts on the most infectious cases outperform random control policies ( Lloyd-Smith et al . , 2005 ) . Second , we analyzed SARS-CoV-2 kinetics during respiratory infection . While heterogeneity remains broad throughout the infectious period , rVL tends to peak at 1 DFSO and be elevated for 1–5 DFSO , coinciding with the period of highest attack rates observed among close contacts ( Cheng et al . , 2020 ) . These results indicate that transmission risk tends to be greatest near , and soon after , illness rather than in the presymptomatic period , which concurs with large tracing studies ( 6 . 4–12 . 6% of secondary infections from presymptomatic transmission ) ( Du et al . , 2020; Wei et al . , 2020 ) rather than early temporal models ( ~44% ) ( He et al . , 2020 ) . Furthermore , our kinetic analysis suggests that , on average , SARS-CoV-2 reaches diagnostic concentrations 1 . 54–3 . 17 days after respiratory infection ( −3 . 84 to −2 . 21 DFSO ) , assuming assay detection limits of 1–3 log10 copies/ml , respectively , for nasopharyngeal swabs immersed in 1 ml of transport media . Third , we assessed the relative infectiousness of COVID-19 subgroups . As a common symptom of COVID-19 ( Guan et al . , 2020 ) , coughing sheds considerable numbers of virions via droplets and short-range aerosols . Thus , symptomatic infections tend to be more contagious than asymptomatic ones , providing one reason as to why asymptomatic cases transmit SARS-CoV-2 at lower relative rates ( Li et al . , 2020b ) , especially in close contact ( Luo et al . , 2020 ) , despite similar rVLs and increased contact patterns . Accordingly , children ( 48–54% of symptomatic cases present with cough ) ( Lu et al . , 2020b; Team and CDC COVID-19 Response Team , 2020 ) may be less contagious than adults ( 68–80% ) ( Guan et al . , 2020; Team and CDC COVID-19 Response Team , 2020 ) based on tendencies of symptomatology rather than rVL . Conversely , coughing sheds few virions via smaller aerosols . While singing and talking loudly , highly infectious cases can shed tens to hundreds of SARS-CoV-2 virions/min via long-range and buoyant aerosols . Our study has limitations . The systematic search found a limited number of studies reporting quantitative specimen measurements from the presymptomatic period , meaning that these estimates may be sensitive to sampling bias . Although additional studies have reported semiquantitative metrics ( cycle thresholds ) , these data were excluded because they cannot be compared on an absolute scale due to batch effects ( Han et al . , 2021 ) , limiting use in compound analyses . In addition , our models considered virion partitioning during atomization to be a Poisson process , which stochastically associates partitioning with particle volume . Partitioning mechanisms associated with surface area , perhaps such as film bursting ( Bird et al . , 2010; Johnson and Morawska , 2009 ) , may enrich the quantities of virions in smaller aerosols , based on their surface area-to-volume ratio . As severe COVID-19 is associated with high , persistent SARS-CoV-2 shedding in the lower respiratory tract ( Chen et al . , 2021 ) and small particles are typically generated there ( Johnson et al . , 2011 ) , severe cases may also expel higher quantities of virions via smaller aerosols . Furthermore , this study considered population-level estimates of the infectious periods , viability proportions and profiles for respiratory particles , which omit individual or environmental variation . Studies differ in their measurements of the emission rates and size distributions of the particles expelled during respiratory activities ( Johnson et al . , 2011; Schijven et al . , 2020 ) . Their characterization methods may prompt these differences , or they may be due to individual variation , including from distinctions in respiratory capacity , especially for young children , and phonetic tendencies ( Asadi et al . , 2020 ) . Some patients shed SARS-CoV-2 with diminishing viability soon after symptom onset ( Wölfel et al . , 2020 ) , whereas others produce replication-competent virus for weeks ( van Kampen et al . , 2021 ) . The proportion of viable SARS-CoV-2 in respiratory particles , and how case characteristics or environmental factors influence it , remains under investigation ( Fears et al . , 2020; Lednicky et al . , 2020; Morris et al . , 2020 ) . Cumulatively , these sources of variation may influence the shedding model estimates , further increasing heterogeneity in individual infectiousness . Taken together , our findings provide a potential path forward for disease control . While talking , singing and coughing , our models indicate that SARS-CoV-2 is shed via droplets ( 55 . 6–59 . 4% of shed virions ) , short-range aerosols ( 30 . 1–34 . 9% ) , long-range aerosols ( 7 . 7–8 . 3% ) and buoyant aerosols ( 0 . 01–6 . 5% ) . Transmission , however , requires exposure . For direct transmission , droplets tend to be sprayed ballistically onto susceptible tissue . Aerosols can be inhaled , may penetrate more deeply into the lungs and more easily facilitate superspreading events . However , with short durations of stay in well-ventilated areas , the exposure risk for both droplets and aerosols remains correlated with proximity to infectious cases ( Liu et al . , 2017a; Prather et al . , 2020 ) . Strategies to abate infection should limit crowd numbers and duration of stay while reinforcing distancing , low-voice amplitudes and widespread mask usage; well-ventilated settings can be recognized as lower-risk venues . Coughing can shed considerable quantities of virions , while rVL tends to peak at 1 DFSO and can be high throughout the infectious period . Thus , immediate , sustained self-isolation upon illness is crucial to curb transmission from symptomatic cases . Collectively , our analyses highlight the role of cases with high rVLs in propelling the COVID-19 pandemic . While diagnosing COVID-19 , qRT-PCR can also triage contact tracing , prioritizing these patients: for nasopharyngeal swabs immersed in 1 ml of transport media , ≥7 . 14 ( 95% CI: 7 . 07–7 . 22 ) log10 copies/ml corresponds to the top 20% of COVID-19 cases for variants before August 2020 . Doing so may identify asymptomatic and presymptomatic infections more efficiently , a key step towards mitigation and elimination as the pandemic continues . We undertook a systematic review and prospectively submitted the protocol for registration on PROSPERO ( registration number , CRD42020204637 ) . Other than the title of this study , we have followed PRISMA reporting guidelines ( Moher et al . , 2009 ) . The systematic review was conducted according to Cochrane methods guidance ( Higgins et al . , 2019 ) . The search included papers that ( i ) reported positive , quantitative measurements ( copies/ml or an equivalent metric ) of SARS-CoV-2 , SARS-CoV-1 or A ( H1N1 ) pdm09 in human respiratory specimens ( endotracheal aspirate [ETA] , nasopharyngeal aspirate [NPA] , nasopharyngeal swab [NPS] , oropharyngeal swab [OPS] , posterior oropharyngeal saliva [POS] and sputum [Spu] ) from COVID-19 , SARS or A ( H1N1 ) pdm09 cases; ( ii ) reported data that could be extracted from the estimated infectious periods of SARS-CoV-2 ( defined as −3 to +10 DFSO for symptomatic cases and 0 to +10 days from the day of laboratory diagnosis for asymptomatic cases ) , SARS-CoV-1 ( defined as 0 to +20 DFSO or the equivalent asymptomatic period ) or A ( H1N1 ) pdm09 ( defined as −2 to +9 DFSO for symptomatic cases and 0 days to +9 days from the day of laboratory diagnosis for asymptomatic cases ) ; and ( iii ) reported data for two or more cases with laboratory-confirmed COVID-19 , SARS or A ( H1N1 ) pdm09 based on World Health Organization ( WHO ) case definitions . Quantitative specimen measurements were considered after RNA extraction for diagnostic sequences of SARS-CoV-2 ( Ofr1b , N , RdRp and E genes ) , SARS-CoV-1 ( Ofr1b , N and RdRp genes ) and A ( H1N1 ) pdm09 ( HA and M genes ) . Studies were excluded , in the following order , if they ( i ) studied an ineligible disease; ( ii ) had an ineligible study design , including those that were reviews of evidence ( e . g . , scoping , systematic or narrative ) , did not include primary clinical human data , reported data for less than two cases due to an increased risk of selection bias , were incomplete ( e . g . , ongoing clinical trials ) , did not report an RNA extraction step before measurement or were studies of environmental samples; ( iii ) reported an ineligible metric for specimen concentration ( e . g . , qualitative RT-PCR or cycle threshold [Ct] values without calibration included in the study ) ; ( iv ) reported quantitative measurements from an ineligible specimen type ( e . g . , blood specimens , pooled specimens or self-collected POS or Spu patient specimens in the absence of a healthcare professional ) ; ( v ) reported an ineligible sampling period ( consisted entirely of data that could not be extracted from within the infectious period ) ; or ( vi ) were duplicates of an included study ( e . g . , preprinted version of a published paper or duplicates not identified by Covidence ) . We included data from control groups receiving standard of care in interventional studies but excluded data from the intervention group . Patients in the intervention group are , by definition , systematically different from general case populations because they receive therapies not being widely used for treatment , which may influence virus concentrations . Interventional studies examining the comparative effectiveness of two or more treatments were excluded for the same reason . Studies exclusively reporting semiquantitative measurements ( e . g . , Ct values ) of specimen concentration were excluded as these measurements are sensitive to batch and instrument variation and , without proper calibration , cannot be compared on an absolute scale across studies ( Han et al . , 2021 ) . We searched , without the use of filters or language restrictions , the following sources: MEDLINE ( via Ovid , 1946 to 7 August 2020 ) , EMBASE ( via Ovid , 1974 to 7 August 2020 ) , Cochrane Central Register of Controlled Trials ( via Ovid , 1991 to 7 August 2020 ) , Web of Science Core Collection ( including Science Citation Index Expanded , 1900 to 7 August 2020; Social Sciences Citation Index , 1900 to 7 August 2020; Arts & Humanities Citation Index , 1975 to 7 August 2020; Conference Proceedings Citation Index – Science , 1990 to 7 August 2020; Conference Proceedings Citation Index – Social Sciences & Humanities , 1990 to 7 August 2020; and Emerging Sources Citation Index , 2015 to 7 August 2020 ) , as well as medRxiv and bioRxiv ( both searched through Google Scholar via the Publish or Perish program , to 7 August 2020 ) . We also gathered studies by searching through the reference lists of review articles identified by the database search , by searching through the reference lists of included articles , through expert recommendation ( by Eric J . Topol and Akiko Iwasaki on Twitter ) and by hand-searching through journals ( Nature , Nat . Med . , Science , NEJM , Lancet , Lancet Infect . Dis . , JAMA , JAMA Intern . Med . and BMJ ) . A comprehensive search was developed by a librarian , which included subject headings and keywords . The search strategy had three main concepts ( disease , specimen type and outcome ) , and each concept was combined using the appropriate Boolean operators . The search was tested against a sample set of known articles that were pre-identified . The line-by-line search strategies for all databases are included in Figure 1—source data 1 , Figure 1—source data 2 , Figure 1—source data 3 , Figure 1—source data 4 , Figure 1—source data 5 . The search results were exported from each database and uploaded to the Covidence online system for deduplication and screening . Two authors independently screened titles and abstracts , reviewed full texts , collected data and assessed risk of bias via Covidence and a hybrid critical appraisal checklist based on the Joanna Briggs Institute ( JBI ) tools for case series , analytical cross-sectional studies and prevalence studies ( Moola et al . , 2020; Munn et al . , 2019; Munn et al . , 2015 ) . To evaluate the sample size in a study , we used the following calculation: ( 1 ) n∗=z2σd2 , where n* is the sample size threshold , z is the z-score for the level of confidence ( 95% ) , σ is the standard deviation ( assumed to be 3 log10 copies/ml , one quarter of the full range of rVLs ) and d is the marginal error ( assumed to be 1 log10 copies/ml , based on the minimum detection limit for qRT-PCR across studies ) ( Johnston et al . , 2019 ) . The hybrid JBI critical appraisal checklist is shown in the Appendix . Studies were considered to have low risk of bias if they met the majority of the items , indicating that the estimates were likely to be correct for the target population . Inconsistencies were resolved by discussion and consensus . The search found 29 studies for COVID-19 ( Argyropoulos et al . , 2020; Baggio et al . , 2020; Fajnzylber et al . , 2020; Han et al . , 2020a; Han et al . , 2020b; Hung et al . , 2020; Hurst et al . , 2020; Iwasaki et al . , 2020; Kawasuji et al . , 2020; L'Huillier et al . , 2020; Lavezzo et al . , 2020; Lennon et al . , 2020; Lucas et al . , 2020; Mitjà et al . , 2020; Pan et al . , 2020; Peng et al . , 2020; Perera et al . , 2020; Shi et al . , 2020; Shrestha et al . , 2020; To et al . , 2020; van Kampen et al . , 2021; Vetter et al . , 2020; Wölfel et al . , 2020; Wyllie et al . , 2020; Xu et al . , 2020; Yonker et al . , 2020; Zhang et al . , 2020a; Zheng et al . , 2020; Zou et al . , 2020 ) , 8 studies for SARS ( Chen et al . , 2006; Cheng et al . , 2004; Chu et al . , 2005; Chu et al . , 2004; Hung et al . , 2004; Peiris et al . , 2003; Poon et al . , 2004; Poon et al . , 2003 ) and 27 studies for A ( H1N1 ) pdm09 ( Chan et al . , 2011; Cheng et al . , 2010; Cowling et al . , 2010; Duchamp et al . , 2010; Esposito et al . , 2011; Hung et al . , 2010; Ip et al . , 2016; Ito et al . , 2012; Killingley et al . , 2010; Launes et al . , 2012; Lee et al . , 2011a; Lee et al . , 2011b; Li et al . , 2010a; Li et al . , 2010b; Loeb et al . , 2012; Lu et al . , 2012; Meschi et al . , 2011; Ngaosuwankul et al . , 2010; Rath et al . , 2012; Rodrigues Guimarães Alves et al . , 2020; Suess et al . , 2010; Thai et al . , 2014; To et al . , 2010a; To et al . , 2010b; Watanabe et al . , 2011; Wu et al . , 2012; Yang et al . , 2011 ) , and data were collected from each study . For preprinted studies that were published as journal articles before the revised submission of this manuscript , we included the citation for the journal article . Descriptive statistics on quantitative specimen measurements were collected from confirmed cases directly if reported numerically or using WebPlotDigitizer 4 . 3 ( https://apps . automeris . io/wpd/ ) if reported graphically . Individual specimen measurements were collected directly if reported numerically or , when the data were clearly represented , using the tool if reported graphically . We also collected the relevant numbers of cases , types of cases , reported treatments , volumes of transport media , numbers of specimens and DFSO ( for symptomatic cases ) or day relative to initial laboratory diagnosis ( for asymptomatic cases ) on which each specimen was taken . Hospitalized cases were defined as those being tested in a hospital setting and then admitted . Non-admitted cases were defined as those being tested in a hospital setting but not admitted . Community cases were defined as those being tested in a community setting . Symptomatic , presymptomatic and asymptomatic infections were defined as in the study . Based on rare description in contributing studies , paucisymptomatic infections , when described , were included with symptomatic ones . Pediatric cases were defined as those below 18 years of age or as defined in the study . Adult cases were defined as those 18 years of age or higher or as defined in the study . In this study , viral concentrations in respiratory specimens were denoted as specimen measurements , whereas viral concentrations in the respiratory tract were denoted as rVLs . To determine rVLs , each collected quantitative specimen measurement was converted to rVL based on the dilution factor . For example , measurements from swabbed specimens ( NPS and OPS ) typically report the RNA concentration in viral transport media . Based on the expected uptake volume for swabs ( 0 . 128 ± 0 . 031 ml , mean ± SD ) ( Warnke et al . , 2014 ) or reported collection volume for expulsed fluid in the study ( e . g . , 0 . 5–1 ml ) along with the reported volume of transport media in the study ( e . g . , 1 ml ) , we calculated the dilution factor for each respiratory specimen to estimate the rVL . If the diluent volume was not reported , then the dilution factor was calculated assuming a volume of 1 ml ( NPS and OPS ) , 2 ml ( POS and ETA ) or 3 ml ( NPA ) of transport media ( Lavezzo et al . , 2020; Poon et al . , 2004; To et al . , 2020 ) . Unless dilution was reported for Spu specimens , we used the specimen measurement as the rVL ( Wölfel et al . , 2020 ) . The non-reporting of diluent volume was noted as an element increasing risk of bias in the hybrid JBI critical appraisal checklist . Specimen measurements ( based on instrumentation , calibration , procedures and reagents ) are not standardized and , as DFSO is typically based on patient recall , there is also inherent uncertainty in these values . While the above procedures ( including only quantitative measurements after extraction as an inclusion criterion , considering assay detection limits and correcting for specimen dilution ) have considered many of these factors , non-standardization remains an inherent limitation in the variability of specimen measurements . To assess the relationship between k and heterogeneity in rVL , we performed a univariate meta-regression ( log⁡k=a*SD+b , where a is the slope for association and b is the intercept ) between pooled estimates of k ( based on studies describing community transmission ) for COVID-19 ( k = 0 . 409 ) ( Adam et al . , 2020; Tariq et al . , 2020; Zhang et al . , 2020b; Laxminarayan et al . , 2020; Bi et al . , 2020; Endo et al . , 2020; Riou and Althaus , 2020 ) , SARS ( k = 0 . 165 ) ( Lloyd-Smith et al . , 2005 ) and A ( H1N1 ) pdm09 ( k = 8 . 155 ) ( Brugger and Althaus , 2020; Roberts and Nishiura , 2011 ) and the SD of the rVLs in contributing studies . Since SD was the metric , we used a fixed-effects model . For weighting in the meta-regression , we used the proportion of rVL samples from each study relative to the entire systematic dataset ( Wi=ni/ntotal ) . All calculations were performed in units of log10 copies/ml . As the meta-regression used pooled estimates of k for each infection , it assumed that there was no correlated bias to k across contributing studies . The limit of detection for qRT-PCR instruments used in the included studies did not significantly affect the analysis of heterogeneity in rVL as these limits tended to be below the values found for specimens with low virus concentrations . The meta-regression was conducted using all contributing studies and showed a weak association . Meta-regression was also conducted using studies that had low risk of bias according to the hybrid JBI critical appraisal checklist and showed a strong association . The p-value for association was obtained using the meta-regression slope t-test for a , the effect estimate . While there is intrinsic measurement error in virus quantitation , based on the systematic review protocol and study design ( as described above ) , this error should similarly increase heterogeneity in rVL for each virus , and the difference in heterogeneity in rVL between viruses should arise from the viruses . Based on the search design and composition of contributing studies , the meta-analysis overall estimates were the expected SARS-CoV-2 , SARS-CoV-1 and A ( H1N1 ) pdm09 rVL when encountering a COVID-19 , SARS or A ( H1N1 ) pdm09 case , respectively , during their infectious period . Pooled estimates and 95% CIs for the expected rVL of each virus across their infectious period were calculated using a random-effects meta-analysis ( DerSimonian and Laird method ) . For studies reporting summary statistics in medians and interquartile or total ranges , we derived estimates of the mean and variance and calculated the 95% CIs ( Wan et al . , 2014 ) . All calculations were performed in units of log10 copies/ml . Between-study heterogeneity in meta-analysis was assessed using Cochran’s Q test and the I2 and τ2 statistics . If significant between-study heterogeneity in meta-analysis was encountered , sensitivity analysis based on the risk of bias of contributing studies was performed . The meta-analyses were conducted using STATA 14 . 2 ( StataCorp LLC , College Station , TX , USA ) . The overall estimate for each subgroup was the expected rVL when encountering a case of that subgroup during the infectious period . Studies reporting data exclusively from a subgroup of interest were directly included in the analysis after rVL estimations . For studies in which data for these subgroups constituted only part of its dataset , rVLs from the subgroup were extracted to calculate the mean , variance and 95% CIs . Random-effects meta-analysis was performed as described above . For meta-analyses of pediatric and asymptomatic COVID-19 cases , contributing studies had low risk of bias , and no risk-of-bias sensitivity analyses were performed for these subgroups . We pooled the entirety of individual sample data in the systematic dataset by disease , COVID-19 subgroups and DFSO . For analyses of SARS-CoV-2 dynamics across disease course , we included estimated rVLs from negative qRT-PCR measurements of respiratory specimens for cases that had previously been quantitatively confirmed to have COVID-19 . These rVLs were estimated based on the reported assay detection limit in the respective study . Probability plots and modified Kolmogorov–Smirnov tests used the Blom scoring method and were used to determine the suitability of normal , lognormal , gamma and Weibull distributions to describe the distribution of rVLs for SARS-CoV-2 , SARS-CoV-1 and A ( H1N1 ) pdm09 . For each virus , the data best conformed to Weibull distributions , which is described by the probability density function ( 2 ) f ( υ ) =αβ ( υβ ) α−1e− ( υ/β ) α , where α is the shape factor , β is the scale factor and υ is rVL ( υ≥0 log10 copies/ml ) . Weibull distributions were fitted on the entirety of collected individual sample data for the respective category . Since individual specimen measurements could not be collected from all studies , there was a small bias on the mean estimate for each fitted distribution . Thus , for the curves shown in Figure 4B , C , the mean of the Weibull distributions summarized in Figure 4—figure supplement 2 was adjusted to be the subgroup meta-analysis estimate for correction; the SD and distribution around that mean remained consistent . For each Weibull distribution , the value of the rVL at the x th percentile was determined using the quantile function , ( 3 ) υx=β[−ln⁡ ( 1−x ) ]1/α . For cp curves , we used Equation ( 3 ) to determine rVLs from the 1st cp to the 99th cp ( step size , 1% ) . Curve fitting to Equation ( 2 ) and calculation of Equation ( 3 ) and its 95% CI was performed using the Distribution Fitter application in Matlab R2019b ( MathWorks , Inc , Natick , MA , USA ) . To model SARS-CoV-2 kinetics during respiratory infection , we used a mechanistic epithelial cell-limited model for the respiratory tract ( Baccam et al . , 2006 ) , based on the system of differential equations: ( 4 ) dTdt=−βTV ( 5 ) dIdt=βTV−δI ( 6 ) dVdt=pI−cV , where T is the number of uninfected target cells , I is the number of productively infected cells , V is the rVL , β is the infection rate constant , p is the rate at which airway epithelial cells shed virus to the extracellular fluid , c is the clearance rate of virus and δ is the clearance rate of productively infected cells . Using these parameters , the viral half-life in the respiratory tract ( t1/2=ln⁡2/c ) and the half-life of productively infected cells ( t1/2=ln⁡2/δ ) could be estimated . Moreover , the cellular basic reproductive number ( the expected number of secondary infected cells from a single productively infected cell placed in a population of susceptible cells ) was calculated by ( 7 ) R0 , c=pβT0cδ , For initial parameterization , Equations ( 4 ) – ( 6 ) were simplified according to a quasi-steady state approximation ( Ikeda et al . , 2016 ) to ( 8 ) dTdt=−βTV ( 9 ) dVdt=rTV−δV , where r=pβ/c , for a form with greater numerical stability . The system of differential equations was fitted on the mean estimates of SARS-CoV-2 rVL between -2 and 10 DFSO using the entirety of individual sample data in units of copies/ml . Numerical analysis was implemented using the Fit ODE app in OriginPro 2019b ( OriginLab Corporation , Northampton , MA , USA ) via the Runge–Kutta method and initial parameters V0 , I0 and T0 of 4 copies/ml , 0 cells and 5 × 107 cells , respectively , for the range –5 to 10 DFSO . The analysis was first performed with Equations ( 8 ) and ( 9 ) . These output parameters were then used to initialize final analysis using Equations ( 4 ) – ( 6 ) , where the estimates for β and δ were input as fixed and variable parameters , respectively . The fitted line and its coefficient of determination ( r2 ) were presented . The estimated half-life of SARS-CoV-2 RNA has a skewed 95% CI ( Figure 4—figure supplement 4 ) . As c is in the denominator of the equation for half-life ( t1/2=ln⁡2/c ) , t1/2 is sensitive to c below 1 , which is the case for its lower 95% CI ( Figure 4—figure supplement 4 ) and the source of the skew . To estimate the average incubation period , we extrapolated the kinetic model to 0 log10 copies/ml pre-symptom onset . To estimate the average duration of shedding , we extrapolated the model to 0 log10 copies/ml post-symptom onset . Unlike in experimental studies , this estimate for duration of shedding was not defined by assay detection limits . To estimate the average DFSO on which SARS-CoV-2 concentration reached diagnostic levels , we extrapolated the model pre-symptom onset to the equivalent of 1 and 3 log10 copies/ml ( chosen as example assay detection limits ) in specimen concentration for NPSs immersed in 1 ml of transport media , as described by the dilution factor estimation above . The average time from respiratory infection to reach diagnostic levels was then calculated by subtracting these values from the estimated average incubation period . The extrapolated time for SARS-CoV-2 to reach diagnostic concentrations in the respiratory tract should be validated in tracing studies , in which contacts are prospectively subjected to daily sampling . To calculate an unbiased estimator for viral partitioning ( the expected number of viable copies in an expelled particle at a given size ) , we multiplied rVLs with the volume equation for spherical particles during atomization and the estimated viability proportion , according to the following equation: ( 10 ) λ=πρvpγυ6d3 , where λ is the expectation value , ρ is the material density of the respiratory particle ( 997 kg/m3 ) , vp is the volumetric conversion factor ( 1 ml/g ) , γ is the viability proportion , υ is the rVL and d is the hydrated diameter of the particle during atomization . The model assumed γ was 0 . 1% as a population-level estimate . For influenza , approximately 0 . 1% of copies in particles expelled from the respiratory tract represent viable virus ( Yan et al . , 2018 ) , which is equivalent to one viable copy in 3 log10 copies/ml for rVL or , after dilution in transport media , roughly one in 4 log10 copies/ml for specimen concentration . Respiratory specimens taken from influenza cases show positive cultures for specimen concentrations down to 4 log10 copies/ml ( Lau et al . , 2010 ) . Likewise , for COVID-19 cases , recent reports also show culture-positive respiratory specimens with SARS-CoV-2 concentrations down to 4 log10 copies/ml ( Wölfel et al . , 2020 ) , including from pediatric ( L'Huillier et al . , 2020 ) and asymptomatic ( Arons et al . , 2020 ) cases . Moreover , replication-competent SARS-CoV-2 has been found in respiratory specimens taken throughout the respiratory tract ( mouth , nasopharynx , oropharynx and lower respiratory tract ) ( Jeong et al . , 2020; Wölfel et al . , 2020 ) . Taken together , these considerations suggested that the assumption for viability proportion ( 0 . 1% ) was suitable to model the likelihood of respiratory particles containing viable SARS-CoV-2 . In accordance with the discussion above , the model did not differentiate this population-level viability estimate based on age , symptomatology or sites of atomization . Based on the relative relationship between the residence time of expelled particles before assessment ( ~5 s ) ( Yan et al . , 2018 ) , we took the viability proportion to be for equilibrated particles . Likelihood profiles were determined using Poisson statistics , as described by the probability mass function ( 11 ) P ( X=k ) =λke−λk ! , where k is the number of virions partitioned within the particle . For λ , 95% CIs were determined using the variance of its rVL estimate . To determine 95% CIs for likelihood profiles from the probability mass function , we used the delta method , which specifies ( 12 ) Var ( g ( θ ) ) ≈σ2g˙ ( θ ) ′Dg˙ ( θ ) , where σ2D is the covariance matrix of θ and g˙θ is the gradient of gθ . For the univariate Poisson distribution , σ2D=λ and ( 13 ) g˙ ( θ ) =λk−1e−λk ! ( k−λ ) . Distributions from the literature were used to determine the rate profiles of particles expelled during respiratory activities . For breathing , talking and coughing , we used data from Johnson et al . , 2011 . For singing , we used data from Morawska et al . , 2009 for smaller aerosols ( da < 20 μm ) and used the profiles from talking for larger aerosols and droplets based on the oral cavity mechanism from Johnson et al . , 2011 . Rate profiles ( particles/min or particles/cough ) were calculated based on the corrected normalized concentration ( dCn/dlogDp , in units of particles/cm3 ) at each discrete particle size , normalization ( 32 size channels per decade ) for the aerodynamic particle sizer used , unit conversion ( cm3 to l ) and the sample flow rate ( 1 l/min ) . For coughing , the calculation assumed that participants coughed 10 times in the 30-s sampling interval . To determine the corrected normalized concentrations for breathing , we used a particle dilution factor of 4 and evaporation factor of 0 . 5 , consistent with the other respiratory activities in Johnson et al . , 2011 . Breathing was taken to expel negligible quantities of larger respiratory particles based on the bronchiolar fluid film burst mechanism ( Johnson et al . , 2011 ) . To account for intermittent breathing while talking and singing , the rate profiles for these activities included the contribution of aerosols expelled by breathing . We compared these rate profiles with those collected from talking loudly and talking quietly from Asadi et al . , 2020 . In our models , we took the diameter of dehydrated respiratory particles to be 0 . 3 times the initial size when atomized in the respiratory tract ( Johnson et al . , 2011; Lieber et al . , 2021; Liu et al . , 2017b ) . Equilibrium aerodynamic diameter was calculated by da=dpρ/ρ01/2 , where dp is the dehydrated diameter , ρ is the material density of the respiratory particle and ρ0 is the reference material density ( 1 g/cm3 ) . Curves based on discrete particle measurements were connected using the nonparametric Akima spline function . To model the respiratory shedding rate across particle size , rVL estimates and the hydrated diameters of particles expelled by a respiratory activity were input into Equation ( 10 ) , and the output was then multiplied by the rate profile of the activity ( talking , singing , breathing or coughing ) . To assess the relative contribution of aerosols and droplets to mediating respiratory viral shedding for a given respiratory activity , we calculated the proportion of the cumulative hydrated volumetric rate contributed by buoyant aerosols ( da ≤ 10 μm ) , long-range aerosols ( 10 μm < da ≤ 50 μm ) , short-range aerosols ( 50 μm < da ≤ 100 μm ) and droplets ( da > 10 μm ) for that respiratory activity . Since the Poisson mean was proportional to cumulative volumetric rate , this estimate of the relative contribution of aerosols and droplets to respiratory viral shedding was consistent among viruses and cps in the model . To determine the total respiratory shedding rate for a given respiratory activity across cp , we determined the cumulative hydrated volumetric rate ( by summing the hydrated volumetric rates across particle sizes for that respiratory activity ) of particle atomization and input it into Equation ( 10 ) . Using rVLs and their variances as determined by the Weibull quantile functions , we then calculated the Poisson means and their 95% CIs at the different cps . To assess the influence of heterogeneity in rVL on individual infectiousness , we first considered transmission of A ( H1N1 ) pdm09 via aerosols ( Cowling et al . , 2013 ) . The 50% human infectious dose ( HID50 ) of aerosolized A ( H1N1 ) pdm09 was taken to be 1–3 virions ( Fabian et al . , 2008 ) . To determine the expected time required for a A ( H1N1 ) pdm09 case to shed one virion via aerosols , we took the reciprocal of the Poisson means and their 95% CIs at the different cps of the estimated shedding rates . The expected time required for a COVID-19 case to shed one virion via aerosols or one virion via droplets or aerosols was determined in a same manner . The systematic dataset and model outputs from this study were uploaded to Zenodo ( https://zenodo . org/record/4658971 ) . The code generated during this study is available at GitHub ( https://github . com/paulzchen/sars2-heterogeneity; Chen , 2020; copy archived at swh:1:rev:06649ccfb6e92918b439332314ebf330abfa3d16 ) . The systematic review protocol was prospectively registered on PROSPERO ( registration number , CRD42020204637 ) .	To understand how viruses spread scientists look at two things . One is – on average – how many other people each infected person spreads the virus to . The other is how much variability there is in the number of people each person with the virus infects . Some viruses like the 2009 influenza H1N1 , a new strain of influenza that caused a pandemic beginning in 2009 , spread pretty uniformly , with many people with the virus infecting around two other people . Other viruses like SARS-CoV-2 , the one that causes COVID-19 , are more variable . About 10 to 20% of people with COVID-19 cause 80% of subsequent infections – which may lead to so-called superspreading events – while 60-75% of people with COVID-19 infect no one else . Learning more about these differences can help public health officials create better ways to curb the spread of the virus . Chen et al . show that differences in the concentration of virus particles in the respiratory tract may help to explain why superspreaders play such a big role in transmitting SARS-CoV-2 , but not the 2009 influenza H1N1 virus . Chen et al . reviewed and extracted data from studies that have collected how much virus is present in people infected with either SARS-CoV-2 , a similar virus called SARS-CoV-1 that caused the SARS outbreak in 2003 , or with 2009 influenza H1N1 . Chen et al . found that as the variability in the concentration of the virus in the airways increased , so did the variability in the number of people each person with the virus infects . Chen et al . further used mathematical models to estimate how many virus particles individuals with each infection would expel via droplets or aerosols , based on the differences in virus concentrations from their analyses . The models showed that most people with COVID-19 infect no one because they expel little – if any – infectious SARS-CoV-2 when they talk , breathe , sing or cough . Highly infectious individuals on the other hand have high concentrations of the virus in their airways , particularly the first few days after developing symptoms , and can expel tens to thousands of infectious virus particles per minute . By contrast , a greater proportion of people with 2009 influenza H1N1 were potentially infectious but tended to expel relatively little infectious virus when the talk , sing , breathe or cough . These results help explain why superspreaders play such a key role in the ongoing pandemic . This information suggests that to stop this virus from spreading it is important to limit crowd sizes , shorten the duration of visits or gatherings , maintain social distancing , talk in low volumes around others , wear masks , and hold gatherings in well-ventilated settings . In addition , contact tracing can prioritize the contacts of people with high concentrations of virus in their airways .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health", "microbiology", "and", "infectious", "disease" ]	2021	Heterogeneity in transmissibility and shedding SARS-CoV-2 via droplets and aerosols
Self-assembly of ESCRT-III complex is a critical step in all ESCRT-dependent events . ESCRT-III hetero-polymers adopt variable architectures , but the mechanisms of inter-subunit recognition in these hetero-polymers to create flexible architectures remain unclear . We demonstrate in vivo and in vitro that the Saccharomyces cerevisiae ESCRT-III subunit Snf7 uses a conserved acidic helix to recruit its partner Vps24 . Charge-inversion mutations in this helix inhibit Snf7-Vps24 lateral interactions in the polymer , while rebalancing the charges rescues the functional defects . These data suggest that Snf7-Vps24 assembly occurs through electrostatic interactions on one surface , rather than through residue-to-residue specificity . We propose a model in which these cooperative electrostatic interactions in the polymer propagate to allow for specific inter-subunit recognition , while sliding of laterally interacting polymers enable changes in architecture at distinct stages of vesicle biogenesis . Our data suggest a mechanism by which interaction specificity and polymer flexibility can be coupled in membrane-remodeling heteropolymeric assemblies . Eukaryotic organelles that are demarcated by membranes undergo continuous remodeling to maintain their integrity and function . Different evolutionarily conserved heteropolymeric machines and scaffolds including ESCRTs ( endosomal sorting complexes required for transport ) , dynamin , clathrin , COP coatmers , BAR-domain containing proteins , etc . control the remodeling at different organelles ( Zimmerberg and Kozlov , 2006 ) . This polymerization-mediated membrane remodeling in eukaryotes is important for various cellular events such as trafficking , cell-division and migration . In bacteria , structural homologs of actin and tubulin ( that include FtsZ and MreB proteins ) also form heteropolymers , which deform and remodel membranes , leading to cell shape maintenance and division ( Eun et al . , 2015 ) . In most cases these proteins are recruited to the membrane from the cytosol as monomers , polymerization then occurs on the 2D membrane , which generates the mechanical force to bend membranes . Given the functional importance of polymerization , elucidating mechanisms of how these various machines recognize their partners to form heteropolymers , and how these polymers adapt to physical and mechanical changes in the cellular environment is critical to our overall understanding of how cells control membrane homeostasis . ESCRTs drive an ever-growing list of cellular processes . These include biogenesis of multivesicular bodies ( MVBs ) ( Katzmann et al . , 2001; Odorizzi et al . , 1998 ) , viral budding ( Garrus et al . , 2001 ) , cytokinesis ( Carlton and Martin-Serrano , 2007 ) , nuclear envelope reformation ( Olmos et al . , 2015; Vietri et al . , 2015 ) , plasma membrane repair ( Jimenez et al . , 2014 ) , surveillance of defective nuclear pore complexes ( Webster et al . , 2014 ) , lysosomal protein degradation ( Zhu et al . , 2017 ) , endolysosomal repair ( Skowyra et al . , 2018 ) , and so on Hurley ( 2015 ) . This machinery consists of five evolutionarily conserved complexes , ESCRT-0 , I , II , III and the AAA ATPase Vps4 . ESCRT-0 , I and II are recruited to the membrane through interactions with ubiquitinated proteins , while ESCRT-0 and ESCRT-II also bind phosphatidyl-inositol-3-phosphate ( PI3P ) at membranes ( Robinson et al . , 1988; Rothman et al . , 1989; Raymond et al . , 1992; Katzmann et al . , 2001; Babst et al . , 1997; Katzmann et al . , 2003; Babst et al . , 2002a; Babst et al . , 2002b ) . ESCRT-II recruits ESCRT-III , also nucleating ESCRT-III polymerization . Polymerization of ESCRT-III is dynamically rearranged and directed towards disassembly via the recruitment of the AAA ATPase Vps4 ( Mierzwa et al . , 2017; Adell et al . , 2017 ) . The requirement of early ESCRTs ( 0 , I and II ) varies between the different membranes . For example , in viral budding processes , the ESCRT-0 complex is not required , while in cytokinetic abscission neither ESCRT-0 nor ESCRT-II are required . ESCRT-III , however , is ubiquitously required in all ESCRT-dependent processes . This membrane-modifying property of ESCRT-III is assumed to be a direct consequence of its self-assembly into meso-scale polymers . The requirement of ESCRT-III function in all ESCRT-dependent processes emphasizes the importance of understanding the molecular details of ESCRT-III co-assembly . ESCRT-III proteins are ~26 KDa charged proteins that assemble into a polymer at membranes . In yeast , the ‘core’ ESCRT-III machinery consists of Vps20 , Snf7 , Vps24 and Vps2 , as the deletion of any of these four genes leads to a strong defect in the MVB pathway ( Babst et al . , 2002b ) . Other ESCRT-III proteins consist of Ist1 , Did2 , Vps60 and Chm7 , which have accessory roles in the MVB pathway , but are essential in mammalian cytokinesis ( Ist1/Did2 ) ( Bajorek et al . , 2009a ) or nuclear envelope sealing ( Chm7 ) ( Webster et al . , 2016 ) . All ESCRT-III proteins consist of an N-terminal helical bundle , which is predominantly basic , and a C-terminal flexible and acidic region ( Muzioł et al . , 2006; Bajorek et al . , 2009b ) . The N-terminal helical bundle is known to be important for ESCRT-III co-assembly ( Henne et al . , 2012 ) , while the C-terminus recruits the AAA ATPase Vps4 ( Obita et al . , 2007 ) . Different architectures of ESCRT-III polymers have been visualized with electron microscopy – flat sheets , rings , spirals and helices , owing to the flexible nature of the polymers ( Henne et al . , 2012; Hanson et al . , 2008; Chiaruttini et al . , 2015; Shen et al . , 2014 ) . The most abundant of the ESCRT-III proteins , Snf7 , forms 2D spirals . Combination of Snf7 with downstream proteins Vps24 and Vps2 changes the architecture of these spirals to form three dimensional helical polymers ( Henne et al . , 2012 ) . The mammalian orthologues of Vps24 and Vps2 ( named CHMP3 and CHMP2 ) co-assemble to form helical structures ( Lata et al . , 2008 ) , whereas Vps24 has been visualized to form linear polymers ( Ghazi-Tabatabai et al . , 2008 ) . Despite our understanding of the individual homo-polymerization mechanism ( Snf7-Snf7 ) , the mechanisms of how ESCRT-III proteins recognize one another to create heteropolymers remain unclear . While recruitment of downstream ESCRT-III proteins Vps24 and Vps2 by Snf7 is established , what roles these partners of Snf7 play in regulating the polymers of Snf7 remain unclear . How these proteins change the architecture of the flat 2D spirals of Snf7 to 3D helical structures also remains undefined . Furthermore , in analyzing ESCRT-III polymerization , most studies only focus on either biochemical/structural approaches , or only cell biological approaches , with a limited number of studies combining both approaches to understand ESCRT-III function in vivo . Here , we apply a combination of in vivo and in vitro approaches to shed light on the hetero-polymerization mechanisms of ESCRT-III proteins Snf7 and Vps24 . In particular , we utilize powerful directed evolution-based approaches in yeast to select for Snf7 mutants that rescue polymerization defects . These approaches allow us to describe in vivo mechanisms that ESCRT-III utilizes to co-assemble in the MVB pathway and provide guidance to understand existing literature on ESCRT-III filament structures . We previously reported the crystal structure of Snf7 ( Tang et al . , 2015 ) , which depicted the ESCRT-III subunit in an ‘active , ’ polymerization-capable conformation . This structure provided important clues regarding how the polymerization of Snf7 is regulated . Here , we identify and characterize a motif in Snf7 critical for recruitment of its downstream partner Vps24 . Mutagenesis analyses suggest that this motif is involved in Snf7/ESCRT-III lateral interaction in vitro . The acidic nature on its solvent-exposed surface , rather than a consensus sequence , is important for the function of this helix , suggesting an electrostatic mechanism that Snf7 , and most likely all ESCRT-III subunits , use to recognize their partners . We hypothesize that ESCRT-III proteins utilize these lateral charge interactions with weak residue-to-residue specificity to slide on the filaments , enabling them to form spiraling polymers of different diameters and architectures . Our structural analysis of Snf7 allowed us to specify helices 1 , 2 and 3 as the core polymeric interface that drives longitudinal polymerization . Structure/function studies of the Drosophila Snf7 ( Shrub ) confirmed the existence of this interface in the polymer , showing nearly identical packing arrangement in the polymer ( Tang et al . , 2015; McMillan et al . , 2016 ) , despite some differences in the side-chain residues . Additionally , the structure of the similarly organized CHMP1B in its helical assembly with IST1 suggests that the same core interface drives ESCRT-III polymerization with an evolutionarily conserved mechanism of ESCRT-III assembly ( McCullough et al . , 2015; McCullough et al . , 2018; Talledge et al . , 2019 ) . Helix-4 of Snf7 lies at the periphery of this core longitudinal interface ( Figure 1A–1B ) and stretches from residue ~120 to~150 . In the crystal structure of Snf7core , which included residues 12–150 , we observed that residues D124 to E138 are structured , while the rest of the amino acids are not visible . Helix-4 is mostly acidic in nature , with the acidic residues falling on one interface ( Figure 1A , Figure 1—figure supplement 1A ) . We had previously observed that deleting helix-4 of Snf7 resulted in a defect in recruiting Vps24-GFP to endosomes , although we did not have a mechanistic understanding behind this phenotype ( Henne et al . , 2012 ) . Interestingly , the previously solved cryo-EM structure of CHMP1B ( which is the mammalian homolog of Did2 ) depicted extensive electrostatic contacts made by CHMP1B’s acidic residues of helices 4 and 5 with the basic helix-1 of IST1 ( McCullough et al . , 2015 ) . The overall acidity of Snf7 helix-4 is conserved in the Snf7 orthologues of Saccharomyces cerevisiae , Homo sapiens , Mus musculus , Xenopus laevis , Drosophila melanogaster , Caenorhabditis elegans , and Schizosaccharomyces pombe ( Figure 1—figure supplement 1B ) , which suggests an important role of the electrostatics mediated by these charged residues . Based on these analyses and the insights provided by the recent atomic structures , we hypothesized that the acidic interface of helix-4 may be important for Snf7’s assembly with its other ESCRT-III partners . We comprehensively analyzed this acidic interface of helix-4 using endocytosis assays in Saccharomyces cerevisiae . One of the well-developed assays utilizes canavanine , a toxic analog of arginine . ESCRT mutant strains are sensitive in their growth in the presence of the drug , as the ESCRT mutants are incapable of downregulating the canavanine transporter Can1 from the plasma membrane ( Lin et al . , 2008 ) ( Figure 1—figure supplement 2A ) . Using this canavanine-sensitivity assay , we found that mutations in the acidic residues of helix-4 have strong sensitivity to canavanine ( Figure 1C , Figure 1—figure supplement 3A ) . Mutations in residues D131 and D127 show the strongest phenotype ( Figure 1C ) . Charge-inversion single mutations ( D to K ) or double-alanine ( D127A D131A ) mutations showed strong effects in cargo sorting , while single mutations to alanine do not show any defects ( Figure 1C ) . In an orthogonal ESCRT dependent cargo-sorting assay , these helix-4 mutants show a similar behavior . We have routinely used the sorting and degradation of the methionine transporter Mup1 fused to the pH sensitive pHluorin to analyze ESCRT mutants ( Henne et al . , 2012; Tang et al . , 2016 ) ( Figure 1—figure supplement 2B ) . The defect of the helix-4 mutants of Snf7 is consistently observed in the methionine-dependent sorting of Mup1-pHluorin ( Figure 1C and Figure 1—figure supplement 3C ) . In our assays , we observed that mutations in residues D127 , D131 , E138 and E142 gave the strongest effects ( Figure 1—figure supplement 3C ) . Importantly , D127 , D131 and E138 lie on the same surface on the structure of helix-4 ( Figure 1A , Figure 1—figure supplement 3C ) . The effects of the helix-4 mutants that we have observed are not due to the instability of the mutated proteins . In in vitro assays using the mutation D131K , the protein behaves similarly to the non-mutated version . We have previously analyzed the activating mutation in Snf7 ( R52E ) , as the mutation lowers the critical concentration for polymerization ( Henne et al . , 2012 ) . This mutation allows us to observe polymers of Snf7 on electron microscopy grids and on lipid monolayers . The helix-4 mutant D131K ( with R52E ) is able to form polymers of Snf7 ( Figure 1—figure supplement 4A ) . One important exception here was that the D131K mutant showed defects in its lateral interaction , as it preferentially makes thinner filaments ( Figure 1—figure supplement 4A -inset ) . A similar fraction ( ~50% ) of Snf7 bound to liposomes with both Snf7R52E and Snf7 R52E D131K proteins ( Figure 1—figure supplement 4B ) . Furthermore , both proteins showed a similar size-exclusion chromatogram ( Figure 1—figure supplement 4C ) . These analyses suggest that the helix-4 mutation does not adversely affect folding of Snf7 . In the crystal lattice of Snf7core , the helix-4 residue D131 makes electrostatic contacts with helix-1 residues K21 and K25 in trans on a laterally interacting polymer strand ( Figure 2—figure supplement 1A ) . In vivo , mutating K21 and K25 individually do not have any defect in cargo sorting , while the double mutation K21E K25E has a mild defect ( ~60% MVB function ( Figure 2—figure supplement 1B–C ) . Although these data do not rule out the possibility that this interface is involved in Snf7-Snf7 lateral interaction , further analyses below suggest that this is an interface mimicking Snf7’s interaction with its partner Vps24 . The requirement of Snf7 to recruit Vps24 at endosomes has been well documented in the literature ( Teis et al . , 2008 ) , and the in vitro binding of these proteins has also been documented ( Mierzwa et al . , 2017; Henne et al . , 2012 ) . However , the molecular description of these interactions is incomplete . We provide several lines of evidence that Snf7 uses helix-4 to recruit Vps24 . First , we found that helices 1–4 of Snf7 can recruit VPS24-GFP to endosomes . However , deleting helix-4 or making specific mutations in helix-4 ( D131K ) reduced the recruitment ( Figure 2A , Figure 2—figure supplement 2A ) . In subcellular fractionation experiments , the helix-4 mutant snf7D131K recruited ~5 fold lower amount of Vps24 than the wild-type protein in the P13 ( endosome-enriched ) fraction ( Figure 2—figure supplement 2B ) . Similarly , in co-immunoprecipitation experiments , the amount of Snf7D131K bound to Vps24 was reduced by >10 fold compared to wild-type ( Figure 2B ) . Furthermore , in in vitro lipid monolayer assays , the Snf7R52E D131K mutant was unable to form 3D helical spirals with Vps24 and Vps2 ( Figure 2C ) . To further analyze ESCRT-III polymerization in vivo , we used rate-zonal velocity gradient assays . In these assays , we found that the Snf7 helix-4 mutant phenocopies the vps24Δ strain . We previously observed that in cells lacking Vps24 , Snf7 sediments to fractions containing higher percent glycerol , indicating formation of higher order polymers of Snf7 ( Teis et al . , 2008 ) . This phenomenon most likely occurs because of the inability of vps24Δ to recruit the AAA-ATPase Vps4 to endosomes , eliminating Vps4-mediated disassembly of ESCRT-III polymers . In the helix-4 mutant snf7D131K strain , we observed a strikingly similar phenotype , as the Snf7D131K protein sediments towards the bottom of the gradient , indicating that snf7D131K VPS24 phenocopies SNF7 vps24Δ ( Figure 2D ) . Altogether , these experiments provide strong evidence that the Snf7 helix-4 mutant is defective in recruiting Vps24 . To identify parameters that can overcome the defect of Snf7 helix-4 mutations and therefore to characterize the interacting surface of Snf7 on Vps24 , we performed unbiased mutagenesis of Vps24 to look for suppressors of snf7D131K . The snf7D131K mutant strain is canavanine sensitive , as ESCRT mutants are defective in endocytosis of Can1 ( Figure 1—figure supplement 2A ) . Taking advantage of the canavanine sensitivity of snf7D131K , we used random mutagenesis to select mutants of vps24 that could rescue the defect of the Snf7 helix-4 mutation . This selection approach identified several mutations on the helix-1 surface of Vps24 ( Q16E , K26E and K33E of Vps24 ) that were canavanine resistant in the snf7D131K background . Of these , Q16E mutation on Vps24 gave the strongest effect in suppressing the defect of snf7D131K , as ~70% of Mup1-pHluorin is sorted with the vps24Q16E mutation in the background of snf7D131K ( Figure 3A , Figure 3—figure supplement 1A–1C ) . Interestingly , Vps24 mutants Q16E and R19E additionally also suppress the defects of other Snf7 mutations D127K and D142K ( Figure 3A ) . Importantly , the expression levels of the Vps24 mutant proteins were similar to wild-type Vps24 ( Figure 3—figure supplement 1B ) . To study whether the residues of helix-1 of Vps24 lie in the vicinity of helix-4 of Snf7 , we used ex vivo crosslinking approaches to crosslink these two proteins . Consistent with the idea that Snf7 helix-4 interacts with the helix-1 surface of Vps24 , the sulfhydryl crosslinker BMOE crosslinks Snf7D131C to cysteine mutations in helix-1 of Vps24 ( Vps24Q16C , Vps24R19C and Vps24K26C ) ( Figure 3B , Figure 3—figure supplement 2B ) . In these assays , we also see Snf7D131C:Snf7D131C crosslinks probably due to the fact that Snf7 spirals are flexible enough to bend and form crosslinks through the longitudinal surface ( Figure 3—figure supplement 2A–B ) . The amount of crosslinked Vps24 decreases with additional mutations in the vicinity of D131 in Snf7 that reduce Vps24 binding ( adding the D127K and E142K mutations - Snf7 D131C D127K and Snf7 D131C D127K E142K ) ( Figure 3—figure supplement 2A ) . With the presence of these D127K and E142K mutants , we see stronger Snf7-Snf7 crosslinking , consistent with the observation that in the absence of Vps24 , Snf7 forms higher amounts of polymers ( Figure 2D ) . Comparison of our data with the cryo-EM structure of CHMP1B and IST1 ( McCullough et al . , 2015 ) provides us with an additional support of the model of Snf7 helix-4 contacting Vps24 helix-1 . In the solved structure ( Figure 2—figure supplement 3A–C ) , the helix-4 residues E106 , S109 and D113 of CHMP1B are in close proximity to form salt bridges with IST1 helix-1 residues N14 and R16 . The polar and acidic residues of helix-4 in CHMP1B and in Snf7 are also conserved ( Figure 2—figure supplement 3C ) . Additionally , the IST1 structure is similar to the homology model of Vps24 ( using the hVps24 crystal structure as the template ) ( Figure 2—figure supplement 3D ) implying that the helix-1 residues probably are similarly positioned in a copolymer . Overall , the specific point mutations on Vps24 that suppress the defect of Snf7 helix-4 mutations also provide additional evidence that helix-4 of Snf7 is involved in recruiting its partner Vps24 . Helix-4 on Snf7 is located at the periphery of the core polymer , indicating that Vps24 binds at the sides of the core polymer of Snf7 . The location of the suppressor mutations on Vps24 also points to helix-1 of Vps24 as the interaction surface . These data are supported by the crosslinking assay , and the comparison with an orthogonal ESCRT-III assembly structure of CHMP1B and IST1 ( McCullough et al . , 2015 ) . The suppression of separate residues D127 and E142 in Snf7 that are >21 Å apart ( D127 and E138 are 21 Å apart and residues beyond E138 are not visible in the structure ) by Vps24 helix-1 residues indicated an interacting surface that possesses weaker specificity . Through mutagenesis approaches , we have previously been able to identify several Snf7 activating mutations that allowed us to decipher how Snf7 is activated by upstream nucleating factors ( Tang et al . , 2016 ) . We hypothesized that we would be able to similarly identify parameters in Snf7 that would enhance the affinity of Snf7 to Vps24 , using the snf7D131K mutant that has defects in Vps24 binding . Therefore , we additionally performed error-prone mutagenesis with Snf7 to look for intragenic suppressors of D131K . We mutagenized snf7D131K on a plasmid and selected for mutants that rescue the canavanine-sensitivity phenotype . Through this genetic selection approach , we found that several additional mutations in helix-4 of Snf7 that balanced the acidity of the helix rescued the defect of the D131K mutation . As described above , charge-inversion mutations on the helix-1 surface of Vps24 rescued the defect of the Snf7 helix-4 mutations . Consistent with this , the following pieces of data suggest that rescuing the acidic defect of the D131K mutation rescues functional phenotype of the Snf7 helix-4 mutations . We found that charge-inversion double mutations D131K R134D and D131K R149D in the helix-4 region of Snf7 completely sorts Mup1-pHluorin ( ~100% ) , unlike the defective single mutant D131K ( Figure 4A–4B , Figure 4—figure supplement 1C ) . Additionally , the double mutant D127A D131A was also rescued by the additional inclusion of the mutations R134D or R149D ( Figure 4—figure supplement 1A and C ) . Upon further analysis , to our surprise , mutating the basic residue R134 or R149 to alanine also rescued the defect of the D131K or the double-mutant D127A D131A ( Figure 4—figure supplement 1A ) . These observations hold true for both model cargos Can1 and Mup1 . Our data suggest that maintaining the acidic nature of one surface of helix-4 that is important for Vps24 recruitment rescues the defective phenotype . In the structure of Snf7 helix-4 , K125 points away from the acidic region ( Figure 4A ) . Mutating this Lys ( K125 ) to Glu in snf7D131K only partially suppressed the defect ( Figure 4—figure supplement 1A ) , unlike the charge-inversion mutants of the same surface of the helix ( R134 , R149 ) . Mutations in the polar amino acid Q136 to Glu on the opposite surface of the helix did not suppress the defect of snf7D131K ( Figure 4—figure supplement 1A ) . These charge-inversion suppression effects were also observed with the defects of other mutations in helix-4 of Snf7 - E142K ( Figure 4—figure supplement 1B ) and that of D127K ( Figure 4—figure supplement 1D ) . Mutations R134 to Glu , or R149 to Glu completely suppressed the defects of both D127K and E142K . These data strongly argue that the overall acidic nature of the helix-4 surface of Snf7 is important for its function . Following the observations that mutations in helix-4 affect binding to Vps24 , the overall acidity of helix-4 appears to mediate recruitment of Snf7’s partner Vps24 to the polymer . ESCRT-III assemblies are unique biological polymers , as they form membrane-bound spirals of different architectures that achieve a topologically unique form of membrane bending . Snf7 alone primarily forms flat spirals in vitro ( Henne et al . , 2012; Chiaruttini et al . , 2015; Shen et al . , 2014 ) . Overexpression of CHMP4 , the mammalian homologue of Snf7 produces flat spirals or 3D helical structures in cells , that are similar to the spirals observed in vitro ( Hanson et al . , 2008 ) . We previously observed that incubating Snf7R52E , Vps24 and Vps2 on lipid monolayers produce architecturally distinct polymers similar to three-dimensional super-helical structures ( Henne et al . , 2012 ) , distinct from the flat , 2D spirals . Upon further analysis in this current study , we find a strong thermodynamic and kinetic dependence on the formation of these helices . At lower concentrations ( 1 μM each of Snf7R52E , Vps24 and Vps2 ) and shorter incubation times on lipid monolayers ( 10 min ) , all three proteins are required for the formation of 3D helices ( Figure 5—figure supplements 1–2 ) . However , at higher concentrations ( 7 μM of Snf7R52E ) and shorter times ( 10 min ) , just Snf7R52E and Vps24 can together form 3D helices , albeit at a lower frequency ( Figure 5—figure supplement 2A and C ) . We occasionally observe 3D helices of Snf7R52E alone at higher concentrations ( 7 μM ) and longer incubation times of 60 min ( Figure 5—figure supplement 2B ) , but these occur at a very low frequency ( Figure 5—figure supplement 2C ) . These data suggest that Snf7 has an intrinsic flexibility to transform into 3D helices , which are accelerated by the addition of its partners Vps24/Vps2 . Similar to this property of Snf7 , all ESCRT-III subunits likely have an ability to form flexible polymers , the architecture of which can be modulated by various properties – conformational changes , lateral strain ( Chiaruttini et al . , 2015 ) , surface density of the filaments and crowding effect of proteins ( cargo/filaments ) . Pure filaments of Snf7 also have an ability to form lateral interactions into bundles , as we and others have observed before ( Henne et al . , 2012; Chiaruttini et al . , 2015 ) . This lateral interaction of Snf7 polymers also seems to depend upon both the kinetics and thermodynamics of polymerization . At lower concentrations ( 500 nM ) and shorter incubation time ( 10 min ) , Snf7R52E can form single stranded filaments ( Figure 5—figure supplement 3 ) . At higher concentrations ( 1 μM and above ) , this mutant predominantly forms bundles through lateral interactions ( Figure 5—figure supplement 3 ) . Therefore , depending upon the exact stage of polymerization , Snf7 ( and likely other ESCRT-III proteins ) can form several laterally associating strands . The activation mutant R52E predominantly forms laterally interacting filaments of Snf7 as its critical concentration of polymerization is appreciably lowered . The difference in thickness of polymers of Snf7R52E with and without Vps24/Vps2 is not easily apparent , especially as this condition represents later stages of polymerization . At these later stages of polymerization , which occur after activation of Snf7 , the polymers may consist of helices that are capable of inducing membrane deformation . To observe the earlier stages of copolymer assembly , in our assays , we used wild-type Snf7 in the presence of Snf7’s nucleator Vps20 . With these assays , we observed predominantly single-stranded filaments of Snf7 ( width of ~4–5 nm ) ( Figure 5 ) . Under these conditions of earlier stages of polymer assembly , Snf7 filaments also do not form complete spirals ( Figure 5 ) . With the addition of Vps24 and Vps2 , thicker filaments were formed , suggesting that Vps24 and Vps2 induce formation of laterally-interacting polymers , consistent with previous observations ( Mierzwa et al . , 2017 ) . Importantly , the helix-4 mutant Snf7D131K is defective in inducing lateral bundles ( Figure 5 ) with Vps24/Vps2 , while the suppressor Vps24Q16E mutant forms a higher density of polymers and also partially rescues the bundling property . With wild-type Snf7 , even with the inclusion of the nucleator Vps20 , we do not observe helices of Snf7-Vps24-Vps2 . With the activated mutant Snf7R52E , along with Vps24 and Vps2 , the presence of Vps20 does not inhibit helix formation ( Figure 5—figure supplement 4 ) . These data suggest that Vps20 alone is insufficient to fully activate Snf7 in our in vitro assays . This most likely occurs because our in vitro protein Vps20 is not myristoylated , as it is normally in vivo ( Teis et al . , 2010 ) . Snf7 is activated by Vps20 , and also by ESCRT-II ( Henne et al . , 2012 ) . Adding an additional component in our assay – an ESCRT-II component ( GST tagged Vps25 ) , induces helicity ( Figure 5—figure supplement 4 ) , consistent with the idea that a higher level of activation of Snf7 in the in vitro assays is necessary to produce helices . Our data are most consistent with the model that Vps24 and Vps2 can laterally associate with Snf7 filaments to induce bundles of ESCRT-III at earlier times and lower activation threshold . Over time and at later stages of polymerization , ESCRT-III filaments mold into 3D helices that are more likely to be the mature polymers , structures that can generate the mechanical force important for vesicle-budding reactions . The lateral interactions are promiscuous and likely flexible enough for the polymers to be able to constrict into different architectures . Our in vitro data imply that Vps24 and Vps2 cooperatively bind to the Snf7 filament , inducing bundling and architectural changes in the polymer . To analyze cooperative assembly of Snf7 , Vps24 and Vps2 in vivo , we utilized the property of helix-4 mutants’ cargo-sorting defects . We reasoned that if simply the affinity of Vps24 to Snf7 is reduced in the charge-inversion mutants of helix-4 in Snf7 , we would be able to partially restore this interaction , and the in vivo defects in cargo-sorting , by over-expressing Vps24 . Since Vps24 and Vps2 get cooperatively recruited by Snf7 ( Babst et al . , 2002b; Mierzwa et al . , 2017; Adell et al . , 2017 ) , the same effect should be observed with overexpressed Vps2 as well , as the bound fraction of the complex of Vps24/Vps2 would increase upon overexpression of one of the components ( Figure 6C ) . We expressed Vps24 and Vps2 under two different expression systems ( Figure 6—figure supplement 1A ) . Over-expressing Vps24 or Vps2 by ~2 fold ( using a CEN plasmid ) did not rescue the defect of the Snf7 helix-4 mutant . However , overexpressing Vps24 or Vps24 by ~16 fold ( with a CMV promoter under a tet-off operator ) completely rescued the snf7D131K defect ( Figure 6A–B ) . These analyses suggest that the overexpression of Vps24 or Vps2 increases the bound fraction of the Vps24/Vps2 complex in the Snf7D131K polymer , rescuing the defect of the mutant Snf7 . Our in-vitro EM data suggest that the complex of Vps24/Vps2 interacts laterally on the Snf7 polymer . The polymer surface of Snf7 would provide multiple binding sites to the complex of Vps24/Vps2 . Under overexpression conditions of Vps24/Vps2 , there would be an increase in the bound fraction through avidity effects of Vps24/Vps2 to the Snf7D131K polymer . The rescue by overexpression is less likely if Vps24/Vps2 were to bind at the end of the polymer , with a lower number of binding sites available to Vps24/Vps2 . Consistent with this idea , overexpression of Vps24 or Vps2 did not rescue several longitudinal polymerization defective mutants of Snf7 ( Figure 6—figure supplement 1B–D ) . Altogether , our data provide strong evidence that Vps24 recognizes helix-4 in Snf7 , which , along with Vps2 , lies at the periphery of the core polymer , allowing for lateral association of these ESCRT-III proteins . This lateral interaction is mediated by promiscuous electrostatics , rather than specific residue-to-residue association . The core components of the ESCRT-III complex are essential for numerous biological reactions . Deletions or mutations in Vps24 and or Vps2 are defective for cargo sorting through the MVB pathway . Despite the importance of these ESCRT-III proteins , the mechanism behind how they associate with other ESCRTs , in particular Snf7 , remains undefined . How heteropolymerization of these proteins creates polymers of different curvatures and architectures remains uncharacterized . The molecular details that allow co-assembly of ESCRT-III proteins in general remains an important mechanistic question in cell biology . Recent structural studies have provided important clues on the mechanism of activation and assembly of the mammalian ESCRT-III subunits IST1 and CHMP1B , the yeast ESCRT-III subunit Snf7 , and the fly Snf7 ortholog Shrub . These structures defined the active , open conformation of the ESCRT-III proteins , and in the case of CHMP1B/IST1 copolymer , an open and a closed conformation . Previous genetic and biochemical experiments had suggested that different stimuli – membrane binding and ESCRT-II interaction – trigger ‘opening’ of the closed ESCRT-III conformation . This conformational change presents a hydrophobic surface and an electrostatic surface ( Tang et al . , 2015; McMillan et al . , 2016 ) for the core Snf7 subunits to assemble into polymeric structures . Here we find that the overall electrostatics of one of the helices of Snf7 ( helix-4 ) is critical for Snf7 to recruit its partner Vps24 . Mutations in this region of Snf7 inhibits degradation of model cargos because of the inability of these mutants to recruit Vps24 . Helix-4 of Snf7 is involved in lateral interaction of ESCRT-III subunits . In vitro , in the absence of another ESCRT-III partner , Snf7 uses this interface to assemble laterally , creating a ~ 9 nm protofilament , but it also can make additional contacts to produce filament bundles ( Chiaruttini et al . , 2015 ) . We observe strong dependence of kinetics and thermodynamics in the formation of laterally interacting polymers , suggesting that over time and with an increase in concentration of the polymerizing species , Snf7 can progressively associate into lateral bundles . In the presence of another ESCRT-III subunit such as Vps24 , this interface recruits the partner Vps24 to create a co-polymer , which then remodels the Snf7 spirals to create 3D helices through sliding of laterally interacting polymers ( Figure 7A ) . Recent evidence suggests that in vivo , the recruitment kinetics of ESCRT-III proteins Snf7 , Vps24 and Vps2 is indistinguishable ( Mierzwa et al . , 2017; Adell et al . , 2017 ) . Therefore , pure polymers of Snf7 , Vps24 or other ESCRT-III proteins , while providing important physical insights into polymerization dynamics , do not function individually in vivo . The function of the co-assembled heteropolymer is more relevant for membrane deformation . Charge-inversion mutations in helix-1 of Vps24 rescue the defects exhibited by different Snf7 helix-4 mutants ( Figure 7A ) . It is possible that these mutations in Vps24 allosterically enhance the affinity of Vps24/Vps2 for Snf7 rather than providing a direct interaction surface for Snf7 . However , the following lines of evidence suggest that these mutants probably lie at the interface of Snf7/Vps24 . First , we find strong evidence that Snf7 helix-4 uses its acidic nature to recruit Vps24 . Interestingly , helix-1 of Vps24 is predominantly basic ( Figure 3A ) , and helix-1 of all ESCRT-III proteins is the most basic surface of these complexes . Second , the D131 ( in helix-4 ) residue in the Snf7 crystal lattice contacts helix-1 in trans of a laterally coexisting polymer . Third , in thiol-based ex vivo crosslinking experiments , placing cysteines at Snf7 helix-4 and Vps24 helix-1 induces crosslinks . Fourth , in the cryo-EM structure of CHMP1B-IST1 , CHMP1B helix-4 contacts the helix-1 in IST1 ( McCullough et al . , 2015 ) . These data are consistent with the idea that Snf7 helix-4 contacts helix-1 of Vps24 . Interestingly , we observe that the overall electrostatics on one surface of helix-4 in Snf7 is important for this recognition , rather than a consensus sequence on that surface . The suppressor mutations in Vps24 lie in the basic helix-1 region . We note that our data point to the R19/K26 region of Vps24 as the binding surface for Snf7 , and do not directly show that the binding surface may be spread out throughout the basic helix-1 region . However , it is possible that on a polymeric surface of Vps24 ( and Vps2 ) where Snf7 is bound , each Snf7 monomer may engage with the same location of Vps24 ( R19 ) through different residues of helix-4 ( D127 , D131 , E142 ) ( Figure 7A ) . This continuous charge-charge interaction among the polymers of ESCRT-III may be an important aspect of ESCRT-III co-assembly ( McCullough et al . , 2015; McCullough et al . , 2018 ) . On the one hand , multiplication of electrostatic interactions among each protomer would enhance the avidity of Snf7 to Vps24/Vps2 in the context of the polymer ( Figure 7B ) . On the other hand , the uninterrupted presentation of charges as a binding surface would allow Vps24/Vps2 to adopt different positions along the filament ( Figure 7A–B , Figure 8A ) . As the polymer constricts , the lack of a requirement for residue-to-residue specificity would enable Vps24/Vps2 to bind at different locations , allowing the polymers to adapt , by sliding side-by-side , to different curvatures in the polymer ( Figure 7A–C , Figure 8A ) . In the possible scenario that the dynamics of Snf7-Snf7 assembly and Vps24/Vps2 assembly are different ( i . e . the on and off rates of Snf7 to the polymer is different from that of Vps24/Vps2 ) ( Chiaruttini and Roux , 2017 ) , the heteropolymer would further be able to embrace variable architectures capable of encircling cargo . Additionally , our crystal structure suggests that helix-4 , and most likely the C-terminal region , which lie at the periphery of the polymer and away from it , can also exist in different conformations ( Tang et al . , 2015 ) . An ability to easily change conformations at the periphery of the polymer could additionally enable the polymer to change its architecture . Our data and the reported structure of CHMP1B/IST1 suggest that , similar to the core longitudinal assembly mechanism that are similar between these ESCRT-III proteins , the recognition of partners through helix-4 or the C-terminal peripheral region could be a general feature in ESCRT-III assembly . It is important to note that CHMP1B also contacts IST1 through the C-terminal regions of CHMP1B ( Talledge et al . , 2019 ) . In the case of the Snf7-Vps24 interaction , we find that Vps24-GFP is recruited to endosomes even with a Snf7 construct in which the C-terminus ( helices-5 and beyond ) is deleted ( Figure 2—figure supplement 2A; Henne et al . , 2012 ) . This could mean that the C-terminus of Snf7 contacts Vps24 only with a weak affinity and therefore we are unable to see an obvious effect in our cellular assays . It is also possible that the C-terminus of Snf7 does not contact Vps24 . Since the topology of membrane vesicles created by the reported CHMP1B-IST1 copolymer and those created by Snf7-Vps24/Vps2 at endosomes are opposite to one another , further analyses of these two systems is necessary for us to fully understand the similarities and differences between them . One possibility on how these two systems could create opposite topologies while possessing similar heteropolymer contacts is provided in Figure 8—figure supplement 1 . In this model , while CHMP1B is bound to the membrane and resides inside of the polymer helix , Snf7 could reside on the outside , with the membrane bound to Snf7 . The combination of this plethora of different data is consistent with the following model: upon assembly of early ESCRTs ( 0 , I and II ) at the endosomes with cargo , Vps20 is recruited , which nucleates Snf7 polymerization . Snf7 forms a scaffold at the membrane which then is required to recruit Vps24/Vps2 , although kinetically this recruitment likely happens simultaneously with Snf7 nucleation ( Figure 8B ) . Interestingly , a recently reported study of the bacterial tubulin homolog FtsZ suggests that weak van der Waals interactions between laterally interacting filaments may be an important property of that polymeric system ( Guan et al . , 2018 ) . Furthermore , previous analyses suggested that partners of FtsZ can bundle FtsZ through lateral associations , giving rise to helical structures ( Goley et al . , 2010 ) . Filament sliding models for FtsZ and its partners have been proposed before ( Szwedziak et al . , 2014 ) . However , as far as we are aware , the molecular mechanisms of how such sliding may be accomplished by the filaments have not been defined . Given the corollary between ESCRT-III and FtsZ proteins ( membrane associated spiraling polymers ) such lateral associations created by non-specific interactions maybe a general way for such polymeric proteins to adopt different curvatures during assembly on malleable membrane surfaces ( Figure 8A ) . Although we observe that non-specific charge-charge interactions is a property of helix-4 in Snf7 , we suspect such promiscuous interactions to be present within other interaction surfaces ( in the longitudinal interface ) as well . We and others have observed heterogeneity in the diameter of Snf7 spirals , suggesting that ESCRT-III polymers can adopt different diameters to be able to organize cargo inside the spirals . A rigid polymer with the same interaction surface will be unable to perform this task . As the inter-subunit angle changes and the polymers constrict , the contacts among the surface residues should also change to allow for generation of curvature . Currently , structural analyses only depict an energetically stable complex with the same interaction surfaces in different lattices and are unable to clarify the dynamic ensemble of different interfaces . More structures of individual ESCRT-III proteins and in protein complexes will help clarify the interfaces that are important for homo and hetero polymerization of these fascinating self-assembling proteins and will allow us to understand how nature controls these ensembles . Strains , plasmids and reagents used in this study are listed in the Key Resources table . Previously used strains , plasmids , reagents were from the following references ( Robinson et al . , 1988; Adell et al . , 2017; Henne et al . , 2012; Ghazi-Tabatabai et al . , 2008; Tang et al . , 2015; Tang et al . , 2016; Buchkovich et al . , 2013; Sikorski and Hieter , 1989; Garí et al . , 1997; Babst et al . , 1998 ) and are also listed and referenced to in the table . Canavanine-sensitivity spot plating assays were performed as described before ( Lin et al . , 2008 ) . Mid-log cells were diluted back to an optical density ( at 600 nm ) of 0 . 1 . 10-fold serial dilutions were made and the dilutions applied to plates with selective drop-out media with different concentrations of canavanine . Images of the plates were taken after three or more days . Experiments were performed at least twice in all cases . Canavanine resistance was used to select for mutations in Snf7 or Vps24 that function as suppressors of helix-4 mutant . To select for Vps24 mutants , a strain harboring a chromosomally integrated snf7D131K and vps24Δ was transformed with a plasmid library of randomly mutagenized VPS24 . Random mutation was performed by error-prone PCR , as previously described ( Tang et al . , 2016 ) , using primers annealing to the 5’ and 3’ ends of the VPS24 ORF . Suppressing mutations on Snf7 were similarly obtained , mutagenizing the whole ORF of snf7D131K on a plasmid and then transforming to a snf7Δ strain . Flow cytometry analysis of Mup1-pHluorin endocytosis and trafficking was performed as described ( Henne et al . , 2012 ) . Briefly , mid-log cells were treated with 20 μg/mL of L-methionine for 90 min . Cells were spun down and resuspended in synthetic dextrose complete minimal medium ( SCD ) . Mean fluorescence of 100 , 000 cells were recorded using a BD Accuri C6 Flow Cytometer . % MVB sorting was calculated by normalizing the WT sorting to 100% and mutant ( ESCRT deletion ) to 0% . At least three independent experiments were performed to calculated standard deviation . Western blots of Mup1-pHluorin was performed as follows . 5 OD equivalent of cells treated with 20 μg/mL methionine were collected by centrifugation at different time points at 4000 xg . Centrifuged cells were then washed with 1 mL of cold H2O , and then centrifuged again at 4000 xg . Cells were then precipitated with 10% TCA for >1 hr on ice . Cells were washed twice with 1 mL of acetone , resuspending pellets between washes by bath sonication . Pelleted cells were then lysed in 100 μL lysis buffer ( 50 mM Tris-HCl , pH 7 . 5 , 8 M urea , 2% SDS , and 1 mM EDTA ) by bead beating for 10 min . 100 μL of sample buffer ( 150 mM Tris-Cl , pH 6 . 8 , 8 M urea , 10% SDS , 24% glycerol , 10% v/v βME , and bromophenol blue ) was then added to the sample and vortexed for 10 min . After centrifugation for 6 min at 21 , 000 xg , supernatant was loaded on an SDS-PAGE gel and transferred onto a nitrocellulose membrane . Rabbit polyclonal GFP antibody ( Torrey Pines ) was used to detect pHluorin . Imaging of the western blots was performed using an Odyssey CLx imaging system and analyzed using the Image Studio Lite 4 . 0 . 21 software ( LI-COR Biosciences ) . Snf7 sequences were aligned using Mafft ( Katoh et al . , 2002 ) . Jalview ( Clamp et al . , 2004 ) was used to visualize the sequences . Homology modeling of the Vps24 structure was performed using Modeller ( Fiser and Sali , 2003 ) , using the CHMP3 ( PDB 3FRT ) structure as the template . Helical wheel analysis was performed using Heliquest ( Gautier et al . , 2008 ) . Structures were viewed and analyzed using UCSF Chimera ( Pettersen et al . , 2004 ) . 1 mL of mid-log cells expressing VPS24-GFP were centrifuged for 2 min at 10 , 000 xg , and resuspended in 25 mL of synthetic media . Microscopy was performed on a Deltavision Elite system with an Olympus IX-71 inverted microscope , using a 100X/1 . 4 NA oil objective . Image extraction and analysis were performed using the FiJi software ( Schindelin et al . , 2012 ) . 15 ODs of mid-log cells were harvested and spheroplasted using zymolyase treatment as previously described ( Buchkovich et al . , 2013 ) . Spheroplasts were lysed by douncing on ice in 50 mM Tris pH 7 . 5 , 200 mM sorbitol with protease inhibitors . Lysates were centrifuged at 500 xg at 4°C . This supernatant ( S5 ) was then centrifuged at 13 , 000 xg for 10 min at 4°C , which provided us P13 ( endosome enriched pellet fraction at 13 , 000 xg ) and S13 ( supernatant fraction ) . The P13 and S13 fractions were then precipitated in TCA , and immunoblotted as described above for Mup1-pHluorin . 30 ODs of mid-log cells were harvested and spheroplasted as done for subcellular fractionation experiments . Lysis was performed by douncing in 50 mM Hepes pH 7 . 5 , 200 mM Sorbitol , 150 mM NaCl , 1 mM EDTA , 1 mM DTT and 1%-TritonX-100 . Lysate was then centrifuged at 13 , 000 xg . Supernatant was then treated with protein G beads ( Dynabeads ) for 30 min at 4°C to clear background binding to beads . After centrifugation at 500 xg for 10 min , supernatant was then incubated with anti-Vps24 antibody ( at 1/250 dilution ) for 2 hr . Protein G beads were then used to pull-down Vps24 bound complexes . After washing three times with PBS buffer at 20 fold excess volume of the beads , the beads were treated with sample buffer ( 150 mM Tris-Cl , pH 6 . 8 , 8 M urea , 10% SDS , 24% glycerol , 10% v/v βME , and bromophenol blue ) . After SDS-PAGE , western blots were performed , and anti-Snf7 and anti-Vps24 antibodies were used to probe bound complexes . 30 ODs of cells were harvested in 50 mM Hepes pH 7 . 5 , 200 mM Sorbitol , 150 mM NaCl , 1 mM EDTA , fresh 0 . 5 mM DTT and Roche’s complete protease inhibitor . Lysis was performed by bead-beating ( Zirconia-Silicon beads ) twice for 30 s , with 30 s intervals on ice . After lysis , the lysate was supplemented with 1% of Triton-X 100 and incubated at 4°C for 20 min . Lysate was cleared by centrifugation at 500 xg for 5 min at 4°C . Supernatant was treated with 3 . 3 mM of BMOE ( bismaleimidoethane ) and 10 mM EDTA and incubated at 4°C for 10 min . Reaction was stopped using 10 mM of DTT and the solution then treated with 10% of TCA . TCA precipitation was performed for >1 hr . Western blots were performed as described above . 30 ODs of cells were harvested in PBS buffer , fresh 0 . 5 mM DTT and Roche’s complete protease inhibitor . Lysis was performed by bead-beating ( Zirconia-Silicon beads ) . Glycerol gradients were made using Gradient Master 108 from Biocomp . Centrifugation was performed at 100 , 000 xg for 4 hr at 4°C . 1 mL fractions were collected from the solutions , TCA precipitated and immunoblotted as described above . Snf7 , Vps24 and Vps2 and Vps20 proteins were expressed from a modified pET28a ( + ) vector expressing His6-SUMO protein at the N-terminus . GST-Vps25 was expressed using the pGEX6p1 vector . Expression of ESCRT-III was performed using the Rosetta E . coli strain . Snf7 , Vps20 , Vps24 and GST-Vps25 were constructs were expressed at 37°C for 4 hr by inducing with 0 . 5 mM IPTG . Snf7-D131K and Vps2 were expressed at 26°C overnight , inducing with 0 . 5 mM IPTG . Harvested cells were lysed by sonication . Affinity purification of the proteins through the His6 tag was performed using Co2+ talon resin . The SUMO tag was cleaved overnight at 4°C on the beads using ULP1 protease . Eluate was subjected to a Hi-trap Q Seph FF column . The anion exchange eluate was concentrated and ran through a Superdex 200increase column . GST-Vps25 was purified using GSH-sepharose beads , eluted using glutathione , concentrated and ran through a Superdex 200 column . Eluted proteins were concentrated , flash-frozen in liquid nitrogen and stored at −80°C . Liposomes were made using a mixture of 60% POPC and 40% POPS . Lipids in chloroform were mixed at the appropriate molar ratios and dried overnight under vacuum . Lipids were hydrated for 3 hr and resuspended in 25 mM Hepes 7 . 5 , 150 mM NaCl , to make a lipid concentration of 1 mg/mL . Large unilamellar vesicles ( LUVs ) were made using extrusion filters of 800 nm pores from Avanti Polar Lipids . Proteins were added to the liposomes at a final concentration of 200 nM and a final lipid concentration of 0 . 5 mg/mL . After incubation for 30 min at room temperature , centrifugation was performed in a TLA-100 ( Beckman Coulter ) at 70 , 000 rpm for 10 min at 4°C . SDS-PAGE was used to determine fraction of protein pelleted with the liposomes . Lipid monolayers were formed using a ratio of 60% POPC , 30% POPS and 10% PI3P in chloroform . Monolayers were formed above an aqueous buffer solution , and lipids were injected underneath the monolayer , using a home-made Teflon apparatus , as described before ( Henne et al . , 2012 ) . Carbon-coated electron microscope grids were applied to the top of the aqueous solution simultaneously with the application of proteins . Incubation of proteins on the monolayers/grids were performed at various times as indicated in the text . Grids were then stained with 2% ammonium molybdate and imaged on FEI Morgagni 268 TEM .	Cells are separated from the outside environment by a fatty layer called the plasma membrane . This layer not only isolates the inside of the cell from the outside , it is also essential for the cell to sense and respond to cues around it . For example , the plasma membrane contains different types of proteins that can act as receptors for signals from outside the cell or as channels to take in essential nutrients . One of the ways that the cell can respond to its environment is by recycling the proteins at the plasma membrane . During a cell’s life , proteins from its membrane are recycled by being pulled into lysosomes , which are sacs or vesicles full of enzymes that digest these molecules . However , before reaching the lysosomes , the molecules pass through another set of vesicles called endosomes . There , ESCRT-III , a flexible scaffold made out of the proteins Snf7 , Vps24 and Vps2 , forms a spiral like-structure that collects the proteins and fats from the membrane . This corkscrew-like shape allows the ESCRT-III scaffold to work , but it is unclear how it is formed . Snf7 is a protein that forms long bending chains , or “polymers” , by linking to itself . Banjade et al . found that Snf7 uses its negatively charged surface to interact with the parallel chain that Vps24 and Vps2 form at its side . However , Vps24 and Vps2 do not fit rigidly into Snf7 like a key fits in a lock . Rather , their interaction is flexible , based on charge . This flexibility may allow Vps24 and Vps2 to slide along the side of the Snf7 chain , helping to create a spiral . Banjade et al . used budding yeast as a model organism and also imaged purified proteins with electron microscopy to come upon these findings . Understanding how ESCRT proteins interact to form complex structures may lead to a better understanding of how other membrane-bound polymers form elsewhere in the cell . ESCRT proteins are also involved in degenerative diseases , such as Alzheimer’s , where proteins that need to be recycled cannot be properly processed , and they are important for viruses such as HIV to spread between cells . Understanding how these proteins interact to form their characteristic spiral structure could potentially lead to the development of new therapies .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology" ]	2019	Electrostatic lateral interactions drive ESCRT-III heteropolymer assembly
Mutations in the fukutin-related protein ( FKRP ) cause Walker-Warburg syndrome ( WWS ) , a severe form of congenital muscular dystrophy . Here , we established a WWS human induced pluripotent stem cell-derived myogenic model that recapitulates hallmarks of WWS pathology . We used this model to investigate the therapeutic effect of metabolites of the pentose phosphate pathway in human WWS . We show that functional recovery of WWS myotubes is promoted not only by ribitol but also by its precursor ribose . Moreover , we found that the combination of each of these metabolites with NAD+ results in a synergistic effect , as demonstrated by rescue of α-dystroglycan glycosylation and laminin binding capacity . Mechanistically , we found that FKRP residual enzymatic capacity , characteristic of many recessive FKRP mutations , is required for rescue as supported by functional and structural mutational analyses . These findings provide the rationale for testing ribose/ribitol in combination with NAD+ to treat WWS and other diseases associated with FKRP mutations . Mutations in the fukutin-related protein ( FKRP ) gene result in a broad spectrum of muscular dystrophy ( MD ) phenotypes , ranging from mild Limb-Girdle MD ( LGMDR9 ) to Walker-Warburg syndrome ( WWS ) , the most severe form of congenital MD ( CMD ) ( Beltran-Valero de Bernabé et al . , 2004; Brockington et al . , 2001a; Brockington et al . , 2001b ) . The biochemical hallmark of FKRP muscle disorders is hypoglycosylation of α-dystroglycan ( α-DG ) , which leads to disruption in the interaction of α-DG with extracellular matrix proteins , in particular laminin-α2 , which is essential for muscle fiber integrity ( Ervasti and Campbell , 1993; Ibraghimov-Beskrovnaya et al . , 1992 ) . Due to its rarity and reduced life expectancy ( <3 years ) , disease pathogenesis and treatment strategies remain elusive for WWS . To date , there is no effective treatment for FKRP-associated MDs ( Ortiz-Cordero et al . , 2021 ) . FKRP is a ribitol-5-phosphate transferase , that in tandem with fukutin ( FKTN ) , adds ribitol 5-phosphate onto the 3GalNAc-β1-3GlcNAc-β1–4 ( P-6 ) Man-1-Thr/Ser modification of α-DG ( Core M3 ) ( Manya et al . , 2004; Yoshida-Moriguchi et al . , 2013 ) , using cytidine diphosphate ( CDP ) -ribitol , which is produced by isoprenoid synthase domain-containing protein ( ISPD ) ( Gerin et al . , 2016; Kanagawa et al . , 2016; Praissman et al . , 2016; Riemersma et al . , 2015 ) . The presence of ribitol-5-phosphate is essential for the subsequent addition of −3Xylα1-3GlcAβ1 and ( 3Xylα1-3GlcAβ1 ) n- and , therefore , for allowing α-DG to bind to extracellular membrane ligands ( Inamori et al . , 2012; Manya et al . , 2016; Praissman et al . , 2014; Willer et al . , 2014 ) . The addition of the pentose phosphate pathway ( PPP ) metabolites ribitol or ribose , a precursor of ribitol ( Gerin et al . , 2016; Huck et al . , 2004 ) , to fibroblasts from patients with ISPD mutations resulted in increased CDP-ribitol levels and rescue of α-DG functional glycosylation . However , rescue varied significantly among samples with different mutations ( Gerin et al . , 2016; van Tol et al . , 2019 ) . Cataldi et al . , 2018 documented that ribitol treatment in FKRP mutant mice modeling severe LGMDR9 partially restored α-DG functional glycosylation . More recently , Nickolls et al . , 2020 showed that ribitol treatment of embryoid bodies from an LGMDR9 patient-specific induced pluripotent stem ( iPS ) cell line partly recovered the glycosylation defect . Taken together , these results raise several possibilities: ( i ) ribitol and ribose may also have a beneficial effect in WWS associated with FKRP mutations , ( ii ) rescue by PPP metabolites might be mutation-specific , and ( iii ) other metabolites may potentiate the recovery of α-DG functional glycosylation . Another metabolite of interest is β-nicotinamide adenine ( NAD+ ) . NAD forms ( NAD+ , NADH , NADP , and NADPH ) are essential cofactors for oxidoreductases in the PPP ( Singh et al . , 2017 ) . Moreover , Bailey et al . , 2019 reported that NAD+ supplementation in FKRP-deficient zebrafish led to decreased muscle degeneration and improved muscle function when administered at gastrulation , before muscle development occurs . Nevertheless , to date , no human model exists to test or validate the therapeutic potential of these or any other compounds for WWS . In this study , we took advantage of the ability of iPS cells to differentiate into skeletal myotubes ( Selvaraj et al . , 2019b ) to establish a novel WWS patient-specific in vitro model . Our results demonstrate that this system recapitulates the major skeletal muscle hallmarks of WWS . Moreover , we find that ribitol and ribose can partially rescue functional glycosylation of α-DG , and that administration of NAD+ along with each of these PPP metabolites significantly potentiates α-DG functional glycosylation rescue . Using an integration-free approach , we generated iPS cells from a 1-year-old WWS male patient ( FP4 ) harboring two mutations in exon 4 of the FKRP gene , dc . 558dupC ( p . A187fs ) and c . 1418T>G ( p . F473C ) ( Kava et al . , 2013 ) . FP4 iPS cells express pluripotency markers , display normal karyotype , and develop teratomas containing cell types from all three germ layers ( Figure 1—figure supplement 1 ) . Using inducible expression of PAX7 ( Darabi et al . , 2012 ) , we differentiated FP4 and control wild type ( WT ) iPS cells into myogenic progenitors and subsequently into terminally differentiated myosin heavy chain ( MHC ) -positive myotubes ( Figure 1A ) . Immunostaining for MHC showed similar differentiation between WT and mutant FP4 myotubes ( Figure 1A , upper panel ) . Staining with IIH6 , a monoclonal antibody specific to the laminin binding domain of α-DG ( Ervasti and Campbell , 1993 ) , showed drastically reduced IIH6 immunoreactivity in FP4 myotubes ( Figure 1A , lower panel ) , which was corroborated by western blot . In accordance with the loss of α-DG functional glycosylation , WWS FP4 myotubes showed decreased molecular weight for α-DG core ( Figure 1B ) , marked reduction in IIH6 ( Figure 1B and C ) , and most importantly , lack of laminin binding , as demonstrated by the laminin overlay assay ( LOA ) following enrichment by wheat germ agglutinin ( WGA ) pull-down ( Figure 1B ) . As proof of concept , we introduced a WT FKRP transgene into FP4 cells to determine whether WT FKRP could restore functional glycosylation of α-DG . FKRP-overexpressing FP4 myogenic progenitors gave rise to MHC-positive myotubes ( Figure 1—figure supplement 2A ) that displayed increased FKRP expression ( Figure 1—figure supplement 2B ) and enhanced immunoreactivity to IIH6 ( Figure 1—figure supplement 2A and C ) , which led to rescue of laminin binding capacity ( Figure 1—figure supplement 2D ) . Having developed this platform , we tested whether ribitol , a precursor for CDP-ribitol ( Figure 2A ) , would be able to increase functional α-DG glycosylation in the human context using the FP4 patient-specific iPS cell-derived model . At the onset of terminal myogenic differentiation , we treated myogenic progenitors with increasing concentrations of ribitol ( 25 mM , 50 mM , 100 mM , and 200 mM ) for 5 days . Following evaluation of cell morphology and IIH6 immunoreactivity ( Figure 2—figure supplement 1A and B ) , the 50 mM concentration was chosen for the studies described here . Treated cells retained differentiation capacity , as shown by MHC levels ( Figure 2B , Figure 2—figure supplement 1C ) , and exhibited rescue of α-DG functional glycosylation ( Figure 2B , C and D ) . Importantly , this increase in functional glycosylation of α-DG was sufficient to increase binding between α-DG and laminin , as shown by the detection of laminin only in ribitol-treated FP4 myotubes ( Figure 2C ) . Since ribitol is endogenously generated by the reduction of ribose via an oxidoreductase ( Figure 2A ) , we hypothesized that supplementing FP4 myogenic cells with ribose might also recover α-DG functional glycosylation . As before , we treated FP4 cultures with increasing concentrations of ribose , ranging from 5 to 100 mM . Because the lowest concentration of ribose able to enhance α-DG functional glycosylation was 10 mM and higher concentrations led to cell death ( >50 mM ) , we chose the concentration of 10 mM for further analysis ( Figure 2—figure supplement 1D to F ) . As shown in Figure 2E , F and G , we found a significant increase in functional glycosylation of α-DG upon 10 mM ribose supplementation . This increase was sufficient to enhance laminin binding capacity in FP4 myotubes ( Figure 2F ) . To determine the effect of ribitol and ribose supplementation on the synthesis of ribitol-5-P and CDP-ribitol , we quantified the levels of these PPP metabolites after 5 days of treatment by liquid chromatography with tandem mass spectrometry ( LC/MS-MS ) . Quantification of each metabolite was determined based on generated standard curves ( Figure 3—figure supplement 1 ) . The data from this analysis revealed that both ribitol and ribose supplementation result in significant increases in ribitol , ribose , ribitol-5-P , and CDP-ribitol compared to untreated cultures ( Figure 3A ) . The reduction of ribose to ribitol has been suggested to be mediated by the sorbinil sensitive aldose reductase ( AKR1B1 ) ( Gerin et al . , 2016 ) . To determine whether inhibition of this aldose reductase would diminish α-DG functional glycosylation and counteract the ribose-mediated rescue , we treated WT cells at the onset of terminal differentiation with sorbinil . This resulted in a 40% reduction in IIH6 levels in WT myotubes ( Figure 3B and C ) . Most importantly , sorbinil treatment counteracted the positive effect of ribose in FP4 myotubes by 63% , as shown by the diminished rescue of IIH6 levels ( Figure 3D and E ) . Since NAD+ has been shown to improve muscle function in the FKRP dystroglycanopathy zebrafish model ( Bailey et al . , 2019 ) , we tested the effect of NAD+ alone or in combination with ribitol or ribose . We treated FP4 cultures with 100 μM of NAD+ , as this concentration has been previously documented ( Goody et al . , 2012 ) . In FP4 myotubes treated with NAD+ alone , we observed a small , yet significant increase in functional glycosylation of α-DG ( Figure 4—figure supplement 1 ) . However , when we combined NAD+ supplementation with PPP metabolites , we observed , on average , a 59% increase in IIH6 positivity in ribitol/NAD+ when compared to ribitol alone ( Figure 4A and C ) . The same synergistic effect was observed when FP4 cells were treated with the NAD+/ribose combination , as levels of IIH6 increased on average 50% compared to ribose alone ( Figure 4B and D ) . Importantly , in both cases , the combination also promoted increased laminin binding capacity ( Figure 4A and B ) . We also tested the effect of ribitol and ribose in combination with NAD+ in established myotubes . For this , we differentiated FP4 myogenic progenitors into myotubes , and 4 days later , added the compounds for 24–72 hr . Again , a synergistic effect was observed upon the combination of NAD+ with PPP metabolites . Ribitol/NAD+ treatment led to an 85% increase in IIH6 immunoreactivity compared to ribitol alone ( Figure 4E and F ) . Likewise , ribose/NAD+ treatment on average doubled functional glycosylation of α-DG at 48 hr when compared to ribose alone ( Figure 4G and H ) . These results support the beneficial effect of combining ribitol or ribose with NAD+ to enhance α-DG functional glycosylation . Whereas ribitol and ribose rescue functional glycosylation of α-DG by increasing the generation of the FKRP substrate CDP-ribitol , the mechanism for NAD+ remains unclear . Since several forms of NAD act as cofactors for oxidoreductases , we hypothesized that NAD+ could enhance the generation of ribitol-5-P and CDP-ribitol . To test this , we quantified the levels of PPP metabolites in NAD+-treated cells by LC/MS-MS , as described above for ribitol/ribose ( Figure 3A ) . We found that NAD+ treatment led to a small increase in ribose levels compared to untreated counterparts , but no significant differences were detected in ribitol-5-P and CDP-ribitol , as shown by comparing untreated vs . NAD+ , ribitol vs . ribitol/NAD+ , and ribose vs . ribose/NAD+ ( Figure 4—figure supplement 2 ) . These results suggest that the synergistic effect of NAD+ is independent of the FKRP substrate CDP-ribitol . Because the null mutation for FKRP is embryonic lethal ( Chan et al . , 2010 ) , most FKRP mutations are thought to have some residual activity . To determine whether such residual activity of FKRP is required for the effects observed upon ribitol and ribose supplementation , we generated an FKRP-deficient pluripotent stem cell line ( FKRP knockout [KO] ) using CRISPR/Cas9 genome editing . Immunostaining for MHC showed similar differentiation between FKRP KO myotubes and respective control WT counterparts ( Figure 5A ) , but as anticipated , FKRP KO myotubes lacked functional glycosylation of α-DG , as evidenced by immunostaining and western blot to IIH6 ( Figure 5A and B ) and absence of laminin binding ( Figure 5B ) . Importantly , treatment of FKRP KO myotubes with ribitol , ribose , or NAD+ did not rescue α-DG functional glycosylation ( Figure 5C ) . To investigate whether other mutations associated with the WWS phenotype are amenable to rescue by these metabolites , we introduced the WWS-clinically associated FKRP-C318Y mutation ( Beltran-Valero de Bernabé et al . , 2004 ) located in the zinc finger loop of the FKRP catalytic domain into WT iPS cells using CRISPR-Cas9 genome editing . Isogenic myotubes generated from FKRP-C318Y iPS cells displayed a similar phenotype to patient-specific FP4 myotubes ( Figure 5D ) , thus confirming the in vitro WWS phenotype . We tested ribitol , ribose , or combinations with NAD+ supplementation in cultures of FKRP-C318Y myotubes , as described above for FP4 , and none of the metabolites were able to rescue α-DG functional glycosylation in these cells ( Figure 5E ) , suggesting that rescue is mutation specific . To understand how WWS-associated mutations ( Figure 6A ) may interfere with the FKRP enzymatic activity , we turned to the recently deciphered crystal structure of FKRP ( Figure 6B; Kuwabara et al . , 2020 ) . The zinc finger loop in the FKRP catalytic domain consists of four conserved cysteine residues ( C289 , C296 , C317 , C318 ) required for Zn2+ ion binding . The C318Y mutation disrupts direct chelation to the Zn2+ ion and sterically hinders the binding site from occupancy . This result suggests that this mutation will lead to significant , if not complete , loss of enzymatic function , and therefore metabolite-mediated rescue is not possible . On the other hand , F473 makes up a small hydrophobic pocket with L348 , I357 , W359 , V477 , and P481 that is essential for CDP-ribitol substrate binding within the catalytic domain . The mutation of F473 to cysteine ( F473C ) present in the FP4 patient sample leads to a free energy change in substrate binding affinity of +4 . 3 kcal/mol . This difference suggests that the F473C mutation results in destabilization of the Michaelis complex formation with diminished enzyme efficiency , and therefore , metabolite supplementation that increases CDP-ribitol levels allows for increased FKRP activity . These results indicate that functional FKRP is required for rescue of α-DG functional glycosylation by ribitol , ribose , and NAD+ . Myoblasts harvested from patients are commonly used to model muscular dystrophies in vitro . However , in cases like WWS , the short lifespan along with the difficulty in obtaining tissue from patients represents major hurdles in establishing patient-specific myoblasts lines . The generation of patient-specific iPS cells circumvents the restricted patient tissue availability and the limited cell proliferation capacity seen in ex vivo expanded primary cells ( Kondo et al . , 2013; McKeithan et al . , 2020; Sampaziotis et al . , 2015; Young et al . , 2016 ) . To date , several experimental studies in animal models have provided evidence supporting the potential therapeutic application of gene therapy ( Gicquel et al . , 2017; Xu et al . , 2013 ) and cell therapy ( Azzag et al . , 2020; Frattini et al . , 2017 ) for FKRP-associated muscular dystrophies , but these studies are still at early stages , and therefore , currently there are no clinical trials underway . In this study , we show that PPP metabolites are able to increase functional glycosylation of α-DG in WWS patient-specific iPS cell-derived myotubes associated with FKRP mutations ( FP4 ) . Besides ribitol , we show that ribose is also able to provide significant increase in IIH6 immunoreactivity in FKRP mutants , which is accompanied by rescue of laminin binding . Our results indicate that both these PPP metabolites increase ribitol-5-P and CDP-ribitol levels in FP4-treated myotubes . The enhanced functional glycosylation of α-DG in FP4 mutant myotubes is hypothesized to be due to increased CDP-ribitol levels leading to increased ribitol-5-P transferase activity in the disease-causing FKRP mutant . We show for the first time that NAD+ can increase functional glycosylation of α-DG in a human WWS FKRP model , and when combined with ribitol or ribose , can significantly potentiate the rescue of the muscle pathology in vitro . Studies in dystroglycan ( dag1 ) and FKRP zebrafish mutants have demonstrated a beneficial effect for NAD+ ( Bailey et al . , 2019; Goody et al . , 2012 ) . Although the mechanism is not entirely elucidated , NAD+ was reported to promote increased ADP-ribosylation of integrin receptors , which in turn increase integrin and laminin binding , increase laminin-111 organization and subcellular localization of paxillin to cell adhesion complexes ( Goody et al . , 2012; Zolkiewska , 2005 ) . Interestingly , overexpression of paxillin rescues muscle structure in dag1 but not in FKRP mutants , suggesting a different mechanism of action ( Bailey et al . , 2019 ) . Previous studies in two Duchenne MD mouse models showed that NAD+ improved muscle function via reduced parylation , as well as increased mitochondria function and expression of structural proteins ( Ryu et al . , 2016 ) . Our data on the quantification of core α-DG in FKRP-C318Y myotubes that had been treated or not with NAD+ revealed a significant increase in core α-DG upon NAD+ treatment , whereas β-DG levels remained unchanged ( Figure 5—figure supplement 1 ) , suggesting that NAD+ supplementation specifically increases α-DG . Although further studies are required to elucidate the mechanism by which NAD+ may lead to increased α-DG , a plausible hypothesis is a post-translational effect . Importantly , we show that functional FKRP mediates the rescue by ribitol , ribose , and NAD+ since FKRP KO and FKRP-C318CY myotubes do not show IIH6 rescue upon treatment with any of these compounds or combinations . Based on the recently reported FKRP crystal structure ( Kuwabara et al . , 2020 ) , C318 is located in the zinc finger loop ( G288 to C318 ) of the FKRP catalytic domain , which has been proposed to be of fundamental importance for the catalytic activity of FKRP ( Kuwabara et al . , 2020 ) . Our results suggest that the ability of PPP metabolites to partially rescue α-DG functional glycosylation is mutation dependent . Although further studies are required to determine which patients could benefit from this potential treatment , our results suggest that phenotypes associated with mutations in the zinc finger loop region may not be rescued by ribitol and ribose , whereas FKRP mutations in other regions of the catalytic domain are amenable to rescue , as shown for FP4 . This is in line with previous studies in ISPD fibroblasts , in which rescue of functional glycosylation of α-DG was found to be mutation dependent ( Gerin et al . , 2016; van Tol et al . , 2019 ) . Dietary interventions can provide a feasible and economically accessible solution for the treatment of MD associated with CDP-ribitol defects . Although ribitol/NAD+ showed promising results in our model , clinical trials to assess the safety of ribitol are still necessary . On the other hand , ribose is a commercially available supplement , and to date , with no major side effects in humans ( Dodd et al . , 2004; Seifert et al . , 2017; Thompson et al . , 2014 ) . Furthermore , NAD+ levels can be increased by several vitamin B3 forms , such as nicotinic acid ( niacin ) and nicotinamide riboside , which have been investigated , showing no major side effects ( Elhassan et al . , 2019; Guyton and Bays , 2007; Pirinen et al . , 2020 ) . Although future research studies are necessary to determine the optimal dosage of the combined approach , the safety record of these compounds justifies using ribose/NAD+ as potential candidates to treat FKRP-associated MD . Together , our results support the use of iPS cell-derived myotubes as a reliable platform for in vitro disease modeling and drug screening . Importantly , our data provide a rationale for the potential use of ribitol/NAD+ and ribose/NAD+ as therapeutics to increase α-DG functional glycosylation in patients with FKRP mutations . FKRP mutant fibroblasts obtained from a 1-year-old male patient ( Kava et al . , 2013 ) were reprogrammed into iPS cells , named FP4 , using the CytoTune-iPS 2 . 0 Sendai Reprogramming Kit ( Thermo Fisher Scientific ) using feeder-free conditions , according to the manufacturer’s instructions . FP4 iPS cells were passaged with ReLeSR ( STEMCELL Technologies ) and cultured on Matrigel-coated dishes using mTeSR1 medium ( STEMCELL Technologies ) . Newly generated and previously described WT iPS/embryonic stem ( ES ) cells ( Darabi et al . , 2012; Selvaraj et al . , 2019b ) are listed in the key resources table . Cell lines were authenticated by verification of genetic mutation by sanger sequencing . All cell lines were negative for mycoplasma contamination . Experiments were carried out according to protocols ( protocol ID 2002-37833A ) approved by the University of Minnesota Institutional Animal Care and Use Committee . NOD scid gamma ( NSG ) mice ( Jackson laboratory ) were used to perform teratoma studies . FP4 cells ( 1 . 5 × 106 ) were suspended in a 1:1 Dulbecco's Modified Eagle Medium: Nutrient Mixture F-12 ( DMEM-F12 , ThermoScientific ) and Matrigel ( Corning ) solution and injected in the quadriceps of NSG mice . The teratoma was harvested 2 months after injection . Inducible PAX7 FP4 human iPS cells were generated by lentiviral transduction of the pSAM2-PAX7-IRES-GFP and pFUGW-rtTA constructs . Inducible Pax7 WT iPS cells were generated previously ( Darabi et al . , 2012 ) , and they were maintained on Matrigel-coated flasks using mTeSR 1 ( STEMCELL Technologies ) . iPAX7-iPS cells were dissociated with Accumax ( Innovative Cell Technologies ) , and 1 × 106 cells were plated on a 6 cm non-adherent Petri dishes using mTeSR1 medium supplemented with 10 µM Y-27632 ( ROCK inhibitor ) and incubated on a shaker at 60 rpm ( day 0 ) . On day 2 , the medium was replaced with embryoid body ( EB ) differentiation medium ( 15% fetal bovine serum ( FBS ) , 10% horse serum , 1% KnockOut Serum Replacement , 1% GlutaMax , 1% penicillin-streptomycin , 50 µg/ml ascorbic acid , and 4 . 5 mM monothioglycerol in Iscove’s modified Dulbecco’s medium ) supplemented with 10 µM CHIR990217 ( GSK3 inhibitor ) . After 2 days of incubation of EBs in suspension , the medium was replaced with fresh EB differentiation medium containing 10 µM SB-431542 and 200 nM LDN-193189 . On day 5 , 1 µg/ml doxycycline was added to promote PAX7 induction . After 24 hr the media was changed with fresh ( EB ) differentiation medium with 1 µg/ml doxycycline . Day 8 EBs were collected and plated as a monolayer on gelatin-coated flasks using EB differentiation medium supplemented with 10 ng/ml human basic fibroblast growth factor and 1 µg/ml doxycycline . On day 12 , GFP+ cells ( PAX7+ myogenic progenitors ) were sorted using a FACS Aria II ( BD Biosciences ) and expanded on gelatin-coated flasks using the same medium . At 90% cell density , cells were passaged using Trypsin-EDTA ( Gibco ) and replated on new gelatin-coated flasks . Myogenic progenitors were terminally differentiated into myotubes by growing them to confluency and then switching to terminal differentiation medium , which consisted of DMEM low glucose supplemented with 2% horse serum , 1% insulin-transferrin-selenium , 1% penicillin-streptomycin , 10 µM SB-431542 , 10 µM LY-374973 , 10 µM Forskolin , and 10 µM dexamethasone ( Selvaraj et al . , 2019b ) . At this point , cultures were exposed to different treatments as follows: ribitol ( A5502 , Sigma-Aldrich ) , D- ( − ) -ribose ( R9629 , Sigma-Aldrich ) , 100 μM NAD+ ( N0632 , Sigma-Aldrich ) , and/or 100 μM sorbinil ( S7701 , Sigma-Aldrich ) . Media was replenished on day 3 of differentiation , and myotubes characterization was performed after 5–8 days of terminal differentiation . Live iPS cells were submitted to the Cytogenomics core at the University of Minnesota Masonic Cancer Center for G-band karyotype analysis . Cells were treated with colcemid for 3 hr to arrest cells , and 20 different metaphases were analyzed at a resolution of 400–450 band level . To generate the FKRP KO pluripotent stem cell line , the previously published gRNA ( CATGCGGCTCACCCGCTGCCAGG ) targeting the start codon of FKRP ( Yagi et al . , 2016 ) was cloned into pSpCas9 ( BB ) −2A-GFP ( PX458; Addgene plasmid # 48138 ) ( Ran et al . , 2013 ) . The ES cell line H9 was nucleofected using the Human Stem Cell Nucleofector Kit 1 ( Lonza ) and sorted for GFP at 48 hr post-nucleofection . ES cells were expanded , and IIH6 negative cells were sorted by FACS . The deletion was confirmed by sequencing . FKRP C318Y mutant iPS cells were generated using an HDR donor vector as previously described ( Selvaraj et al . , 2019a ) . FKRP exon 4 carrying the c . 953 G>A ( p . 318 C>Y ) mutation was cloned upstream of GFP-2A-neoR cassette ( Dhoke et al . , in prep ) . Gene editing was carried out using a ribonucleoprotein based delivery of guide RNA ( Synthego ) and Hifi Cas9 protein ( IDT ) . Following antibiotic selection , FACS purified IIH6 negative cells were expanded and subjected to single cell cloning . The FKRP vector was generated by cloning the full-length FKRP coding sequence from Dharmacon ( clone 3160297 ) into pSAM-ires-mCherry vector ( Bosnakovski et al . , 2008 ) . Plasmids were prepared using an Endofree Midiprep kit ( Nucleobond ) . Lentiviruses were produced by co-transfection of the transfer vector and the packaging constructs ( pVSV-G and pΔ8 . 74 ) into HEK 293 T cells . Transfections were performed using Lipofectamine LTX with Plus Reagent ( Invitrogen ) following manufacturer instructions . Supernatants containing the lentiviral particles were collected 36 hr after transfection and passed through a 0 . 45 μm filter . Myogenic progenitors were transduced with pSAM-ires-mCherry ( empty-LV FP4 ) or pSAM-FKRP-iresmCherry ( FKRP FP4 ) , and subsequently mCherry-positive cells were purified by FACS . IIH6 staining for FACS was performed as previously described with minor modifications ( Rojek et al . , 2007 ) . iPS cells were washed once with phosphate buffer saline ( PBS ) and then harvested using enzyme-free cell dissociation buffer ( Gibco ) following the manufacturer’s instructions . Cells were collected , centrifuged , washed with PBS , and then resuspended in PBS supplemented with 10% FBS ( PBSF ) in the presence of Fc Block ( 1 μl/million cells – BD Bioscience ) and incubated for 5 min . Staining was performed by adding 1 μl of anti-α-DG antibody IIH6C4 ( Millipore ) or normal mouse IgM ( Santa Cruz Biotechnology ) antibody per million cells followed by 20 min incubation on ice . Cells were then washed with PBS and labeled with 488- or 555-conjugated secondary antibodies ( 1:500 in FACS buffer ) for 20 min on ice in the dark . Cells were washed with PBS and filtered through a 70 μm strain to remove cell clumps , then resuspended in PBSF . Samples were sorted using a FACS Aria II ( BD Biosciences ) . Frozen cells were homogenized in Tris-Buffer Saline ( TBS , 50 mM Tris-Cl , pH 7 . 5 , 150 mM NaCl ) with 1% Triton X-100 and a cocktail of protease inhibitors ( Complete – Millipore-Sigma ) at 4°C by vortexing and then centrifuged for 30 min at 30000 g . Solubilized proteins from the supernatant were quantified with Bradford reagent ( Millipore-Sigma ) . Protein samples were prepared in Laemmli Sample Buffer ( LSB , BioRad ) . WGA pull-downs were performed using 350–600 μg of protein lysate that was loaded on 35–60 μl of WGA-bound agarose beads ( Vector Laboratories , Inc ) and incubated with end-over-end mixing at 4°C overnight . After three washes with PBS ( 150 mM NaCl , 8 mM NaH2PO4 , 42 mM Na2HPO4 , pH 7 . 5 ) with 0 . 1% Triton X-100 , bound protein was eluted with 2x LSB and incubated at 100°C for 5 min . Protein samples were separated on 4–15% using precast polyacrylamide gel ( BioRad ) by electrophoresis and then transferred to Immobilon-FL PVDFmembranes ( Millipore ) for detection with the indicated antibodies using Licor’s Odyssey Infrared Imaging System . Total protein detection using was preformed using LI-COR REVERT kit according to the manufacturer's instructions . Used antibodies are described in the key resources table . The LOA was performed as previously described with minor modifications ( Pall et al . , 1996 ) . Briefly , 20 μl of WGA purified samples were separated on 4–15% SDS-polyacrylamide gels by electrophoresis and then transferred to Immobilon-FL PVDF membranes . Transfers were blocked with PBS and 5% nonfat dry milk for 1 hr at room temperature , and then briefly rinsed with TBS and incubated for 2 hr at room temperature in TBS containing 1 mM CaCl2 , 1 mM MgCl2 ( TBSS ) , 3% bovine serum albumin ( BSA ) , and 1 mg/ml native laminin ( L2020 , Sigma ) . Transfers were washed twice for 10 min in TBSS and incubated overnight at 4°C with TBSS 3% BSA and anti-laminin ( L9393 , Sigma ) . Afterward , the membrane was washed with TBSS twice for 10 min and incubated with anti-rabbit DyLight 680 for 45 min at room temperature . Finally , membranes were washed with TBSS and visualized using Licor’s Odyssey Infrared Imaging System . As a negative control , TBSS without 1 mM CaCl2 was used during incubation and washes . Ribitol-5-phosphate and CDP-ribitol were synthesized by Z Biotech ( Aurora , CO ) . Myogenic progenitors were serum-starved after changing to differentiation medium only or supplemented with ribitol , ribose , ribitol/NAD+ , ribose/NAD+ , or NAD+ for 5 days , washed with cold PBS three times and harvested by scrapping the cells . In a blinded manner , samples were subjected to the following procedures . Cells were homogenized with 300 μl of MeOH:acetonitrile ( 1:1 ) and then centrifugated for 5 min at 11 , 000 rpm . The supernatants were removed , transferred to individual wells of 96-well plate , and analyzed by LC/MS-MS . An Applied Biosystems Sciex 4000 ( Applied Biosystems , Foster City , CA ) equipped with a Shimadzu HPLC ( Shimadzu Scientific Instruments , Inc , Columbia , MD ) and Auto-sampler ( LEAP Technologies , Carrboro , NC ) were used to detect ribitol , ribose , ribitol-5-P , and CDP-ribitol . The analysis of metabolites was performed by Z Biotech as described previously ( Cataldi et al . , 2018 ) . Modeling of FKRP with its CDP-ribitol and M3 substrates ( PDB: 6KAM ) ( Kuwabara et al . , 2020 ) was carried out using the Schrodinger modeling suite package ( Schrödinger Release 2018-4 , 2018 ) . All missing side chains and hydrogens atoms were added according to the default protein preparation protocol at pH 7 . 0 , followed by energy minimization using OPLS2005 force field ( Jorgensen et al . , 1996 ) to optimize all hydrogen-bonding networks . The crystallographic Ba2+ ion was replaced by its native Mg2+ ion . The relative change in substrate-binding free energy due to the effect of mutation , ΔΔGbind ( F473C ) , was performed based on the molecular mechanics generalized Born solvent accessible method ( Still et al . , 1990 ) . It is evaluated as the difference in the protein stability between the unbound and bound states of FKRP and its F473C mutant . Immunofluorescence staining was performed by fixing cells with 4% paraformaldehyde in PBS for 10 min at 4°C , followed by permeabilization with 0 . 1% Triton in PBS and blocking with 3% BSA in PBS , before incubation with the primary antibodies . Samples were rinsed with PBS , blocked with 3% BSA in PBS , and then incubated with DAPI and respective secondary antibodies . Antibodies used in this study are described in the key resources table . Samples were collected with TRIzol Reagent ( Invitrogen ) , and RNA was purified using a Direct-zol RNA Miniprep Plus Kit ( Zymo Research ) . Purified RNA was quantified with NanoDrop 2000 ( Thermo Fisher Scientific ) and retrotranscribed using SuperScript VILO cDNA Synthesis Kit ( Invitrogen ) following the manufacturer’s instructions . Gene expression analyses were performed using the cDNA corresponding to 12 . 5 ng of starting RNA for each reaction . The RT-qPCR analysis was performed using TaqMan Universal PCR Master Mix and TaqMan probes ( Applied Biosystems ) . For comparisons of two independent samples , we used the unpaired or paired Student’s t test . For comparisons of multiple groups , we used the two-way ANOVA followed by the Tukey’s multiple comparisons test . The one-way ANOVA followed by the Sidak’s multiple comparisons test was used when measuring one variable . p-values < 0 . 05 were considered significant . Statistical comparisons were performed using GraphPad Prism software .	Healthy muscles are complex machines that require a myriad of finely tuned molecules to work properly . For instance , a protein called alpha-DG sits at the surface of healthy muscle cells , where it strengthens the tissue by latching onto other proteins in the environment . To perform its role correctly , it first needs to be coated with sugar molecules , a complex process which requires over 20 proteins , including the enzyme FKRP . Faulty forms of FKRP reduce the number of sugars added to alpha-DG , causing the muscle tissue to weaken and waste away , potentially leading to severe forms of diseases known as muscular dystrophies . Drugs that can restore alpha-DG sugar molecules could help to treat these conditions . Previous studies on mice and fish have highlighted two potential candidates , known as ribitol and NAD+ , which can help to compensate for reduced FKRP activity and allow sugars to be added to alpha-DG again . Yet no model is available to test these molecules on actual human muscle cells . Here , Ortiz-Cordero et al . developed such a model in the laboratory by growing muscle cells from naïve , undifferentiated cells generated from skin given by a muscular dystrophy patient with a faulty form of FKRP . The resulting muscle fibers are in essence identical to the ones present in the individual . As such , they can help to understand the effect various drugs have on muscular dystrophies . The cells were then put in contact with either NAD+ , ribitol , or a precursor of ribitol known as ribose . Ortiz-Cordero et al . found that ribitol and ribose restored the ability of FKRP to add sugars to alpha-DG , reducing muscle damage . Combining NAD+ with ribitol or ribose had an even a bigger impact , further increasing the number of sugars on alpha-DG . The human muscle cell model developed by Ortiz-Cordero et al . could help to identify new compounds that can treat muscular conditions . In particular , the findings point towards NAD+ , ribose and ribitol as candidates for treating FKRP-related muscular dystrophies . Further safety studies are now needed to evaluate whether these compounds could be used in patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "cell", "biology" ]	2021	NAD+ enhances ribitol and ribose rescue of α-dystroglycan functional glycosylation in human FKRP-mutant myotubes
Past experiments with reconstituted proteoliposomes , employing assays that infer membrane fusion from fluorescent lipid dequenching , have suggested that vacuolar SNAREs alone suffice to catalyze membrane fusion in vitro . While we could replicate these results , we detected very little fusion with the more rigorous assay of lumenal compartment mixing . Exploring the discrepancies between lipid-dequenching and content-mixing assays , we surprisingly found that the disposition of the fluorescent lipids with respect to SNAREs had a striking effect . Without other proteins , the association of SNAREs in trans causes lipid dequenching that cannot be ascribed to fusion or hemifusion . Tethering of the SNARE-bearing proteoliposomes was required for efficient lumenal compartment mixing . While the physiological HOPS tethering complex caused a few-fold increase of trans-SNARE association , the rate of content mixing increased more than 100-fold . Thus tethering has a role in promoting membrane fusion that extends beyond simply increasing the amount of total trans-SNARE complex . A hallmark of eukaryotic cells is their internal organization by distinct , highly specialized , membrane-enclosed compartments . Vesicular traffic functionally connects these individual compartments while preserving their integrity . The constant vectorial exchange between organelles is ensured by a complex machinery that facilitates the formation of vesicles with selective cargo , vesicle movement , specific tethering of each vesicle at its correct target membrane , and fusion of the vesicle and organelle membranes . Selective tethering and fusion are sequential events , but it remains unclear how tightly they are coupled . Tethering is commonly performed by specific proteins , often in large multi-subunit complexes that bind Rab family GTPases ( Stenmark , 2009; Yu and Hughson , 2010 ) . Fusion is mediated by soluble N-ethylmaleimide-sensitive attachment protein receptor ( SNARE ) proteins ( Jahn and Scheller , 2006 ) and Sec1/Munc18 ( SM ) proteins ( Rizo and Sudhof , 2012 ) . We study membrane fusion with the vacuoles ( lysosomes ) of Saccharomyces cerevisiae ( Wickner , 2010 ) . The vacuole is a large organelle that continuously undergoes fission and homotypic fusion with other vacuoles . At steady state , wild-type strains of yeast have just a few large vacuoles , but mutations which block vacuole fusion allow unimpeded fission , resulting in cells with numerous small vacuoles . This phenotype allowed the identification of the genes encoding vacuole fusion proteins ( Wada et al . , 1992 ) . Vacuoles are readily isolable ( Bankaitis et al . , 1986 ) , and a simple colorimetric assay of their fusion ( Haas et al . , 1994 ) permitted definition of the consecutive stages of fusion and the catalysts of each stage ( Ostrowicz et al . , 2008 ) . Vacuole tethering requires the Rab family GTPase Ypt7p and the heterohexameric Rab effector complex HOPS ( homotypic vacuole fusion and protein sorting ) , which is comprised of Vps11p , Vps16p , Vps18p , Vps33p , Vps39p , and Vps41p ( Seals et al . , 2000; Wurmser et al . , 2000 ) . Two of these subunits , Vps39p and Vps41p , have direct affinity for Ypt7p , suggesting that one HOPS complex could tether two vacuoles , each bearing Ypt7p ( Brett et al . , 2008; Ostrowicz et al . , 2010 ) . Fusion of tethered membranes requires vacuolar SNAREs . Each vacuole has four SNAREs that are required for homotypic fusion: the R-SNARE Nyv1p , the Qa-SNARE Vam3p , the Qb-SNARE Vti1p , and the Qc-SNARE Vam7p ( Nichols et al . , 1997; Ungermann et al . , 1999 ) . Vam7p has the unusual property of lacking a hydrophobic membrane anchor while having an N-terminal membrane targeting domain with affinity for phosphatidylinositol 3-phosphate , HOPS , and acidic lipid ( Cheever et al . , 2001; Lee et al . , 2006; Stroupe et al . , 2006; Karunakaran and Wickner , 2013 ) . Vacuolar cis-SNARE complexes are disassembled by Sec17p and Sec18p in an ATP-dependent manner , allowing the SNAREs to reassemble into trans-complexes , which are essential for fusion ( Haas and Wickner , 1996; Mayer et al . , 1996 ) . HOPS fulfills several functions: Rab-dependent tethering ( Hickey and Wickner , 2010 ) , catalysis of the assembly of Vam7p into trans-SNARE complexes ( Zick and Wickner , 2013 ) , and protecting trans-SNARE complexes from Sec17p/Sec18p-mediated disassembly ( Xu et al . , 2010 ) . The reconstitution of vacuolar fusion in vitro has provided an important complementary approach for examining the individual steps of membrane fusion . In an early study , vacuole detergent extracts were reconstituted into proteoliposomes that could fuse with the purified organelle . Both the SNARE complex and vacuolar Rab GTPase Ypt7p were shown to be essential components of such detergent extracts ( Sato and Wickner , 1998 ) . Shortly thereafter , it was demonstrated that proteoliposomes bearing the vacuolar R-SNARE and two fluorescent lipids that exhibit fluorescence quenching , NBD-PE and rhodamine-PE , will interact with non-fluorescent proteoliposomes bearing the three Q-SNAREs to give fluorescence dequenching ( Fukuda et al . , 2000 ) . This finding has since been confirmed in several additional studies ( Mima et al . , 2008; Izawa et al . , 2012; Furukawa and Mima , 2014 ) . A very recent study ( Furukawa and Mima , 2014 ) has shown that vacuolar SNAREs are apparently unique among the yeast SNAREs , in that the purified SNAREs of the other yeast organelles will not mediate the dequenching of fluorescent lipids on proteoliposomes without additional proteins or tethering factors . In striking contrast , when fusion was assayed by the mixing of lumenally entrapped proteins while they remain inaccessible to an external competitor ( Zucchi and Zick , 2011 ) , very little fusion was seen without a tethering agent ( Zick and Wickner , 2013 ) . Thus , there are apparently contradictory reports as to whether SNAREs alone can efficiently mediate each stage of the fusion process . We now report that vacuolar SNAREs alone will only support membrane fusion with an extremely low efficiency . Strikingly , fluorescent lipid dequenching can apparently occur due to trans-SNARE interactions that do not necessarily go on to fusion , emphasizing the pivotal importance of content mixing experiments in membrane fusion studies . A tethering/docking step is strictly required for rapid fusion , whether by the main physiological tethering system of Ypt7p and HOPS or by a vacuole-specific tethering mechanism mediated by the interaction of Vam7p's PX domain and PI ( 3 ) P in trans . Does the dequenching assay actually measure symmetric lipid mixing between the two proteoliposome populations ? Proteoliposomes were prepared with both MB-PE and NBD-PE on the R-SNARE proteoliposomes or on the 3Q-SNARE proteoliposomes . Strikingly , the dequenching signal was seen when the fluorescent lipids were both on the R-SNARE proteoliposomes , but not when they were on the 3Q-SNARE proteoliposomes ( Figure 2A , dark gray bars ) . For these direct comparisons of dequenching when the fluorophores are on one or the other proteoliposome , the proteoliposomes were mixed at a 1:1 ratio ( Figure 2A ) . As in most reported dequenching assays , increasing the proportion of non-fluorescent acceptor 3Q-SNARE proteoliposomes enhances the signal such that up to 60% of maximal dequenching is achieved in 10 min ( Figure 2B , light gray columns ) , whereas little dequenching of fluorescent 3Q-SNARE proteoliposomes is seen when incubated with even a 16-fold excess of non-fluorescent 1R-proteoliposomes ( Figure 2B , dark gray columns ) . Providing an excess of 3Q-RPLs , but not 1R-RPLs , also promoted content mixing ( see Figure 5—figure supplement 2 ) beyond what was seen at a 1:1 ratio of both RPLs , suggesting that only a fraction of the 3Q-RPLs are fusion competent . Dequenching does require trans-SNARE interactions: it is blocked by antibody to Vam3p , by recombinant soluble domain of Nyv1p , or by Sec17p/Sec18p ( Figure 2C , light gray columns ) , which are reported ( Xu et al . , 2010 ) to disassemble trans-SNARE complexes . Each of the 14 compositions of SNAREs other than 3Q on the acceptor , non-fluorescent proteoliposomes did not support dequenching ( Figure 2C , the average and standard deviation of these is shown for simplicity as a dark gray column ) . Thus , proteoliposomes can undergo trans-SNARE interactions that lead to asymmetric lipid dequenching without symmetrical lipid mixing or proportionate fusion . 10 . 7554/eLife . 03251 . 006Figure 2 . Asymmetric lipid dequenching . ( A ) Lipid dequenching reactions of 1R- and 3Q-RPLs ( 250 µM lipid each ) , with both fluorescent lipids ( Marina-Blue-phosphatidylethanolamine [MB-PE] and NBD-PE , 1 . 5 mol% each ) on either the 1R-RPLs or the 3Q-RPLs . Incubations contained either Sec17p ( 600 nM ) and Sec18p ( 200 nM ) ( light gray ) or no additional proteins ( dark gray ) . ( B ) Lipid dequenching reactions of MB/NBD-labeled 1R- ( light gray ) or 3Q-RPLs ( dark gray ) with increasing amounts of non-labeled complementary RPLs ( 3Q or 1R; 1:1 corresponds to 50 µM lipid of each ) . No additional proteins were added . ( C ) Sensitivity of lipid dequenching reactions of 1R ( MB/NBD ) -RPLs and 3Q-RPLs to various inhibitors that might prevent trans-SNARE complex formation . Incubations received: ( 1 ) no addition , ( 2 ) Sec17p/Sec18p ( 400 nM each ) , ( 3 ) antibodies directed against Vam3p ( 1 µM ) , or ( 4 ) soluble Nyv1p ( 1-231 ) ( 1 µM ) . Column 5: 1R ( MB/NBD ) -RPLs were incubated with non-fluorescent RPLs bearing various combinations of SNAREs other than 3Q: QabcR , QbcR , QacR , QabR , QcR , QbR , Qbc , QaR , Qac , Qab , R , Qc , Qb , Qa . The extent of these reactions was comparably low for all conditions , and is shown as average for these 14 conditions with standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 00610 . 7554/eLife . 03251 . 007Figure 2—figure supplement 1 . Representative kinetic data for panels A–C in figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 007 Does the presence of the lipidic probes on the R-SNARE or 3Q-SNARE proteoliposomes affect the actual fusion reaction , as measured by protected lumenal content mixing ? The asymmetry of lipid dequenching , controlled by the distribution of the probes ( Figure 3A ) , is not reflected in asymmetry of fusion per se in the very same incubations ( Figure 3B ) . Even lipid quenching assays , in which MB-PE and NBD-PE are initially incorporated separately into R-SNARE or 3Q-SNARE proteoliposomes , show an asymmetry in quenching ( Figure 3C ) that is not reflected in fusion per se ( Figure 3D ) . This asymmetry could not be attributed to differences in RPL size ( Figure 3—figure supplement 2 ) , or fluorophore or protein incorporation ( Figure 3—figure supplement 3 ) . 10 . 7554/eLife . 03251 . 008Figure 3 . Discrepancies between lipid mixing and content mixing assays in 1R-3Q RPL fusion reactions . Incubations contained Sec17p ( 600 nM ) , Sec18p ( 200 nM ) , and HOPS ( 100 nM ) , as indicated . ( A and B ) Incubations contained 1R-RPLs ( containing Biotin-PhycoE ) and 3Q-RPLs ( containing Streptavidin-Cy5 ) ( 250 µM lipid each ) , with the two fluorescent lipids Marina-Blue-phosphatidylethanolamine ( MB-PE ) and NBD-PE ( 1 . 5 mol% each ) being together on either the 1R-RPLs or the 3Q-RPLs . Lipid dequenching ( A ) and protected lumenal content mixing ( B ) were recorded from the same reactions . ( C and D ) Incubations bore 1R-RPLs ( containing Biotin-PhycoE ) and 3Q-RPLs ( containing Streptavidin-Cy5 ) ( 250 µM lipid each ) , with each RPL bearing a single fluorescent lipid , either Marina-Blue-phosphatidylethanolamine ( MB-PE ) ( 1 mol% ) or NBD-PE ( 3 mol% ) . Lipid quenching ( C ) and protected lumenal content mixing ( D ) were recorded from the same reactions . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 00810 . 7554/eLife . 03251 . 009Figure 3—figure supplement 1 . Representative kinetic data for panels A–D in figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 00910 . 7554/eLife . 03251 . 010Figure 3—figure supplement 2 . Characterization of liposome size by dynamic light scattering . ( A–H ) The size distribution of various proteoliposome preparations was analyzed by dynamic light scattering with a Zetasizer nano ZS ( Malvern Instruments , Westborough , MA ) through non-invasive back-scatter at 173° . Samples of 400 μl at a lipid concentration of 20 μM were measured in low volume disposable sizing cuvettes at 25°C . The Z averages ( r . nm ) were: 101 . 5 for 1R ( MB/NBD ) ( A ) , 90 . 63 for 1R ( ) ( B ) , 99 . 18 for 1R ( MB ) ( C ) , 91 . 07 for 1R ( NBD ) ( D ) , 89 . 78 for 3Q ( ) ( E ) , 95 . 36 for 3Q ( MB/NBD ) ( F ) , 86 . 93 for 3Q ( NBD ) ( G ) , and 89 . 40 for 3Q ( MB ) ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 01010 . 7554/eLife . 03251 . 011Figure 3—figure supplement 3 . Characterization of dye and protein incorporation . ( A ) Samples ( 20 µl ) of 1R ( MB/NBD ) and 3Q ( MB/NBD ) RPLs ( 25 µM lipid ) were incubated for 30 min at 27°C in the presence or absence of Thesit ( 0 . 2% [wt/vol] ) in 384-well plates in a SpectraMax Gemini XPS , and the fluorescence of Marina-Blue ( Ex: 370 nm , Em: 465 nm , cutoff: 420 nm ) and NBD ( Ex: 460 nm , Em: 538 nm , cutoff: 515 nm ) was recorded . The average of three repeats ± standard deviations is shown . ( B ) RPLs of various composition were analyzed for their protein composition by SDS-PAGE and Colloidal Coomassie staining . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 011 Might dequenching that does not seem to represent fusion reflect a peculiar physical property of one or the other fluorophore ? 3Q- and 1R-proteoliposomes were prepared with three sets of fluorescent PE probes , either MB and NBD , NBD and rhodamine , or Oregon-Green and Texas-Red . In the absence of HOPS or Sec17p/Sec18p , the same asymmetric dequenching signal was seen in each case ( Figure 4A–C , no addition ) . We do not know how trans-SNARE interactions lead to this dequenching , but it is not symmetric , as expected for true lipid mixing , and is not proportionate to the content mixing for each reaction condition; thus , it cannot be relied on as a faithful reporter of fusion . 10 . 7554/eLife . 03251 . 012Figure 4 . Asymmetric lipid dequenching is seen with different pairs of fluorescent lipidic markers . ( A–C ) Lipid dequenching reactions of 1R- and 3Q-RPLs ( 250 µM lipid each ) had combinations of Sec17p ( 600 nM ) , Sec18p ( 200 nM ) , and HOPS ( 100 nM ) , as indicated . Either the 1R-RPLs ( light gray ) or the 3Q-RPLs ( dark gray ) bore pairs of fluorescently labeled lipids ( 1 . 5 mol% each ) : Marina-Blue ( MB ) and NBD ( A ) , NBD and rhodamine ( Rh ) ( B ) , or Oregon-Green ( OG ) and Texas-Red ( TR ) ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 01210 . 7554/eLife . 03251 . 013Figure 4—figure supplement 1 . Representative kinetic data for panels A–C in figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 013 While there was far less protected lumenal content mixing between 3Q- and 1R-SNARE proteoliposomes in the absence of HOPS , Sec17p , and Sec18p than in their presence ( Figure 1G , columns 1 vs 8 ) , even this low level ( column 1 ) was suppressed by Sec17p/Sec18p ( column 5 ) and thus presumably relied on the SNAREs . Was this fusion independent of tethering , that is , did it rely on direct formation of stable proteoliposome adherence through 4-SNARE coiled-coil domain assembly , or did another kind of interaction mediate tethering ? The Vam7p N-terminal PX domain has direct affinity for PI ( 3 ) P ( Cheever et al . , 2001 ) , which is important for fusion ( Fratti and Wickner , 2007 ) . 3Q- and 1R-proteoliposomes were prepared with or without PI ( 3 ) P . They were incubated in each combination without further protein addition , with HOPS , with Sec17p/Sec18p , or with HOPS/Sec17p/Sec18p , and their protected lumenal compartment mixing was assayed ( Figure 5A ) . Strikingly , with PI ( 3 ) P only on the R-SNARE proteoliposomes , fusion was as vigorous without HOPS as in its presence ( Figure 5 , columns 5 and 7 ) . This fusion was blocked by Sec17p/Sec18p ( column 6 ) and was not seen when the PI ( 3 ) P was entirely absent ( column 1 ) or when PI ( 3 ) P was only on the 3Q-SNARE proteoliposomes ( column 9 ) . The content mixing that was seen when PI ( 3 ) P was on the R-SNARE proteoliposomes alone was strongly suppressed when PI ( 3 ) P was present on the 3Q-RPLs as well ( columns 5 vs 13 ) , presumably because the PX domain is engaged in an interaction with PI ( 3 ) P in cis , rendering a trans interaction less likely . The delay in reaction kinetics for conditions with Sec17 , Sec18 , and HOPS and no PI ( 3 ) P on the 3Q-RPLs ( Figure 5—figure supplement 1A , B , purple lines ) probably reflects the requirement for phosphoinositides for the synergy between Sec17/Sec18 and HOPS ( Mima and Wickner , 2009 ) . The fusion that is supported by the presence of PI ( 3 ) P on the R-SNARE proteoliposomes can be ascribed to its binding in trans by the PX domain of Vam7p , as 3Q-SNARE proteoliposomes which were made with Vam7pΔPX will not fuse with R-SNARE proteoliposomes which bear PI ( 3 ) P ( column 17 ) . After 10 min assays , the very low lumenal compartment mixing values observed with the RPLs alone were not significantly above those seen with inhibition by Sec17p and Sec18p ( Figure 5A , lower panel , columns 1 and 2 ) . To determine whether SNARE-bearing RPLs alone can undergo any full fusion at all on their own , we performed an extended fusion assay with 3Q- and 1R-proteoliposomes without PI ( 3 ) P in the presence or absence of added HOPS ( Figure 5B ) . Detectable fusion could in fact be supported by SNAREs alone ( blue ) , albeit at less than a 100th the rate as seen in the presence of HOPS ( red ) . 10 . 7554/eLife . 03251 . 014Figure 5 . Modulation of fusion activity through asymmetric distribution of PI ( 3 ) P . ( A ) 1R-RPLs ( with 0 . 5 mol% Marina-Blue-phosphatidylethanolamine [PE] , containing biotinylated R-phycoerythrin ) and 3Q-RPLs ( with 3 mol% NBD-PE , containing Cy5-labeled streptavidin ) , bearing PI ( 3 ) P as indicated , were assayed for protected lumenal compartment mixing . Where indicated ( columns 17–20 ) , 3Q-RPLs bore a truncated version of the Qc-SNARE Vam7p that lacks the PI ( 3 ) P binding PX domain ( Xu and Wickner , 2012 ) . Reactions contained Sec17p ( 600 nM ) , Sec18p ( 200 nM ) , and HOPS ( 100 nM ) , as indicated . The lower panel provides a magnified view of the bottom 5% of the data ( indicated by dotted box ) . ( B ) Kinetics of 1R-3Q content mixing assays with RPLs not bearing PI ( 3 ) P . Reactions contained HOPS ( 100 nM; red curve ) , Sec17p and Sec18p ( 600 and 200 nM; gray curve ) , or no added proteins ( blue curve ) . The averages of five experiments are shown . The reactions containing Sec17p/Sec18p were normalized to zero , and these values were subtracted from all other conditions . Light-colored areas indicate standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 01410 . 7554/eLife . 03251 . 015Figure 5—figure supplement 1 . Representative kinetic data for panel A in figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 01510 . 7554/eLife . 03251 . 016Figure 5—figure supplement 2 . Effect of varying the ratio of 1R to 3Q RPLs . Protected lumenal compartment mixing was measured as the decrease in fluorescence intensity of Biotin-PhycoE when it binds to Streptavidin-Cy5 . Incubations either contained Sec17p ( 600 nM ) and Sec18p ( 200 nM ) ( light gray ) or no additional proteins ( dark gray ) . All 1R-RPLs contained PI ( 3 ) P ( 1 mol% ) , while all 3Q-RPLs did not . ( A ) Content mixing reactions of 1R-RPLs ( 50 µM lipid; containing the lumenal marker Biotin-PhycoE ) and 3Q-RPLs ( 0 , 50 , 150 , or 450 µM lipid; containing the lumenal marker Streptavidin-Cy5 ) . ( B ) Content mixing reactions of 3Q-RPLs ( 50 µM lipid; containing the lumenal marker Biotin-PhycoE ) and 1R-RPLs ( 0 , 50 , 150 , or 450 µM lipid; containing the lumenal marker Streptavidin-Cy5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 016 Does this dramatic difference in fusion rates reflect a proportionate inability of the three Q-SNAREs on proteoliposomes to engage with R-SNAREs in the absence of a tether ? To study how much trans-interaction occurs between membrane-bound 3Q-SNAREs and membrane-bound R-SNAREs in the presence or absence of HOPS , we performed a content mixing reaction ( Figure 6A ) and also subjected samples which had been detergent-solubilized to immunoprecipitation with anti-Vam3p antibodies to determine the amount of Nyv1p that co-precipitated as a measure of trans-SNARE interaction ( Figure 6B ) . The detergent bore a large excess of GST-Nyv1p , sufficient to suppress any association between the 3Q-SNAREs and wild-type Nyv1p after detergent-solubilization ( Figure 6 , Figure 6—figure supplement 1 ) . The presence of HOPS led to an increase in the trans-association of membrane anchored SNAREs ( Figure 6B , compare lanes 3 and 4 ) . However , this increase was less than fourfold , and thus insufficient to explain the difference in content mixing rates ( Figure 6A ) . Quantitation of replicate experiments ( Figure 6C ) shows that about one third as much trans-SNARE complex formed in the absence of HOPS ( column 3 ) as in its presence ( column 4 ) , whereas fusion was over 100 times faster in the presence of HOPS ( columns 1 , 2 ) . Might this reflect an inability of the 3Q-SNAREs to interact with Nyv1p without HOPS ? When incubations bore the recombinant soluble domain of Nyv1p at the same 100 nM concentration as was otherwise present on RPLs in our assay , the soluble Nyv1p readily bound to the Q-SNAREs regardless of the presence or absence of HOPS ( Figure 6B , compare lanes 7 and 8 ) . Furthermore , 100 nM soluble Nyv1p can suppress most of the content mixing reaction ( Figure 6D ) , just as fusion was blocked by the mixed soluble domains of the three Q-SNAREs ( Figure 6—figure supplement 3 ) , and inhibition by added soluble domain of Nyv1p was reversible by including the SNARE complex disassembly chaperones Sec17/Sec18 ( Figure 6E ) . This indicates that the 3Q-SNARE complexes were capable of engaging in complex formation with membrane-bound R-SNAREs , but in a manner that did not lead to fusion . To test whether the highly disproportional increase in trans-SNARE association and content mixing can be attributed to a possible requirement for high cooperativity of SNARE complexes , we examined the levels of content mixing and trans-SNARE interactions when R-RPLs were used that had been reconstituted with lower amounts of Nyv1p ( Figure 7 ) . Lower levels of Nyv1p did not lead to a drastic diminution of fusion activity ( Figure 7A , C ) , but resulted in lower amounts of trans-SNARE association ( Figure 7B , D ) . Since the amount of trans-SNARE interaction that was detected at low Nyv1p levels with HOPS was lower than the amount seen with high Nyv1p levels without HOPS ( Figure 7B , D , lane 2 vs lane 7 ) , yet the amount of fusion was drastically higher in the sample containing HOPS ( Figure 7A , C , lane 2 vs lane 7 ) , the requirement for high cooperativity can be ruled out as an explanation for the observed discrepancy between increases in trans-SNARE interaction and fusion . 10 . 7554/eLife . 03251 . 017Figure 6 . HOPS increases the rate of content mixing substantially more than the formation of trans-SNARE associations . ( A ) Kinetics of 1R-3Q content mixing assays . RPLs ( reconstituted with a protein:lipid ratio of 1:2500 for SNAREs ) were incubated for 30 min at 27°C before and after the addition of HOPS ( 100 nM ) or its buffer . ( B ) Additional 40 µl reactions that were performed in parallel were analyzed for trans-SNARE association by co-immunoprecipitation of Nyv1p ( R-SNARE ) with Vam3p ( Qa-SNARE ) . Lanes 3 and 4 correspond to the conditions as shown in ( A ) . Lanes 7 and 8 are from equivalent reactions , but containing soluble Nyv1p instead of R-RPLs . For lanes 1 , 2 , 5 , and 6 , the components were mixed directly in detergent solution without prior incubation . ( C ) Quantification of content mixing and trans-SNARE association ( as seen in panels A and B ) from three independent repeat experiments . The rate of fusion was determined as the slope for the first minute of the content mixing reaction after addition of HOPS or its buffer ( between minutes 30 and 31 ) . Trans-SNARE interactions were quantified as the amount of Nyv1p that co-precipitated with Vam3p ( lanes 3 and 4 in panel B ) using UN-SCAN-IT gel 5 . 3 software ( Silk Scientific , Orem , UT ) . ( D ) 1R- and 3Q-RPLs were mixed with increasing concentrations of soluble Nyv1p ( sR ) , and incubated for 30 min at 27°C . HOPS ( 100 nM ) was added , and the reactions were assayed for protected lumenal compartment mixing . ( E ) A mixture of 1R- and 3Q-RPLs was pre-incubated with or without 1 µM soluble Nyv1p for 30 min at 27°C . Sec17p ( 600 nM ) , Sec18p ( 200 nM ) , and HOPS ( 100 nM ) or their respective buffers were added as indicated , and the reactions were assayed for protected lumenal compartment mixing . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 01710 . 7554/eLife . 03251 . 018Figure 6—figure supplement 1 . Suppression of the formation of 1R-3Q SNARE complexes in detergent by increasing concentrations of GST-tagged Nyv1p . The 40 µl fusion reactions were incubated at 27°C as indicated , and analyzed for trans-SNARE association by co-immunoprecipitation of Nyv1p ( R-SNARE ) with Vam3p ( Qa-SNARE ) . The co-immunoprecipitation was carried out in the presence of increasing concentrations of GST-Nyv1p . Lanes 1 and 2 contained only R- or 3Q-RPLs . Lanes 3-6 contained R- and 3Q-RPLs . Lanes 7 and 8 contained 3Q-RPLs and soluble Nyv1p . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 01810 . 7554/eLife . 03251 . 019Figure 6—figure supplement 2 . Representative kinetic data for panels D–E in figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 01910 . 7554/eLife . 03251 . 020Figure 6—figure supplement 3 . Reversible inhibition by soluble Q-SNAREs . ( A ) 1R- and 3Q-RPLs were mixed with increasing concentrations of soluble Vam3p , Vti1p , and Vam7p , and incubated for 30 min at 27°C . HOPS ( 100 nM ) was added , and the reactions were assayed for protected lumenal compartment mixing . ( B ) A mixture of 1R- and 3Q-RPLs was pre-incubated with or without 4 µM soluble Vam3p , Vti1p , and Vam7p for 30 min at 27°C . Sec17p ( 600 nM ) , Sec18p ( 200 nM ) , and HOPS ( 100 nM ) or their respective buffers were added as indicated , and the reactions were assayed for protected lumenal compartment mixing . ( C ) Reactions as in ( A ) , but with increasing concentrations of either soluble Vam3p , or soluble Vti1p , or Vam7p individually . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 02010 . 7554/eLife . 03251 . 021Figure 7 . High cooperativity does not explain the disproportion between trans-SNARE interaction and fusion rate resulting from HOPS addition . ( A ) Kinetics of content mixing reactions of 3Q-RPLs ( reconstituted with a protein:lipid ratio of 1:2000 for Vam3p , Vti1p , and Vam7p ) and 1R-RPLs ( reconstituted with a protein:lipid ratio of either 1:2000 [blue] , 1:4000 [red] , 1:8000 [green] , or 1:16 , 000 [purple] for Nyv1p ) . RPLs were incubated for 30 min at 27°C before and HOPS ( 100 nM ) or its buffer were then added . ( B ) Equivalent 40 µl reactions were analyzed for trans-SNARE association by co-immunoprecipitation of Nyv1p ( R-SNARE ) with Vam3p ( Qa-SNARE ) . Lanes 1–8 correspond to the conditions as shown in ( A ) with odd numbers representing conditions with no addition and even numbers representing conditions with HOPS ( 100 nM ) . For lanes a , b , c , and d , the components were mixed directly in detergent solution without prior incubation . ( C ) Quantification of content mixing ( as seen in panel A ) from three independent repeat experiments . The rate of fusion was determined as the slope for the first 10 min of the content mixing reaction ( between minutes 0 and 10 ) for samples that did not receive HOPS or as the slope for the first minute of the content mixing reaction after addition of HOPS ( between minutes 30 and 31 ) . ( D ) Quantification of trans-SNARE association ( as seen in panel B ) . Trans-SNARE interactions were quantified as the amount of Nyv1p that co-precipitated with Vam3p ( lanes 1–8 in panel B ) using UN-SCAN-IT gel 5 . 3 software ( Silk Scientific , Orem , UT ) , and are represented as percentage of the amount of Nyv1p that co-precipitated in the reaction that contained HOPS and R-RPLs with a SNARE:lipid ratio of 1:2000 ( lane 8 in panel B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03251 . 021 Our knowledge of the initial steps of membrane fusion is limited , and the terminology to describe these processes is vague . ‘Tethering’ often refers to trans-interactions of membranes that are of low affinity , reversible , or do not involve SNAREs , and ‘docking’ is reached when trans-SNARE complexes have formed . These terms are used loosely , even interchangeably , and it has remained unclear whether and how the functions of Rab GTPases , their effectors , SM proteins , and SNAREs are coupled or even related . The steps that lead to trans-SNARE complex formation are dynamic and interwoven , and we are only beginning to understand how they lead to membrane fusion . Several biochemical assays have been developed to monitor the fusion process in vitro . Most studies with isolated organelles have assayed fusion as the mixing of lumenal contents . Studies with reconstituted proteoliposomes , in contrast , have relied heavily on lipid dequenching assays to infer the merging of two lipid bilayers . Discrepancies of lipid mixing and content mixing have repeatedly been noted in the past: for example , neuronal SNAREs alone do not produce much content mixing yet readily promote lipid dequenching ( Kyoung et al . , 2011; Diao et al . , 2012 ) , content mixing occurs seconds after initial lipid mixing in influenza virus-induced fusion ( Floyd et al . , 2008 ) , and content mixing occurs with a delay of several minutes after lipid mixing in vacuolar fusion ( Jun and Wickner , 2007 ) . We have recently described an assay that allows concurrent measurement of both lipid and content mixing ( Zucchi and Zick , 2011 ) . We have reconstituted a fusion reaction that mimics the homotypic fusion of yeast vacuoles , that is , with the four SNAREs in a cis-complex on both membranes , depending on Sec17p/Sec18p for ATP-dependent priming and Ypt7p:HOPS for tethering ( see Wickner , 2010 for review ) . Both content and lipid mixing assays show that this reaction requires Sec17p , Sec18p , and HOPS ( Figure 1C , D ) . A modified reaction scheme , in which one population of RPLs bears the R-SNARE Nyv1p , while the other population bears the three Q-SNAREs Vam3p , Vti1p , and Vam7p , allows study of a sub-reaction that bypasses the need for Sec17p/Sec18p mediated priming . This condition has repeatedly been reported to allow lipid mixing to occur without the need for any additional components ( Fukuda et al . , 2000; Mima et al . , 2008 ) . While we could reproduce this finding ( Figure 1F , column 1 ) , we observed a substantially lower amount of content mixing for the same condition ( Figure 1G , column 1 ) . Furthermore , among the several sets of SNAREs of yeast organelles , only the vacuolar SNAREs showed a purported fusion signal ( Izawa et al . , 2012; Furukawa and Mima , 2014 ) . Lipid mixing without proportional content mixing may indicate hemifusion ( see Figure 1A ) . However , the lipid dequenching signal showed an asymmetric character , depending on the two fluorescent lipids being present on the R-SNARE-bearing RPLs ( Figure 2A , B ) . It is nevertheless possible to suppress this signal with well-established inhibitors of fusion that interfere with the formation of trans-SNARE complexes ( Figure 2C ) . The fluorescent lipids did not appear to modulate the capacity of RPLs to fuse , as content mixing was not affected by the presence or distribution of fluorescent lipids ( Figure 3 ) . The effect was also not limited to a specific set of fluorescent lipids ( Figure 4 ) . We cannot explain this asymmetric behavior , but it may be related to an undefined interaction between vacuolar SNAREs and the fluorescent lipids . The latter might be selectively transferred to another membrane that contains a particular SNARE via an interaction with that SNARE; transfer of lipid dyes between membranes has been reported for rhodamine B lipids ( Ohki et al . , 1998 ) . Basic residues in the juxtamembrane regions of a Q-SNARE might facilitate transfer of the fluorescent lipids from the R-RPLs when the two membranes are brought into proximity by trans-SNARE complex assembly , thus catalyzing transfer from the R- to the 3Q-RPLs . In any case , the lipid mixing assay needs to be employed with great caution , and should ideally only serve as a complement to content mixing experiments . Lipid dequenching has provided many valuable lessons , but can apparently report events other than true fusion . A vigorous content mixing signal , which is always accompanied by a lipid mixing signal ( dequenching or quenching ) , is only seen in the presence of HOPS ( Figure 3 ) , indicating that tethering is a vital component of the fusion process . The small amount of content mixing in 1R-3Q reactions that was seen without added protein was suppressed by the addition of Sec17p/Sec18p ( Figure 1G , compare columns 1 and 5 ) , indicating that it did indeed represent SNARE-dependent true membrane fusion . A previous report has shown that the trans-interaction of PI ( 3 ) P and the PX domain of the Qc-SNARE Vam7p could promote tethering ( Xu and Wickner , 2010 ) . This interaction was partially responsible for the SNARE-dependent content mixing seen without additional proteins , since the complete absence of PI ( 3 ) P , its asymmetric disposition on 3Q-RPLs , or a truncated form of Vam7p lacking its PX domain further reduced the small amount of content mixing seen without HOPS ( Figure 5A ) . In contrast , when PI ( 3 ) P was only present on R-SNARE RPLs , rapid fusion did not require HOPS ( Figure 5A , columns 5 and 6 ) , establishing a condition with reduced complexity in which an SM function is not absolutely required . It remains to be determined whether the Vam7p:PI ( 3 ) P tethering plays any role on intact vacuoles , where priming releases Vam7p early in the fusion reaction ( Boeddinghaus et al . , 2002 ) , or if it is merely a phenomenon seen in in vitro reconstitutions with purified components . In this context , we note that the content mixing supported by Vam7p:PI ( 3 ) P tethering is fully suppressed in the presence of Sec17p/Sec18p ( Figure 5A , column 6 ) . The ubiquitous presence of Sec17p/Sec18p in the cell may underlie a strict requirement for HOPS and other large tethering complexes in vivo . HOPS has the capacity to recruit Vam7p ( Zick and Wickner , 2013 ) , to proofread trans-SNARE complexes ( Starai et al . , 2008 ) , and to protect them from disassembly by Sec17p/Sec18p ( Hickey and Wickner , 2010; Xu et al . , 2010 ) . While the requirement for a tether was not absolute , the addition of HOPS stimulated the rate of the reaction more than 100-fold ( Figure 5B , Figure 6C ) . Colliding SNARE-bearing membranes form spontaneous trans-SNARE complexes ( Figure 6D , lane 3 ) . These can occasionally trigger fusion , but the rate achieved under such a condition is of questionable physiological relevance . This uncertainty is reinforced by the fact that mutations in Rab GTPases or tethering complexes have as strong a phenotype as mutations in SNARE proteins ( Wada et al . , 1992 ) . The addition of HOPS stimulates the fusion reaction more than 100-fold , yet it only increases the amount of trans-SNARE interaction a few-fold ( Figure 6C , D ) . This shows that the role of HOPS and tethering goes beyond merely facilitating trans-SNARE complex formation . The trans-SNARE complexes that form spontaneously are very inefficient in catalyzing membrane fusion . This indicates that it is not just a matter of quantity , but that additional factors are important to produce trans-SNARE interaction of the right quality to facilitate membrane fusion . Fusion-incompetent SNARE complexes have been observed in other systems ( reviewed in Brunger , 2005; Rizo and Sudhof , 2012 ) , and it will be crucial to determine what differentiates a productive from a non-productive complex , and what aids the formation of one over the other . Our findings do not vitiate the central role of trans-SNARE complexes in fusion; rather , they reveal that a distinct tethering step is a critical upstream event that substantially increases the rate of SNARE-mediated fusion , as suggested ( Smith and Weisshaar , 2011 ) , and its mode of action is not limited to increasing the quantity of trans-SNARE pairings . Reconstitution studies using lipid dequenching assays and endosomal SNAREs have also suggested that SNAREs alone suffice to drive fusion efficiently . Lipid mixing was reported to occur as a result of fairly promiscuous endosomal trans-SNARE pairing ( Brandhorst et al . , 2006 ) and in multiple topologies ( Zwilling et al . , 2007 ) without auxiliary proteins . However , another study ( Ohya et al . , 2009 ) , which employed a content mixing assay to measure fusion , found that a Rab GTPase and its effectors are essential for endosomal SNARE-mediated fusion . Neuronal fusion has also been extensively studied through in vitro reconstitution , and led to the concept that SNAREs alone suffice as the machinery of fusion ( Jahn and Fasshauer , 2012 ) . While SNAREs are undoubtedly essential components , their spontaneous assembly into a four-helical bundle only occurs very slowly ( Pobbati et al . , 2006 ) . Vital components that accelerate this assembly are more recently being included in the definition of a core machinery in this system . Rizo's group has provided compelling evidence for the critical roles of Munc18 and Munc13 during neuronal SNARE-mediated fusion , together with NSF/SNAP and the soluble domain of synaptotagmin 1 ( Ma et al . , 2013 ) , while Brunger's group has shown that complexin has a large effect on Ca2+ triggered synchronized fusion in the presence of neuronal SNAREs and full-length synaptotagmin 1 ( Diao et al . , 2012; Lai et al . , 2014 ) . Further study of these components can broaden our understanding of the complex mechanisms that regulate and facilitate membrane fusion . Tethering may regulate vacuolar and other fusion systems . The highly selective distribution of each Rab GTPase , which serves as a receptor for its cognate tethering complex ( Hutagalung and Novick , 2011 ) , could impose specificity on otherwise promiscuous SNARE complex assembly ( Yang et al . , 1999 ) . Membrane repulsion and the conformation of unpaired , membrane-anchored SNAREs may limit the spontaneous formation of trans-SNARE complexes . Bringing membranes into close apposition and stabilizing such a tethered intermediate drastically enriches the local concentration of SNAREs and facilitates the interactions that promote trans-SNARE complex formation . Recent studies suggested that some multi-subunit tethering complexes could also directly modulate the assembly of functional SNARE complexes ( reviewed in Hong and Lev , 2014 ) . Nevertheless , even in a tethered state , only a minor fraction ( a few percent ) of SNAREs seem to engage in such complexes ( Mima et al . , 2008 ) and lowering the SNARE concentration only a few-fold can take a substantial toll on fusion activity ( Zick et al . , 2014 ) . Other factors that promote fusion , like Sec17p and Sec18p , further stimulate HOPS-dependent reactions ( Figure 1G; Mima et al . , 2008; Mima and Wickner , 2009 ) , but their mode of action is only poorly understood . The observation that Sec17p/Sec18p relieved the inhibition by soluble SNARE proteins ( Figure 6E and Figure 6—figure supplement 3B ) demonstrates the importance of dynamic cycling of SNARE interactions , and highlights the ability of HOPS to selectively protect productive trans-SNARE pairs ( Starai et al . , 2008; Hickey and Wickner , 2010; Xu et al . , 2010 ) . Tethering is integral to our working model of vacuolar homotypic fusion . ATP-dependent priming entails phosphoinositide synthesis ( Mayer et al . , 2000 ) and the disassembly of cis-SNARE complexes by Sec17p/Sec18p/ATP ( Haas and Wickner , 1996; Mayer et al . , 1996 ) , providing SNAREs for later assembly of trans-SNARE complexes . HOPS mediates tethering ( Hickey and Wickner , 2010 ) , possibly supported by the trans-interaction of the PX domain of Vam7p with PI ( 3 ) P ( Xu and Wickner , 2010 ) . Neuronal SNAREs can associate in multiple conformations ( Weninger et al . , 2003 ) that can participate in docking ( Bowen et al . , 2004 ) . Similarly , the vacuolar SNAREs may associate in several conformations when pairing in trans . Additional factors that guide the formation of fusion-competent complexes , which might differ from fusion-incompetent complexes in composition , conformation , or associations , are required for rapid fusion . Our data indicate that tethering via HOPS or the binding of the Vam7p PX domain to PI ( 3 ) P in trans can provide such guidance . Earlier studies of intact vacuoles have also shown that the rate of fusion is not directly proportional to the level of trans-SNARE complexes ( Ungermann et al . , 1998 ) , suggesting that only a fraction of trans-SNARE interactions might also contribute to fusion activity in vivo . In addition to its role in facilitating the formation of fusion-competent trans-SNARE complexes , tethering may play a direct role in lowering the energy barrier for fusion . Tethering could provide the activation energy required for partially assembled SNARE complexes to fully zipper ( Min et al . , 2013 ) . It could also contribute by inducing membrane bending at the edges of tightly apposed membranes ( Wang et al . , 2002 ) . Domains like the vertex ring in vacuole fusion ( Wang et al . , 2002 , 2003; Fratti et al . , 2004 ) require tethering complexes for their formation and stability . Fusogenic lipids , such as the non-bilayer lipids phosphatidylethanolamine , diacylglycerol , and sterol , are required for fusion ( Zick et al . , 2014 ) and are dependent on other lipids and docking proteins for their enrichment at the vertex ring ( Fratti et al . , 2004 ) . We propose that tethering may thus make several direct contributions to fusion , as well as its role in promoting trans-SNARE complexes in conjunction with additional auxiliary proteins . Lipids were obtained from Avanti Polar Lipids ( Alabaster , AL ) , except for ergosterol which was from Sigma-Aldrich ( St . Louis , MO ) , PI ( 3 ) P was from Echelon Biosciences ( Salt Lake City , UT ) , and the fluorescent lipids ( Marina-Blue-DHPE , NBD-DHPE , Rhodamine-DHPE , Oregon-Green-488-DHPE , Texas-Red-DHPE ) were from Life Technologies ( Carlsbad , CA ) . Biotinylated R-phycoerythrin was purchased from Life Technologies , Cy5-derivatized streptavidin from KPL ( Gaithersburg , MD ) , and unlabeled streptavidin from Thermo Scientific ( Waltham , MA ) . Sec18p ( Haas and Wickner , 1996 ) , Sec17p ( Schwartz and Merz , 2009 ) , Ypt7p ( Zick and Wickner , 2013 ) , HOPS ( Zick and Wickner , 2013 ) , and vacuolar SNARE proteins ( Mima et al . , 2008; Schwartz and Merz , 2009; Zucchi and Zick , 2011 ) were purified as described . Vti1p and Nyv1p were exchanged into octylglucoside buffer as described ( Zucchi and Zick , 2011 ) . Soluble Nyv1p ( sR ) was purified as GST-TEVsite-Nyv1 ( Δtm ) as described ( Thorngren et al . , 2004 ) , and cleaved with TEV protease prior to use . GST-His6-3Csite-Nyv1p was purified as described ( Izawa et al . , 2012 ) . Proteoliposomes were prepared by detergent dialysis ( 20 kDa cutoff membrane ) in RB150/Mg2+ ( 20 mM HEPES-NaOH , pH 7 . 4 , 150 mM NaCl , 1 mM MgCl2 , 10% glycerol [vol/vol] ) as described ( Zick and Wickner , 2013 ) , with modifications . Lipids dissolved in chloroform were mixed at proportions that mimic the vacuolar lipid composition: 44 . 6–47 . 6 mol% POPC ( 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine ) , 18 mol% POPE ( 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine ) , 18 mol% Soy PI ( L-α-phosphatidylinositol ) , 4 . 4 mol% POPS ( 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine ) , 2 mol% POPA ( 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphate ) , 1 mol% 16:0 DAG ( 1 , 2-dipalmitoyl-sn-glycerol ) , 8 mol% ergosterol , and 1 mol% ( unless specified otherwise ) di-C16 PI ( 3 ) P ( 1 , 2-dipalmitoyl-sn-glycero-3-phospho- ( 1′-myo-inositol-3′-phosphate ) ) . Different fluorescent lipids ( Marina-Blue-DHPE , NBD-DHPE , rhodamine-DHPE , Oregon-Green-488-DHPE , Texas-Red-DHPE ) were included to allow assays of lipid dequenching or lipid quenching . Concentrations are indicated in the figure legends of the respective experiments . Molar protein:lipid ratios were 1:1000 for SNAREs and 1:2000 for Ypt7p . Isolation after reconstitution was achieved by floatation on a three-step Histodenz gradient ( 35% , 25% Histodenz [wt/vol] , and RB150/Mg2+ ) ; Histodenz ( Sigma-Aldrich ) solutions were prepared as 70% stock solution in modified RB150/Mg2+ with a reduced concentration ( 2% [vol/vol] ) of glycerol to compensate for the osmotic activity of the density medium; lower concentration solutions were obtained by dilution with RB150/Mg2+ . The proteoliposomes used in Figures 6 and 7 were essentially prepared the same way , but with the dilinoleoyl-forms of PC , PE , PS , and PA , and with molar protein:lipid ratios for SNAREs as indicated . Fusion reactions of 20 µl were assembled from three pre-mixes: two mixes ( 5 µl each ) of RPLs ( 250 µM lipid each ) in RB150/Mg2+ with 5 µM streptavidin ( for content mixing reactions only ) , and one mix of auxiliary factors ( e . g . , Sec17p , Sec18p , Mg2+:ATP , HOPS , anti-Vam3p , sNyv1p ) or their respective buffers . All components were incubated individually at 27°C for 10 min , and then combined in wells of 384-well plates to initiate the reaction . The plates were incubated at 27°C in a fluorescence plate reader for 30–90 min and content and/or lipid mixing signals were recorded at intervals of 5–60 s in a SpectraMax Gemini XPS ( Molecular Devices , Sunnyvale , CA ) fluorescent plate reader ( PhycoE:Cy5-FRET , Ex: 565 nm; Em: 670 nm; cutoff: 630 nm; Marina-Blue quenching or dequenching , Ex: 370 nm; Em: 465 nm; cutoff: 420 nm; NBD dequenching , Ex: 460 nm; Em: 538 nm; cutoff: 515 nm; Oregon-Green dequenching , Ex: 504 nm; Em: 534 nm; cutoff: 515 nm ) . For content mixing reactions , maximal values were determined after addition of 0 . 1% ( wt/vol ) Thesit to samples that had not received streptavidin . To estimate the extent of lipid mixing , liposomes mimicking 100% fusion were prepared by mixing respective ‘donor’ and ‘acceptor’ mixed micellar solutions 1:1 prior to liposome formation by dialysis . To estimate the amount of trans-SNARE association that formed during a reaction , the amount of Nyv1p that co-immunoprecipitated with Vam3p was determined . A 40 μl fusion reaction was incubated for 40 min at 27°C , placed on ice , and diluted 10-fold in RIPA buffer ( 20 mM HEPES-NaOH , pH 7 . 4 , 150 mM NaCl , 0 . 2% [wt/vol] bovine serum albumin , 1% [vol/vol] Triton X-100 , 1% [wt/vol] sodium cholate , 0 . 1% [wt/vol] sodium dodecyl sulfate ) containing 50 μg/ml of affinity-purified anti-Vam3p antibody and 10 µM GST-Nyv1p . After 20 μl of RIPA-buffer washed protein A magnetic beads ( Thermo Scientific , Portsmouth , NH ) were added , the mix was incubated while nutating at room temperature for 2 hr . After the beads were washed three times with 1 ml of RIPA buffer , samples were eluted in 100 μl of reducing SDS sample buffer at 95°C for 5 min . Aliquots ( 20 μl ) of each sample were subjected to SDS–PAGE and immunoblotting with anti-Nyv1p antibody .	Cells of higher organisms contain compartments called organelles and structures called vesicles that transfer molecules and proteins between these organelles . Each organelle and each vesicle is enclosed within a membrane , and these membranes must fuse together to allow these transfers to take place . A certain group of proteins , called SNAREs , have a central role in these fusion events . Since membrane fusion is difficult to observe directly , many researchers have used a method called ‘fluorescent lipid dequenching’ to study it indirectly . In this approach , one fraction of vesicles is labeled with two fluorescent molecules , with one of these molecules quenching the fluorescence of the other . However , when a labeled vesicle fuses with an unlabeled vesicle , the surface concentrations of the fluorescent molecules are diluted . This reduces the amount of quenching and the resulting increase in fluorescence can be measured . Experiments utilizing this technique had suggested that SNARE proteins are sufficient for fusion to take place , and that no other protein complexes need to be present . However , when a different assay method called ‘lumenal compartment mixing’ was used , little fusion was seen when the only proteins present were the SNAREs . The lumenal compartment mixing approach relies on measuring the degree of mixing between the contents of two vesicles . To address these conflicting results , Zick and Wickner used both methods to study fusion in a yeast-based system . The lumenal compartment mixing approach , which is the more reliable method , revealed that rapid and efficient membrane fusion in fact requires another protein complex , called HOPS , to hold the two membrane vesicles together . Zick and Wickner found that the HOPS complex does not enable fusion by just increasing the amount of interactions between the SNARE proteins . Rather , it seems to facilitate the formation of a particular quality of SNARE interactions . Future work is needed to work out how the SNARE complexes become ‘fusion-competent’ , and to explore the mechanism that allows the HOPS complex to assist in the formation of fusion-competent SNARE complexes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2014	A distinct tethering step is vital for vacuole membrane fusion
A common strategy by which developing neurons locate their synaptic partners is through projections to circuit-specific neuropil sublayers . Once established , sublayers serve as a substrate for selective synapse formation , but how sublayers arise during neurodevelopment remains unknown . Here , we identify the earliest events that initiate formation of the direction-selective circuit in the inner plexiform layer of mouse retina . We demonstrate that radially migrating newborn starburst amacrine cells establish homotypic contacts on arrival at the inner retina . These contacts , mediated by the cell-surface protein MEGF10 , trigger neuropil innervation resulting in generation of two sublayers comprising starburst-cell dendrites . This dendritic scaffold then recruits projections from circuit partners . Abolishing MEGF10-mediated contacts profoundly delays and ultimately disrupts sublayer formation , leading to broader direction tuning and weaker direction-selectivity in retinal ganglion cells . Our findings reveal a mechanism by which differentiating neurons transition from migratory to mature morphology , and highlight this mechanism’s importance in forming circuit-specific sublayers . In the developing nervous system , neurons form selective synapses to generate circuits comprised of cell-type-specific connections . This selectivity is important for circuit function because it ensures connectivity between neurons specialized for particular information-processing tasks . Despite its importance , basic questions about selective synapse formation remain unanswered . For example , we do not know how cell types fated to form synapses coordinate their growth to establish contact with each other . This is a significant cell biological challenge , because the neurons that comprise a single circuit are often born at disparate times and physical locations . In many tissues , notably the insect and vertebrate visual systems , synaptic specificity is facilitated by laminar specificity , the phenomenon whereby circuit partners project their axons and dendrites to narrow strata within a laminated neuropil ( Sanes and Zipursky , 2010 ) . The inner plexiform layer ( IPL ) of the vertebrate retina comprises at least 10 distinct sublayers built from the axons and dendrites of different amacrine , bipolar , and retinal ganglion cell ( RGC ) types ( Baier , 2013 ) . By projecting to the same IPL sublayer , circuit partners can be assured of encountering each other . The developmental events that create sublayers and guide circuit partners to converge upon them are therefore essential for establishment of retinal circuitry . At later developmental stages , when rudimentary IPL sublayers have already formed , neurons rely on molecular cues localized to those sublayers for guidance to the appropriate IPL strata ( Duan et al . , 2014; Matsuoka et al . , 2011; Sun et al . , 2013; Yamagata and Sanes , 2008; Visser et al . , 2015 ) . However , a crucial question remains unresolved: How do sublayers form in the first place ? Understanding the mechanisms that initiate creation of sublayers will provide significant insight into the earliest step in circuit formation . To learn how members of a single circuit create layers and converge upon them to achieve synapse specificity , we studied the direction-selective ( DS ) circuit of mouse retina ( Figure 1A ) . This circuit reports the direction of image motion to the brain through the spiking activity of distinct DS ganglion cell ( DSGC ) types that are tuned to prefer stimuli moving in particular directions ( Demb , 2007; Vaney et al . , 2012 ) . The DS circuit comprises a limited number of well-described cell types amenable to genetic marking and manipulation ( Kay et al . , 2011; Huberman et al . , 2009; Duan et al . , 2014 ) : ( 1 ) DSGCs; ( 2 ) GABAergic/cholinergic interneurons called starburst amacrine cells ( SACs ) ; and ( 3 ) four subtypes of glutamatergic bipolar cells ( Chen et al . , 2014; Duan et al . , 2014; Greene et al . , 2016; Kim et al . , 2014 ) . These DS-circuit cell types project to two IPL sublayers , ON and OFF , named for the light response profiles of the neurons that project to them . ON-OFF DSGCs ( ooDSGCs ) send dendrites to both sublayers , while SACs and bipolar cells project to one or the other , depending on their subtype ( Figure 1A ) . Several molecular perturbations have been described that influence ON vs . OFF laminar targeting in the mouse DS circuit ( Sun et al . , 2013; Duan et al . , 2014 ) , but in these cases , IPL sublayers still form in the right place; errors are limited to choosing the wrong DS sublayer . Thus , neither the establishment of the DS circuit sublayers nor their positioning in the appropriate IPL region depends on molecules that have been studied to date . Here , we seek to understand the earliest events leading to formation of the DS circuit IPL sublayers . Two lines of evidence suggest that SACs may take the lead in assembling this circuit . First , SACs are among the first cells to stratify the IPL: Even though other neurons innervate it contemporaneously , SACs are precocious in restricting their arbors into sublayers ( Stacy and Wong , 2003; Kay and Sanes , 2013 ) . Second , in mutant mice that entirely lack RGCs or bipolar cells , SAC IPL projections are largely normal , indicating SACs can form sublayers in the absence of their circuit partners ( Moshiri et al . , 2008; Green et al . , 2003 ) . Thus , we set out to test the hypothesis that SACs orchestrate assembly of the DS circuit sublayers . We find evidence supporting this hypothesis , and we identify a surprising cellular mechanism initiating SAC lamination: Rather than immediately innervating the IPL , newborn SACs first produce a transient homotypic arbor network outside the IPL . These early homotypic contacts serve as a cue promoting SAC dendrite development and circuit integration upon conclusion of their radial migration to the inner retina . When deprived of homotypic contacts , SAC IPL innervation – and consequent sublayer formation – is impaired . We identify the SAC cell-surface protein MEGF10 as the molecular mediator of IPL innervation upon homotypic contact . In the absence of MEGF10 , SACs persist in growing arbors outside the IPL , delaying IPL innervation . This in turn delays formation of the DS circuit sublayers and leads to SAC sublaminar targeting errors that persist to adulthood . We further show that impaired SAC sublayer formation has consequences for laminar targeting of their circuit partners: While partnering remains intact , lamination is disrupted , leading to spatial inhomogeneity in the DS circuit network . Finally , we show that these MEGF10-dependent anatomical changes both broaden and weaken direction tuning across the population of ooDSGCs . These results demonstrate that SACs orchestrate DS circuit assembly , first by initiating sublayer formation via homotypic contact , and then by using their laminated dendrites as a scaffold that guides projections of their circuit partners . To explore how the DS circuit creates its IPL sublayers , we began by determining when the sublayers first emerge in mouse . This analysis focused on SACs and ooDSGCs because bipolar cells develop later ( Morgan et al . , 2006 ) . Previous estimates of layer emergence vary widely ( Stacy and Wong , 2003; Sun et al . , 2013 ) due to the lack of adequate markers to study dendrite development in neonatal SACs . We therefore assembled a suite of mouse lines and antibody markers for this purpose , enabling anatomical studies of the full SAC population as well as individual cells ( Figure 1B–C; Figure 1—figure supplement 1; Figure 2—figure supplement 1 ) . These markers revealed that SAC dendrites form two continuous well-defined laminae by P1 ( Figure 1B , E ) . Some dendrites were stratified already at P0 , even though the P0 IPL neuropil is less than one-cell diameter wide ( Figure 1B; Figure 1—figure supplement 1 ) . Further supporting this timeline , individual P1 SACs made lamina-specific projections ( Figure 1C ) : 96% of OFF SACs in the inner nuclear layer ( INL ) , and 99% of ON SACs in the ganglion cell layer ( GCL ) , stratified within the expected IPL sublayer ( n = 49/51 OFF; 78/79 ON; four mice ) . By contrast , ooDSGCs projected rudimentary and unstratified dendrites at P1 ( n = 18 cells , three mice , none were stratified; Figure 1E; Figure 1—figure supplement 2; also see Peng et al . , 2017 ) . Even at P2 , only 30% of ooDSGCs co-fasciculated with SAC arbors; the rest projected diffusely within the IPL ( n = 23 cells , two mice; Figure 1D , E; Figure 1—figure supplement 2 ) . These results indicate that SACs form IPL sublayers at P0-P1 , and are joined later by their synaptic partners . To gain insight into how SACs form their sublayers , we next investigated the cell-cell interactions that immediately precede SAC dendrite stratification . Because SACs stratify early – before any other cell type investigated to date ( Figure 1; Kay and Sanes , 2013; Stacy and Wong , 2003 ) – they are unlikely to form strata by following pre-existing laminar cues . Instead , we hypothesized that SACs create their sublayers by engaging in homotypic interactions . To test this idea , we examined embryonic retina to determine if and when SACs establish homotypic contact . SACs exit the cell cycle at the apical retinal surface and migrate radially through the outer neuroblast layer ( ONBL ) . They next arrive at the inner neuroblast layer ( INBL ) , where postmitotic neurons reside ( Hinds and Hinds , 1978 ) ; Figure 2A , B ) . Then they begin to innervate the nascent IPL , which begins to appear in some retinal regions at E16 ( Figure 2A ) . To reveal SAC morphology throughout these steps , the early SAC marker Isl1 ( Galli-Resta et al . , 1997 ) was used to drive Cre-dependent expression a membrane-targeted GFP ( mGFP ) reporter ( Isl1mG mice ) . We also examined the orientation of SAC dendrite projections using antibodies to internexin , a marker of SAC primary dendrites ( Figure 2—figure supplement 1 ) . Staining was performed at E16 , when SACs at all stages of their early development could be discerned ( Figure 2A–D ) . Since mature SACs contact each other in the IPL , we expected that the onset of SAC homotypic contact would occur around the time of their earliest IPL projections . Surprisingly , however , this analysis revealed that SACs begin to contact each other within the INBL cell body layer upon the conclusion of their radial migration . Migrating SACs rarely interacted , but on arrival at the INBL , SAC arbors were observed touching the soma or primary dendrite of neighboring SACs ( Figure 2A–D ) . The majority of INBL SACs engaged in these soma-layer contacts , such that a GFP+ arbor network connected them ( Figure 2G ) . Analysis of primary dendrite orientation indicated that soma-layer contacts likely arose due to projections targeted within this layer: Unlike mature SACs , which exclusively project their primary dendrites toward the IPL , many E16 SACs projected tangentially through the INBL – that is , toward neighboring somata ( Figure 2E , F ) . We even noted cases where SACs appeared to project directly towards each other ( Figure 2E ) . These observations suggest that post-migratory SACs initiate contact with each other by generating an arbor network in the INBL cell body layer . Many E16 SACs also innervate the nascent IPL , raising the question of whether the soma- or IPL-layer projection establishes the first homotypic contact . We concluded that soma-layer SAC contact precedes IPL innervation , for three reasons . First , soma contacts were found in retinal regions where the IPL had not yet emerged ( Figure 2—figure supplement 2 ) . Second , soma contacts were observed among cells that still showed migratory morphological features , such as apical and/or basal processes ( Deans et al . , 2011; Hinds and Hinds , 1978 ) , and did not yet project into the IPL ( Figure 2D; Figure 2—figure supplement 2 ) . Third , SAC dendrite polarization in the tangential plane was highly transient: By P1 , the vast majority of SAC primary dendrites were oriented toward the IPL ( Figure 2E , F ) . These three observations suggest that INBL SACs transiently seek out homotypic soma contact before shifting to target the IPL . We next sought to determine how long the soma-layer SAC arbor network persists . To this end , we examined SAC anatomy at early postnatal ages using Isl1mG and ChatmG ( Figure 1—figure supplement 1 ) mice . At P0-1 , although SAC arbors within the soma layers no longer express internexin ( Figure 2—figure supplement 1 ) , the arbor network remained remarkably prominent ( Figure 2G ) . Most OFF SACs assumed a bi-laminar morphology , with one set of arbors in the IPL and another set targeting neighboring SACs in the INL ( Figure 2H–J , L; Figure 2—figure supplements 2–3 ) . INL contacts were highly SAC-selective: 88 . 8% of branches terminated homotypically ( n = 122 arbor tips from 22 cells ) , significantly greater than the contact rate expected by chance ( Figure 2—figure supplement 3 ) . By P2-3 , however , this dense INL network was mostly gone ( Figure 2G , L; Figure 2—figure supplement 2 ) . ON SACs also made soma layer projections between P0 and P3 that contacted neighboring SAC somata ( Figure 2K , L; Figure 2—figure supplement 3 ) . Together , these observations demonstrate that both ON and OFF SACs make transient soma-layer homotypic contacts that arise prior to IPL dendrite elaboration , and are disassembled at P2-3 after SAC sublayers have formed ( Figure 2M ) . SAC homotypic contacts arise at a time when they could serve as a cue for IPL innervation and sublayer formation . To test this idea , we developed a genetic strategy to prevent SACs from contacting each other in vivo . Ptf1a encodes a transcription factor required for progenitor cells to assume an amacrine fate ( Fujitani et al . , 2006; Nakhai et al . , 2007 ) ; Figure 3—figure supplement 1 ) . We crossed conditional Ptf1aflox mutant mice ( Krah et al . , 2015 ) to a Cre line ( Six3-Cre; Furuta et al . , 2000 ) that drives widespread recombination in central retina but spares some progenitors from Cre activity in peripheral retina ( Figure 3A; Figure 3—figure supplement 1 ) . In Six3-Cre; Ptf1aflox/flox mice ( abbreviated Ptf1a-cKO ) , only these spared Cre– progenitors were capable of giving rise to SACs , indicating that any SACs produced in these mutants are wild-type at the Ptf1a locus ( Figure 3C ) . Therefore , the Ptf1a-cKO mutant creates a situation where otherwise-normal SACs are present at significantly lower density than in wild-type retina ( Figure 3B , C ) . In P1-2 mutants , some SACs were effectively segregated from their neighbors – these were termed ‘solitary’ SACs – while others had neighbors sufficiently nearby that they touched ( Figure 3B–F; Figure 3—figure supplement 2 ) . Comparing solitary to touching SACs in Ptf1a-cKO retinas revealed a role for homotypic contacts in promoting IPL innervation and sublayer formation . At P1-2 , touching SACs projected normally to the IPL , similar to SACs from Ptf1a+ littermates ( Figure 3D , E , G ) . This suggests that any changes in retinal cell type composition caused by loss of Ptf1a ( Figure 3—figure supplement 1 ) are not by themselves sufficient to perturb SAC sublayer formation . By contrast , solitary SACs largely failed to innervate the IPL ( Figure 3F , G ) . This was not caused by abnormal migration: Solitary SACs were properly positioned at the IPL border , but sent only rudimentary arbors into it ( Figure 3F; Figure 3—figure supplement 2 ) . Solitary SACs were also more likely to project processes into the soma layers ( Figure 3G ) , and when they did so , the projections were typically more elaborate than those observed in wild-type retina ( Figure 3D , F; Figure 3—figure supplement 2 ) . Thus , solitary SACs overgrew arbors directed toward neighboring somata instead of growing IPL dendrites . Both types of projection errors were also seen at P15 , indicating that early errors persist to retinal maturity ( Figure 3—figure supplement 2 ) . Misprojecting SACs were still closely apposed to numerous other amacrine cells , and their arbors were intermingled in the IPL , strongly suggesting that generic amacrine interactions are not sufficient to ensure normal dendrite targeting ( Figure 3—figure supplement 2 ) . Instead , homotypic interactions are specifically required for IPL innervation and sublayer formation . To understand how SACs initiate IPL innervation upon homotypic contact , we next sought to identify the molecular cues that SACs use to recognize that contact has occurred . The cell-surface protein MEGF10 ( Figure 4A ) is a strong candidate to mediate homotypic recognition in this context , for four reasons . First , it is selectively expressed by SACs during the perinatal period ( Figure 1B; Figure 1—figure supplement 1 ) . Second , the onset of its expression coincides with onset of SAC homotypic contact at the conclusion of radial migration ( Figure 4B ) . Third , MEGF10 protein is present on soma-layer SAC arbors , making it available to transduce signals arising on these arbors ( Figure 4C ) . Finally , MEGF10 mediates SAC-SAC interactions in a separate context – during formation of the orderly ‘mosaic’ among SAC cell bodies across the retina ( Kay et al . , 2012 ) . Thus , we tested whether MEGF10 also mediates SAC-SAC recognition to initiate IPL innervation . If so , SACs from mice lacking Megf10 gene function should have phenotypes similar to solitary Ptf1a-cKO SACs – that is , reduced IPL innervation and increased arborization in cell body layers . To test this prediction , we examined SAC anatomy in Megf10 null mutants ( Kay et al . , 2012 ) and littermate controls at P0-1 , when sublayers are first forming . We found a striking effect on sublayer formation: Both ON and OFF strata were absent or severely disrupted in mutants ( Figure 5A ) . The cause of sublayer absence was investigated using pan-SAC labeling ( Figure 5A , B ) and single-cell analysis ( Figure 5C; Figure 6D ) . These studies revealed a severe deficit in IPL dendrite arborization: Most Megf10–/– SACs made only rudimentary , unstratified IPL projections at P0-1 ( n = 1/15 OFF SACs were stratified ) . Other amacrine cell types showed normal dendritic morphology in Megf10 mutants ( Figure 5—figure supplement 1 ) , indicating that the phenotype was specific to SACs . Loss of IPL innervation was not due to aberrant SAC radial migration , because , at P0 , mutant SACs had reached the inner retina in normal numbers ( wild-type , 2600 ± 287 SACs/mm2; mutant , 3153 ± 145 SACs/mm2; p=0 . 144 , 2-tailed t-test; n = 3 each group ) , and were positioned adjacent to the IPL , similar to littermate controls ( Figure 5A ) . Furthermore , most mutant SACs sent at least some arbors into the IPL at P0-1 ( Figure 5A , C; Figure 6D ) , suggesting that they migrated to a location from which IPL innervation was feasible . However , the mutant SAC arbors that reached the IPL appeared undifferentiated , with a lack of space-filling branches ( Figure 5A , C ) . As a result , not only did their arbors enclose a significantly smaller IPL territory , but they also failed to sample as much of their enclosed territory as control SACs ( Figure 5C; also compare to control cell in Figure 2H ) . By P3 some ON SAC IPL innervation was evident , but OFF SAC arbors remained largely confined to the soma layer; those that did reach the IPL remained undifferentiated ( Figure 5B; Figure 6A , D ) . These observations indicate that deletion of MEGF10 causes an IPL innervation phenotype strongly reminiscent of Ptf1a-cKO solitary SACs: Both manipulations profoundly impair SAC dendrite arborization within the IPL , preventing timely sublayer formation . In contrast to their underinnervation of the IPL , Megf10 mutant SACs arborized exuberantly in the soma layers ( Figure 6A ) . Both ON and OFF SACs were affected ( Figure 6D , E; Figure 6—figure supplement 1 ) , but the OFF SAC phenotype was particularly striking: Starting at P1 , the mutant INL network became much more elaborate than the control network of any age ( Figure 6A , C ) . INL arbor density increased in mutants from P0 to P1 and remained high at P3; by contrast , control SACs largely eliminated their INL projections over the same period ( Figure 5A , B; Figure 6A , E ) . To understand how mutant SACs generate a denser and more persistent soma-layer network , we assessed single SAC morphology ( Figure 6A , D ) . From this analysis , we determined that one reason for the denser mutant network , particularly at P2-3 , was that a larger number of mutant cells projected to the soma layers ( Figure 6E ) . However , this reason was not sufficient to explain the denser mutant INL network at P1 ( Figure 6C ) , because at that age the number of cells projecting to the INL was similar in mutants and littermate controls ( Figure 6E ) . Therefore , to account for this increase in INL arbor density , we surmised that individual mutant SAC must , on average , overinnervate INL . Supporting this conclusion , we found that mutant SACs frequently had more extensive INL arbors than littermate control SACs ( Figure 6B ) . Further , mutant SACs continued to grow primary dendrites tangentially within soma layers at P1 , when the vast majority of control SACs only targeted the IPL ( Figure 6F; also see Figure 2F ) . These observations indicate that mutant SACs continue to expand their soma layer arbor network at P1 . Thus , as with solitary Ptf1a-cKO SACs , soma layer projections were both more frequent and more exuberant for Megf10–/– SACs . Together , these data suggest that MEGF10 governs a developmental transition from soma-layer to IPL-layer dendrite growth ( Figure 6G ) : Whereas control SACs have only a brief period of soma-layer growth , switching to IPL ramification around P0 , Megf10 mutant SACs do not make this transition and instead persist in soma-layer innervation . As a result of this failed transition , many individual mutant SACs ramify extensively in the INL but underinnervate the IPL , causing the dendrite targeting phenotypes that were observed at the population level ( Figure 6A–C ) . We conclude that , because MEGF10 regulates IPL innervation in this way , MEGF10 is required for initial formation of SAC IPL sublayers . Given the similar phenotypes of Megf10 mutant and solitary Ptf1a-cKO SACs , we hypothesized that MEGF10 is the molecular cue that triggers IPL innervation upon SAC-SAC contact . A key prediction of this model is that SACs should require MEGF10 signals from their neighbors to target their dendrites properly . To test this prediction , we generated a conditional Megf10flox allele and used it to create a situation where Megf10+ SACs were surrounded by Megf10– mutant cells . This was accomplished via the same Six3-Cre strategy that we employed in our Ptf1a-cKO studies ( Figure 3A–C ) . In central retina of Six3-Cre; Megf10flox/lacZ ( Six3-Megf10-cKO ) animals , the vast majority of cells expressed a Cre-dependent GFP reporter , indicating that they lacked Megf10 function ( Figure 7A ) . Accordingly , SACs projected exuberantly to the INL and sublayer formation was disrupted , as in null mutants ( Figure 7B; Figure 7—figure supplement 1 ) . In peripheral retina , some SACs escaped Cre activity , leading to absence of the GFP reporter and continued MEGF10 protein expression ( Figure 7A , B; Figure 7—figure supplement 1 ) . Our model predicts that these cells should have mutant dendrite phenotypes despite retaining MEGF10 . To test this prediction , we imaged βgal-stained OFF SACs from Six3-Megf10-cKO and littermate control mice at P2 . This age was chosen because wild-type and null mutant mice showed a large difference in SAC INL projection frequency ( Figure 6E ) . In littermate controls , we found that βgal+ SACs rarely projected to the INL ( Figure 7C , D ) ; therefore , they behaved like control SACs from earlier experiments ( Figure 6E ) . By contrast , Megf10+ SACs surrounded by mutant SACs in Six3-Megf10-cKO retina showed a high rate of INL projections , nearly identical to their Megf10– neighbors ( Figure 7B , D; Figure 7—figure supplement 1 ) . Thus , when Megf10+ SACs are deprived of MEGF10 signal from adjacent SACs , they make exuberant soma-layer projections . This finding implicates MEGF10 as a transcellular signal that controls SAC dendrite targeting ( Figure 7K ) . Next , we investigated how SACs receive this MEGF10 signal from their neighbors . Given that MEGF10 can function as a receptor in other contexts ( Chung et al . , 2013; Kay et al . , 2012 ) , we speculated that MEGF10 might act as its own receptor . In support of this idea , co-immunoprecipitation experiments using intracellularly truncated Megf10 constructs showed that MEGF10 can interact with itself through its extracellular domain ( Figure 7I , J; Figure 7—figure supplement 2 ) . Thus , MEGF10 appears biochemically capable of acting as both ligand and receptor . If MEGF10 is indeed a receptor in this context , SACs should require it to detect contact with MEGF10-expressing homotypic neighbors . To test this prediction , we asked whether removal of Megf10 from a single SAC , during the period of soma-layer homotypic contact , would impair its IPL innervation despite normal MEGF10 expression by surrounding cells . We used ChatCre to achieve sparse recombination in SACs of neonatal mice , as in the anatomy experiments described above ( Figure 2H–K; Figure 6D ) . In Chat-Megf10-cKO animals , MEGF10 immunostaining was used to identify SACs that lost MEGF10 protein prior to P3 – that is , during the period when soma-layer arbors are present ( Figure 7F , G ) . MEGF10– cells constituted a small minority of SACs at P3 , meaning that they were generally surrounded by MEGF10+ neighbors ( Figure 7—figure supplement 1 ) . In this context , MEGF10– SACs produced more exuberant soma-layer arbors than neighboring MEGF10+ cells , while sending only minimal arbors into the IPL ( Figure 7E–H ) . Thus , single MEGF10– SACs had phenotypes similar to SACs from mice entirely lacking Megf10 ( Figure 7G , H; compare to Figure 6D ) . By contrast , adjacent MEGF10+ cells in the same Chat-Megf10-cKO retinas were indistinguishable from littermate control SACs ( Figure 7E , F , H ) . Therefore , when Megf10 is lost during dendro-somatic contact ( but not after; see below ) , SACs make projection errors typical of neurons deprived of homotypic interactions , and they do so even if their neighbors express MEGF10 and are developing normally . Together , these experiments support the conclusion that MEGF10 is a receptor through which SACs detect each other to terminate soma-layer growth and initiate IPL innervation ( Figure 7K ) . We next asked whether neonatal MEGF10-mediated interactions influence the anatomy of SAC IPL sublayers at maturity . We found that SAC sublayers eventually formed ( by P5; Figure 8H ) , and were present in the mature Megf10–/– retina , but they were marred by numerous errors . Sporadically , and at apparently arbitrary retinal locations , two kinds of local laminar disruptions were apparent . First , there were discontinuities in the ON and OFF strata , such that mutant SACs did not completely innervate their sublaminae ( Figure 8A–C ) . These discontinuities diminished retinal coverage within each mutant sublayer by ~15% ( OFF decrease , 15 . 0 ± 0 . 9%; ON decrease , 13 . 7 ± 4 . 0%; mean ±SD; n = 9 fields of view/2 mice per genotype ) . Innervation gaps were not observed for other amacrine cells , indicating that SACs were selectively affected ( Figure 8—figure supplement 1 ) . Examination of single SACs revealed that while dendritic patterning substantially recovered between P1 and adulthood , SAC arbor territories remained significantly smaller in mutants ( Figure 8D ) . These phenotypes suggest that mutant SACs never fully made up for their initial IPL innervation deficit , thereby contributing to gaps in the dendritic plexus . The second type of SAC error in mature Megf10–/– IPL was dendrite mistargeting to ectopic IPL strata ( Figure 8A , B , E ) . Both ON and OFF SACs were affected; in each case , ectopic arbors were mostly found in IPL regions inappropriately close to the soma layers ( Figure 8A , B ) . En-face images of mutant IPL revealed that ectopic OFF arbors formed a patchy but extensive fascicle network connecting many of the cells ( Figure 8E , F; 78 . 5 ± 3 . 5% of SACs participated in the network , mean ±95% CI ) . This IPL network was morphologically similar to the ectopic INL network observed in mutants at earlier ages ( Figure 6C ) , raising the possibility that the early network gives rise to the adult network by shifting location from the INL to the IPL . Supporting this view , we found that a soma layer-to-IPL transition occurs at P5 , when mutant SACs began projecting to ectopic IPL locations in addition to the soma layers ( Figure 8G , H; Figure 8—figure supplement 1 ) . This transition occurred without a significant change in the number of mutant SACs projecting into the ectopic network ( Figure 8F; Figure 8—figure supplement 1 ) , suggesting that the same cells continued to participate in the network but simply altered their anatomy to target the IPL . Thus , early exuberant soma-layer projections appear to give rise to adult IPL ectopias , starting between P3 and P5 . Together , these two adult mutant phenotypes demonstrate that DS circuit sublayer formation is delayed and imperfect in the absence of MEGF10 . While other mechanisms appear to partially compensate for MEGF10 in generating the sublayers , such mechanisms are not sufficient to prevent persistence of innervation gaps and laminar targeting errors . Thus , MEGF10 is essential for normal formation of the mature SAC IPL projection . Next , we sought to directly test the idea that MEGF10 is required early – at the time of initial SAC homotypic contact – to ensure normal SAC IPL lamination at maturity . To this end , we used Megf10flox mice to delete MEGF10 at different times . Deletion prior to the onset of homotypic contact , using the Six3-Cre line , fully phenocopied Megf10–/– adult IPL errors ( Figure 9A ) , suggesting a requirement for MEGF10 at the time of contact . To remove MEGF10 from SACs that had already established homotypic contact , we used ChatCre . In this line , the number of SACs expressing Cre gradually increases over the first postnatal days to encompass the full SAC population ( Xu et al . , 2016 ) . Therefore , Chat-Megf10-cKO mice can be used both for early , sparse MEGF10 deletion ( Figure 7F–H ) and for later , broad MEGF10 deletion . MEGF10 immunostaining revealed that this late , broad deletion occurs between P3 and P5 ( Figure 7—figure supplement 1 ) , such that MEGF10 expression is largely preserved during the period when homotypic soma-layer contacts exist ( Figure 2L ) , but is eliminated shortly thereafter . In this ChatCre-mediated deletion regime , SAC laminar targeting and gap errors were exceedingly rare ( Figure 9A ) . These experiments therefore define a time window for MEGF10 function ( Figure 9C ) : Adult IPL targeting phenotypes require absence of MEGF10 during the soma-layer projection phase of SAC development – that is , prior to P3 . Any additional activity of MEGF10 after P3 is dispensable for the adult IPL phenotype . These findings strongly support a model whereby the functions of MEGF10 during early homotypic contact – that is , promoting IPL innervation and terminating soma-layer arbor growth – are necessary for development of normal SAC IPL innervation at maturity . In addition to laminar targeting errors , Megf10 mutants also show disruptions in the mosaic spacing of SAC cell bodies across the retina: Instead of a regular , uniform distribution , mutant SAC positioning is random ( Kay et al . , 2012 ) . We considered the possibility that SAC IPL errors might arise due to MEGF10 effects on soma spacing . Two lines of evidence suggest that this is not the case . First , the two phenotypes were not well correlated at the individual SAC level: Regardless of the severity of their mosaic spacing defects , SACs made IPL targeting errors at a constant rate ( Figure 9—figure supplement 1 ) . This finding suggests that disturbed cell positioning does not influence the probability of making an IPL error . Second , using our Megf10flox allele , we were able to dissociate the IPL and mosaic phenotypes: Deletion of MEGF10 after P3 in Chat-Megf10-cKO mice caused mosaic patterning deficits , but IPL projections were largely normal ( Figure 9A , B ) . This finding demonstrates that IPL laminar perturbations are not an inevitable consequence of altered soma positioning . Altogether , these experiments support the notion that altered SAC position makes at best a minor contribution to IPL phenotypes; instead , delayed IPL innervation and exuberant soma-layer arborization are likely the major sources of perturbed SAC projections at maturity . Next , we asked whether MEGF10 , and its effects on SAC sublayer formation , are important for assembly of the broader DS circuit . To this end , we tested the impact of SAC IPL stratification errors on laminar targeting by their circuit partners . First , we examined ooDSGC IPL projections using the Hblx9-GFP ( referred to as Hb9-GFP; Figure 10 ) and Drd4-GFP ( Figure 10—figure supplement 1 ) transgenic lines , which label ooDSGC subtypes with different preferred directions ( Trenholm et al . , 2011; Huberman et al . , 2009 ) . In littermate control mice ( n = 9 ) , ooDSGC dendrites were tightly and selectively associated with SAC arbors , as shown previously ( Vaney and Pow , 2000 ) . This association was maintained in Megf10 mutants: Both normal and ectopic SAC IPL arbors reliably recruited ectopic ooDSGC projections ( Figure 10A–C; Figure 10—figure supplement 1; n = 240 ectopias from five mutants , >97% contained ooDSGC arbors ) . Further , when SAC gaps were present in the mutant IPL , ooDSGC dendrites typically grew around the gap edges and failed to enter them ( Figure 10D; Figure 10—figure supplement 1; n = 325 gaps from five mutants , >95% devoid of ooDSGC arbors ) . Thus , SACs provide both permissive cues required for ooDSGC IPL innervation , and also attractive cues sufficient to recruit ooDSGCs to the wrong IPL sublayer . Next we determined the impact of altered SAC lamination on the axons of bipolar cells that participate in the DS circuit ( Figure 11A ) . We examined the four cell types ( BC2 , BC3a , BC5 , and BC7 ) that make extensive monosynaptic connections with SACs and ooDSGCs ( Duan et al . , 2014; Ding et al . , 2016; GreeneGreene et al . , 2016; KimKim et al . , 2014; Chen et al . , 2014 ) . Bipolar axons were marked with type-specific antibodies and mouse lines reported previously ( Wässle et al . , 2009; Duan et al . , 2014 ) , as well as a novel transgenic marker of BC5 ( Gjd2-GFP; Figure 11—figure supplement 1 ) . In wild-type retina , DS-circuit bipolar cells arborized in close contact with SAC dendrites; however , unlike ooDSGCs , they remained adjacent to SACs rather than overlapping them ( Figure 11A–D; Figure 11—figure supplement 1 ) . This arrangement was preserved in Megf10 mutants: Axons of all four bipolar cell types were recruited to ectopic IPL locations by mistargeted SAC arbors , where they stratified adjacent to SACs ( Figure 11B–D , F; Figure 11—figure supplement 1 ) . For example , BC5 and BC7 terminals always sandwiched SAC arbors , regardless of their IPL location – even when doing so required formation of a supernumerary BC axon field between the normal and ectopic SAC sublayers ( Figure 11C , D ) . To quantify the mistargeting effect , we measured the position of BC5 and BC7 terminals adjacent to ON SAC ectopias . Their arbors were pushed farther apart by SAC arbor clumps ( Figure 11C–E ) , which shifted BC7 terminals significantly toward the GCL by ~4 µm ( 69 ± 0 . 8% of IPL depth in control regions to 74 ± 1 . 9% in affected regions; mean ± S . E . M . ; n = 21 control , n = 6 affected; 2-tailed t-test , p=0 . 0024 ) . These observations indicate that DS-circuit bipolar cells , like ooDSGCs , respond to SAC attractive cues . However , in contrast to ooDSGCs , bipolar cell projections were minimally affected by SAC IPL gaps . While BC5 and BC7 terminals were slightly mispositioned in the absence of SAC arbors – they were closer together – innervation of gap regions was otherwise normal ( Figure 11C–F ) . Thus , DS-circuit bipolar axons either do not require SAC-derived signals for IPL innervation , or the relevant signals are capable of acting over larger distances than the typical SAC IPL gap size ( 35–45 µm maximum diameter ) . Altogether , these analyses of DS circuit anatomy in Megf10 mutants support the notion that early-stratifying SACs form a scaffold that directs IPL laminar targeting of their circuit partners using multiple guidance strategies . Finally , we investigated the extent to which developmental events controlled by MEGF10 affect DS circuit function . We sought to determine whether the anatomical perturbations caused by loss of MEGF10 – that is , SAC laminar targeting and mosaic spacing errors – alter direction coding by ooDSGCs . To do this , we recorded from wild-type and Megf10–/– retinas on a large-scale multielectrode array ( Field et al . , 2007; Yu et al . , 2017 ) . ooDSGCs were identified based on their responses to drifting gratings and moving bars ( see Materials and methods ) , which unambiguously distinguished them from other recorded RGCs ( Figure 12A ) . Because MEGF10 is not expressed in the adult DS circuit ( Kay et al . , 2012 ) , we could be confident that any mutant physiological phenotypes reflect anatomical changes that arose during development . These experiments revealed that ooDSGCs with robust direction selectivity were present in both wild-type and Megf10–/– retinas ( Figure 12A , B ) , and constituted a similar fraction of the RGC population in both strains ( wild-type: 80/609 , 13 . 1%; mutant: 74/551 , 13 . 4% ) . Furthermore , loss of Megf10 did not alter the organization of ooDSGC preferred directions along cardinal axes ( Oyster and Barlow , 1967 ) , or the fraction of ooDSGCs preferring each direction ( Figure 12—figure supplement 1 ) . These results are consistent with the observation that mutant SACs remain paired with ooDSGC dendrites and bipolar cell axons even when normal lamination and arbor spacing are disrupted . They indicate that the qualitative functional properties of the circuit are still present . However , a more careful examination of DS tuning properties in Megf10–/– retinas revealed clear quantitative differences in ooDSGC responses . Moving bars were used to measure the width and strength of direction tuning for each identified ooDSGC across the populations recorded on the electrode array ( Figure 12C ) . Tuning width was measured as the circular standard deviation of the tuning curve , while tuning strength was measured as the normalized response difference to motion in the preferred and null directions ( see Materials and methods ) . These experiments revealed systematic shifts toward broader ( Figure 12D ) and weaker ( Figure 12E ) direction tuning across the population of ooDSGCs in Megf10 mutant retinas . This was mainly due to higher null direction spiking among ooDSGCs in mutants ( Figure 12B , C , E ) . Furthermore , these effects on tuning width and strength persisted across a broad range of stimulus contrasts ( Figure 12—figure supplement 1 ) . These results demonstrate that disruption of MEGF10-dependent developmental patterning degrades the precision and strength of ooDSGC direction tuning . They further suggest that perturbations to the anatomical regularly of the circuit across space ( e . g . laminar uniformity and SAC spacing ) may effectively introduce noise in the DS circuit that broadens and weakens direction tuning ( see Discussion ) . This idea led us to consider additional functional properties of ooDSGCs that might depend on the spatial regularity of the DS circuit , and therefore might be perturbed in Megf10 mutants . One such property is the generation of symmetric DS responses to stimuli that are darker or brighter than the background ( Figure 12F , G ) . This ON-OFF symmetry allows the DS response to be largely insensitive to contrast reversals ( Amthor and Grzywacz , 1993 ) ; it arises because ooDSGCs receive highly symmetric SAC inputs in both ON and OFF sublayers ( Figure 1A ) . In Megf10 mutants , ON-OFF anatomical symmetry is disturbed , because ON and OFF SAC errors are not spatially correlated ( Figure 8A–C ) . We hypothesized that this might lead to disparities in the direction tuning of individual cells’ ON and OFF responses . Indeed , Megf10–/– ooDSGCs exhibited greater separation ( i . e . less coherence ) between their ON and OFF preferred directions than wild-type ooDSGCs , across a broad range of contrasts ( Figure 12H; Figure 12—figure supplement 1 ) . These results support the idea that MEGF10 serves to establish a highly uniform and regular network of SAC dendrites ( via controlling both the precise timing of INL lamination and through regularizing inter-SAC spacing ) , the net effect of which is to allow greater precision and coherence in the direction tuning of ooDSGCs . During radial migration , newborn central nervous system neurons have a multipolar morphology , but on arrival at their final position within the tissue they become highly polarized ( Nadarajah et al . , 2001; Tabata and Nakajima , 2003; Cooper , 2014; Chow et al . , 2015; Krol et al . , 2016; Hinds and Hinds , 1978 ) . This morphological change enables elaboration of dendrites and integration into local circuitry . If dendrite differentiation begins early , migration is impaired ( Hoshiba et al . , 2016 ) , suggesting that the transition from migratory to mature morphology must be highly regulated to ensure that neurons only differentiate once they arrive at their final position . The extracellular cues that signal arrival are poorly understood in most nervous system regions . Here , we show that SACs use homotypic recognition , mediated by MEGF10 , to initiate IPL-directed dendrite morphogenesis . When deprived of homotypic neighbors or MEGF10 , SACs at the IPL retain a multipolar morphology ( compare Figure 2C to Figures 3F and 6A ) instead of polarizing arbors toward the IPL . This indicates that the transition from migratory to mature morphology is impaired in the absence of SAC homotypic recognition . We show that migrating SACs first establish homotypic contact upon arrival at the inner retina . At this stage , they are still multipolar ( Figure 2D ) , but they orient primary dendrites tangentially within the INBL to ultimately contact their SAC neighbors . These contacts occur prior to IPL innervation , and are required for it to occur in a timely manner . SACs lacking neighbors or the molecular means to detect them ( i . e . MEGF10 ) appear to persist in this multipolar soma-layer-targeting phase , causing over-innervation of the INL/GCL and delaying IPL innervation ( Figure 6G ) . Thus , establishment of homotypic contact is a key checkpoint for the progression of SAC dendrite differentiation and IPL sublayer morphogenesis . We propose that the function of this checkpoint is to ensure that SACs elaborate dendrites only when they have arrived adjacent to the IPL . The presence of other SACs that have already completed their migration is a reliable indicator of arrival in the proper location . Because soma-layer SAC contacts appear earliest , and because MEGF10 selectively influences IPL innervation during the period when they exist , we favor the notion that the key homotypic interactions occur through these arbors . However , we cannot exclude that IPL-based interactions also play a role . INL-directed arbors resembling those we describe can be discerned in many developing zebrafish amacrine cells ( Godinho et al . , 2005; Chow et al . , 2015 ) , raising the possibility that this mechanism applies across species and across other amacrine cell types . Because most neurons require a way to control when and where they differentiate , we anticipate that this homotypic contact strategy , or variations upon it , may have important roles in the differentiation of other CNS neurons at the completion of their radial migration . We conclude that MEGF10 is the molecule responsible for homotypic recognition during SAC IPL innervation . Four key results support this conclusion . First , MEGF10 is expressed at the right time and place to assume this role: It is expressed selectively in SACs ( Figure 1 ) , upon conclusion of their radial migration , and in the soma-layer arbors that we propose mediate recognition ( Figure 4 ) . Second , Megf10 null mutant SACs phenocopy the dendrite polarization errors seen in solitary Ptf1a-cKO SACs , suggesting that homotypic recognition requires Megf10 . Third , co-immunoprecipitation experiments indicate that MEGF10 interacts with itself via its extracellular domain , suggesting it could act as both ligand and receptor . While this biochemical interaction may take place in the cis configuration , the fourth line of evidence indicates that MEGF10 interacts in trans as well: Using a conditional-null Megf10 allele in vivo , we show that MEGF10 is required on the cell that sends homotypic signals as well as the cell receiving those signals . Loss of MEGF10 on either side leads to dendritic phenotypes resembling solitary SACs and Megf10 null mutants . Together , these data are consistent with a model whereby SAC-SAC contact initiates a transcellular MEGF10 homophilic interaction , in which MEGF10 serves as both receptor and ligand to trigger the switch from migratory to mature morphology ( see model , Figure 7K ) . This homophilic model of MEGF10 function is consistent with its role during establishment of mosaic cell body patterning ( Kay et al . , 2012 ) . In that context , MEGF10 acts as ligand and receptor to mediate cell-cell repulsion , thereby spacing SAC somata evenly across the retina . Here we discover a second MEGF10 function in SAC IPL innervation . Because the two SAC phenotypes have different underlying cell biology ( soma movement vs . dendrite dynamics ) , and separable temporal requirements for MEGF10 function ( Figure 9 ) , it seems unlikely that they reflect disruption of a single biological event . Instead , MEGF10 appears to act at distinct , albeit partially overlapping times , to control different aspects of SAC development , each of which are regulated by contact with homotypic neighbors ( see model , Figure 9C ) . Our results shed light on the mechanisms controlling SAC dendrite lamination . While repulsion mediated by Sema6a and PlexinA2 prevents OFF SACs from straying to the ON sublayer ( Sun et al . , 2013 ) , molecules required for formation of the SAC sublayers have not been identified . We show that SACs deprived of homotypic neighbors or MEGF10 initially fail to form IPL sublayers , and when they eventually do so , their strata are riddled with errors . Both the lack of sublayers at early stages and the dendritic mistargeting to inappropriate sublayers at maturity are novel SAC phenotypes; they implicate MEGF10 as a key player in forming SAC IPL sublayer-specific projections . It is generally assumed that sublayer formation has two basic molecular requirements: 1 ) Attractive/adhesive molecules that mediate co-fasciculation of stratified arbors; and 2 ) repulsive cues that prevent straying of arbors into other sublayers ( Lefebvre et al . , 2015; Sanes and Yamagata , 2009 ) . Our MEGF10 studies suggest an additional , earlier requirement for cell-cell interactions that occur prior to neuropil innervation . The purpose of this surprisingly early SAC-SAC interaction , we propose , is to ensure that SACs grow dendrites at the right time and place to co-fasciculate with their SAC neighbors . The molecular basis of this homotypic co-fasciculation – clearly another essential player in sublayer formation – remains to be determined . MEGF10 is probably not involved; the co-fasciculation system appears intact in Megf10 mutants given that sublayers do eventually form . Perhaps this system is part of the mechanism that compensates for loss of MEGF10 to ultimately generate the sublayers . When IPL arborization is delayed by loss of Megf10 , two SAC errors ensue . First , SACs generate mistargeted dendritic material that appears to persist as ectopic IPL sublayers . Second , SACs never completely innervate their sublayers , resulting in fragmented IPL strata . These two errors are caused by delays rather than an ongoing requirement for MEGF10 during later stages of arbor growth , as shown by conditional mutant experiments ( Figure 9 ) . Thus , our findings support the idea that timing is critical to the sequential lamination of the IPL: When SAC dendrites arrive in the IPL too late , they encounter a different cellular and molecular milieu that may not support the proper development of their arbors . In this view , the normal role of MEGF10 in DS circuit assembly is to instigate SAC dendrite outgrowth at the crucial time when laminar self-assembly can occur . SACs may face an additional obstacle to overcoming their delayed IPL innervation in Megf10 mutants: abnormal soma positioning . While mosaic spacing errors do not account for the Megf10 mutant ectopic IPL phenotype , we cannot exclude the possibility that the placement of IPL arbor gaps might be at least partly explained by soma position . If SACs are struggling to make up for their delayed IPL innervation , it is plausible that increasing the distance between SACs ( as happens sporadically due to random positioning ) might further hinder the development of complete retinal coverage . Because of their early stratification , SAC dendrites have been proposed to act as a scaffold that guides assembly of the DS circuit ( Stacy and Wong , 2003 ) . A key prediction of this model is that laminar targeting of later-stratifying cell types should depend on the existence of this scaffold . We show using a SAC-specific manipulation – removal of Megf10 – that disruption of SAC stratification causes their bipolar and ooDSGC circuit partners to make corresponding projection errors . Based on the kinds of errors we observed , SACs appear to provide attractive , permissive , and possibly even repulsive arbor sorting cues to influence the laminar positioning of their circuit partners . This work thus constitutes the first critical test of the scaffolding model , and provides strong support for it . We find that SACs use homotypic interactions to initiate formation of their circuit sublayers , and then heterotypic interactions to recruit circuit partners to join them . SACs might achieve their scaffolding functions directly , by providing guidance cues to their partners; or they may do so indirectly , by patterning the IPL projections of an intermediary cell type that in turn guides later-arriving projections . Direct scaffolding may be mediated in part by Cadherins 8 and 9 , which regulate interactions between SAC dendrites and DS circuit bipolar cell axons ( Duan et al . , 2014 ) . Molecular mediators of ooDSGC-SAC dendrite interactions remain to be identified . Evidence that the SAC scaffold can be repulsive – or at least can exclude bipolar arbors from certain IPL regions – came from our observations of BC axon anatomy . In wild-type retina , we were surprised to note how completely the BC3a , BC5 , and BC7 axon terminals were excluded from the SAC territory – they contacted it but did not enter ( Figure 11B–D; Figure 11—figure supplement 1 ) . This behavior stands in stark contrast to the behavior of ooDSGC dendrites , which completely overlapped SACs ( Figure 10A–B; Vaney and Pow , 2000 ) . Moreover , in Megf10 mutants , the laminar distance between BC5 and BC7 terminals was reduced in the absence of SAC arbors , and increased in the presence of SAC ectopias , further suggesting the existence of local SAC-BC repulsion . The finding that SACs exclude bipolar circuit partners from their sublayers appears at first counterintuitive . But given that no bipolar cell type is exclusively devoted to the DS circuit ( Wässle et al . , 2009; Greene et al . , 2016; Kim et al . , 2014 ) , a mechanism must exist to ensure that they can also contact non-DS partners . We speculate that SACs initially recruit their bipolar partners using long-range attractive cues , and then use contact-repulsion ( or an equivalent arbor sorting mechanism ) to displace bipolar arbors such that they remain in contact with the SAC layers but also innervate adjacent layers . This model is consistent with bipolar arbor phenotypes in Megf10 mutants , but will require further study . We found that impairment of SAC interactions in the perinatal retina causes permanent functional DS circuit deficits . In Megf10 mutants , direction tuning of ooDSGCs becomes broader and weaker , and their ON/OFF preferred directions are less aligned . Direction tuning is degraded in large part because mutant ooDSGCs have aberrant spiking responses to null-direction stimuli . This suggests that impaired null-direction inhibition – which arises from SACs – is a key contributor to the phenotype . Broader ooDSGC tuning curves have been shown , in modeling studies , to degrade population-level coding of directional information , and the ability of downstream neurons to extract such information ( Fiscella et al . , 2015 ) . Thus , the physiological phenotypes we identified are likely sufficient to impair the ability of mutant retina to appropriately relay visual information . Dysfunctional DS circuit physiology in Megf10 mutants is almost certainly a consequence of its effects on development , because neurons do not express MEGF10 beyond the second postnatal week ( Kay et al . , 2012 ) . Further , even though MEGF10 is expressed by Müller glia in adulthood , we have been unable to detect any changes in Müller glia anatomy or interactions with DS circuit synapses upon loss of Megf10 function ( Wang et al . , 2017 ) ; J . W . and J . N . K . , unpublished observations ) . We therefore conclude that anatomical changes to the DS circuit arising during development are responsible for circuit dysfunction . The fundamental change to DS circuit anatomy in Megf10 mutants is altered distribution of arbors and synapses , unlike other manipulations which simply serve to destroy SAC radial morphology or disrupt synaptic partnering among DS circuit cells ( Sun et al . , 2013; Duan et al . , 2014; Kostadinov and Sanes , 2015; Peng et al . , 2017 ) . In Megf10 mutants , the combined effect of mosaic spacing defects and IPL laminar targeting errors is to disturb the regularity of SAC IPL innervation . As a result , some parts of the visual map become over-innervated ( e . g . Figure 10C ) while others are uninnervated ( Figure 10D ) . In turn , ooDSGCs are recruited to the over-innervated regions and excluded from uninnervated gaps , likely causing sporadic local inhomogeneity in synapse density across visual space . According to some models of DS , which posit that the total amount of SAC inhibition is the key factor underlying DS responsiveness , these relatively small-scale changes would be considered unlikely to change circuit function ( Taylor and Vaney , 2002; Demb , 2007 ) . A more recent alternate view is that the fine spatial arrangement of glutamatergic inputs to SACs , and the synaptic balance of SAC and bipolar input onto ooDSGC dendrites , are both important for DS responses ( Ding et al . , 2016; Vlasits et al . , 2016; Poleg-Polsky and Diamond , 2016; Sethuramanujam et al . , 2016; Sethuramanujam et al . , 2017 ) . The finding that Megf10 mutants have DS tuning phenotypes suggests that local synaptic arrangements are indeed important for the DS computation . More broadly , this finding shows that the developmental mechanisms we describe here are important for enabling circuit function , raising the possibility that other circuits throughout the retina and CNS may use similar developmental mechanisms to establish their functional connectivity . All animal experiments were reviewed and approved by the Institutional Animal Care and Use Committee of Duke University . The animals were maintained under a 12 hr light-dark cycle with ad lib access to food and water . Retinas from adult ( 4–8 weeks old ) Megf10–/– mutant mice and wild-type control mice with same genetic background were used for experiments performed on the multielectrode array ( MEA ) . Animals were dark-adapted overnight prior to the experiment . For this study , the following transgenic and mutant mouse lines were used: ( 1 ) Megf10tm1b ( KOMP ) Jrs ( Kay et al . , 2012 ) , referred to as Megf10– or Megf10lacZ; ( 2 ) Ptf1atm3Cvw ( Krah et al . , 2015 ) , referred to as Ptf1aflox or ( when crossed to Cre mice ) Ptf1a-cKO; ( 3 ) Isl1tm ( cre ) Sev ( Yang et al . , 2006 ) , referred to as Isl1Cre; ( 4 ) Tg ( Hlxb9-GFP ) 1Tmj/J ( Trenholm et al . , 2011 ) , referred to as Hb9-GFP; ( 5 ) Chattm2 ( cre ) Lowl ( Rossi et al . , 2011 ) , referred to as ChatCre; ( 6 ) Tg ( Six3-cre ) 69Frty ( Furuta et al . , 2000 ) referred to as Six3-Cre; ( 7 ) Kcng4tm1 . 1 ( cre ) Jrs ( Duan et al . , 2014 ) referred to as Kcng4Cre; ( 8 ) Tg ( Drd4-EGFP ) W18Gsat ( Huberman et al . , 2009 ) , referred to as Drd4-GFP; ( 9 ) Tg ( Gjd2-EGFP ) JM16Gsat , referred to as Gjd2-GFP; ( 10 ) Tg ( Gad1-EGFP ) G42Zjh , referred to as Gad1-GFP . Two Cre reporter strains were used that express membrane-targeted green fluorescent protein ( mGFP ) upon Cre recombination: ( 1 ) Gt ( ROSA ) 26Sortm4 ( ACTB-tdTomato , -EGFP ) Luo , also known as mT/mG ( Muzumdar et al . , 2007 ) ; ( 2 ) Rosa26fGFP ( Rawlins et al . , 2009 ) . An additional Cre reporter strain was used that expresses tdTomato fluorescent protein upon Cre recombination: Gt ( ROSA ) 26Sortm14 ( CAG-tdTomato ) Hze ( Madisen et al . , 2010 ) . See Key Resources table for repository stock numbers where applicable . To produce Megf10flox mice , Megf10tm1a ( KOMP ) Jrs mice ( Kay et al . , 2012 ) were crossed to germline Cre strain B6;SJL-Tg ( ACTFLPe ) 9205Dym/J , thereby generating a functional allele ( also known as Megf10tm1c ) in which exon four was flanked by loxP sites . HEK293T cells were obtained from , validated by , and mycoplasma tested by ATCC . The cells were cultured in Dulbecco’s Modified Eagle’s Medium ( DMEM ) with 10% bovine growth serum , 4 . 5 g/L D-glucose , 2 . 0 mM L-glutamine , 1% Penicillin/Streptomycin in 10 cm cell culture dishes . Cells were passaged every 2–3 days to reach confluence . Before splitting , culture media were removed and Dulbecco’s phosphate-buffered saline ( D-PBS ) was used to rinse cell layers as well as removing residual serum . Cells were detached from dish with 4 ml of 0 . 05% Trypsin and incubated at 37°C until cell layer is dispersed ( about 5 min ) . Equal volume of complete culture media was added to the dish to inhibit protease activity . The suspension was centrifuged at 200 x g for 5 min . Supernatant was aspirated and the cells were suspended with appropriate amount of media and plated ( 1:4-1:8 ) . Cells used for experiments were passaged no more than 10 times . Cell stocks were stored as 2 million cells per vial in complete culture media with 10% DMSO in liquid nitrogen . To study SAC anatomy during embryonic stages , Isl1Cre was crossed to lox-stop-lox-mGFP Cre reporter mice ( mT/mG or Rosa26GFPf; see Key Reagents ) to generate Isl1mG animals . Timed-pregnant dams were sacrificed at E16 and eyes collected from embryos ( n = 11 mice from three litters ) . Tissue was processed as described for postnatal eyes , except fixation time was 60 min . Cross-sections were stained with anti-GFP to reveal the morphology of Isl1mG-expressing neurons , as well as Sox2 to distinguish Isl1mG-positive SACs from RGCs . ( All cells shown in Figure 2B–G were confirmed to be SACs by Sox2 co-labeling . ) In combination with these markers , anti-internexin staining was used to assess orientation of primary dendrites . Location and/or presence of the IPL was determined using Hoechst nuclear staining , which revealed cell body-free neuropil regions , and/or by Isl1mG labeling of neuronal processes , which filled these neuropil regions ( Figure 2—figure supplement 2 ) . We assessed anatomy of mGFP+ migrating SACs in the ONBL , as well as SACs in the INBL that were concluding their migration . Morphology of ON SACs in the GCL could not be discerned due to Isl1 expression by RGCs ( Figure 2A , B ) , but because displaced amacrine cells pause at the INL-IPL border before crossing to the GCL ( Chow et al . , 2015 ) , the population of cells available to analyze might have included both ON and OFF SACs . To measure the orientation of primary dendrites at E16 and P1 , the angle ROI function in ImageJ was used . This function outputs an angle degree measurement ( absolute value ) between two line segments . The first line segment of the angle was drawn to follow the trajectory of the internexin+ primary dendrite; the endpoint was at the cell body . The second line segment of the angle was a plumb line to the IPL ( i . e . it was drawn to intersect the IPL at ~90˚ ) . As such , dendrites oriented exactly toward the IPL were assigned an angle of 0˚ . At E16 the IPL was occasionally not present yet; in this case the second line segment was a plumb line to the inner limiting membrane . In cases where the internexin+ dendrite curved , we traced the initial trajectory of the dendrite as it emerged from the cell body . Dendrites were classified as projecting ( 1 ) towards the IPL; ( 2 ) toward the ONBL; or 3 ) tangentially , according to the angle scheme delineated in Figure 2—figure supplement 1E . Image stacks were randomly selected for analysis from a larger library of images; within each selected stack every SAC was traced . Isl1mG and Sox2 were used to confirm the SAC identity of each measured cell , as well as the trajectory of the internexin+ dendrite . The homotypic nature of SAC soma-layer contacts was investigated by imaging single ChatmG-labeled OFF SACs in mice also carrying a single copy of the Megf10lacZ allele ( Figure 2I , J ) . Anti-βgal staining was used to reveal the full SAC population , including arbors . En-face images were captured in Z-stacks spanning the INL and IPL; slices corresponding to each layer were separately Z-projected for display in Figure 2 and Figure 2—figure supplement 3 . To quantify the frequency of SAC-SAC contacts , we used Z-stacks from P1 tissue to examine the trajectory and termination site of each dendritic tip in three dimensions . The fraction of ChatmG-labeled dendrites terminating on the βgal-positive soma or arbor of a neighboring SAC was quantified . To be counted , the putative contact needed to be confirmed in a single Z-stack slice; where necessary , 3D reconstructions and orthogonal views were used to confirm contact . We also performed the same analysis on Z-stacks in which one channel had been flipped about the horizontal and vertical axes . This served as a negative control to measure the frequency with which GFP and βgal arbors interact by chance , given their density and geometry in the P1 retina . Sample sizes are given in main text and in Figure 2—figure supplement 3 . Single SACs labeled in ChatmG and ChatmG;Megf10–/– mice were morphologically assessed in cross-sections . GFP signal was amplified with anti-GFP antibody staining . All GFP+ SACs on any given slide were imaged and analyzed , to avoid cell selection bias , with the exceptions of: 1 ) cells severed by the cryosectioning process; 2 ) cells with arbors that could not clearly be distinguished from those of their neighbors; 3 ) cells in the far retinal periphery , where sections were oblique to retinal layers , obscuring IPL strata . In experiments analyzing Megf10 mutants , littermates were always used as controls to avoid complications arising from the fact that the precise state of retinal development at the time of birth might vary from litter to litter . A cell was scored as innervating the IPL if it ramified branched dendrites within the neuropil . Dendrites that entered the neuropil but did not branch or stratify ( e . g . Figure 6D ) were not sufficient . A cell was scored as projecting to the soma layer if arbors emanating from the cell soma or primary dendrite terminated or arborized in the INL ( for OFF SACs ) or GCL ( for ON SACs ) . The arbor was required to be ~≥1 cell radius in length ( i . e . small fine arbors were not counted ) . One other important exception that was not counted: We observed that many SACs at young ages had single unbranched arbors extending ~180˚ away from the IPL ( e . g . Figure 2J , K – all four cells have such arbors , even the ones that do not project towards neighboring SAC somata ) . These processes were not counted for two reasons . First , their trajectory was such that they were unlikely to join the soma-layer dendrite network or contact neighboring somata . Second , these 180˚ arbors were sometimes still present in P5 SACs ( Figure 2—figure supplement 2 ) and therefore they did not appear to be subject to the same developmental regulation as tangentially-directed arbors ( Figure 2L ) . This observation suggests they are fundamentally different , and likely serve a different ( as yet uncharacterized ) purpose . No obvious difference in their frequency was observed between wild-type and Megf10 mutants . To produce graphs in Figures 2L , 6E and 8F , the fraction of cells making ectopic projections – either to the soma layer or to inappropriate IPL sublayers – was calculated for each genotype and each time point . To determine whether a GFP+ IPL arbor was located in normal or abnormal IPL strata , Megf10:βgal was used as a counterstain . ChatCre was rarely expressed in OFF SACs at P0 , making it difficult to obtain large sample sizes at this age . For this reason , and because soma-layer projection frequency did not appear to differ much between P0 and P1 , the data from each time point was pooled for analysis of Megf10 litters . Sample sizes for Figure 2L: P0 , n = 25 OFF , 63 ON; P1 , n = 51 OFF , 79 ON; P2 , n = 46 OFF , 55 ON; P3 , n = 33 OFF , 49 ON; P5 , n = 15 OFF , 26 ON; P7 , n = 23 OFF , 34 ON . Data were from four litters of mice , each of which was assessed at no less than two of these time points . Sample sizes for Megf10; ChatmG experiments ( Figure 6E; Figure 8F ) : Megf10 heterozygous littermate controls: P0/1 , n = 11 OFF , 25 ON; P2 , n = 25 OFF , 23 ON; P3 , n = 17 OFF , 22 , ON; P5 , n = 16 OFF , 16 ON . Megf10 mutants: P0/1 , n = 6 OFF , 25 ON; P2 , n = 14 OFF , 20 ON; P3 , n = 34 OFF , 41 ON; P5 , n = 48 OFF , 54 ON . Data were from two litters of mice . For the adult data reported in Figure 8F , a different procedure was used; see ‘Quantification of Mosaic Spacing Phenotypes’ section below . P1-P2 retinas carrying Megf10lacZ and Hb9-GFP were co-stained for βgal and GFP . RGCs with dendrites that co-fasciculated with βgal-positive IPL strata were counted . Cells that projected to βgal-positive regions , but also filled non-SAC-projecting IPL regions , were not counted as co-fasciculated . To judge co-fasciculation , we used two criteria: 1 ) inspection of dendrite anatomy across the confocal stack; 2 ) fluorescence profiles of GFP and βgal channels across IPL ( see next section below ) . Examples of cells falling into each category are provided in Figure 1 and Figure 1—figure supplement 2 . See Results for sample sizes . Images of retinal cross sections were processed in ImageJ . A vertical ROI ( 12 . 5 µm wide ) was drawn to perpendicularly bisect the IPL strata , from the edge of the INL to the edge of the GCL . IPL stratification levels were reported as percentage of IPL width . Intensity was calculated for each pixel along the length of the ROI as an average across its width . Background ( minimum pixel value ) was subtracted; then , all pixel intensity values were normalized to the maximum value of that ROI . Location of fluorescent peaks was calculated as the pixel with maximum intensity; if multiple pixels had the same intensity the peak was defined as the center of the plateau . The procedure was typically performed on single confocal optical sections , but for some P1-2 cells , which have much smaller arbors , it was necessary to use a maximum-intensity projection of a small number of slices in order to fully capture dendrite morphology . For BC5-BC7 arbor distance measurements ( Figure 9F ) , distances as percentage of total IPL width were compared by one-way ANOVA/Tukey’s post-hoc test . n = 14 measurements from two control mice; n = 7 normal IPLs , 11 SAC clumps , 11 SAC gaps from 3 Megf10–/– mice . The MEGF10-ΔICD-GFP construct was reported previously ( Kay et al . , 2012 ) , which was originally made from pUbC-MEGF10-GFP ( Addgene #40207 ) . It encodes a version of MEGF10 in which the cytoplasmic domain is truncated after the 9th amino acid and replaced by GFP . Inclusion of those nine amino acids was necessary to achieve plasma membrane localization . For this study , it was subcloned into the pEGFPN3 plasmid , containing the CMV promoter , to make pCMV-MEGF10-ΔICD-GFP . To make the MEGF10-ΔICD-Flag construct , Megf10 ( truncated after the 9th intracellular domain amino acid as above ) was PCR amplified from pUbC-MEGF10-GFP vector using M10flag_Fwd forward primer and Cyto9_flag_Rev1 reverse primer . Resulting PCR products were digested with NotI and AscI restriction enzymes and ligation cloned into pEGFPN3 vector linearized with corresponding restriction enzymes . Statistical analysis was performed using GraphPad Prism software ( anatomy/development studies ) or using custom JAVA-based software and MATLAB software ( physiology studies ) . This software is available , together with the primary data it was written to analyze , at a public repository ( Roy and Field , 2017; https://github . com/Field-Lab/megf10-dstuning; copy archived at https://github . com/elifesciences-publications/megf10-dstuning ) . Statistical tests used for each experiment are given in the appropriate Materials and methods section above , and/or in the figure legends . Sample sizes for each experiment are given in the appropriate Methods section above or else in the Results . p-Values ( α = 0 . 05 ) are given in figure legends , or in the Results if no figure is shown . Error bars are defined in figure legends . Exact p-values are reported unless the value was less than 1 . 0 × 10−7 .	Our experience of the world relies on circuits spanning the sense organs and the brain that process information received through our senses . These circuits are made up of many different types of nerve cells that form connections with each other while the brain is developing . For these circuits to be set up properly , nerve cells have to be selective about how they connect with each other . However , researchers know little about how exactly nerve cells form the right connections , or about which genes and proteins are involved . One of the better understood circuits in the body is known as the ‘direction-selective circuit’ . Found in the retina at the back of the eye of all backboned animals , this circuit’s task is to detect the direction that objects are moving . In the case of mice , scientists have identified all of the cells that make up the circuit , and know how they are all supposed to be connected together . This is a useful starting point for researchers to look in more detail at how nerve cells make the right connections during development to set up a working circuit . Ray et al . looked at how the direction-selective circuit forms in the retinas of young mice by genetically engineering cells to carry fluorescent proteins , or staining them with chemicals . This allowed the cells to be examined under a microscope at different points in their development . It turns out that one type of cell , known as the ‘starburst amacrine cell’ because of its firework-like shape , coordinates the formation of the whole direction-selective circuit . First , starburst cells branch out and touch each other . Next , they build a scaffold for the circuit with their branch-like extensions . Finally , other cell types follow this scaffold to form connections and complete the circuit . Ray et al . identified a protein called MEGF10 on the surface of starburst cells that tells the cells when they have made contact with each other . When starburst cells had MEGF10 taken away , or were prevented from contacting each other , they did not build a scaffold properly , and the circuit was less effective at detecting movement . It is possible that cells in other brain circuits use a similar method to form connections . Understanding more about how nerve cells form circuits will help researchers to work out what goes wrong in developmental disorders that affect vision , memory and learning . This knowledge would be helpful for designing new treatments for these conditions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Formation of retinal direction-selective circuitry initiated by starburst amacrine cell homotypic contact
Cancer is a clonal evolutionary process , caused by successive accumulation of genetic alterations providing milestones of tumor initiation , progression , dissemination , and/or resistance to certain therapeutic regimes . To unravel these milestones we propose a framework , tumor evolutionary directed graphs ( TEDG ) , which is able to characterize the history of genetic alterations by integrating longitudinal and cross-sectional genomic data . We applied TEDG to a chronic lymphocytic leukemia ( CLL ) cohort of 70 patients spanning 12 years and show that: ( a ) the evolution of CLL follows a time-ordered process represented as a global flow in TEDG that proceeds from initiating events to late events; ( b ) there are two distinct and mutually exclusive evolutionary paths of CLL evolution; ( c ) higher fitness clones are present in later stages of the disease , indicating a progressive clonal replacement with more aggressive clones . Our results suggest that TEDG may constitute an effective framework to recapitulate the evolutionary history of tumors . Cancer is a complex , Darwinian , adaptive clonal evolutionary process , driven by the accumulation of genetic alterations that confer high proliferative and survival advantage ( Merlo et al . , 2006; Greaves and Maley , 2012 ) . Recent advances in sequencing technologies have allowed uncovering the most common genetic alterations of many tumors , but the temporal order of most of these alterations is still unknown ( Futreal et al . , 2004; Greenman et al . , 2007; Santarius et al . , 2010; Lawrence et al . , 2014 ) . Temporal patterns of genetic alterations may indicate the fate of tumor progression , allowing early diagnosis of tumor subtypes and improving the choice of therapeutic strategies . To understand the evolutionary history of tumors , several experimental and computational strategies have been used . Longitudinal strategies require samples at multiple time points spanning the clonal tumor evolution process ( Fearon and Vogelstein , 1990; Campbell et al . , 2010; Ding et al . , 2010; Notta et al . , 2011; Gerlinger et al . , 2012; Turajlic et al . , 2012; Landau et al . , 2013 ) . Landau et al . sampled leukemia cells from 18 CLL patients at two time points , revealing that SF3B1 and TP53 mutations are late events in subclonal tumor cells ( Landau et al . , 2013 ) . The study of different stages of colorectal carcinogenesis showed the sequence of genetic events to be APC , KRAS , and then TP53 ( Fearon and Vogelstein , 1990 ) . Another alternative approach is a cross-sectional strategy , which makes use of a large cohort of patients to computationally predict the preferred orders . RESIC is a stochastic process model to identify the order of mutations ( Attolini et al . , 2010 ) , which successfully confirmed the results in colorectal cancer , suggesting that cross-sectional data is informative for the prediction of mutation order . However , RESIC does not consider a critical aspect of carcinogenesis that most tumors are heterogeneous ( Parsons , 2011 ) . Following the assumption that different tumors proceed through related temporally ordered alterations , we propose to summarize tumor histories using a newly developed analytical approach that integrates the genomic information from different longitudinally characterized patients . Our method , termed tumor evolutionary directed graphs ( TEDG ) , proceeds in two steps to ensemble in a simplified way cancer clonal evolutionary histories of large number of patients: first , by merging the evolutionary history of each patient , and second , by removing indirect relationships using spectral techniques for network deconvolution ( Feizi et al . , 2013 ) . The resulting TEDG is a directed graph with nodes representing driver genes and arrows representing temporal order of gene lesions . A non-randomly distributed TEDG shows that cancer proceeds in an orchestrated fashion and indicates the main paths and the alternative routes of cancer evolution . In this study , we have applied TEDG to study the dynamics of the acquisition of alterations in chronic lymphocytic leukemia ( CLL ) , which represents the most common adult leukemia in Western countries ( Hallek et al . , 2008; Müller-Hermelink et al . , 2008 ) . CLL is an ideal model for studying clonal dynamics because it is possible to collect highly purified sequential samples over time , and its clinical course is well characterized by serial cycles of response , remissions , and relapse ending in some instances with the development of lethal complications such as chemoresistant progression or transformation into an aggressive lymphoma ( Richter syndrome ) ( Pasqualucci et al . , 2011; Zenz et al . , 2012; Fabbri et al . , 2013 ) . No systematic approach has been followed to disentangle and characterize the ensemble of evolutionary histories of this disease . For this purpose , we envision a dual cross-sectional and longitudinal strategy by collecting genomic information from the most common alterations in a cohort of 70 CLL patients spanning over a period of 12 years ( 2001–2012 ) . To recapitulate and compare the history of genetic alterations in many patients , we propose a framework to infer TEDG by integrating longitudinal and cross-sectional genomic data of cancer patients . First , we reconstruct the sequential network of genetic alterations in each patient by analyzing genomic data from different time points . Specifically , the techniques of high-depth next generation sequencing ( NGS ) and fluorescence in situ hybridization ( FISH ) are separately carried out to assess the mutation allele frequency ( MAF ) and copy number abnormalities ( CNA ) of selected driver genes . To unify both types of data , and to adjust the MAF of mutations in genes with CNA , we introduce mutation cell frequency ( MCF , defined as the fraction of tumor cells with a particular alteration ) for quantification of genetic lesions ( ‘Materials and methods’ , Figure 1—figure supplement 1 ) . Based on MCF , we investigate alterations represented in at least 5% of leukemic cells ( see examples of CLL patients in Figure 1—figure supplement 2 ) . First , if a given genetic lesion is observed to be temporally earlier than another lesion , we connect them with a directed edge to represent their sequential order of development ( Figure 1A ) . Second , we pool many sequential networks from different patients to construct an Integrated Sequential Network ( ISN ) . Third , we infer TEDG from ISN by removing indirect associations with spectral techniques and minimal spanning tree algorithm . TEDG is the backbone of ISN , representing an optimal explanation of the mutation order across many patients ( Figure 1B ) . 10 . 7554/eLife . 02869 . 003Figure 1 . Tumor Evolutionary Directed Graph ( TEDG ) framework . ( A ) A toy example of clonal evolution of one patient . The evolutionary history of four alterations A , B , C , and D is shown in the left panel . We then sample different time points and analyze genomic data . Specifically , for each patient , tagged-amplicon library next generation sequencing ( NGS ) and fluorescence in situ hybridization ( FISH ) analyses are carried out at different time points to evaluate the presence and quantify the clonal abundance of possible driver genetic lesions . Then we use mutation cell frequency ( MCF ) to adjust and unify the data ( middle panel ) . Based on this longitudinal data , we build sequential network of one patient ( right panel ) . CNA: copy number abnormalities . ( B ) Sequential networks derived from different patients ( left panel ) are further pooled to generate Integrated Sequential Network ( ISN ) , which is a cross-sectional integration of longitudinal data ( middle panel ) . We then infer TEDG by removing the indirect interactions with network deconvolution and simplification algorithms ( ‘Materials and methods’ ) . To construct TEDG , we calculate a minimal spanning tree-based on the deconvolution scores ( right panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 00310 . 7554/eLife . 02869 . 004Figure 1—figure supplement 1 . Adjustment of MAF based on copy number data . ( A ) Definition of mutation cell frequency . The black lines within the circles represent DNA copies , and the crosses represent point mutations . The contingent table shows the difference between MAF and MCF . MAF: mutation allele frequency; MCF: mutation cell frequency; NAN: not available . ( B ) Optimization of Hill function by grid-search method . z-axis indicates the objective function F , x-axis and y-axis are parameters of the Hill function . ( C ) The optimal Hill function and the simple piecewise function . ( D ) MAF and MCF of the cancer two-hit model . ( E ) Justification of MCF . x-axis indicates the fraction of CD19+CD5+ cells assessed by FACS analysis , and y-axis indicates the maximal mutation fraction of all targeted driver genes of each sample calculated by different methods . One blue dot represents one sample , and contours indicate the density of dots . A suitable calculation of maximal driver mutation fraction will approximate but not exceed the fraction of cancer nuclei . The upper red line indicates CD19+CD5+ cell fraction , and the lower red line indicates a 20% lower interval of it . Apparently , tumor purities of 55 samples are properly assessed by the Hill function MCF , which is better than both MAF without adjustment ( 10 samples ) and simple piecewise MCF ( 27 samples ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 00410 . 7554/eLife . 02869 . 005Figure 1—figure supplement 2 . Relative timing of mutations of 70 CLL patients . Each column represents one patient with at least two time points . Magenta ( MCF > 5% , present ) and blue ( absent ) mutations are ordered by time information , indicating the status of the corresponding alterations . For one patient , if the presence of alteration A is earlier than alteration B , we assert that A predates B . DOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 005 To test TEDG method and also to show how many patients are required to approximate the ground truth , we employ artificial examples by both simulating linear evolution and branching evolution of cancer , where the longitudinal data are generated by one-step Markov process and Nordling's multi-mutation model ( Nordling , 1953 ) ( Figure 2A , B , ‘Materials and methods’ ) . For example , in a cohort of 15 patients with linear evolution , we start the Markov process of each case from no mutations at time zero . Mutation status at three time points 10 , 20 , and 30 are then successively updated based on the Markov chain transition probability described in ‘Materials and methods’ . We finally pool all temporal information of those 15 patients to generate ISN ( left panel of Figure 2D ) . The TEDG is further deduced by deconvoluting ISN . Suppose Gdir is the adjacent matrix of all direct interactions/orders , the observed ISN should be a summary of direct and indirect orders . The deconvolution is formulated by Gdir = Gobs ( I + βGobs ) −1 , where , Gobs is the observed weighted adjacent matrix , I is the identify matrix , and β is the scaling factor between zero and one , indicating the degree of deconvolution . To evaluate TEDG , we define the accuracy by how frequently its results match the input model . To calculate the accuracy , we generate 10 simulated datasets of N patients ( with three sequential samples per patient ) . In each simulation , we apply the TEDG analysis to reconstruct the sequential order and compare the result with the input model . We then apply the concept of accuracy to find a reasonable β for our simulation , by calculating the accuracy of TEDG with different β ∈ ( 0 , 1 ) when N ∈ {1 , 2 , … , 100} . Figure 2C indicates that the optimal value of β is related to the number of samples , but the wide range of high accuracy region suggests that the TEDG results are robust to the parameter selection . With 30 patients , TEDG's accuracy for a linear model is 80% , and for a branching model , it reaches 90% ( Figure 2E–G , ‘Materials and methods’ ) . 10 . 7554/eLife . 02869 . 006Figure 2 . The calibration of TEDG framework on the simulation of two basic evolutionary models . ( A ) The representation of linear model and branching model showing the sequential orders of four alterations . ( B ) The probability of observing different mutation profiles by Nordling's multi-mutation model . Specifically , if patient i harbored k0 mutations at time point t1 , the probability to observe k more mutations at time point t2 is ( 1−e−f· ( t2−t1 ) ) k , where f represents the fitness of the new mutation . ( C ) The selection of parameter β . The color of heat map represents the accuracy of TEDG analysis , which is defined by the probability of TEDG reconstructing the input model . ( D ) An example showing the deconvolution algorithm on linear model . ( F ) An example showing the deconvolution algorithm on branching model . The left panel represents the ISNs of simulations of 15 patients . The weight of the edges represents the number of patients supporting the corresponding edges . The figures in the middle show the results of network deconvolution , where the numbers on the edge indicate deconvoluted weights . ( E and G ) The accuracy of TEDG framework on linear model and branching model at β = 0 . 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 006 In order to investigate the evolutionary history of CLL , we apply the TEDG framework to the driver genetic lesions of this leukemia . We study the most common alterations of 164 temporally sequential samples from 70 patients by high-depth next generation sequencing ( NGS ) and fluorescence in situ hybridization ( FISH ) analysis ( ‘Materials and methods’ ) . Half ( 35 out of 70 ) of the patients have at least one subclonal genetic lesion with mutation frequency less than 20% at diagnosis ( Figure 3—figure supplement 1 ) . We firstly build sequential networks for each patient based on this longitudinal genetic data . Then , we pool all sequential networks to construct the ISN of CLL . We ask whether the genetic lesions in CLL are temporally ordered or randomly accumulated and reason that if the genetic alterations driving CLL progression follow a preferential order , there exists a well ordered directed flow in ISN . Figure 3A represents a hierarchical layout of ISN , which depicts an ordered structure of lesions in genes represented by sources ( nodes with more outgoing arrows ) and sinks ( nodes with more incoming arrows ) ( p-value < 0 . 0001 , chi square distribution ) . 10 . 7554/eLife . 02869 . 007Figure 3 . Evolutionary network analysis of CLL genetic lesions . ( A ) Network representing the sequential order of genomic alterations in CLL . The nodes in the network represent genetic alterations and the oriented edges ( arrows ) represent sequential events in different patients where alterations in one gene predate alterations in other genes . The size of the nodes represents the recurrence of alterations in our cohort . The thickness and the color codes of the edges represent the number of patients showing a specific connection between nodes . ( B ) TEDG of CLL , which is the deconvolution of ISN by removing indirect interactions , representing the optimal tree to explain observed orders in CLL patients . ( C ) The order of CLL alterations . We calculate both the sequential in-degree ( number of arrows to a node ) and out-degree ( number of arrows from the node ) of each genetic lesion in ISN and use the binomial test to assess the significance by assuming that the number of in-degree and out-degree is randomly distributed . Our null assumption is that if there is no preferential time ordering , there should be the same number of alterations in gene A occurring before alterations in gene B and vice versa , up to statistical fluctuations . Deviations from the null assumption indicate a preferential order in the development of the alterations . All events are sorted by fold change between out-degree and in-degree and all significant early or late events are labeled by * ( p-value < 0 . 05 ) or ** ( p-value < 0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 00710 . 7554/eLife . 02869 . 008Figure 3—figure supplement 1 . Summary of longitudinal data in 70 patients . ( A ) The 70 patients are selected from a large cohort of 1403 CLL patients with no-bias screening . ( B ) The 70 patients are ranked according to their minimal cell frequency of all available genetic lesions at diagnosis . Patients with minimal cell frequency less than 20 are in red , the others are in green . DOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 00810 . 7554/eLife . 02869 . 009Figure 3—figure supplement 2 . The order of CLL alterations with and without treatments . All events are sorted by fold change between in-degree and out-degree , and all significant early or late events are labeled by * ( p-value < 0 . 05 ) or ** ( p-value < 0 . 01 ) . The upper panel represents the results of patients with only watch and wait , while the lower panel represents the results of patients with treatment . DOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 009 To understand the evolutionary pattern of genetic alterations in CLL , we infer TEDG by removing indirect orders in ISN . The TEDG of CLL ( Figure 3B ) shows a clear pattern of the flow of alterations , revealing that genetic alterations of CLL are sequentially ordered in a branching mode . To statistically assess the temporal pattern of the genetic lesions , we use the binomial test to assess significance by assuming that the number of in-degree and out-degree of each alteration are randomly distributed . Consistent with TEDG , this analysis suggests the following temporal order of the lesion: mutations of MYD88 , deletion 13q14 , +12 , mutations of NOTCH1 , RB1 deletion , 17p13 deletion , mutations of SF3B1 , 11q22-q23 deletion , BIRC3 deletion , mutations of TP53 , and mutations of BIRC3 ( Figure 3C ) . Deletion of 13q14 , +12 and mutations of NOTCH1 are significant early events in CLL development , while mutations of TP53 and BIRC3 are significant late events . Though the sample size prevents statistical considerations , MYD88 mutations might occur even before 13q14 deletion ( one of six patients ) , indicating that activation of the Toll-like receptor pathway could be important in the initiation of a fraction of CLL tumors ( Arvaniti et al . , 2011 ) . The sequential order of the genetic lesions remains consistent ( Pearson's correlation = 0 . 9 , p-value < 1e-3 ) when evolutionary networks are constructed with patients who received chemotherapy ( Figure 3—figure supplement 2 ) , suggesting that the order of development of the genetic lesions may be affected by therapy ( the Pearson's correlation of the order in patients without chemotherapy is 0 . 4 with p-value > 0 . 1 ) . By considering each single type of mutation affecting the same gene as a distinct and independent node , we construct the comprehensive ISN of CLL mutations ( Figure 4A ) . It is very difficult to capture useful information directly from ISN , while TEDG simplified the topology by capturing the backbone of tumor evolution ( Figure 4B ) . We observe that the monoallelic 13q14 deletion , RB1 deletion , and +12 are significant early events ( p-value < 0 . 01 ) , while the BIRC3 E537fs and the TP53 R248Q are significant late lesions ( p-value < 0 . 05 ) ( Figure 4—figure supplement 1 ) . Also , the analysis of ISN and TEDG shows that different lesions affecting the same gene may occur in distinct branches and stages . For example , mutations K700E and K666E of SF3B1 are late events in cases harboring 13q14 deletion , while mutations R273C of TP53 are late events in cases with +12 . Though the sample size prevents statistical considerations , TEDG reveals the H179R and Y234C missense substitutions in TP53 may be early events , while the R248Q , R273C , P152fs , and N239T substitutions are late events . 10 . 7554/eLife . 02869 . 010Figure 4 . TEDG analysis of specific genetic lesions . ( A ) Network representing the sequential order of specific genomic alterations . Two alterations are connected by a directed edge if they are observed to successively appear in at least one patient . The patient ID labels the corresponding edges , which are further colored by the number of patients . Nodes are colored by the fold change between out-degree and in-degree , indicating temporal order of corresponding alterations . The size of the nodes represents the recurrence of the alteration . ( B ) Tumor Evolutionary Directed Graph of 32 specific mutations . Some nodes and arrows in this figure are manually merged or adjusted for the purpose of illustration . DOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 01010 . 7554/eLife . 02869 . 011Figure 4—figure supplement 1 . Statistical test of in-degree and out-degree of specific alterations . All alterations are ranked by fold change between in-degree and out-degree . DOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 011 Based on pivotal NGS studies , two different evolutionary models have been proposed in CLL , namely gradual linear and branching evolution ( Knight et al . , 2012; Schuh et al . , 2012 ) . The analysis of our cohort ( excluding patients with Richter's transformation , refer to ‘Materials and methods’ for details ) shows that a minority of patients ( n = 3/60 , 5%; FDR = 0 . 1 ) are characterized by a significantly decreased or undetectable representation of the founding clone , coupled with a significant increase of a second subclone that represented a small subpopulation at an earlier time point , consistent with a branching evolution model ( highlighted by yellow circle in Figure 5A–B , and Table 1 ) . Interestingly , in all three cases clonal replacement events involve SF3B1 mutations and occur after treatment ( Figure 5C ) , suggesting that the branching evolution model is closely connected to the combination of treatments and the emergence of SF3B1 mutations ( p-value = 0 . 0016 by Fisher's exact test ) . This observation implies that , at the time of treatment requirement , limiting the knowledge of disease genetics to the dominant clone will likely be uninformative for accurate therapeutic decisions . Of particular interest in this scenario is the development of therapeutic strategies to prevent the branching evolution of the tumor , with the goal of eradicating dominant as well as minor clones ( Anderson et al . , 2011; Notta et al . , 2011; Ding et al . , 2012; Egan et al . , 2012; Keats et al . , 2012; Walker et al . , 2012; Rossi et al . , 2014 ) . 10 . 7554/eLife . 02869 . 012Figure 5 . Fitting the evolution models . ( A ) Distribution of the clonal evolution pattern in the 60 non-Richter cases . Three of the 60 cases show replacement during tumor progression . ( B ) Scatter plot showing p-values of observing increased and decreased subclones in the 80 samples of the 60 multi-time point patients . Samples with evidence of clonal replacement are located in the right-up corner ( highlighted by yellow circle ) . ( C ) Number of patients following linear vs branched evolution pattern according to SF3B1 mutational emergence and previous treatment ( p-value 0 . 0016 by Fisher's exact test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 01210 . 7554/eLife . 02869 . 013Table 1 . Patients with branching evolutionDOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 013Patient IDSampling timeTreatmentIncreased alterationsDecreased alterationsDetectable alterations#12001None––SF3B1 ( K666E ) , SF3B1 ( K700E ) , del13q2005FCSF3B1 ( K666E ) NoneSF3B1 ( K666E ) , SF3B1 ( K700E ) , del13q2008FCR/CAM/RBENSF3B1 ( K700E ) , del11qNoneSF3B1 ( K666E ) , SF3B1 ( K700E ) , del13q , del11q2011BENCAMSF3B1 ( K700E ) , del11q , del13q , delBIRC3SF3B1 ( K666E ) SF3B1 ( K700E ) , del13q , del11q , delBIRC3#142004RF––+12 , del13q2007FC/CAMNoneNone+12 , del13q2010BENCAMTP53 ( R248Q ) , del17p , SF3B1 ( K666E ) None+12 , del13q , TP53 ( R248Q ) , del17p , SF3B1 ( K666E ) 2012BENDOFATP53 ( R248Q ) , del17p , SF3B1 ( K666E ) , T12del13q+12 , del13q , TP53 ( R248Q ) , del17p , SF3B1 ( K666E ) #332002None––NOTCH1 ( P2514- ) 2004FCRSF3B1 ( K700E ) NoneNOTCH1 ( P2514- ) , SF3B1 ( K700E ) 2009FCRSF3B1 ( K700E ) NOTCH1 ( P2514- ) SF3B1 ( K700E ) FC , fludarabine , cyclophosphamide; FCR , fludarabine , cyclophosphamide , rituximab; CAM , Campath; RBEN , rituximab , bendamustine; BENCAM , bendamustine , Campath; RF , rituximab , fludarabine; BENDOFA , bendamustine , ofatumumab . To further investigate the relationship between driver genetic lesions in TEDG , we assess their associations and anti-associations in a cross-sectional cohort of 1403 CLL patients , of which 1054 cases are informative in at least one lesion ( Figure 6A , Table 2 ) . Most of the co-mutations ( connected in red in Figure 6B ) are experimentally confirmed by previous studies , including: co-occurrence of 17p13 deletion and TP53 mutations , reflecting a typical two-hit model for tumor suppressor genes; co-occurrence of BIRC3 deletion , 11q22-q23 deletion , and BIRC3 mutations; and the relationship between NOTCH1 mutations and +12 ( Balatti et al . , 2012; Del Giudice et al . , 2012 ) . In addition to previously reported associated lesions , this large cohort of patients allows the statistical power to reveal other significant and previously unreported co-occurrence interactions , including the co-occurrence of BIRC3 abnormalities with +12 and NOTCH1 mutations and the co-occurrence of 13q14 deletion and BIRC3 deletion . This large cohort also reveals mutually exclusive relationships between +12 and 13q14 deletion and between NOTCH1 mutations and 13q14 deletion ( connected in blue in Figure 6B ) . The statistical association analysis is consistent with the prediction of TEDG and indirectly validates the ability of TEDG in successfully capturing the major information in tumor evolution . 10 . 7554/eLife . 02869 . 014Figure 6 . Association network of CLL lesions in a larger cohort . ( A ) The mutation status of 1054 CLL patients , which is informative in 11 most common genetic lesions in this leukemia . ( B ) The association analysis of CLL genetic lesions in TEDG . Lesion pairs are connected if they are significantly co-mutated ( red ) or mutually exclusive ( blue ) . ( C ) Box plot of growth rate of all genetic lesions . DOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 01410 . 7554/eLife . 02869 . 015Figure 6—figure supplement 1 . Box plot to show the difference of Richter and non-Richter , and mutations of TP53 in Maximal Mutation Frequency Slope . DOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 01510 . 7554/eLife . 02869 . 016Figure 6—figure supplement 2 . Survival analysis of fittest genomic alterations . ( A ) Kaplan–Meier curve showing the cumulative probability of overall survival ( OS ) for patients with high and low MMFS ( maximal mutation frequency slope ) . ( B ) Kaplan–Meier curve showing the cumulative probability of Richter syndrome ( RS ) transformation for patients with high and low MMFS . p-values are based on log-rank test . CI ( confidential interval ) is estimated by the Greenwood method . DOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 01610 . 7554/eLife . 02869 . 017Figure 6—figure supplement 3 . Comparison of the growth rates of subclonal alterations between early and late events . p-value is calculated by Wilcoxon Rank-Sum Test . Growth rate indicates the change of mutation frequency per year . DOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 01710 . 7554/eLife . 02869 . 018Table 2 . Clinical features at presentation of the CLL cohortsDOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 018TEDG analysisSA analysisNTotal%NTotal%IGHV homology >98%376953 . 6532138038 . 6del13q14357050 . 0682140348 . 6+12197027 . 1205140314 . 6del11q22-q2387011 . 413614039 . 7del17p13207028 . 611414038 . 1TP53 mutation227031 . 411414018 . 1NOTCH1 mutation227031 . 4150140310 . 7SF3B1 mutation177024 . 310014037 . 1MYD88 mutation6708 . 65214033 . 7BIRC3 mutation107014 . 34114032 . 9BIRC3 deletion6708 . 67914035 . 6TEDG analysis , Tumor Evolutionary Directed Graphs analysis; SA analysis , Statistical Association analysis; IGHV , immunoglobulin heavy variable gene; FISH , fluorescence in situ hybridization . By integrating the topology of TEDG to the association analysis , two distinct and mutually exclusive evolutionary paths of CLL evolution are disclosed . The first evolutionary path involves CLL harboring +12 and NOTCH1 mutations as early driver events . In this path , clonal evolution proceeds toward the development of TP53 and BIRC3 abnormalities . The second evolutionary path involves 13q14 deletion as early driver lesion and proceeds toward the development of SF3B1 mutations and BIRC3 abnormalities . Overall , our results are consistent with the different clinico-biological phenotype of +12 CLL and 13q14 deleted CLL and further support the hypothesis that at least two distinct genetic subtypes of CLL exist . We reason that if a subclone replaces the major clone during tumor evolution , it is because of its higher fitness . Although fitness of a clone is not directly measurable , the growth rate of mutation frequency of drivers in this clone can serve as an indicator . Given this , mutations related to high fitness clones can be identified by extracting the alterations that rapidly change their allele/cell frequency within the tumor . We define the growth rate of each genetic lesion as the average increasing speed of mutation cell frequency per year ( ‘Materials and methods’ ) . Interestingly , the growth rates of late events ( 17p13 deletion , mutations of SF3B1 , 11q22-q23 deletion , BIRC3 deletion , mutations of TP53 , and mutations of BIRC3 ) are found higher than those of early events ( mutations of MYD88 , 13q14 deletion , +12 , mutations of NOTCH1 , and RB1 deletion ) with p-value = 0 . 005 by Wilcoxon Rank-Sum Test ( Figure 6C ) , revealing that late events may drive higher fitness or more aggressive clones . Note that the initiating lesions are usually clonal , presenting in most of the tumor cells , and therefore do not increase in frequency at the same magnitude as subclonal mutations . However , after eliminating clonal genetic lesions with MCF > 20% in the above analysis , we still find the growth rates of late and early events to be significantly different with p-value = 0 . 0114 by Wilcoxon Rank-Sum Test ( Figure 6—figure supplement 3 ) . We define maximal mutation frequency slope ( MMFS ) as the rate of allele frequency change of the fastest-growing clone in a patient ( Table 3 , ‘Materials and methods’ ) . MMFS is a function that aims at characterizing the relative fitness of a particular clone carrying a particular mutation . In our cohort , only genetic lesions affecting the TP53 gene show a statistically significant association with MMFS ( p-value = 0 . 0164 , by Wilcoxon Rank-Sum Test ) . Clinically , the fastest-growing clones are strongly correlated with poor survival and Richter syndrome transformation , consistent with the fact that CLL transformed to Richter syndrome presents significantly higher MMFS values ( p-value = 0 . 002 , by Wilcoxon Rank-Sum Test ) ( Figure 6—figure supplement 1 ) . Indeed , by survival analysis , having a clone with high MMFS associates with an approximately threefold significant increase in the hazard of death ( HR: 3 . 17; 95% CI: 1 . 38–7 . 40; p-value = 0 . 005 ) and a significant shortening of overall survival ( 47% at 5 years ) ( Figure 6—figure supplement 2 ) . Most deaths ( 77% ) in patients carrying a clone with high MMFS are due to Richter syndrome transformation . Consistently , patients having a clone with high MMFS show an approximately fourfold increased risk of Richter syndrome development ( HR: 4 . 61; 95% CI: 1 . 54–13 . 80; p-value = 0 . 003 ) and an ∼50% of them are projected to develop Richter syndrome at 5 years ( Figure 6—figure supplement 2 ) . 10 . 7554/eLife . 02869 . 019Table 3 . Patients showing high maximal mutation frequency slope ( MMFS ) DOI: http://dx . doi . org/10 . 7554/eLife . 02869 . 019PatientMMFSFast growing mutationsMCF1 ( % ) MCF2 ( % ) dT ( moths ) TreatmentRichter syndrome transformation#52117 . 7TP53 ( P152+ ) 2 . 111 . 4<1NoneYes#689 . 1TP53 ( R273C ) 0 . 0100 . 011NoneNo#516 . 7SF3B1 ( I704F ) 12 . 268 . 95RCVPYes#376 . 4TP53 ( R248Q ) 3 . 890 . 010RDHAOXYes#575 . 9NOTCH1 ( P2514- ) 0 . 894 . 212NoneYes#474 . 6del13q37 . 6100 . 013NoneNo#143 . 9del17p18 . 990 . 016ANo#423 . 4TP53 ( N239T ) 0 . 0100 . 016RDHAOXYes#43 . 3del17p53 . 6100 . 015CLB-ONo#543 . 0NOTCH1 ( P2415- ) 100 . 0†100 . 022FCONo#222 . 9BIRC3 deletion0 . 093 . 929FCRNo#202 . 8del13q0 . 0100 . 036CLBNo#632 . 5BIRC3 ( M388V ) 0 . 086 . 024FCRYes#62 . 3del17p0 . 092 . 134CLBNo#382 . 3NOTCH1 ( P2514- ) 41 . 742 . 94CVPYes#132 . 2TP53 ( G136H ) 0 . 0100 . 045RDHAOXNo#12 . 1del11q11 . 8100 . 041FCR/A/BRNoMMFS , maximal mutation frequency slope ( in standard deviation per year ) ; MCF1 , mutation cell frequency of selected mutation at the first time point; MCF2 , mutation cell frequency at the second time point; dT , the elapsed time between two samples; RCVP , rituximab , cyclophosphamide , vincristine , prednisone; RDHAOX , rituximab , dexamethasone , high dose cytarabine , oxaliplatin; A , alemtuzumab; CLB-O , chlorambucil , ofatumumab; FCO , fludarabine , cyclophosphamide , ofatumumab; FCR , fludarabine , cyclophosphamide , rituximab; CLB , chlorambucil; CVP , rituximab , cyclophosphamide , vincristine , prednisone; BR , bendamustine , rituximab . †Total number of the cancer cells with NOTCH1 alteration does not change , but the allele frequency of the mutation increases because of the deletion of the wild-type allele . It is not known whether there is a preferred order of mutations in the development of cancer and how the order of mutations may impact clinical outcomes . We propose a TEDG framework , which is able to integrate longitudinal and cross-sectional genomic data into a directed graph of tumor evolution . The flow in this graph reveals underlying paths of tumor progression . Starting from time-series genomic data in one patient , a sequential network is reconstructed to capture the possible historical events of the evolution in the particular tumor . Integrating sequential data from many patients , we collect the ensemble of tumor histories by ISN , which presents a comprehensive topology of evolutionary landscape . A recent technology of network deconvolution is able to distinguish direct and indirect interactions using spectral methods , which reduce the weights of indirect connections and remove low weighted edges by selecting appropriate cut-offs ( Feizi et al . , 2013 ) . To adapt this method to our particular problem depending on the number of samples , we introduce a degree parameter , β . To gain insights of the selection of beta , we propose a strategy to simulate tumor evolution by a one-step Markov process , with transition probability derived from Nordling's multi-mutation model . A linear and a branching evolution model are separately simulated in a study of four alterations . The proof-of-concept simulation shows that the TEDG strategy can obtain excellent performance in capturing the evolutionary history when the number of cases is beyond 30 . We apply TEDG to CLL and: ( i ) reconstruct an evolutionary network representing the sequential order of genetic lesions occurring during the course of this disease; ( ii ) investigate statistical associations and competitiveness of driver genetic lesions to uncover evolutionary paths; and ( iii ) correlate the order of alterations with the kinetics of changes in their allelic abundance , to identify the molecular alterations associated with the fastest growth of a subclone and their clinical impact on outcome . Phylogenetic trees are often employed to infer temporal order of gene mutations by assuming that common ancestors are early events , but phylogenetic methods usually require higher number of alterations than the ones available in our cohort . Standard statistical techniques to assess the significance of different branches rely on bootstrapping segregating sites , and the statistical power of these techniques using longitudinal data from cancer patients in few selected driver genes is extremely limited . None of the standard phylogenetic methods such as distance matrix methods , Bayesian phylogenetic methods , or parsimony methods can produce branches with enough bootstrap support ( >80% ) . On the other hand , TEDG works with aggregate data from several patients allowing statistical power for robust estimates . According to TEDG analysis , the molecular lesions of CLL are temporally ordered in a specific fashion rather than being randomly accumulated . Among recurrent lesions , 13q14 deletion and +12 are initiating events , while mutations of TP53 and BIRC3 are late . This observation is consistent with the notion that del13q and +12 occur at a similar prevalence in all CLL phases , including monoclonal B-cell lymphocytosis , a condition that often anticipates overt CLL , thus suggesting that they are early events ( Rawstron et al . , 2008; Nieto et al . , 2009; Rossi et al . , 2009b; Fazi et al . , 2011; Pasqualucci et al . , 2011; Kern et al . , 2012 ) . Also , 13q14 deletion has been directly implicated in CLL initiation ( Klein et al . , 2010 ) . On the other hand , late-onset abnormalities are known to accumulate in more advanced phases of the disease , thus suggesting that they are second-hit lesions ( Zenz et al . , 2009; Rossi et al . , 2012 ) . In our model , mutations of SF3B1 and 11q22-q23 deletions appear to be acquired at an intermediate time point . This observation is consistent with the clinical observation that mutations of SF3B1 and 11q22-q23 deletion co-segregate with an intermediate prognostic group of CLL ( Döhner et al . , 2000; Oscier et al . , 2013; Rossi et al . , 2014 ) . By integrating the cross-sectional data of association and anti-association between genetic lesions at CLL diagnosis with longitudinal data of clonal evolution from TEDG , two distinct and mutually exclusive evolutionary paths of evolution emerge . The first evolutionary path involves CLL initially harboring +12 and NOTCH1 mutations . In this path , clonal evolution proceeds toward the development of TP53 and BIRC3 abnormalities . The second evolutionary path involves CLL initially harboring 13q14 deletion and proceeds toward the development of SF3B1 mutations and BIRC3 abnormalities . Our data , as well as previous analyses ( Rossi et al . , 2013 ) , show that deletion of 13q14 and +12 are mutually exclusive in CLL . From a clinical standpoint , +12 CLL is known to stand out of typical 13q14 deleted CLL because of their atypical cytomorphology and phenotype , the more intense expression of CD20 , the preferential nodal presentation and the higher risk of transformation to Richter syndrome . These notions along with our novel observation that +12 and 13q14 deleted CLL proceed through distinct paths of clonal evolutions further support the hypothesis that at least two distinct genetic subtypes of CLL exist . The presence at diagnosis of fully clonal genetic lesions that are considered late genetic events in CLL evolution , such as TP53 abnormalities , is already known to have an adverse impact on disease outcome ( Zenz et al . , 2008 , 2010; Dicker et al . , 2009; Malcikova et al . , 2009; Rossi et al . , 2009a; Gonzalez et al . , 2011 ) . We observe that a fraction of TP53 abnormalities , though subclonal at presentation , lead to expansion of fitter subclones that progressively predominate with time in the tumor architecture . The clinical impact of this observation is supported by the evidence that harboring higher fitness alterations correlates with poor survival and increased risk of CLL transformation into an aggressive lymphoma , an often-lethal complication known as Richter syndrome . In this study , we have considered some of the most important and clinically relevant drivers of CLL ( Rossi et al . , 2013 ) . The ATM mutations are also commonly found in CLL; however , the ATM gene is large and highly polymorphic without well-known hotspots . Therefore , the distinction of its driver mutations from constitutional variants is challenging . Also , from a clinical standpoint , the prognostic relevance of ATM mutations in CLL is still controversial , so we have excluded this gene from the current study ( Lozanski et al . , 2012; Ouillette et al . , 2012; Skowronska et al . , 2012 ) . In conclusion , the application of TEDG to CLL provides the proof-of-principle that this method is able to: ( i ) improve cancer classification and dissection into genetic subgroups following different paths of clonal evolution; and ( ii ) anticipate the genetic composition of the progressive/relapsed disease according to the genetic composition of the tumor clone at the time of diagnosis , including the development of genetic lesions associated with chemorefractoriness . TEDG provides a general framework that could be used to study and compare the evolutionary histories of other tumors . We collected 202 CLL patients provided with at least 2 sequential samples and followed for at least 2 years after presentation ( median interval between baseline and last sequential sample: 62 . 8 months , range 24–150 months ) . 70 out of the 202 patients were informative for TP53 , NOTCH1 , SF3B1 , MYD88 , or BIRC3 mutations ( which are the most common mutations in this leukemia ) , overall accounting for 164 paired sequential samples collected at diagnosis , progression , and last follow-up . To analyze the dynamics of those mutations , we carried out high-depth next generation sequencing ( NGS ) to quantify the mutation allele frequencies at each time point of the disease course and to establish their modifications during leukemia progression . Additionally , copy number abnormalities at 13q14 , chromosome 12 , 11q22-q23 , 17p13 , as well as at the RB1 , and BIRC3 loci were investigated by FISH . Inclusion criteria for the longitudinal analysis of clonal evolution were: ( i ) having at least two years of follow-up after diagnosis; and ( ii ) availability of >2 sequential samples collected at: ( a ) diagnosis; ( b ) each progression requiring treatment; ( c ) last follow-up . CLL diagnosis was according to 2008 IWCLL-NCI criteria and confirmed by a flow cytometry score >3 in all cases . Monoclonal B-cell lymphocytosis ( MBL ) was excluded . The study was approved by the institutional ethical committee of the Azienda Ospedaliero-Universitaria Maggiore della Carità di Novara affiliated with the Amedeo Avogadro University of Eastern Piedmont , Novara , Italy ( Protocol Code 59/CE; Study Number CE 8/11 ) . Patients provided informed consent in accordance with local IRB requirements and Declaration of Helsinki . CLL samples were extracted from fresh or frozen peripheral blood mononuclear cells ( PBMC ) isolated by Ficoll-Paque gradient centrifugation . In all cases , the fraction of tumor cells corresponded to 70–98% as assessed by flow cytometry . Matched normal DNA from the same patient was obtained from saliva or from purified granulocytes and confirmed to be tumor-free by PCR of tumor-specific IGHV-D-J rearrangements . High-molecular-weight ( HMW ) genomic DNA was extracted from tumor and normal samples according to standard procedures . DNA was quantified by the NanoDrop 2000C spectrophotometer ( Thermo Scientific , Wilmington , DE ) . To validate TEDG , 1403 newly diagnosed and previously untreated CLL were enrolled in the study , of whom 931 ( 66% ) were provided with clinical data and regular follow-up ( Table 2 ) . The frequency of alterations was higher in TEDG analysis group , because only informative patients with known mutated driver genes were included . Cross-sectional investigation of the associations and anti-associations between the most recurrent genetic lesions at diagnosis was based on the entire CLL cohort of 1403 patients . Survival analysis was based on CLL cases provided with clinical data ( n = 931 ) . Mutation screening of the IGHV , TP53 , NOTCH1 , SF3B1 , MYD88 , and BIRC3 genes were performed by Sanger sequencing . Among cases that were informative for TP53 , NOTCH1 , SF3B1 , MYD88 , or BIRC3 mutations at any time point in the disease course , we carried out high-depth NGS to quantify the variant allele frequencies at each stage of their disease . Positions known to harbor TP53 , NOTCH1 , SF3B1 , MYD88 , or BIRC3 mutations by Sanger sequencing were amplified from genomic DNA by oligonucleotides containing the gene-specific sequences , along with 10-bp MID tag for multiplexing and amplicon library A and B sequencing adapters . The obtained amplicon library was subjected to deep sequencing on the Genome Sequencer Junior instrument ( 454 Life Sciences ) . In order to obtain at least 700-fold coverage per amplicon , no more than 100 amplicons/run were analyzed . The obtained sequencing reads were mapped to reference sequences and analyzed by the Amplicon Variant Analyzer software ( Roche ) to establish the mutant allele frequency . The sequencing depth in this study is on average 1200× , which is sufficient for a highly sensitive detection of mutations with allele frequency >1% out of the background error noise ( Rossi et al . , 2014 ) . Probes used for FISH analysis were: ( i ) LSID13S319 ( 13q14 deletion ) , CEP12 ( trisomy 12 ) , LSIp53 ( 17p13/TP53 deletion ) , and LSIATM ( 11q2-q23/ATM deletion ) ( Abbott , Rome , Italy ) ; and ( ii ) the RP11-177O8 ( BIRC3 ) BAC clone . The labeled BIRC3 BAC probe was tested against normal control metaphases to verify the specificity of the hybridization . For each probe , at least 400 interphase cells with well-delineated fluorescent spots were examined . Nuclei were counterstained with 4′ , 6′-diamidino-2-phenylindole ( DAPI ) and antifade reagent , and signals were visualized using an Olympus BX51 microscope ( Olympus Italia , Milan , Italy ) . The count of CD19+CD5+ cells is a standard assay to define the representation of CLL cells in a diagnostic or research sample as measured by Fluorescence-activated cell sorting ( FACS ) analysis . A FACSCalibur flow cytometer ( Becton–Dickinson ) was utilized for the analysis . Expression of CD5 and CD19 was analyzed by combining Peridinin-Chlorophyll-Protein–Cyanine-5 . 5 ( PerCP–Cy5 . 5 ) -conjugated anti-CD19 mAbs and fluorescein isothiocyanate-conjugated anti-CD5 mAbs . In order to estimate the proportion of cells co-expressing CD19 and CD5 , for each sample events were acquired by gating on low forward and side scatter ( FSC/SSC ) CD19+ cells , which were further divided into CD5− and CD5+ subsets . Irrelevant isotype-matched antibodies ( Becton–Dickinson ) were used to determine background fluorescence . FACS data of all samples are listed in Supplementary file 2 . To consider tumor content , we performed CD19/CD5 FACS analysis to quantify the fraction of tumor cells . To unify FISH and NGS data , and also to adjust MAF of mutations in genes with copy number abnormalities , we introduced mutation cell frequency ( MCF ) . As shown in Figure 1—figure supplement 1A , MCF represents the fraction of cancer cells containing particular alterations . Unlike MAF , copy numbers or tumor content do not affect MCF of a point mutation . For instance , if tumor purity is 70% and MAF of a heterogeneous variance in a diploid region is 0 . 2 , then MCF is approximated by 0 . 2 × 2/0 . 7 ≈ 0 . 57 . To infer MCF of different types of lesions , we applied the following strategy:i . MCF of a copy number abnormality ( i . e . , T12 , del11q , del13q_x1 , del13q_x2 , del17p , delBIRC3 , delRB1_x1 , delRB1_x2 ) was calculated by FISH analysis divided by the fraction of CD19/CD5 cells based on FACS , correcting for the tumor purity . ii . Genes whose copy numbers were not frequently changed , such as MYD88 and SF3B1 , were considered as diploid . So the MCF should be close to twofolds of MAF . However , MAF of some mutations might exceed 0 . 5 , because of the random noise introduced in PCR . Here , MCF could be simply defined as a Piecewise function:MCF={2×MAF , MAF<0 . 51 , MAF≥0 . 5 . But this function was not smooth and it was not able to distinguish MAF = 0 . 5 and MAF = 0 . 9 . To smooth this function , a Hill function was introduced , assumingMCF=K ( MAF ) n[1+K ( MAF ) n] , where K and n are parameters . To find optimal parameters , such that MCF ≈ 2 × MAF , we optimize the following objective function:minK , nF ( K , n ) =[∫00 . 5 ( K ( MAF ) n[1+K ( MAF ) n]−2×MAF ) 2dMAF] . A grid-search method was applied to exhaust potential values of K and n . Particularly , for K ∈ {20 , ⋯ , 29} and n ∈ {1 , ⋯ , 10} , F was calculated by the numerical integration of the above equation . Figure 1—figure supplement 1B shows that the optimal solution is K^=16 and n^=2 . So for mutations of MYD88 and SF3B1 , MCF=16 ( MAF ) 2[1+16 ( MAF ) 2] . This Hill function , which was used to smooth the piecewise function , showed a more smooth correlation between MAF and MCF ( Figure 1—figure supplement 1C ) . Additionally , by comparing FACS measurement of tumor purity , this Hill function form of MCF is more robust than either MAF without adjustment or simplistic piecewise function in assessing the fraction of cancer nuclei ( Figure 1—figure supplement 1E ) . iii . Genes that are frequently deleted in CLL , such as TP53 and BIRC3 , were analyzed in the model without considering copy-neutral LOH or homozygous deletions ( Figure 1—figure supplement 1D ) . We assume four different cell types: wild type ( upper left ) , heterozygous deletion ( upper right ) , mutation ( lower left ) , and both mutation and deletion ( lower right ) . The cell fractions were represented by xi ( i = 1 , ⋯ , 4 ) , respectively . SoMAFmut=x3+x42x1+x2+2x3+x4=x3+x42− ( x2+x4 ) =MCFmut2−MCFdel and then MCF of mutation depends on MAF of mutation , as well as MCF of the deletion:MCFmut=MAFmut ( 2−MCFdel ) . To smooth it , we optimizedminK , nF ( K , n;MCFdel ) =[∫00 . 5 ( K ( MAF ) n[1+K ( MAF ) n]−2×MAF+MAF×MCFdel ) 2dMAF] . Here if we use K^ and n^ to represent the optimal solution given MCFdel , for mutations of TP53 and BIRC3 , MCF=K^ ( MAF ) n^[1+K^ ( MAF ) n^] . The MCF values were then divided by the fraction of CD19+CD5+ cells based on FACS . To generate the artificial data for TEDG , we simulate cancer clonal evolution as a one-step Markov process , that is , the transition probability of mutation profile P ( ψtk\|ψtk−1 , ⋯ , ψt1 ) =P ( ψtk\|ψtk−1 ) , where ψtk is the observed mutation profile at time tk , which depended on the mutation profile at time tk−1 . To define the transition probability , we focused on two simple models of four mutations x1 , x2 , x3 , x4 . In the linear evolution model , we assumed that all mutations were mutated in a linear order shown in the left panel of Figure 2A . All possibilities of mutation status are{π0=ϕ , π1= ( x1 ) , π2= ( x1 , x2 ) , π3= ( x1 , x2 , x3 ) , π4= ( x1 , x2 , x3 , x4 ) } . According to Nordling's multi-mutation model ( Nordling , 1953 ) , the transition probability was defined asP ( ψtk=πj\|ψtk−1=πi ) ={[1−e−f· ( tk−tk−1 ) ]j−i−[1−e−f· ( tk−tk−1 ) ]j−i+1 , j>i;e−f· ( tk−tk−1 ) , j=i;0 , j<i . , where f represents fitness of the new mutation . The transition probability is shown in Figure 2B . The longitudinal data were then generated for each patient following the above model . To simplify the model , we fixed the number of time points ( three ) and the length of interval between time points ( ten ) . We asked whether or not TEDG framework can reconstruct the order of mutations and how many patients are required . To answer this question , we applied a grid-search strategy to optimize parameter β and number of patients . Specifically , all possible combinations of parameter β from 0 to 1 ( step by 0 . 1 ) and patient number from 1 to 100 ( step by 1 ) are exhausted . For each combination , we randomly simulated the cancer patients and applied TEDG framework to reconstruct the order for 10 times . The frequency of reconstructing the exact order was defined as the accuracy of TEDG . Figure 2D shows an example with β equals 0 . 2 and the number of patients equals 15 . Edge weights of ISN present the number of simulated patients with one mutation happening before another . The techniques of deconvolution and minimal spanning tree reconstruct the real structure by removing indirect interactions . Figure 2E summarizes the correlation between number of patients and optimal accuracy of TEDG . Different from linear model , the branching model assumes that x3 and x4 are independently following the mutation of x2 . All possibilities of mutation status are{π0=ϕ , π1= ( x1 ) , π2= ( x1 , x2 ) , π3= ( x1 , x2 , x3 ) , π4= ( x1 , x2 , x4 ) } . According to Nordling's multi-mutation model ( Nordling , 1953 ) , when j ≤ 2 , the transition probability is the same as linear model:P ( ψtk=πj\|ψtk−1=πi ) ={[1−e−f· ( tk−tk−1 ) ]j−i−[1−e−f· ( tk−tk−1 ) ]j−i+1 , j>i;e−f· ( tk−tk−1 ) , j=i;0 , j<i . When j > 2 , the transition probability isP ( ψtk=πj\|ψtk−1=πi ) ={[1−e−f· ( tk−tk−1 ) ]3−i , i<3;1 , i=j;0 , otherwise . Figure 2F shows an example of branching model with β equals 0 . 2 and number of patient equals 15 . Figure 2G summarized the correlation between number of patients and optimal accuracy of TEDG in branching model . To construct the sequential network of alterations for each tumor , we monitored the presence or absence of each genetic lesion in each sample . By collecting the clonal representation of each lesion from NGS and FISH analysis , we defined the status of each lesion with a cut-off of 5% . If the frequency of a genetic lesion is larger than 5% , we will name it as present; otherwise absent . If event A predates event B , we added a directed link between A and B . The ISN was constructed by pooling all sequential networks from different patients . To simplify ISN , we removed self-loops by subtracting the weight of weaker direction . To hierarchically layout the simplified ISN , we use yFiles Hierarchical Layout algorithm in Cytoscape 2 . 8 . 2 , which well represents main direction or ‘flow’ in a directed network . With this method , nodes were placed in hierarchically arranged layers and the nodes within each layer are ordered in such a way that minimizes the number of edge crossings ( Smoot et al . , 2011 ) . To statistically test whether the alterations are temporally ordered or randomly accumulated , we assume that in-degree ( the number of incoming arrows ) dini is equal to out-degree ( the number of outgoing arrows ) douti for each node i , and then ∑i ( dini−douti ) 2douti follows a chi square distribution with degree of freedom n − 1 , where n is the number nodes in the network . To test the statistical association of lesions in TEDG , we counted the number of samples carrying each pair of lesions and then calculated the p-value based on hypergeometric distribution to test whether the two genetic lesions are independent or not . To consider the effect of multi-hypothesizes , we corrected p-values with Bonferroni method and made a cut-off of 0 . 05 . If two lesions were significantly co-mutated , a red link was added , while if they were significantly mutually exclusive , a blue edge was added . Suppose Gdir is the adjacent matrix of all direct interactions/orders , the simplified ISN should be a summary of direct and indirect orders in the deconvolution formula Gdir = Gobs ( I + βGobs ) −1 , where , Gobs is the observed weighted adjacent matrix , I is the identify matrix , and β is a scaling factor between zero and one indicating the degree of deconvolution . The resulting matrix of deconvolution formula indicates the score of an edge to be a direct interaction ( edge weights in the middle panel of Figure 2D , F ) . We then introduced a minimal spanning tree-based method to determine the final TEDG , in which we transformed the deconvoluted weights with a negative exponential function , and then calculated the minimal spanning tree with Prim's algorithm . We developed Fit the Evolutionary Model ( FEM ) to properly fit the evolutionary model by systematically identifying clonal replacement in all sample pairs . FEM defines Z-scores for variant allele frequency by NGS and percentage of nuclei harboring abnormalities by FISH to represent normalized changes of the frequency of the lesions . For ith mutation in jth patient at the kth time point tkZij ( tk ) =fij ( tk ) −fij ( tk−1 ) σseqfij ( tk ) +fij ( tk−1 ) 2 ( 1−fij ( tk ) +fij ( tk−1 ) 2 ) , where f is the mutation frequency of some particular gene mutation and σseq is its standard variation . Similarly , for the ith copy number change in the jth patient at the kth time point tkZij ( tk ) =fij ( tk ) −fij ( tk−1 ) σFISHfij ( tk ) +fij ( tk−1 ) 2 ( 1−fij ( tk ) +fij ( tk−1 ) 2 ) , where f is the frequency nuclei harboring the abnormality of some particular gene mutation and σFISH is its standard variation . The change of genetic lesion frequency is a synergic effect of treatment , tumor progression , and experimental noises , such as sequencing error or change of tumor purity . To obtain the variance caused by background noises , FEM robustly fit σseq and σFISH by eliminating data obviously affected by treatments or tumor progression . Based on this , we could calculate p-values of each genetic lesion in all sample pairs . To assess whether there was a significantly increased ( or decreased ) subclone in a given sample pair , we use Fisher's combinational test to combine p-values of all increased ( or decreased ) genetic lesions . The resulting p-values were separately defined as P_increase and P_decrease . Sample pairs that were significant in both P_increase and P_decrease were defined as sample pairs containing replacement events , which were further fit to the branching evolution model . All the others were compatible with the gradual linear model . To investigate the fitness of genetic lesions , we defined growth rate and the maximal mutation frequency slope ( MMFS ) . For the ith mutation in the jth patient at the kth time point tk , mutation frequency slope is defined as sij ( tk ) =zij ( tk ) tk−tk−1 , where z is the Z-scores defined above . The growth rate of ith mutation in the jth patient at the kth time point tk is defined as gij ( tk ) =max{sij ( tk ) , 0} . Furthermore , for the jth patient , MMFS is defined as sj=maxi{maxk[sij ( tk ) ]} . Overall survival ( OS ) was measured from date of initial presentation to date of death from any cause ( event ) or last follow-up ( censoring ) . The cumulative probability of Richter syndrome transformation was measured from date of initial presentation to date of the biopsy documenting Richter syndrome transformation ( event ) , death or last follow-up ( censoring ) . Survival analysis was performed by the Kaplan–Meier method . The crude association between time-fixed exposure variables at diagnosis and survival was estimated by Cox proportional hazard regression . The analysis was performed with the Statistical Package for the Social Sciences ( SPSS ) software v . 20 . 0 ( Chicago , IL ) . The mutation hotspots of TP53 ( exons 4–9 , including splicing sites; RefSeq NM_000546 . 5 ) , NOTCH1 ( exon 34 , including splicing sites; RefSeq NM_017617 . 2 ) , SF3B1 ( exons 14 , 15 , 16 , 18 , including splice sites; RefSeq NM_012433 . 2 ) , MYD88 ( exons 3 , 5 , including splicing sites; RefSeq NM_002468 . 4 ) , and BIRC3 ( exons 6–9 , including splicing sites; RefSeq NM_001165 . 4 ) genes were analyzed by PCR amplification and DNA direct sequencing of high-molecular weight genomic DNA . Sequences for all annotated exons and flanking splice sites were retrieved from the UCSC Human Genome database using the corresponding mRNA accession number as a reference . PCR primers , located ∼50 bp upstream or downstream to target exon boundaries , were either derived from previously published studies or designed in the Primer 3 program ( http://frodo . wi . mit . edu/primer3/ ) and filtered using UCSC in silico PCR to exclude pairs yielding more than a single product . All PCR primers and conditions are listed in Supplementary file 1 . Purified amplicons were subjected to conventional DNA Sanger sequencing using the ABI PRISM 3100 Genetic Analyzer ( Applied Biosystems ) and compared to the corresponding germline sequences using the Mutation Surveyor Version 4 . 0 . 5 software package ( SoftGenetics ) after automated and/or manual curation . Of the evaluated sequences , 99% had a Phred score of 20 or more and 97% had a score of 30 or more . Candidate variants were confirmed from both strands on independent PCR products . The following databases were used to exclude known germline variants: Human dbSNP Database at NCBI ( Build 136 ) ( http://www . ncbi . nlm . nih . gov/snp ) ; Ensembl Database ( http://www . ensembl . org/index . html ) ; The 1000 Genomes Project ( http://www . 1000genomes . org/ ) ; five single-genome projects available at the UCSC Genome Bioinformatics resource ( http://genome . ucsc . edu/ ) . Synonymous variants , previously reported germline polymorphisms and changes present in the matched normal DNA were removed from the analysis . PCR amplification of IGHV-IGHD-IGHJ rearrangements was performed on high molecular weight genomic DNA using IGHV leader primers or consensus primers for the IGHV FR1 along with appropriate IGHJ genes , as previously described . PCR products were directly sequenced with the ABI PRISM BigDye Terminator v1 . 1 Ready Reaction Cycle Sequencing kit ( Applied Biosystems ) using the ABI PRISM 3100 Genetic Analyzer ( Applied Biosystems ) . Sequences were analyzed using the IMGT databases and the IMGT/V-QUEST tool ( version 3 . 2 . 17 , Université Montpellier 2 , CNRS , LIGM , Montpellier , France ) .	A historical event is often the culmination of the preceding circumstances . The same can be said of cancer as a disease . Cancer results from genetic mutations that disrupt the normal biological processes within a cell , removing the fail-safes that prevent it from growing and reproducing uncontrollably . Cancer is not caused by just one mutation , and once one gene is malfunctioning , other genes become much more likely to mutate . Although modern sequencing methods have revealed many of the genes that mutate in several different kinds of cancer , uncovering when each of these mutations occurs has been more difficult . Knowing when each mutation occurs could make it easier to predict how the cancer will progress and could also help target cancer treatments more effectively . Wang , Khiabanian , Rossi et al . have devised a new method of studying the history of genetic mutations of cancer patients . This combines a ‘longitudinal’ method that looks at how mutations develop in a single tumor by taking samples from it at different times and ‘cross-sectional’ methods that make predictions based on data collected from a large number of patients . Wang , Khiabanian , Rossi et al . call this method ‘tumor evolutionary directed graphs’ ( TEDG ) , as it produces a graph that shows how different gene mutations are related to each other . Initial tests showed that the TEDG method could accurately decipher the main chain of events in cancer evolution when used on data collected from at least 30 patients . Wang , Khiabanian , Rossi et al . then used TEDG on data from 164 tumor samples collected over 12 years from 70 patients with chronic lymphocytic leukemia , the type of leukemia that is most widespread amongst adults in Western countries . This uncovered two separate ways that this cancer may develop , one of which has a higher risk of life-threatening complications . Knowing which of the two ways chronic lymphocytic leukemia is progressing in a patient could help treat the disease , as each pathway responds differently to different treatments . In addition , understanding the paths that cancer progression follows could also provide early warning signals of the mutations that will occur next . This could help to develop alternative , targeted cancer treatments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology" ]	2014	Tumor evolutionary directed graphs and the history of chronic lymphocytic leukemia
Borophagine canids have long been hypothesized to be North American ecological ‘avatars’ of living hyenas in Africa and Asia , but direct fossil evidence of hyena-like bone consumption is hitherto unknown . We report rare coprolites ( fossilized feces ) of Borophagus parvus from the late Miocene of California and , for the first time , describe unambiguous evidence that these predatory canids ingested large amounts of bone . Surface morphology , micro-CT analyses , and contextual information reveal ( 1 ) droppings in concentrations signifying scent-marking behavior , similar to latrines used by living social carnivorans; ( 2 ) routine consumption of skeletons; ( 3 ) undissolved bones inside coprolites indicating gastrointestinal similarity to modern striped and brown hyenas; ( 4 ) B . parvus body weight of ~24 kg , reaching sizes of obligatory large-prey hunters; and ( 5 ) prey size ranging ~35–100 kg . This combination of traits suggests that bone-crushing Borophagus potentially hunted in collaborative social groups and occupied a niche no longer present in North American ecosystems . F:AM Frick Collection of the American Museum of Natural History , New York , New York; FMNH , Field Museum of Natural History , Chicago , Illinois; LACM , Natural History Museum of Los Angeles County , Los Angeles , California; UCMP , Museum of Paleontology at University of California , Berkeley , California . The large number of bones inside most Mehrten coprolites rules out herbivores as their producers . The size of the coprolites further indicates large carnivorans as their original makers . For medium to large carnivorans from the Mehrten Formation , Wagner ( 1976 , 1981 ) listed a bone-crushing dog Borophagus secundus ( =Osteoborus cyonoides ) , a small coyote-sized Eucyon davisi , an ancestral badger Pliotaxidea garberi , an early wolverine Plesiogulo marshalli , and an ancestral cat Pseudaelurus near P . hibbardi . Most recently , Balisi et al . ( 2018 ) added a fox , Vulpes stenognathus , to the list . Of the above , Vulpes , Eucyon , Pliotaxidea , and Plesiogulo can be ruled out as being too small to produce scats of the size of the Mehrten coprolites , whereas the true nature of Mehrten felids is poorly known . Of the large Mehrten canids , Balisi et al . ( 2018 ) recognized two bone-crushing canids , B . secundus and B . parvus , which are the only wolf-sized taxa large enough to be the producers of the Mehrten coprolites . Of these two species , B . secundus is rare , represented by two fragmentary jaws and teeth plus 1–2 questionably referred teeth , whereas B . parvus is far better represented by 27 specimens . At the main coprolite-producing locality ( see Materials and methods ) , LACM locality 3937 ( =Dennis Garber T-34 locality ) , an isolated P2 or P3 ( UCMP 235515 ) is questionably referred to B . secundus ( Balisi et al . , 2018 ) , whereas in LACM locality 3935 , no identifiable carnivoran is found ( Figure 9 ) . The Mehrten coprolites are comparable in size ( Table 1 ) to scats from extant wolves and are generally larger than those from living coyotes , despite significant overlap between scat diameters of the wolves ( average 27 mm , range 13–47 mm ) and the coyotes ( average 21 mm , range 7–34 mm ) ( Weaver and Fritts , 1979; Reid , 2015 ) . In extant African carnivores , Harrison ( 2011 ) documented scat diameters of 20–35 mm from African hunting dogs , Lycaon pictus , and striped hyena , Hyaena hyaena . Therefore , with a maximum diameter of 31 . 2 mm , the Mehrten coprolites are more likely produced by a wolf-sized Borophagus than a coyote- to fox-sized Eucyon . Of the two species of Mehrten Borophagus , B . parvus was the more likely producers of Mehrten coprolites based on their body size and far better representation of body fossils , although the possibility of B . secundus cannot be excluded . We adopt a modified scheme for characterizing hyaenid coprolite aggregate pellets introduced by Diedrich ( 2012 ) , but we use different terminologies for orientations ( Figure 1A ) . Although scat morphology of extant wolves and hyenas may be somewhat different—depending on length of retention in digestive tract , fiber and water content of feces , and hardness of ground on which scats were dropped—our Mehrten coprolites ( Figures 2 and 3 ) appear to share substantial similarities to those of living hyenas ( Figure 1B ) . Of the 14 individually catalogued coprolites , five probably are a first dropping due to their bluntly constricted terminal on at least one of their ends and their relatively greater diameter ( LACM 158707 , 158708 , 158709 , 158711 , and 158712 ) . However , only one , LACM 158709 , has the typical shape of a conical pellet ( Figure 1A ) , although LACM 158707 represents a variation of the conical-disk pellet combination that failed to separate after dropping . LACM 158708 has tapering on both ends , suggesting that the modern hyena pellet terminology by Diedrich ( 2012 ) does not completely apply to the Mehrten canids . The rest of the nine pieces are all incomplete pellets , and their exact position within the scat string is difficult to determine . If the above assessment is correct , the Mehrten coprolite sample probably consists of individual pellets from multiple dropping events possibly by multiple individuals . This is also suggested by different degrees of desiccation among different coprolite pellets ( Figure 5E ) , that is , they were not defecated at the same time . If this is the case , and assuming that the coprolites have not been transported ( there is no sign for transportation ) , the LACM 3937 locality may have been an ancient ‘latrine’ ground for social defecating and scent-marking for territorial boundaries . Such locations have been well documented in extant spotted hyenas ( Kruuk , 1972 ) , coyotes ( Gese and Ruff , 1997 ) , and wolves ( Asa et al . , 1985; Harrington and Asa , 2003 ) . While such behavior is common among social carnivorans , it has not been documented in extinct carnivorans . Mehrten coprolites maintain nearly perfectly rounded cross sections , showing no sign of post-defecation settling or flattening , nor is there any sign of deformation during the initial impact of dropping . This suggests that the original feces were able to maintain their integrity either because of a relatively hard , moisture-free matrix , and/or because the bones inside plus the high-calcareous contents of the matrix resulted in relatively rigid feces at defecation . Nor do the coprolites show major signs of post-defecation alteration , suggesting fast burial after dropping . Bones are abundant in all coprolites , consisting of 5% of total volume of all coprolites ( range 2–25%; see Table 1 for individual volume estimates ) . As examples , we describe two complete coprolites below . The majority of bones inside the coprolites , even when fully exposed , are too small and too fragmentary to be identified to a particular element or to a particular taxon beyond mammals or even vertebrates . Such difficulty can also be compounded by digitally segmented microCT reconstructions . These digitally separated bones are often an inexact replication of the actual shapes , mostly due to high similarity in X-ray opacity between bones and surrounding matrix . With the exception of a single rib fragment in LACM 158708 , all other virtually segmented bones lack sufficient morphological detail to be unambiguously identified . Generally , there is a lack of clear orientation relative to the long axis of each coprolite ( Figure 4 ) . This randomness may be a result of several factors . With the exception of the rib fragment—which , because of its length , must be aligned along the long axis of the coprolite ( Figure 3C , D ) —most bones are relatively small , and intestine diameter is not a limiting factor in their orientation . Lack of a longitudinal orientation may also be due to a relatively viscous ( low water content ) matrix and compaction during the last ( dehydration ) journey of feces through the large intestine . Surface modifications on bones include rounding of corners , polishing of surface , and acid etching . The external surface of a small bone ( red dashed line in Figure 3B ) exposed to the intestine wall has experienced visible polishing; this polished surface was also stained a darker color than the unpolished parts . Polishing is known to occur in 80% of bones in extant wolf scat ( [Esteban-Nadal et al . , 2010]:Figure 18 ) . Etching and flaking are seen on an exposed bone in LACM 158707 ( Figure 2D , E ) ; this is relatively uncommon in the scat of extant wolves , occurring in only 0 . 9% of bones contained in wolf scat ( [Esteban-Nadal et al . , 2010]:Figure 18 ) . About 5% of bones recovered from living wolf scat can be identified to their prey species ( Fosse et al . , 2012 ) . Four bone fragments from Mehrten coprolites , consisting of 8% of the total number of bone fragments ( Table 1 ) , preserve enough original morphology to be narrowed to more specific taxa or anatomic structure . They are described below . Experimental data on modern gray wolf diet and their scat permit a certain measure of quantifying scat contents and identifying prey items . However , most of these methods are based on sorting soft matter in wolf feces ( e . g . [Floyd et al . , 1978; Weaver , 1993] ) , which are typically not preserved in coprolites . A study on bone fragments preserved in extant Iberian wolf ( Canis lupus signatus ) scat provides a valuable basis for comparison ( Esteban-Nadal et al . , 2010 ) . In the Spanish samples , the numbers of skeletal fragments per scat vary from one to 96 ( [Esteban-Nadal et al . , 2010]:Table 2 ) , the upper limit being substantially more than those in our fossil samples . These higher numbers of bones can probably be explained by two factors . First , although Esteban-Nadal et al . did not specifically state it , their count of a scat almost certainly includes the entire ejected feces in a single dropping event , in contrast to our own treatment of a single individual piece of coprolite . ( our disarticulated pieces of coprolites correspond to individual pellets of a long series of scat described in [Diedrich , 2012]:Figure 4 ) ( Figure 1 ) . Second , bones from extant wolf scat are exhaustively sampled ( picked through dry samples and/or screened after chemical treatments ) in contrast to our visual inspections in microCT-scanned images . Small bones that have similar radio-opacity as the surrounding matrix can potentially be missed in the counts ( Table 1 ) . If we discount the above two factors , fossil coprolites from the Mehrten possibly contain numbers of bone fragments comparable to those in extant Iberian wolf scat . More than 80% of bone fragments in Iberian wolf scat are not identifiable to a particular bone or taxon . A study of Polish wolves had a 95% rate of unidentified bones ( Fosse et al . , 2012 ) . The same is true for Mehrten canid coprolites: four relatively large bones are identified among 48 in total ( i . e . 92% unidentified bones ) . Finally , sizes of individual bone fragments in scat of extant wolves are also roughly comparable to those in our fossils . The digested bones have a rather uniform size range of 1–2 cm to a few mm in diameter . Borophagus and bone-cracking hyaenids such as Crocuta share several craniodental features that have been interpreted as adaptations for a durophagous diet . These include robust cheek teeth often exhibiting heavy cusp wear ( the lower p4 and m1 in Borophagus and lower p3 and p4 in Crocuta; Figure 6A ) . Upper and lower dentitions of both taxa also exhibit specialized enamel microstructure ( Hunter-Schreger Bands ) of the cheek teeth interpreted to represent evolutionary responses to resisting increasingly hard and abrasive foods ( Rensberger and Wang , 2005; Tseng , 2011 ) . In both dental morphology and enamel microstructure , Borophagus and Crocuta share more similarities to each other than to Canis ( Figure 6A , C ) . However , macrowear analyses of the lower carnassial tooth ( m1 ) in population samples of the three carnivorans demonstrate that the extant bone-cracking Crocuta exhibits much more extreme cusp wear on average than either Canis or Borophagus ( DeSantis et al . , 2017 ) ( Figure 6B ) . In terms of cranial shape , Crocuta is intermediate between Borophagus and Canis in having a moderately elongate rostrum and a moderately smooth forehead , whereas Borophagus has the combination of a relatively short rostrum with a more 'stepped' appearance of the forehead ( Figure 6D–E ) . Nevertheless , within the phylogenetic context of their respective lineages , Borophagus and Crocuta represent similar extremes along an evolutionary morphological continuum , with Canis located beyond the morphospace occupied by either borophagine canids or hyaenids ( Tseng and Wang , 2011; Balisi et al . , 2018 ) ( Figure 6E ) . Lastly , comparisons of overall stress distributions during unilateral carnassial ( P4 ) bite simulations using finite element analysis indicate that the crania of Crocuta and Borophagus are more similar to each other in exhibiting lower and more dissipated stress patterns than Canis ( Tseng , 2011 ) ( Figure 6F ) . These functional morphological characteristics ( except for the macrowear data of Borophagus , newly presented here ) have been used to justify classifications of both Borophagus and Crocuta as specialized bone-cracking ecomorphs . The gastrointestinal system of hyenas has apparently evolved to handle large quantities of bones . Hyaenid feces , particularly those of the spotted hyena ( Crocuta crocuta ) , are known to contain highly digested calcium phosphates in the form of white powders and bone residues ( Figure 1B ) ( Estes , 1991 ) . To a lesser extent , the scat of striped hyena ( Hyaena hyaena ) is also white or light gray ( Macdonald , 1978; Hulsman et al . , 2010 ) . These white powders consist of calcium and phosphate salts , Ca3 ( PO4 ) 2·1 . 5Ca ( OH ) 2 , similar to hydroxyapatite , the main inorganic component in bones ( Kruuk , 1972 ) . Assuming that the common ancestor of Crocuta and Hyaena acquired the bone-dissolving gastrointestinal system , such a trait must have existed more than 8 . 6 Ma if the molecular divergence time of these two genera is considered ( Koepfli et al . , 2006 ) . However , despite inferences that the spotted hyena has a highly acidic environment within its gastrointestinal tract , no published measurement is available ( Beasley et al . , 2015 ) . Extant spotted hyenas are also known to regurgitate indigestible contents , such as skin and hair ( Kruuk , 1972; Silvestre et al . , 2000 ) . Living domestic dogs have a gastric pH of 1 . 08–2 . 07 ( Sagawa et al . , 2009 ) ; this is comparable to scavengers with highly acidic stomachs for protection against foreign microbes , such as the turkey vulture ( an obligate scavenger; 1 . 3 ± 0 . 08 ) and red-tailed hawk ( a facultative scavenger; 1 . 8 ± 0 . 27 ) ( Beasley et al . , 2015 ) . Thus , this hyperacidity in dogs is mainly attributed to scavenging . It is not clear if a linear relationship exists between stomach pH value and the amount of bone residual in scats . Without detailed studies of the digestive process in extant hyenas , it is unknown whether a combination of chemical and mechanical differences in the digestive system is responsible for differences in bone residual size observed between Borophagus and living wolves , on one hand , and spotted hyenas on the other . However , examination of the stomach contents of striped hyenas indicates that they can digest some bones to similar degrees as spotted hyenas ( Kingdon , 1977 ) . Despite the high concentration of carbonates , modern spotted hyena scat is easily softened and dissolved in the rainy season ( Kruuk , 1972 ) , and it is not surprising that hyaenid coprolites are rarely preserved in the fossil record . When they are , those from cave hyenas ( Crocuta crocuta spelaea ) are the most common ( Diedrich , 2012; Fourvel et al . , 2015; Sanz et al . , 2016 ) . Highly concentrated and trampled feces can result in ‘white phosphatic beds’ , such as in Pleistocene caves in Europe with known cave hyena activities ( Diedrich , 2012 ) . In North America , records for canid coprolites are similarly scarce . At the Pipestone Springs Main Pocket site ( late Eocene Renova Formation , Jefferson County , Montana ) , small coprolites have been attributed to Hesperocyon ( Lofgren et al . , 2015 ) . Bone fragments inside the coprolites belong mostly to small vertebrates , including marsupials , lizards , lagomorphs , and squirrel-sized rodents , suggesting a diet of mostly small prey . Teeth of Hesperocyon have also been reported to occur in some Oligocene coprolites in the Brule Formation from the Big Badlands of South Dakota ( Parris and Holman , 1978 ) . Modern spotted hyenas and wolves are social hunters , and meals are shared by the clans and packs , respectively . Spotted hyenas consume the entire skeleton , bones included , usually in one feeding session ( Kruuk , 1972 ) . In contrast , wolves are often unable to crack large limb bones , such as those of European bison , and leave substantial parts of the skeleton intact; however , skeletons of smaller prey , such as red deer , suffer far more damage and fewer bones are left uneaten ( Fosse et al . , 2012 ) . In this regard , the bone-processing abilities of wolves are closer to those of brown and striped hyenas than either group is to spotted hyenas . Brown and striped hyenas are solitary foragers and hunters in most observations , although they do have social structures associated with bone accumulations at dens ( Watts and Holekamp , 2007 ) . Bone-cracking borophagines , such as Borophagus , are equipped with far more robust teeth and sturdy jaws than those of extant grey wolves ( Balisi et al . , 2018 ) , although as a clade they did not reach the degree of morphological specialization observed in hyaenids ( Van Valkenburgh et al . , 2003 ) . It is thus reasonable to assume that Borophagus is capable of cracking larger bones than living wolves do , possibly comparable to hyaenids . Whether or not Borophagus would systematically consume an entire skeleton is still a matter of speculation , but this is likely to depend on the competitiveness of their group feeding . Van Valkenburgh et al . , 2003 considered large borophagine canids—such as Epicyon saevus , E . haydeni , Borophagus secundus , Aelurodon ferox , and A . taxoides ( B . parvus was not included in their study ) —to be hunters due to their craniodental morphometrics and abundance in the fossil record , as well as energetic considerations . In contrast to felids that commonly develop a sharp retractile claw as an effective weapon for prey capture ( Gonyea and Ashworth , 1975 ) , canids never developed a retractile claw ( with the possible exception of their arboreal ancestors; [Wang , 1993] ) . Vanvalkenburgh and Hertel ( 1993 ) and Van Valkenburgh et al . , 2003 thus argued that these large borophagines were likely social hunters in order to overcome their inability to capture large prey by a single individual . Furthermore , Carbone et al . ( 1999 ) demonstrated an empirical relationship between the body size of carnivorans and their prey size: extant predators of 21 . 5–25 kg or greater in body mass tend to prey on animals of their own body mass or greater , possibly due to energetic considerations . Our estimate of body mass for Mehrten B . parvus is 18 . 9 ± 1 . 6 kg based on lengths of the first lower molar or 24 . 3 ± 3 . 7 kg based on limb bone circumference and cortical area ( see Materials and methods ) . The latter is generally considered to be more accurate because long bones , as direct weight-bearers , are proportional to body size ( e . g . , [Anyonge , 1993] ) . Mehrten B . parvus thus is comparable in body size to the modern maned wolf Chrysocyon brachyurus ( 23 kg ) and African wild dog Lycaon pictus ( 24 kg ) . Due to some dental and postcranial parallels between borophagines and modern hyenas , the ‘hyaenoid dogs , ’ as borophagines were earlier known ( VanderHoof and Gregory , 1940; Simpson , 1945 ) , were frequently dismissed as mere scavengers ( Munthe , 1989 ) and as such were not presumed to have been able to directly drive the evolution of their prey . Such misconceptions , however , are as much a popular myth about hyenas as a reflection of the fossil dogs . Up to 80% of food consumed by the modern spotted hyena is obtained by their own hunting efforts ( Kruuk , 1972 ) , in contrast to brown and striped hyenas that are primarily omnivorous scavengers of large prey with less than 5% of food consumed from fresh kills ( Macdonald , 1978; Rieger , 1981; Mills , 1982 ) . As active hunters not dependent on the availability of carrion , spotted hyenas typically have a far greater population density and wider distribution than their scavenging relatives . Some large borophagines , such as Borophagus secundus , have a continent-wide distribution and abundant fossil record that strongly suggest that they , too , were hunters ( Wang et al . , 1999; Wang et al . , 2008 ) . In contrast , brown and striped hyenas maintain both smaller species geographic ranges and lower population densities , both of which are likely associated with their solitary hunting of prey smaller than the preferred prey of spotted hyenas ( Wagner , 2006 ) . From the new coprolite evidence alone , it is unclear whether B . parvus from the Mehrten crossed the size threshold and became an obligate predator of large prey . Our rough body size estimates based on the largest rib fragment inside one of the coprolites ( LACM 158708 ) suggest that the Turlock Lake Borophagus probably preyed on ungulates equivalent in size to a modern mule deer Odocoileus hemionus ( 45 to 150 kg ) , vicuña Vicugna vicugna ( 35 to 65 kg ) , and guanaco Lama guanicoe ( 90 to 140 kg ) : animals substantially larger than their own size ( see Materials and methods ) . However , remains of similarly large prey are known from spotted , striped , and brown hyena scats and could represent either scavenged ( more likely for striped and brown hyenas ) or actively hunted ( more likely for spotted hyena ) sources ( Kruuk , 1972; Wagner , 2006 ) . Combined with other evidence presented in the preceding paragraphs , the presence of large prey is consistent with—although does not exclusively support—Borophagus as social hunters of large mammalian prey . Morphologically hyena-like borophagine canids evolved in and were restricted to North America during their entire fossil record . Around the time of Borophagus' extinction towards the end of the Pliocene , and marking the end of hyena-like canid species in North America , a single lineage of hyaenids dispersed to North America ( Berta , 1981; Tseng et al . , 2013 ) . One ( potentially two ) species of the hyaenid Chasmaporthetes , like spotted hyenas in their craniodental biomechanical capability ( Tseng et al . , 2011 ) but with much more cursorially adapted postcranial skeletons ( Berta , 1981 ) , left a widespread but rare fossil record . Rare fossils of Chasmaporthetes from Arizona , Florida , and the Pacific coast of Mexico from otherwise productive localities suggest that either preservational environments were significantly different between Borophagus and Chasmaporthetes localities , or Chasmaporthetes were much less abundant in population density at those localities . Regardless of the reasons for the apparent rarity of the North American hyaenids compared to Borophagus , the bone-cracking ecomorphology went extinct in North America no later than the end-Pleistocene megafaunal extinctions . Although there is evidence that another canid , the dire wolf Canis dirus , had some degrees of morphological adaptation for consuming hard foods such as bone ( Figueirido et al . , 2015 ) , the selective pressure for such dietary habits may have been short-lived and sensitive to local environmental conditions rather than a long-term macroevolutionary trend ( Van Valkenburgh and Koepfli , 1993; DeSantis et al . , 2015 ) . The distinctive morphological traits associated with the bone-cracking ecomorphology ( robust and bulbous premolars , deepened zygomatic arches , arched frontal region , and expanded frontal sinus ) are either poorly developed or absent in extant carnivorans ( coyotes , foxes , cougars ) found today in the geographic regions previously occupied by Borophagus ( Werdelin , 1989; Tseng and Wang , 2010; Tseng and Wang , 2011 ) . This difference suggests that there is no ecological morphological equivalent of Borophagus in modern-day North American food webs . Therefore , the new data and re-interpretation of the functional morphology of Borophagus support the inference that their extinction marked the end of a widespread bone-cracking ecomorphology in North America . Combined with the potentially significant role of megafaunal ( as opposed to microbial ) decomposers such as extant spotted hyenas in influencing or accelerating nutrient cycling pathways and rates by bypassing invertebrate and microbe decomposers in the detrital food web ( Wilson and Wolkovich , 2011 ) , the extinction of Borophagus may have had a much more significant impact on food web dynamics than previously recognized . The above morphological and behavioral comparisons suggest that , regardless of whether Borophagus was ecologically equivalent to the top predator spotted hyena or to the small-prey-hunting and large-prey-scavenging brown or striped hyena , such a bone-cracking ecological niche is no longer present in modern-day North American ecosystems . Furthermore , evidence suggests that this change in ecological community composition is a relatively recent phenomenon . ( 1 ) Frequent bone consumption in Borophagus is supported by both craniodental structure and biomechanics and now ( in this study ) also by coprolite evidence , suggesting that Borophagus may have influenced energy flow in North American food webs similar to what vultures and hyenas may do in Africa today ( DeVault et al . , 2003; Wilson and Wolkovich , 2011 ) . ( 2 ) Bone digestion in Borophagus , as evinced by composition of the coprolites , is less similar to that in extant spotted hyenas and more similar to that in extant wolves and , to some degree , in brown and striped hyenas , suggesting that—top predator or not—Borophagus is similar to spotted hyenas in craniodental morphology more than in gastrointestinal physiology , representing a unique combination of traits . ( 3 ) The coprolite record of other canids and hyaenids shows that Borophagus evolved to consume more bone than earlier canids but did not reach the degree of bone digestion evinced by fossil or living hyaenids . ( 4 ) The presence of bone fragments of large mammalian prey is consistent with the interpretation of Borophagus as hunters of large prey , like extant wolves and spotted hyenas , but does not preclude a large-prey-scavenging interpretation more similar to the ecological role observed in extant brown and striped hyenas . ( 5 ) Borophagus fossil sites from the Miocene and Pliocene Epochs cover the area that is , today , nearly the entire continental U . S . into northern Mexico , overlapping with current ranges of predatory canids such as coyotes and foxes and felids such as cougars; these living species are all top predators with little or none of the bone-cracking craniodental morphological characteristics observed in Borophagus . Given these findings , an important future research direction is to examine whether the pre-Ice Age extinction of the hyena-like , bone-eating scavenger represented by Borophagus had a fundamental effect on the evolution of food web dynamics ( via energy flow modification ) during the Ice Age . Borophagus was not replaced with a similar ecological morphology on the temporal cusp of the establishment of modern day North American ecosystems . Understanding the impact of such permanent exclusion of a predator/decomposer would be important to understanding sympatric modern food webs . Contents from a new sample of coprolites attributed to Borophagus parvus from end-Miocene ( 5 . 3–6 . 4 Ma ) sediments in northern California provide firsthand insight into the diet of this North American group of bone-cracking top predators . The broad range of bone fragment sizes discovered inside the coprolites suggests that these predators consumed small vertebrate prey as well as deer-sized mammals . Incomplete digestion of prey bones in the coprolites also suggests that , despite a comparable degree of craniodental adaptation for durophagy , canid bone-crackers still possessed a digestive process different from spotted hyenas—which are able to completely break down bone into powder—and were more similar to striped hyenas in this regard . These findings suggest that these bone-cracking canids were potentially social hunters with a unique mixture of typical canid features and hyena-like characteristics . The ecological niche occupied by the common and widespread Borophagus was not replaced by other carnivorans or other mammals after their Pliocene extinction , potentially indicating a fundamental change in food web dynamics in North America as the Ice Age began . VanderHoof ( 1933 ) was first to report a fossil horse , Pliohippus tantalus , from near Oakdale in Stanislaus County , California . Although the Oakdale locality has produced only a few fossils since then , it signaled the potential for discovery of vertebrate fossils in the Mehrten Formation , as well as associated plants ( VanderHoof , 1933; Axelrod , 1944 ) . A partial skull of Megalonyx mathisi was described subsequently from Black Rascal Creek in ‘Upper Mehrten Formation’ ( Hirschfeld and Webb , 1968 ) . Systematic collecting of fossils in the Turlock Lake area , mostly by one of us ( DG ) , was carried out in as early as the 1950 s . In an unpublished Ph . D . dissertation , Wagner ( 1981 ) reviewed the geologic setting and laid out a biostratigraphic framework of the Mehrten Formation as related to the vertebrate fossils . More recently , Sankey et al . ( 2015 ) reinvestigated the Turlock Lake fossil sites and began a process of integrating the Mehrten fossils in a modern geologic context ( stratigraphic information archived in LACM and UCMP ) ( see also [Sankey and Biewer , 2017] ) . The vertebrate fauna from the Modesto Reservoir Member of Mehrten Formation ( as defined in [Wagner , 1981] ) was poorly disseminated and with adequate descriptions of only a few forms: a new ‘saber-toothed’ salmonid fish Smilodonichthys rastrosus ( Cavender and Miller , 1972; Sankey et al . , 2016 ) , a bony fish Orthodon microlepidotus ( Casteel and Hutchison , 1973 ) , an extinct New World badger Pliotaxidea garberi ( Wagner , 1976 ) , two plethodontid salamanders Aneides lugubris and Batrachoseps sp . ( Clark , 1985 ) , and most recently , a giant tortoise Hesperotestudo orthopygia ( Biewer et al . , 2014; Biewer et al . , 2015; Biewer et al . , 2016 ) . Wang et al . , 1999 and Tedford et al . , 2009 listed selected borophagine and canine canids from the Mehrten Formation without description or illustration . A systematic revision of Mehrten canids was completed by Balisi et al . ( 2018 ) that recognized four species: Borophagus parvus , B . secundus , Vulpes stenognathus , and Eucyon davisi . See Wagner ( 1981 ) for a preliminary faunal list of the rest of the unpublished mammals . The above four canids are all known in the Hemphillian North American Land Mammal age ( Wang et al . , 1999; Tedford et al . , 2009 ) . Borophagus parvus , however , is the most restrictive both in geographic ( southwestern United States ) and chronologic ( late Hemphillian ) ranges , offering the best potential for age assessment . Wang et al . ( 1999 ) commented on the slightly more derived dental characteristic of the Mehrten B . parvus , as compared to the topotype materials from the Redington Local Fauna in the lower member of the Quiburis Formation in Pima County , southeastern Arizona , which has been magnetically constrained within Chron 3An . 2n ( 6 . 436–6 . 733 Ma ) ( Lindsay et al . , 1984; Hilgen et al . , 2012 ) . If those characters are the result of a chronocline , the Mehrten B . parvus may be slightly younger than their Arizona counterpart . Wagner ( 1981 ) considered the Modesto Reservoir Local Fauna equivalent in age to Pinole Local Fauna in the San Francisco Bay area , which is overlain by a dated tuff ( 5 . 3 ± 0 . 1 Ma ) within Pinole Formation and placed in the latest Hemphillian ( Hh4 ) ( Tedford et al . , 2004 ) . If the above comparisons are correct , the Modesto Reservoir Local Fauna should fall in the latest Hemphillian ( Hh4 ) , possibly within 5 . 3–6 . 4 Ma . From two localities , T-14 ( LACM locality 3917 , Cement Goose Pit Island = UCMP V6878=V90008 ) and T-20 ( LACM locality 3923 , Leaf Island ) , on two small islands in the western part of Turlock Lake ( the former shown in Figure 9A ) , Axelrod ( 1980 ) listed 25 species of fossil plants from what he called Turlock Late Flora consisting of 8 trees ( including one conifer ) , 13 shrubs , 3 herbaceous perennials , and 1 or 2 vines . In particular , aquatic taxa , such as Cyperus ( flatsedges ) , Juncus ( rushes ) , and Typha ( cattails ) , are known to live along the margins of streams , lakes , and ponds , and the fossil plant localities were proposed to be lacustrine deposits ‘some distance from the shore’ ( Axelrod , 1980 ) . This flora was characterized as an oak woodland-savanna and a floodplain assemblage , and comparisons to modern vegetation from nearby regions suggested a paleoclimate of slightly cooler ( mean annual temperature 15 . 5°C ) and considerably wetter ( precipitation 635 mm ) than the present-day Turlock Lake area ( 17 . 5°C and 335 mm for corresponding measurements ) . This shift toward a more continental climate in modern day Turlock Lake was suggested to be brought about by the uplift of the Coastal Range and its rain shadow effects during the Pleistocene ( Axelrod , 1980 ) . The majority of the coprolites complete enough to be assigned a Dennis Garber field number are produced from a single locality , LACM locality 3937 ( =UCMP locality V68134 , Dennis Garber T-34 locality ) , whereas only one coprolite is from LACM locality 3935 ( =Dennis Garber T-32 locality ) ( Figure 9A ) . Both are located on the northwestern corner of Turlock Lake; they are within 300 m of each other and are from approximately the same stratigraphic horizon . T-32 was later subsumed within T-34 as a single locality . Fossil-producing exposures are in a large area forming an elbow shape . At the north end , there are two layers of white volcanic ash sandwiching a brown silty clay ( Figure 9B ) . This ash exposure continues to the eastern end of the area , which has a similar lithology to those to the west , although there seems to be a higher ratio of clay to silt and the contact between the ash and clay seems sharper . ( [Retallack , 1997]:color photo 24 ) remarked that carnivore coprolites are common in sequences of well-drained soils because of their phosphatic composition and enclosed bones . Several regression equations relate skeletal or dental measurements to body mass in extant canids and other carnivorans ( Van Valkenburgh , 1990; Anyonge , 1993; Anyonge and Roman , 2006 ) , enabling prediction of the body mass of extinct canids based on measurements of isolated elements . Body mass proxies and their reliability differ slightly , with measures of cross-sectional area of proximal weight-bearing limb bones generating more accurate estimates than dental predictors do . Dental predictors are still useful , however , because teeth tend to be more abundantly preserved than postcrania . Using the equation of Van Valkenburgh ( 1990 ) , we generated a distribution of body masses from the lengths of 76 lower first molars ( carnassials ) of B . parvus compiled by Wang et al . ( 1999 ) and Balisi et al . ( 2018 ) . We also measured two well preserved B . parvus humeri ( F:AM 75903-B , F:AM 67955 ) and one femur ( F:AM 63008-A ) at the American Museum of Natural History , using the canid equations from Anyonge ( 1993 ) to calculate body mass from humeral circumference , cortical cross-sectional area , and second moments of area . Based on lengths of lower first molars , B . parvus has a median body mass of 18 . 9 ± 1 . 6 kg ( Figure 10 ) . The Arizona population , with a median mass of 19 . 2 ± 1 . 6 kg , tends to be larger in body size than the California population , with a median mass of 18 . 1 ± 2 . 0 kg . Equations using measurements of the humerus and femur , all specimens from the Arizona population , generated higher estimates than dental estimates of both Arizona and California populations . For F:AM 75903-B , a distal humerus , we calculated body mass using an approximation of the circumference ( 22 . 8 kg ) , cortical area ( 25 . 829 kg ) , second moment of area in the anteroposterior plane ( 20 . 898 kg ) , and second moment of area in the mediolateral plane ( 29 . 33 kg ) ; these four estimates produced a median measurement of 24 . 315 ± 3 . 656 kg . For F:AM 67955 , a complete humerus , we obtained a body-size estimate of 32 . 4 kg using an approximation of the circumference . For F:AM 63008-A , a proximal femur , we estimated 20 . 2 kg . These dental and postcranial estimates of body mass place B . parvus in the same size class as the dingo Canis lupus dingo ( 20 kg ) , maned wolf Chrysocyon brachyurus ( 23 kg ) , African wild dog Lycaon pictus ( 24 kg ) , red wolf Canis rufus ( 30 kg ) , and striped hyena Hyaena hyaena ( 35 kg ) ( Nowak , 1999; Macdonald , 2006 ) . These extant species ( except the omnivorous Chrysocyon ) are carnivorous to hypercarnivorous . Several species of fossil ungulates have been recorded at Turlock Lake ( Wagner , 1981 ) , providing a pool of potential prey taxa and a starting point for our analysis . We assembled a comparative rib collection of 14 ungulate individuals belonging to 12 extant species at the LACM , spanning as much as possible the familial diversity preserved at Turlock Lake ( Table 2 ) . Each specimen comprised a full complement of ribs on at least one side of the body . We attempted to sample the ungulate families recorded by Wagner ( 1981 ) or , if extinct , the most closely related extant family ( e . g . extant Cervidae as proxy for Palaeomerycidae ) . Extant perissodactyls were not sampled because the perissodactyls at Turlock Lake tend to be either prehistoric equids smaller than modern equids , for which smaller extant artiodactyls could serve as a proxy , or the rhinocerotid Teleoceras , which is likely too large to generate the rib fragment preserved in the coprolite . The coprolite rib fragment has an anteroposterior width of 9 . 1 mm and mediolateral thickness of 5 . 2 mm . Using Mitutoyo calipers to the nearest 0 . 01 mm , we recorded two measurements on each of 13 ribs per species: ( 1 ) the mediolateral thickness at the point where it measured 9 . 1 mm anteroposteriorly , and ( 2 ) the anteroposterior width at the point where it measured 5 . 2 mm mediolaterally ( Figure 11 ) . The 13 rib measurements per species were visualized using line plots . Species represented by lines intersecting the horizontal line that marked the corresponding coprolite rib measurement were interpreted to be close in size to the prey animal represented by the rib . Because the specimens lacked metadata including body mass , we obtained species body mass estimates from the literature . Figure 12 tracks the anteroposterior width or mediolateral thickness of each rib among the extant taxa in comparison to the corresponding measurements in the coprolite rib fragment . Gaps in the data indicate ribs that were either wider or narrower for much of their length than the two fixed measurements , and so were not measured . Given its relatively flat morphology , the coprolite rib fragment is unlikely to have come from ribs 1 or 2 , which tend to be round in cross-section , despite lines representing these ribs in Figure 12 intersecting the black line marking the fragment . In general , the ribs examined begin to flatten around rib 3 or 4 , and become roughly square around rib 7 to rib 10 before narrowing and rounding again into rib 11 to the end . Therefore , we focused on the species lines that intersect the fragment line only around rib 7 to rib 10 . Given the points of intersection of the species lines with the fragment line , the three ungulate species closest in size to the animal whose rib is preserved in the coprolite are the mule deer Odocoileus hemionus ( 45 to 150 kg ) , vicuna Vicugna vicugna ( 35 to 65 kg ) , and guanaco Lama guanicoe ( 90 to 140 kg ) ( Nowak , 1999; Macdonald , 2006 ) .	Living hyenas are infamous for crushing the bones of their prey to extract the nutritious marrow inside . This feeding ability is rare today , and African and Asian hyenas , particularly the spotted hyena , are the only true ‘bone-crackers’ in our modern ecosystems . Yet , between 16 to 2 million years ago , the common , but now extinct North American dogs also crushed bone . Their skeletal features – such as highly robust skulls and jaws , teeth to withstand high stress , and large muscle-attachment areas for a powerful bite –share many similarities with the spotted hyena . It is therefore likely that these extinct North American dogs played a similar role in the ecosystem as living hyenas do now . The last of these bone-cracking dogs , Borophagus , vanished approximately 2 million years ago . In a recent study in 2018 , researchers discovered fossilized feces , also known as coprolites , which presumably belong to Borophagus parvus that lived in central California between 5 to 6 million years . These coprolites preserve ingested bone and so provide more evidence of what this species of dogs ate . Now , Wang et al . – including some of the researchers involved in the previous study – analyzed the fossil coprolites and their ingredients in great detail using computer tomography , measurements and comparisons with living predators and their prey . The results show that Borophagus parvus weighed around 24 kg and hunted large prey of 35 kg up to 100 kg: the size of a living mule deer . Its skull structure was similar to the spotted hyena , but its digestive system resembled that of striped and brown hyenas . Spotted hyenas have chalk white feces containing digested bone matter , presumably due to a highly acidic digestive system , but the coprolites of Borophagus contained undissolved bones ( which they ate regularly ) . Wang et al . also discovered that these dogs dropped feces in clusters , which is how the spotted hyena and wolves mark territory . This suggests that Borophagus were also social animals . Bone-crackers ( modern and extinct ) act as apex predators and providers of free organic material needed for decomposition , which are essential roles for maintaining a healthy ecosystem . The extinction of Borophagus likely modified the dynamics of the food web over the past few million years . It remains unclear why this way of feeding is absent in all living animals of North America . Future studies could investigate how the disappearance of Borophagus may have influenced the establishment of modern environments , eventually setting the scene for human habitation of the continent .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "ecology" ]	2018	First bone-cracking dog coprolites provide new insight into bone consumption in Borophagus and their unique ecological niche
The widely conserved ParABS system plays a major role in bacterial chromosome segregation . How the components of this system work together to generate translocation force and directional motion remains uncertain . Here , we combine biochemical approaches , quantitative imaging and mathematical modeling to examine the mechanism by which ParA drives the translocation of the ParB/parS partition complex in Caulobacter crescentus . Our experiments , together with simulations grounded on experimentally-determined biochemical and cellular parameters , suggest a novel 'DNA-relay' mechanism in which the chromosome plays a mechanical function . In this model , DNA-bound ParA-ATP dimers serve as transient tethers that harness the elastic dynamics of the chromosome to relay the partition complex from one DNA region to another across a ParA-ATP dimer gradient . Since ParA-like proteins are implicated in the partitioning of various cytoplasmic cargos , the conservation of their DNA-binding activity suggests that the DNA-relay mechanism may be a general form of intracellular transport in bacteria . Perpetuation of life depends on the faithful segregation of genetic material upon division , such that each progeny inherits a full complement of the parental genome . In eukaryotic cells , chromosome segregation is driven by the microtubule-based mitotic spindle in a process for which there is considerable mechanistic understanding ( McIntosh et al . , 2012 ) . In contrast , much less is known in bacteria . Yet a robust segregation mechanism must be in place to keep up with the fast-paced proliferation of bacterial cells . Such segregation mechanism is also critical for maintaining the precise organization of the chromosome and the reproducible positioning of gene loci over generations ( Reyes-Lamothe et al . , 2012; Wang et al . , 2013 ) . Recent studies on various bacteria have shown that chromosome segregation is a multi-step process that initiates with the segregation of the duplicated chromosomal origin regions shortly following their replication ( Shebelut et al . , 2010; Wang et al . , 2013 ) . How the newly replicated chromosomal origin regions segregate remains poorly understood . In recent years , the ParABS systems , which are conserved among most bacterial species ( Livny et al . , 2007 ) , have been recognized as important active transport systems of chromosomal origin regions in diverse bacteria ( Mohl and Gober , 1997; Kim et al . , 2000; Godfrin-Estevenon et al . , 2002; Lewis et al . , 2002; Fogel and Waldor , 2006; Saint-Dic et al . , 2006; Lasocki et al . , 2007; Jakimowicz et al . , 2007a , 2007b; Toro et al . , 2008; Bartosik et al . , 2009; Donovan et al . , 2010; Ptacin et al . , 2010; Schofield et al . , 2010; Shebelut et al . , 2010; Harms et al . , 2013; Iniesta , 2014 ) . ParABS systems , first identified on plasmids for their role in stable plasmid inheritance ( Austin and Abeles , 1983 ) , consist of three components: the DNA sequence parS , the DNA-binding protein ParB , and the deviant Walker A-type ATPase ParA ( Ebersbach and Gerdes , 2005 ) . ParB specifically recognizes parS sequences , which are typically found near the origin of replication of most bacterial chromosomes ( Livny et al . , 2007 ) . Upon binding to parS , ParB is thought to spread on flanking sequences to form the so-called ParB/parS partition complex ( Rodionov et al . , 1999; Murray et al . , 2006; Breier and Grossman , 2007 ) . ParA dimerizes upon ATP binding , which in turn promotes nonspecific DNA binding ( Leonard et al . , 2005; Hester and Lutkenhaus , 2007 ) . Various in vitro studies have also observed that ParA-ATP dimers can further assemble into filaments ( Barilla et al . , 2005; Ebersbach et al . , 2006; Barilla et al . , 2007; Machon et al . , 2007; Ptacin et al . , 2010 ) . By itself , ParA has a weak ATPase activity but this activity is generally stimulated by an interaction with ParB ( Davis et al . , 1992; Easter and Gober , 2002; Leonard et al . , 2005; Barilla et al . , 2007; Ah-Seng et al . , 2009; Scholefield et al . , 2011 ) . While these biochemical properties have been documented for many ParABS systems , how they give rise to directional transport remains a hot topic of debate ( Howard and Gerdes , 2010; Szardenings et al . , 2011; Vecchiarelli et al . , 2012 ) . ParABS-mediated chromosomal segregation has probably been most studied in Caulobacter crescentus where ParA and ParB are essential for viability ( Mohl and Gober , 1997 ) . In this bacterium , the single , densely packed circular chromosome spans the entire cell and is spatially arranged such that the ParB/parS partition complex and the nearby replication origin are located at the ‘old’ cell pole while the replication terminus is localized at the opposite , ‘new’ pole ( Jensen and Shapiro , 1999 ) . Epifluorescence microscopy studies in live cells have shown that prior to DNA replication , ParA forms a cloud-like localization pattern that spans from the new pole to about midcell ( Ptacin et al . , 2010; Schofield et al . , 2010; Shebelut et al . , 2010 ) . Replication of the origin region results in two physically separated copies of the ParB/parS complex . The ParB/parS complex closer to the old pole remains there while the other one , upon contact with the edge of the ParA cloud , migrates toward the new pole in the wake of the receding ParA cloud , as if retraction of ParA was ‘pulling’ the partition complex ( Ptacin et al . , 2010; Schofield et al . , 2010; Shebelut et al . , 2010 ) . This correlated spatial dynamics between ParA and ParB/parS is a common characteristic of ParABS systems involved in chromosome or plasmid partitioning ( Ebersbach et al . , 2006; Fogel and Waldor , 2006; Hatano et al . , 2007; Ringgaard et al . , 2009; Harms et al . , 2013; Iniesta , 2014 ) . The physical mechanism that underlies this correlated dynamics is generally thought to be analogous to the eukaryotic spindle-based mechanism that segregates chromosomes during mitosis ( Gerdes et al . , 2010 ) . According to this popular spindle-like model , ParA polymerizes into a thin filament bundle upon ATP binding . Depolymerization of ParA filaments through ParB-induced ATP hydrolysis then pulls the ParB/parS complex and its associated chromosomal origin region . However , the significance of ParA DNA-binding activity remains unclear , even though this activity is essential for the segregation process based on mutational analysis ( Hester and Lutkenhaus , 2007; Castaing et al . , 2008; Ptacin et al . , 2010; Schofield et al . , 2010 ) . Recent in vitro studies have proposed an alternative ‘Brownian-ratchet’ mechanism for the partitioning of P1 and F plasmids ( Vecchiarelli et al . , 2010; Vecchiarelli et al . , 2012 , 2013 , 2014; Hwang et al . , 2013 ) . However , it is unclear whether the proposed mechanism can support plasmid translocation under physiological conditions . Apart from chromosome and plasmid segregation , ParA-like proteins have been implicated in the positioning of other cellular components such as metabolic microcompartments and cytosolic chemotaxis clusters ( Savage et al . , 2010; Ringgaard et al . , 2011; Roberts et al . , 2012 ) , highlighting the versatility of the ParABS systems . Interestingly , while the chromosomally encoded Bacillus subtilis ParA ( Soj ) and ParB ( Spo0J ) orthologs have been implicated in chromosome partitioning in sporulating cells ( Ireton et al . , 1994; Sharpe and Errington , 1996; Wu and Errington , 2003; Lee and Grossman , 2006 ) , they are involved in the regulation of DNA replication in vegetative cells ( Murray and Errington , 2008 ) . B . subtilis Spo0J binds to parS sites proximal to the origin of replication similar to ParB orthologs involved in chromosome segregation ( Lin et al . , 1997; Lin and Grossman , 1998; Murray et al . , 2006 ) . Like ParA proteins involved in cargo partitioning , Soj forms ATP-dependent dimers that bind DNA in vitro ( Scholefield et al . , 2011 ) and display nucleoid-associated localization in cells lacking Spo0J ( Murray and Errington , 2008 ) . However , in wild-type cells , Soj mostly displays a diffuse distribution in the B . subtilis cytoplasm ( with a weak accumulation at the Spo0J/parS location ) ( Murray and Errington , 2008 ) . This pattern is in stark contrast to the cloud-like localization pattern characteristic of ParA orthologs dedicated to cargo transport . Why ParA localization in B . subtilis differs remains enigmatic . In this study , we combined biochemical and cell biological experiments with computational modeling to investigate how the ParABS system drives chromosomal segregation in C . crescentus . Our findings are inconsistent with a spindle-like or Brownian ratchet mechanism . Rather , our experiments and simulations support a novel physical mechanism in which chromosome dynamics and the rate of ATP hydrolysis are critical for efficient and robust ParA-dependent translocation of the ParB/parS complex . Our data also suggest that the difference in localization and possibly function between B . subtilis Soj and C . crescentus ParA can be explained by a biochemical difference in ParB's ability to stimulate ParA ATPase activity . For the in vitro studies , we purified recombinant C . crescentus ParA and ParB proteins from the soluble fraction of Escherichia coli lysates ( Figure 1—figure supplement 1 ) . Obtaining stable preparations of chromosomal ParA proteins is notoriously difficult because of the tendency of these proteins to precipitate ( Howard and Gerdes , 2010 ) . Through troubleshooting , we found that addition of Mg-ATP to all buffers ( including the cell lysis and storage buffers ) as well as potassium glutamate ( 150 mM ) after chromatography was crucial to maintain long-term stability of purified C . crescentus ParA . Such preparations of ParA were used to investigate the oligomerization state of the protein upon ATP binding by size exclusion chromatography . Prior to chromatography , Mg-ATP was removed from a ParA aliquot via incubation with EDTA followed by buffer exchange ( 'Materials and methods' ) . The resulting sample was then split into two samples: one sample remained without ATP , while Mg-ATP was added to the other sample . In the absence of ATP , ParA eluted from the column as a single peak at an elution volume corresponding to the ParA monomer ( ∼28 kDa ) ( Figure 1A ) . In the presence of Mg-ATP ( 2 . 5 mM ) , ParA eluted earlier , at an elution volume consistent with a dimeric form ( ∼56 kDa ) ( Figure 1A ) . Thus , C . crescentus ParA dimerizes in the presence of ATP concentrations expected to be found inside bacterial cells ( Bennett et al . , 2009 ) . 10 . 7554/eLife . 02758 . 003Figure 1 . Biochemical analysis of ParA ATPase cycle . ( A ) Gel filtration analysis of purified wild-type ParA ( left ) and DNA-binding deficient mutant ParAR195E ( right ) in the presence and absence of ATP . ( B ) Specific activity of ParA ( concentration fixed at 6 µM ) measured as a function of DNA concentration ( 0–2 . 0 mg/ml ) . Michaelis–Menten equation fit to the data ( black ) gives KDNA = 0 . 15 mg/ml . ( C ) Effects of DNA and ParB on ATPase activities of ParA ( cyan ) and ParAR195E ( yellow ) . ( D ) ADP production rate as a function of ParA concentration measured with fixed ATP ( 2 . 5 mM ) and DNA ( 1 . 5 mg/ml ) concentrations . Linear fit to the data ( black line ) gives kcat = 5 . 8 hr−1 . ( E ) Dependence of ATPase activity of ParA upon ATP concentration measured with fixed ParA ( 10 µM ) and DNA ( 1 . 5 mg/ml ) concentrations . Fitting the Michaelis–Menten equation ( black line ) gives kcat = 5 . 9 hr−1 and KM = 150 µM . ( F ) Dependence of ParA ( 3 µM ) ATPase activity on ParB concentrations . Michaelis–Menten equation fit to the data ( black line ) gives kcat = 120 hr−1 and KM = 80 µM . ( G ) Model of ParA ATPase cycle . ATP-binding promotes ParA dimerization . Upon binding DNA , ParA-ATP dimers presumably adopt an ATP-hydrolysis competent state . A high concentration of ParB is required to stimulate ATP hydrolysis of DNA-associated ParA-ATP dimers . The cycle reinitiates through nucleotide exchange . SDS-PAGE images showing the purity of the purified protein preparations and the results from experiments comparing the stimulatory effects of wild-type ( WT ) ParB and mutant ParBL12A on ParA ATPase activity are presented in Figure 1—figure supplements 1 and 2 . Experiments performed to correlate the biochemical activities measured in vitro with in vivo conditions are presented in Figure 1—figure supplements 3–8 . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 00310 . 7554/eLife . 02758 . 004Figure 1—figure supplement 1 . SDS-PAGE analysis of purified protein preparations used in this study . ( A ) ParA , ( B ) ParAR195E , ( C ) ParB-His6 , ( D ) ParBL12A-His6 , and ( E ) ParAG16V-YFP . Approximately 5–15 µg of proteins were loaded in each lane . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 00410 . 7554/eLife . 02758 . 005Figure 1—figure supplement 2 . L12A mutation , which prevents ParA–ParB interaction , severely compromises ParB ability to activate ParA ATPase activity . ATPase rates were measured from reactions containing 3 µM ParA and 1 . 5 mg/ml salmon sperm DNA in the absence or presence of 90 µM of either wild-type ParB or ParBL12A . The measured rates were compared to the rate of the reaction lacking ParB . Error bars represent the standard deviations ( SD ) of results measured in duplicates . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 00510 . 7554/eLife . 02758 . 006Figure 1—figure supplement 3 . Measurement of GFP-ParB distribution at the subcellular level . ( A ) Representative fluorescence image showing autofluorescence in wild-type ( CB15N ) cells and subcellular distribution of GFP-ParB fluorescence in MT174 cells , which produce GFP-ParB from the native parB promoter on the chromosome as the only ParB copy . ( B ) Single-cell total fluorescence and cell areas were quantified using MicrobeTracker and a histogram of normalized fluorescence ( total fluorescence divided by cell area ) is shown . The two-peak distribution was fit by a bimodal Gaussian ( black ) , representing fluorescence detected in two populations of cells: wild-type ( magenta ) and GFP-ParB-producing cells ( cyan ) . The mean of the magenta curve was designated as the mean normalized autofluorescence ( AFmean ) , which was used for background subtraction . ( C ) Histogram showing the fraction of GFP-ParB associated with parS in single cells . The experimental distribution ( grey ) was fitted with a Gaussian function ( dashed black curve ) , giving an average of 80% ParB molecules associated with the partition complex . ( D ) Western blot of a lysate made from an MT174 culture using an anti-GFP monoclonal antibody ( JL-8 ) or an anti-ParB polyclonal antibody ( data not shown ) . GFP-ParB and a degradation fragment were detected . ( E ) The fraction of GFP-ParB fluorescence associated with parS stays constant as a function of cell length . Single-cell raw data ( grey circles ) and average fraction GFP-ParB fluorescence associated with parS ( black circles ) ± SD across 0 . 25 μm intervals are shown . ( F ) Subcellular volume occupied by a ParB/parS complex as determined by PALM microscopy . mEos3 . 2-ParB synthesis in CJW4978 cells was induced with 0 . 03% xylose for 1 hr in M2G before imaging using PALM . Fluorescent emitters within six polar partition complexes were analyzed . The histogram was fitted with a Gaussian function with σ = 39 nm ( solid line ) . Radial distribution of mEos3 . 2-ParB from the foci were then combined and plotted as a histogram ( n = 440 localizations ) . This means that 95% of the molecules were localized within 78 nm ( 2σ ) from the center of the spot . We used 78 nm as the radius of the spherical volume occupied by ParB dimers within a partitioning complex . Such a sphere occupied a volume of 2 × 10−3 fL . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 00610 . 7554/eLife . 02758 . 007Figure 1—figure supplement 4 . Quantification of ParB and ParA abundance in cell lysates by quantitative Western blotting . ( A ) Top , a Western blot of a wild-type swarmer cell lysate ( CB15N producing endogenous ParB ) and indicated amounts of purified ParB-His6 using a polyclonal antibody against ParB . Bottom , the optical densities ( OD ) of the bands corresponding to ParB-His6 standards were quantified ( black circles ) and were fit by a linear function ( black line ) . The amount of ParB in the cell lysate ( red circle ) was calculated from the calibration . This measurement was repeated twice . The number of ParB per cell was calculated using this formula: ParB amount in lysate×Avogadro numberParB moleculr weight×#cells in lysate=720±80 ParB molecules per cell . ( B ) Top , a Western blot of a swarmer cell lysate of CJW3010 strain ( in which parA-yfp replaces parA at the native location in the chromosome ) and indicated amounts of purified ParAG16V-YFP-His6 using a monoclonal antibody against GFP ( JL-8 ) . Bottom , the ODs of the bands corresponding to ParAG16V-YFP-His6 standards were quantified ( black circles ) and fit with a linear function to create a calibration curve ( black line ) . Red circle represents the calculated amount of ParA-YFP in the cell lysate . This measurement was repeated three times . The number of ParA-YFP per cell was calculated using the formula ParA−YFP amount in lysate×Avogadro numberParA−YFP moleculr weight×#cells in lysate=170±20 ParA-YFP molecules per cell . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 00710 . 7554/eLife . 02758 . 008Figure 1—figure supplement 5 . His6-ParB forms a dimer in solution . Purified His6-ParB was subjected to HPLC size exclusion chromatography ( SEC ) coupled with UV , on-line laser light scattering ( LS ) and refractive index ( RI ) detectors ( SEC-UV/LS/RI ) to measure its absolute MW . The dashed lines indicate the elution profile of His6-ParB from an RI detector while the dots are MW calculated from LS measurements at 1-s intervals . The MW measurement showed an average MW of 69 kDa , which is consistent with His6-ParB ( monomeric MW = 34 kDa ) forming a dimer in solution under this experimental condition . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 00810 . 7554/eLife . 02758 . 009Figure 1—figure supplement 6 . Fluorescence-based quantification of ParA-YFP abundance in C . crescentus cells . Purified ParAG16V-YFP-His6 molecules were immobilized on a poly-Lysine-coated cover slip and imaged by stream acquisition using 2-s integration time . ParAG16V-YFP-His6 molecules were observed as immobile diffraction-limited spots . Shown in ( A ) are time lapse sequences representing the three most observed spot behaviors: ( i ) one-step photobleaching , ( ii ) two-step photobleaching and ( iii ) blinking . ( B ) Intensity profiles of single diffraction-limited spots shown in ( A ) showing 1-step photobleaching , 2-step photobleaching and blinking of YFP molecules . ΔI was calculated as the difference between the averaged fluorescence intensities before and after each stepwise photobleaching . ( C ) Histogram showing the magnitude of fluorescence loss following 1-step photobleaching ( n = 219 ) . The mean of a Gaussian fit to the histogram gives the fluorescence value of a single YFP molecule . ( D ) Number of YFP molecules per cell as a function of cell length . Swarmer cells from CJW3010 strain , which expresses parA-yfp from the native promoter of parA , were imaged on the same pad as wild-type swarmer cells . The number of YFP molecule per cell was calculated by dividing single-cell YFP fluorescence values by single YFP fluorescence . The cells were color-coded based on the YFP count using 50 as a cut-off ( blue: parA-yfp-expressing cells , n = 251; red: wild-type , n = 248 ) . The black lines represent linear fits to the red ( y1 = 1 . 6x; autofluorescence in wild type cells on average equals to 1 . 6 YFP molecules per 1 µm cell length ) and blue ( y2 = 56x; fluorescence in parA-yfp cells on average equals to 56 YFP molecules per 1 µm cell length ) data points , respectively . Typically , C . crescentus is ∼2 . 3 µm in length when parS translocation commences . Thus , there are 2 . 3 × ( 56–1 . 6 ) = 120 ParA-YFP molecules at the beginning of parS translocation . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 00910 . 7554/eLife . 02758 . 010Figure 1—figure supplement 7 . Estimation of the relative abundance of ParA and ParB in vivo . ( A ) Cultures ( 1 . 5 ml ) of CJW3010 ( parA-yfp-expressing ) and MT174 ( gfp-parB-expressing ) cells at OD660 = 0 . 25 were harvested and lysed by boiling in equal volumes of SDS buffer followed by sonication . Duplicates of 10 µl ParA-YFP lysate were resolved on the same gel as GFP-ParB lysates ( in varied volumes as indicated ) and subjected to Western blot analysis using the JL-08 monoclonal α-GFP antibody ( Clontech ) . The first lane separated by the dashed line was moved from a different region of the same blot for easier comparison . ( B ) All bands were quantified using ImageJ and the intensities of the GFP-ParB bands were plotted as a function of volume of lysate loaded . The solid line represents the best linear fit , which we used to calculate that 10 µl of ParA-YFP lysate ( the red dot represents the mean intensity of two lanes on the same blot ) contained approximately the same amount of GFP fusion molecules as those in ∼3 µl of GFP-ParB lysate . In other words , assuming that YFP/GFP fusions do not affect protein levels , the ParA:ParB ratio in swarmer/early stalked cell stage is approximately 1:3 . This method estimated that there are 720 ÷ 3 = 240 ParA molecules per cell . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 01010 . 7554/eLife . 02758 . 011Figure 1—figure supplement 8 . Maximum lengths of different possible ParA polymer configurations . ( A ) A surface representation of the Thermus thermophilus Soj dimer crystal structure showing the dimensions of the dimer along the x- and y- axes ( Leonard et al . , 2005 ) . The red areas represent the positions of arginine 182 which corresponds to arginine 195 in C . crescentus ParA and which has been shown to be critical for DNA-binding ( Ptacin et al . , 2010; Schofield et al . , 2010 ) . ( B ) Bottom view of the T . thermophilus Soj dimer showing the dimensions of the dimer along the x- and z- axes . ( C ) Three possible polymer configurations considering polymerization along the x- , y- and z-axes of a ParA dimer . The maximal achievable length of each polymer ( L ) was calculated as the product of the physiological abundance of ParA ( 90 dimers ) and the unit length of a ParA dimer along each axis . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 011 The absence of peaks at lower elution volumes ( including the void volume ) suggests that ParA does not form oligomers larger than dimers upon ATP binding under our experimental conditions ( Figure 1A ) . Electron microscopy ( EM ) analysis of ParA ( 1 µM ) in the presence ( or absence ) of Mg-ATP revealed no evidence of filament formation ( data not shown ) . Next , we determined the biochemical conditions that affect the ATPase cycle of C . crescentus ParA in vitro using an NADH-coupled ATPase assay ( De La Cruz et al . , 2000 ) . In such an assay , ADP produced by ParA ATPase activity is constantly converted to ATP by a pyruvate kinase , continuously replenishing ATP and thus avoiding substrate depletion and potential product inhibition . We determined the specific ATPase activity by measuring the ADP production rate ( in µM h−1 ) and by dividing the result by the concentration of ParA ( in µM ) . By itself , ParA had a very weak ATPase activity that was below the baseline of our assay ( 0 . 5 hr−1 ) . Like other ParA orthologs , C . crescentus ParA is known to have an affinity for DNA ( Easter and Gober , 2002; Ptacin et al . , 2010; Schofield et al . , 2010 ) . Addition of increasing concentrations of non-specific DNA resulted in a dose-dependent increase of ParA ATPase activity in vitro ( Figure 1B ) . Fitting a Michaelis–Menten curve to the data gave a maximum activity kcat = 6 . 7 hr−1 and a half saturating DNA concentration KDNA = 0 . 15 mg/ml . Note that DNA failed to stimulate the ATPase activity of the ParAR195E mutant ( Figure 1C ) , which can still dimerize upon ATP binding ( Figure 1A ) , but does not bind DNA ( Ptacin et al . , 2010; Schofield et al . , 2010 ) . This suggests that the DNA-dependent activation of ParA ATPase activity requires direct binding between ParA and DNA . In the presence of near saturating concentration of DNA ( 1 . 5 mg/ml ) , the ATPase activity was linearly dependent on ParA concentration ( Figure 1D ) . ATPase activity was dependent on the concentration of ATP ( Figure 1E ) , resulting in a KM of 150 µM . We estimated the concentration of DNA in C . crescentus cells to be about 17 mg/ml ( 'Materials and methods’ ) , which is above the KDNA value of 0 . 15 mg/ml . This suggests that the DNA concentration inside cells is sufficient to stimulate ParA ATPase activity . However , this ATPase activity of 6 . 7 hr−1 remains slow relative to the time scale of ParB/parS migration towards the distal pole , which occurs in the minute range ( Shebelut et al . , 2010; Ptacin et al . , 2010; Schofield et al . , 2010; Hong and McAdams , 2011 ) . We therefore anticipated that further stimulation by ParB would be required to obtain significant ATPase activity from ParA . Indeed , ParB ( 60 µM ) further increased ParA ATPase activity by about 10-fold in a DNA-dependent fashion ( Figure 1C ) . The increased activity was likely due to an interaction between ParB and ParA ( as opposed to a contaminating ATPase activity in the ParB preparation ) since substitution of ParB by the ParBL12A mutant ( purified under the same conditions as ParB , Figure 1—figure supplement 1D ) , which is unable to interact with ParA ( Ptacin et al . , 2010 ) , failed to stimulate ATPase activity ( Figure 1—figure supplement 2 ) . The synergistic effect of ParB and DNA on ParA ATPase activity indicates that significant ATP turnover only occurs when ParA , ParB and DNA interact . In the presence of DNA , ParA ATPase activity depended on ParB concentrations ( Figure 1F ) with a surprisingly high apparent KParB of 80 µM . Equimolar concentrations ( 3 µM ) of ParA and ParB had virtually no effect on ATP turnover; instead much higher concentrations of ParB than ParA were required to obtain significant ATPase stimulation ( Figure 1F ) . This is in marked contrast to the B . subtilis ParABS system in which equimolar concentrations ( 2 µM ) of Soj ( ParA ) and Spo0J ( ParB ) are sufficient for a 70-fold activation of Soj ATPase activity in the presence of DNA ( Scholefield et al . , 2011 ) . Inside cells , there are two pools of ParB molecules: the free pool that diffuses in the cytoplasm and the associated pool that is bound to the parS region . In C . crescentus , the high concentration of ParB required for substantial ParA ATPase rate may restrict the ParB-dependent stimulation of ATP hydrolysis to the ParB molecules associated with the parS locus . To test this idea , we measured the concentration of parS-associated ParB and diffusing ParB inside cells using a strain ( MT174 ) in which native parB has been substituted by a functional parB-gfp fusion ( Thanbichler and Shapiro , 2006 ) . Quantitative analysis of the GFP-ParB fluorescent signal ( 'Materials and methods' ) showed that GFP-ParB/parS foci contain ∼80% of the total GFP-ParB signal in the cells while the remaining ∼20% correspond to diffusing GFP-ParB plus a potential degradation product ( Figure 1—figure supplement 3A–D ) . This distribution of GFP-ParB signal does not vary with cell lengths ( Figure 1—figure supplement 3E ) . Using quantitative Western blotting ( 'Materials and methods' , Figure 1—figure supplement 1C , Figure 1—figure supplement 4A ) , we determined the total amount of ParB to be about 720 ± 80 molecules/cell ( Table 1 ) , or ∼360 dimers/cell since C . crescentus ParB dimerizes in solution ( Figure 1—figure supplement 5 ) . 10 . 7554/eLife . 02758 . 012Table 1 . Estimation of ParA and ParB concentrations inside cellsDOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 012Molecules per cellMole x 10−21Volume ( fL ) Concentration ( μM ) ParB720 ± 801 . 20 . 251 . 8ParB ( diffusing ) 1400 . 20 . 250 . 8ParB ( parS-associated ) 58010 . 002500ParA180 ± 300 . 30 . 251 . 2This table summarizes the abundance and concentrations of ParA and ParB in swarmer/early stalked C . crescentus cells . Subcellular distribution of ParB was determined by quantitative fluorescence measurements in cells expressing GFP-ParB ( MT174 , Figure 1—figure supplement 3 ) . ParB abundance was determined by quantitative Western blotting ( Figure 1—figure supplement 4A ) . The result showing dimerization of purified ParB in solution is shown in Figure 1—figure supplement 5 . ParA abundance shown here is the average value determined by three independent methods: ( 1 ) Quantitative Western blotting of ParA-YFP abundance in CJW3010 cells in which parA-yfp functionally replaces wild-type parA in the chromosome ( Figure 1—figure supplement 4B ) , ( 2 ) Calibrated fluorescence measurement ( Figure 1—figure supplement 6 ) and ( 3 ) Comparison between ParB-GFP and ParA-YFP amounts by quantitative Western blotting ( Figure 1—figure supplement 7 ) . The cell volume represents the cytoplasmic volume calculated from a cryoelectron tomograph of a swarmer cell ( Briegel et al . , 2008 ) . The volume occupied by parS-associated ParB molecules was inferred from super-resolution PALM images of cells expressing mEos3 . 2-ParB ( CJW4978 strain , Figure 1—figure supplement 3F ) . The C . crescentus chromosome carries two parS sites separated by 42 nucleotides ( Livny et al . , 2007; Toro et al . , 2008 ) . Each site has the potential of binding two ParB molecules . Yet , our measurements indicate that about 580 molecules of ParB ( 80% of the total 720 ParB molecules ) are associated to the parS DNA regions ( Table 1 ) . This is consistent with the notion that upon binding to a parS site , ParB spreads to flanking DNA regions to form a large ParB/parS assembly ( Rodionov and Yarmolinsky , 2004; Murray et al . , 2006; Breier and Grossman , 2007 ) . Assuming that the ParB/parS assembly is localized within a radius of 78 nm ( based on super-resolution microscopy results , Figure 1—figure supplement 3F ) , we estimated a local concentration of ParB to be at least 500 µM at the ParB/parS complex ( Table 1 ) . Based on our in vitro measurements ( Figure 1F ) , we expect robust activation of ParA ATPase activity by parS-associated ParB inside cells and negligible stimulatory effect by the diffusing ParB ( ∼1 µM ) . From these results , we conclude that in C . crescentus , only the highly concentrated ParB associated with parS ( i . e . , the ParB/parS complex ) is capable of stimulating ParA ATPase activity , unlike in B . subtilis where even the low concentration of diffusing ParB is likely sufficient to stimulate ATP hydrolysis ( Scholefield et al . , 2011 ) . This explains why ParA ( Soj ) is mostly monomeric and has a diffused localization in vegetative B . subtilis cells because monomers cannot bind DNA ( Scholefield et al . , 2011 ) . Conversely , in C . crescentus , ParA can accumulate in a DNA-bound ParA-ATP dimeric form away from the partition complex . Thus , the difference in ParB concentrations required for stimulation of ParA ATPase activity provides an explanation for the difference in ParA localization and possibly function between B . subtilis and C . crescentus ( 'Discussion' ) . Apart from the biochemical cycle of ParA ( summarized in Figure 1G ) , a key element of the translocation mechanism is the organization of ParA inside the cells . In the prevailing spindle-like model of segregation in C . crescentus ( Ptacin et al . , 2010; Banigan et al . , 2011; Shtylla and Keener , 2012 ) , ParA-ATP dimers form a filament bundle that depolymerizes upon interaction with ParB/parS , pulling the partition complex toward the new pole . If the partition complex tracks the end of a ParA filament bundle during translocation , we expected that the trajectory of the migrating ParB/parS complex would follow a line , reflecting the shape of the underlying ParA filament bundle . So far , the translocation of the partition complex has been characterized along only one cell dimension , the long cell axis ( Shebelut et al . , 2010 ) . Along the cell length , parS/ParB motion is known to consist of distinct steps , beginning with a slow , ParA-independent phase during which the two sister partition complexes are separated presumably by ‘bulk’ segregation mechanisms ( such as DNA replication , transcription , entropic unmixing , etc ) . This is followed by a fast , ParA-dependent phase , and completed by the anchoring of the partition complex at the new pole through a physical interaction between ParB and the PopZ matrix ( Bowman et al . , 2008; Ebersbach et al . , 2008; Shebelut et al . , 2010 ) . To acquire two-dimensional ( 2D ) trajectories ( i . e . , along both short and long cell axes ) at high temporal resolution , we tracked the motion of the partition complex ( labeled with GFP-ParB ) inside cells at 20-s intervals ( Figure 2A ) . Plotting the distance of the segregating complex from the old pole as a function of time for each cell showed the multiphasic movement of the segregating GFP-ParB/parS reported before ( Shebelut et al . , 2010 ) , with the sequential slow , fast and anchored phases ( Figure 2B ) . We found that the period of the slow , ParA-independent phase increased with faster image acquisition rates ( data not shown ) , indicating sensitivity to phototoxicity . In contrast , the fast , ParA-dependent phase was insensitive to changes in time intervals between image acquisitions ( data not shown ) . We measured that the partition complex covered a distance of 1 . 0 ± 0 . 2 µm along the cell length ( mean ± standard deviation , SD , n = 141 cells ) ( Figure 2C ) within 4 . 7 ± 1 . 2 min ( Figure 2D ) during this fast ParA-dependent phase , with an average velocity ( run length/run period ) of 220 ± 50 nm/min ( Figure 2E ) . 10 . 7554/eLife . 02758 . 013Figure 2 . Two-dimensional dynamics of the partition complex . ( A ) Representative 2-D trajectory ( Δt = 20 s ) of duplicated GFP-ParB/parS complexes in a CJW4762 cell . Positions of each partition complex detected in consecutive frames were joined with solid lines while dotted lines were used to connect positions interrupted by frame ( s ) with failed localization . ( B ) Trajectory of GFP-ParB/parS complexes in ( A ) along the long cell axis ( black , old pole-proximal ParB/parS complex; for the distal partition complex: blue , ‘slow’ phase , red , ‘fast’ phase , and green , the anchored phase ) . x0 is the position of the distal partition complex at the moment the second GFP-ParB/parS spot appeared in a cell . xfinish is the position of the distal partition complex at the moment it became anchored at the new pole . xstart is the position when the partition complex transitions from the ‘slow’ phase into the ‘fast’ phase . Run length is the distance between xfinish and xstart along the long cell axis . Run period is the time taken to travel from xstart to xfinish along the long axis . Also shown is a table summarizing the mean ± SD values for x0 , xstart and xfinish . There were more counts for xstart and xfinish ( n = 141 ) than for x0 ( n = 96 ) because in some cells the GFP-ParB/parS complex had already duplicated prior image acquisition . These values were used to define the start position for the ParB/parS complex in our computer simulations and the fast phase in the analysis . ( C ) Histogram showing the distribution of run lengths ( n = 141 ) . ( D ) Histogram showing the distribution of run periods ( n = 141 ) . ( E ) Histogram showing the distribution of translocation speeds of partitioning ParB/parS complexes during the fast phase ( n = 141 ) . ( F ) Histogram showing the distribution of short-axis displacements of segregating partition complexes during the slow ( blue ) , fast ( red ) and pole-anchored ( green ) phases . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 013 Surprisingly , the partition complex did not follow a straight line ( or smooth curve ) during the fast translocation phase ( Figure 2A ) . Moreover , when we examined its motion along the short cell axis , we found that the displacement distributions were virtually identical for the fast and slow phases ( Figure 2F ) . These observations are at odds with the idea that a ParA filament bundle serves as a stable bridge guiding the movement of the partitioning complex across the cell . That is , unless the ParA filament bundle rapidly moves across the cell width , which is highly unlikely given the large size of the presumed ParA bundle . We therefore considered two alternative explanations . First , during the process of translocation , the partitioning complex detaches from the ParA filament bundle and diffuses before reattaching to the ParA filament bundle . Second , ParA-ATP does not form a filamentous structure inside the chromosome-packed cell; instead ParA-ATP dimers bind to the chromosomal DNA away from the ParB/parS complexes . Since the distance covered during the fast , ParA-dependent phase is about 1-µm long ( Figure 2C ) , we would expect this distance to approximate the length of the ParA filament bundle if such structure were involved in the transport mechanism . To examine whether the cellular concentration of ParA could accommodate the formation of a micron-long filament bundle , we determined the number of ParA molecules per cell using three independent methods: quantitative Western blotting ( Figure 1—figure supplement 4B ) , calibrated fluorescence microscopy ( Figure 1—figure supplement 6 ) , and stoichiometric comparison with ParB ( Figure 1—figure supplement 7 ) . Each method gave a similar result with an average of about 180 ParA molecules or 90 ParA dimers per cell ( Table 1 ) . ParA2 from Vibrio cholerae has been shown to polymerize on DNA in vitro , forming a nucleoprotein filament with a dimension of 4 . 4 dimer units per helical pitch of 12 nm ( Hui et al . , 2010 ) . If we assume that ParA forms a similar nucleoprotein filament inside C . crescentus , this filament would be at most 240 nm in length . Even if we ignore DNA binding and consider that ParA dimers polymerize back-to-back along their largest dimension ( as determined from the crystal structure of the Thermus thermophilus Soj/ParA dimer; Leonard et al . , 2005 ) , the filament could be at most 580 nm long ( Figure 1—figure supplement 8 ) . Thus , the cellular concentration of ParA appears too low to accommodate the formation of even a single micron-long filament . To gain further insight into the spatial arrangement of ParA , we performed photo-activated localization microscopy ( PALM ) . For this , we generated a functional parA-dendra2 fusion ( 'Materials and methods' ) , which replaced the native parA in the chromosome without disrupting the operon structure . In addition , to increase emitter density , we generated a second strain in which ParA-Dendra2 was expressed from a xylose-inducible promoter on the chromosome in addition to being expressed from the native promoter . To specifically examine ParA-Dendra2 localization prior to parS replication and segregation , cells in G1 cell cycle phase ( swarmer cells ) were isolated and imaged by 2D PALM at an averaged localization precision of 18 nm in both the x- and y-directions . Consistent with epifluorescence images ( Ptacin et al . , 2010; Schofield et al . , 2010; Shebelut et al . , 2010 ) , ParA-Dendra2 was asymmetrically distributed along the long cell axis ( Figure 3A ) . However , we found no evidence of a filament-like localization . Rather , ParA-Dendra2 was widely distributed along the short cell axis ( Figure 3A ) . We also imaged cells at a later stage of the cell cycle , and observed strong accumulation of ParA-Dendra2 at the cell pole ( Figure 3A ) , consistent with the new-pole localization of ParA accumulation following ParB/parS segregation due to ParA's interaction with the polarity factors TipN and PopZ ( Schofield et al . , 2010; Laloux and Jacobs-Wagner , 2013 ) . 10 . 7554/eLife . 02758 . 014Figure 3 . Subdiffraction visualization of ParA-Dendra2 in live cells . ( A ) Representative super-resolution PALM images of cells producing ParA-Dendra2 at native levels in place of ParA ( strain CJW4915 ) , ParA-Dendra2 at xylose-inducible levels in addition to ParA-Dendra2 expressed from the native promoter ( CJW5154 ) , crescentin-Dendra2 at native levels ( CJW4902 ) and ribosome-associated L1-Dendra2 at native levels ( CJW5156 ) . Synthesis of ParA-Dendra2 in CJW5154 cells was induced with 0 . 3% xylose for 1 hr before swarmer cells were isolated for imaging . Scale bar = 1 µm . ( B ) Positional distributions of single emitters along the short-cell axis for cells expressing ParA-Dendra2 ( CJW5154 , green ) , crescentin-Dendra2 ( CJW4902 , blue ) and L1-Dendra2 ( CJW5156 , red ) . For each cell , a 0 . 5–0 . 8 μm segment containing at least 600 emitters was selected . Emitter positions along the short cell axis were calculated relative to the mean position . Additional super-resolution microscopy images and analysis are presented in Figure 3—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 01410 . 7554/eLife . 02758 . 015Figure 3—figure supplement 1 . Comparison of ParA and crescentin localization by super-resolution imaging . ( A ) Left , super-resolution image of CJW5154 cells producing ParA-Dendra2 from both the native and xylose promoters . Cells were imaged at 100 frames/s for 30 s using a custom-built microscope setup equipped with a sCMOS camera ( Huang et al . , 2013 ) . ParA-Dendra2 synthesis was induced with 0 . 3% xylose for 1 hr before swarmer cells were isolated for imaging . The data set was processed using a sCMOS-specific algorithm ( Huang et al . , 2013 ) . Right , line profiles of emitter positions in the white boxes show the spread of ParA-Dendra2 across the cell width near the new pole ( i and ii ) and the distribution of ParA-Dendra2 localized at the cell pole ( iii ) . ( B ) Same as A , except for the CJW4902 strain , which produces crescentin-Dendra2 from the native creS promoter , were imaged . Scale bars = 500 nm . For visualization purpose , the upper bounds in the color map have been adjusted for A and B separately . AU , arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 015 For comparison , we also imaged crescentin-Dendra2 in C . crescentus cells . Crescentin is a bacterial intermediate filament protein that assembles into an inner membrane-associated filamentous structure in C . crescentus ( Ausmees et al . , 2003 ) and is responsible for C . crescentus' characteristic crescent-shaped morphology . When crescentin is fused to a fluorescent protein , the crescentin fusion fails to attach to the cell membrane ( resulting in loss of cell curvature ) but still polymerizes into a single filamentous structure ( Ausmees et al . , 2003 ) . PALM visualization of crescentin-Dendra2 indeed revealed a clear filament-like localization ( Figure 3A ) , indicating that our method can identify protein filaments . As a second control , we imaged L1-Dendra2-tagged ribosomes using PALM . In C . crescentus , ribosomes spread throughout the cytoplasm , as shown by both cryo-electron tomography and conventional epifluorescence microscopy ( Briegel et al . , 2006; Montero Llopis et al . , 2010 ) . Super-resolution images of L1-Dendra2-labeled ribosomes were consistent with these observations , showing near homogenous cytoplasmic localization ( Figure 3A ) . Importantly , the localization of ParA-Dendra2 ( spread along the cell width σ = 146 nm ) was similar to L1-Dendra2-labeled ribosomes ( σ = 150 nm ) but not to crescentin-Dendra2 ( σ = 32 nm ) ( Figure 3B ) . We obtained similar super-resolution images whether we used a commercial N-STORM microscope ( see above ) or a custom-built microscope ( Figure 3—figure supplement 1 ) . Altogether , these results suggest that ParA does not form a thin filamentous structure . Instead , they support a model in which the asymmetric ParA cloud observed by epifluorescence microscopy consists of sparse ParA-ATP dimers ( or small oligomers ) bound to the chromosome away from the ParB/parS complex . How can the properties we observed lead to a robust directional motion of ParB/parS complex toward the new pole ? We considered the possibility that ParB/parS is simply diffusing and that its binding to the DNA-bound ParA-ATP dimers results in a biased diffusion along the ParA-ATP dimer gradient , as we previously suggested ( Schofield et al . , 2010 ) . This mechanism would be similar to the Brownian ratchet proposed for the P1 and F plasmids ( Hwang et al . , 2013; Vecchiarelli et al . , 2013 , 2014 ) . In this proposed mechanism ( hereafter referred to as ‘diffusion-binding’ mechanism , Figure 4A ) , the DNA serves as a matrix to tether ParA-ATP dimers and the partition complex diffuses until it binds to one or more DNA-bound ParA-ATP dimers . Stimulation of ATP hydrolysis results in monomerization of ParA , causing the release of the ParB/parS complex and a local depletion of DNA-bound ParA-ATP dimers . The partition complex then encounters new DNA-bound ParA-ATP dimers by diffusion , reinitiating the biochemical cycle . The hypothesis is that repeated cycles of this process would result in biased diffusion of ParB/parS toward a higher concentration of ParA-ATP dimers . 10 . 7554/eLife . 02758 . 016Figure 4 . The DNA-relay model results in robust translocation of the ParB/parS complex . ( A ) Left , graphical representation of the ‘diffusion-binding’ model: the diffusing ParB/parS complex ( green disk ) interacts with DNA-immobilized ParA-ATP dimers ( filled red disks ) , stimulates ATPase activity , resulting in the dissociation of ParA from the DNA ( open red disks ) . Right , graphical representation of the ‘DNA-relay model’: same as in ( A ) except that DNA-bound ParA-ATP dimers fluctuate randomly according to the movement of the associated DNA . When the ParB/parS complex is associated with one or more DNA-bound ParA-ATP dimers , it experiences the elastic force governing the dynamics of the DNA loci associated with ParA dimers . ( B ) Averaged positions of the simulated ParB/parS complex as a function of time . ( C ) Fraction of trajectories that completed translocation as a function of time . A summary of all parameters used in the simulations is presented in Table 2 . Additional experiments performed to obtain these parameter values are presented in Figure 4—figure supplements 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 01610 . 7554/eLife . 02758 . 017Figure 4—figure supplement 1 . Average ParA-YFP fluorescence profiles during ParB/parS segregation . Cells expressing ParA-YFP and CFP-ParB ( CJW3367 strain ) were imaged , analyzed with MicrobeTracker and SpotFinderY , and sorted based on the distance of the translocating partition complex from the new pole at 100-nm intervals . Shown in the figure are the average fluorescence profiles of ParA-YFP between the translocating CFP-ParB/parS partition complex and the new pole ( red curves ) of cells sorted into the 2 . 0 ± 0 . 05 and 1 . 6 ± 0 . 05 µm groups . The grey bars mark the average distance of the partition complex from the new pole . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 01710 . 7554/eLife . 02758 . 018Figure 4—figure supplement 2 . Estimation of the diffusion coefficient of the ParB/parS complex , DPC . ( A ) Three segments were identified in the trajectories of GFP-ParB/parS when the partition complexes were not engaged in the fast , ParA-dependent movement ( see ‘Materials and methods’ for details ) . ( B ) The displacement along the long cell axis within each segment was collected from all cells and the distributions were plotted . Closed circles indicate data while solid lines indicate Gaussian fits to displacement distributions for each of the segments . ( C ) Mean square displacement ( MSD ) was plotted for each of the segments ( closed circles ) . Solid lines represent linear fits ( MSD = DPC t + C , with a variable offset C , as described previously [Bakshi et al . , 2011; English et al . , 2011] ) to the data points . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 018 To test this hypothetical mechanism , we developed a 'diffusion-binding' model ( 'Materials and methods' ) that was constrained by experimentally-determined parameters ( Table 2 ) to best reflect the in vivo situation . The ParB-rich partition complex was modeled as a sphere with a radius of 50 nm to simulate the interaction radius of the partition complex that we estimated from our super-resolution images ( Figure 1—figure supplement 3 ) . The ParB sphere was allowed to interact with multiple DNA-bound ParA-ATP dimers at any given time to reflect the high number of ParB molecules ( i . e . , ParA binding sites ) within the partition complex ( Table 1 , Figure 1—figure supplement 3 ) . The rate constant of ParB-stimulated ATPase activity was 0 . 03 s−1 , as measured in our biochemical study ( Figure 1F ) . The concentration of ParA-ATP dimers was 90 per cell ( Table 1 ) . Their spatial distribution was modeled to reproduce the gradient of ParA-ATP dimers bound to the DNA matrix inside cells . The shape of the ParA gradient was determined by quantitative analysis of cells expressing ParA-YFP at different stages of segregation ( Figure 4—figure supplement 1 ) . The diffusion coefficient of the partition complex ( DPC ) was estimated to be 0 . 0001 μm2 s−1 from tracking the motion of detached GFP-ParB/parS prior to the directed ParA-dependent phase ( Figure 4—figure supplement 2 and 'Materials and methods' ) . 10 . 7554/eLife . 02758 . 019Table 2 . Default parameters and values used in simulations of mathematical modelsDOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 019ModelParameterValueCommentsSource1 , 2 , 3Time step of simulationsΔt0 . 001 sNo significant differences with simulations using smaller steps1 , 2 , 3Number of simulated trajectoriesnruns10241 , 2 , 3Duration of simulationstfin2000 s∼3 × time scale of translocation1 , 2 , 3Cell lengthl02 . 6 µmAverage cell length and width of cells with first appearance of 2 ParB fociThis studyCell widthw00 . 4 µm1 , 2 , 3Initial coordinates of ParB/parS complex , relative to the long cell axis ( 0 = old pole; l0 = new pole ) and short axis ( cell walls at −w0/2 and w0/2 ) x00 . 8 µmAverage coordinate of the distal ParB focus in cells at the first appearance of 2 ParB fociThis studyy00 . 0 µm1 , 2 , 3Start of ‘fast’ phase ( used only in analysis of the simulations ) xstart1 . 5 µmCalculated from xfinish - run length ( from Figure 2D ) This study1 , 2 , 3End point of translocationxfinish2 . 5 µmAverage coordinate at which distal ParB focus became anchoredThis study1 , 2 , 3Radius of the disk for ParB/parS complexRParB50 nmValue close to an estimate from super-resolution imagesThis study2 , 3Radius of the disk for ParA dimerRParA2 nmValue close to the dimension of the crystal structure of a Soj dimer ( PDB: 2BEK ) ( Leonard et al . , 2005 ) 2 , 3Number ParA dimersnParA90Average of three measurements by different techniquesThis study2 , 3Rate of ParB-stimulated hydrolysis of ATP by ParA dimerskcat0 . 03 s−1Best fit value to ParB dependence curveThis study2 , 3Rate of ParA dimer rebinding to the DNAkdb0 . 03 s−1Results do not depend on the exact values ( 0 . 01–1 s−1 range tested ) 2 , 3Spatial distribution of DNA-bound ParA dimersPParA-DNAEquation 6Measured from ParA-YFP fluorescence profile during segregationThis study1 , 2 , 3Diffusion coefficient of the translocating ParB/parS complexDPC0 . 0001 µm2s−1Estimated from non-directional phases of ParB trajectoriesThis study3Diffusion coefficient of DNA-bound ParA dimersDA0 . 01 µm2s−1Calculated from the time-dependent generalized diffusion coefficient ( Weber et al . , 2010; Javer et al . , 2013 ) 3Standard deviations of fluctuating DNA-bound ParA dimers used to define elastic constants ( 1/σ2 = ksp/kT ) σlong0 . 06 µmMeasured from the positional fluctuation of the groESL , 139_lac and 165_lac DNA loci . This studyσshort0 . 04 µmModel 1: diffusion , Model 2: diffusion-binding , Model 3: DNA-relay . We generated over 1000 trajectories in virtual cells using Brownian dynamics simulation . We focused our analysis of the simulated trajectories to the region between 1 . 5 µm ( xstart ) and 2 . 5 µm ( xfinish ) from the old pole of virtual cells since this 1-µm region corresponds to the active , ParA-dependent motion ( fast phase ) in real cells ( Figure 2B ) . We found that the averaged trajectory showed little ( if any ) net translocation toward the new pole ( Figure 4B ) . In fact , none of the simulated trajectories were able to cross this 1-µm region within 30 min ( Figure 4C; Video 1 for a representative trajectory ) . This is in marked contrast to experimental observations where most cells complete translocation of ParB/parS within that time frame ( Figure 2D , Figure 4C ) . Moreover , the diffusion-binding model did not work better than a simple diffusion model ( 'Materials and methods' ) in which the partition complex diffuses without interacting with ParA-ATP dimers ( Figure 4B , C; Video 2 ) . 10 . 7554/eLife . 02758 . 020Video 1 . Example of a simulated trajectory for the diffusion-binding model . Trajectories were generated by Brownian dynamics simulations with time step dt = 1 ms and the following parameters: diffusion coefficient of the partition complex DPC = 0 . 0001 µm2 s−1 , total number of ParA dimers nParA = 90 , ParB-stimulated rate of ParA ATPase activity kcat = 0 . 03 s−1 , rate constant for DNA-binding kDB = 0 . 03 s−1 . The partition complex and ParA-ATP dimers are shown as green and red spheres , respectively . Shown here is a representative trajectory in absolute cell coordinate ( 0 µm = old pole; 2 . 5 µm = new pole ) as a function of time in a virtual cell . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 02010 . 7554/eLife . 02758 . 021Video 2 . Example of a simulated trajectory for the diffusion model . Trajectories were generated by Brownian dynamics simulations with time step dt = 1 ms and a diffusion coefficient of partition complex DPC = 0 . 0001 µm2 s−1 . The partition complex is shown in green . Shown here is a representative trajectory in absolute cell coordinate ( 0 µm = old pole; 2 . 5 µm = new pole ) as a function of time in a virtual cell . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 021 The diffusion-binding model did not lead to directional motion , indicating that diffusion of the ParB/parS complex coupled with its biochemical interplay with DNA-bound ParA-ATP dimers does not create a Brownian ratchet . This is because this mechanism is only governed by the diffusion of the partition complex . There is no force involved and nothing prevents the partition complex from diffusing in the wrong direction when the interaction between ParB and ParA is lost following ATP hydrolysis ( Video 2 ) . The transient interactions with DNA-bound ParA-ATP dimers only intermittently stall the motion of the partition complex . These results suggested that something was missing from the diffusion-binding model . So far , our diffusion-binding model only considered the chromosome as a static matrix for the attachment of ParA-ATP dimers . Although chromosomal loci keep their average physical position inside bacterial cells ( Viollier et al . , 2004; Wiggins et al . , 2010 ) , they are mobile and exhibit discernible motion independent of chromosome segregation ( Espeli et al . , 2008; Weber et al . , 2010; Hadizadeh Yazdi et al . , 2012; Javer et al . , 2013 ) . This suggests that the DNA-bound ParA-ATP dimers inside cells are not static; instead , they fluctuate locally due to their associations with a dynamic DNA matrix . To realistically describe chromosome dynamics in our model , we tracked the dynamics of a LacI-CFP-labeled chromosomal locus ( groESL ) positioned near the middle of the C . crescentus cell prior to replication and segregation . We performed time-lapse imaging at 2-s interval and only analyzed cells with a single fluorescent locus to eliminate motions due to segregation . Each locus ( n = 641 ) moved randomly around an equilibrium position within individual cells ( Figure 5A ) . To measure temporal fluctuations in position at the single-cell level , we determined the deviations of the locus position from the equilibrium point ( i . e . , mean position for each trajectory ) at each time-point . The distribution of groESL position deviations was well approximated by an asymmetric two-dimensional ( 2-D ) Gaussian distribution ( Figure 5B ) . This indicates that the DNA locus moves in an asymmetric 2-D harmonic potential , which implies elastic dynamics . This is consistent with the recent proposal that bacterial chromosomes behave like elastic filaments ( Wiggins et al . , 2010 ) . 10 . 7554/eLife . 02758 . 022Figure 5 . Chromosomal loci exhibit elastic dynamics . ( A ) Representative trajectory of a LacI-CFP-labeled groESL locus tracked in live CJW2966 cells showing the dynamics of a chromosomal locus . The DNA locus was imaged every 2 s for 180 s . The displacements were colored as a function of time and overlaid with the cell outline ( green ) . The zoomed trajectory in the inset shows the magnitude of displacements . To induce LacI-CFP expression , cells were incubated for 1 hr with 0 . 03% xylose prior to imaging . ( B ) Left , two-dimensional distribution of the groESL locus positions relative to the mean position of each trajectory . The scatter plots were generated from locus positions of 641 trajectories while the mesh surface is the best 2-D asymmetrical Gaussian fit ( with σlong = 0 . 06 µm and σshort = 0 . 04 µm ) . The lower plane is a heatmap of the experimental data . Right , the same data set is represented as 1-D distributions of the groESL locus positions relative to the mean position of each trajectory along the long and short cell axes in single cells . Experimental data ( filled circles , long axis; open circles , short axis ) and Gaussian fits ( solid line , long axis; dashed line , short axis ) are shown . ( C ) 1-D distributions of the 139_lac and 165_lac locus positions ( Viollier et al . , 2004 ) relative to the mean position of each trajectory along the long and short cell axes in single cells . Left , asymmetrical Gaussian fit gives σlong = 0 . 06 µm and σshort = 0 . 03 µm for the 139_lac locus in which the lac operators are inserted at position 1599540 on the C . crescentus chromosome ( CJW5466 strain ) . Right , asymmetrical Gaussian fit gives σlong = 0 . 05 µm and σshort = 0 . 03 µm for the 165_lac locus in which the lac operators are inserted at position 2481399 on the C . crescentus chromosome ( CJW5468 strain ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 022 Gaussian fitting along long and short cell axes yielded standard deviations σlong = 0 . 064 ± 0 . 002 µm ( best fit ± error of fitting ) and σshort = 0 . 038 ± 0 . 001 µm ( Figure 5B ) , which were converted ( using Equation 2 , 'Materials and methods' ) into effective spring constants ( ksp ) of 0 . 001 pN/nm and 0 . 003 pN/nm along the long and short axes , respectively . We also measured the fluctuations of two other chromosomal loci ( 139_lac and 165_lac ) at positions 1599540 and 2481399 on the chromosomal map ( Viollier et al . , 2004 ) and observed similar elastic behaviors ( Figure 5C ) . We used the mean σlong and σshort values of these three chromosomal loci ( Table 2 ) to quantitatively incorporate the intrinsic fluctuating motion of DNA-bound ParA-ATP dimers into the diffusion-binding model . Simulation of this revised model , named the ‘DNA-relay’ model hereafter ( Figure 4A ) , showed that considering the DNA as a dynamic matrix has a dramatic effect , yielding fast and robust translocation of partition complexes ( Figure 4B , C; Video 3 ) . 10 . 7554/eLife . 02758 . 023Video 3 . Example of a simulated trajectory for the DNA-relay model . Trajectories were generated by Brownian dynamics simulations with time step dt = 1 ms and following parameters: diffusion coefficient of partition complex DPC = 0 . 0001 µm2 s−1 , total number of ParA dimers nParA = 90 , ParB-stimulated rate of ATPase activity kcat = 0 . 03 s−1 , rate constant for DNA binding kDB = 0 . 03 s−1 and spring constant ksp/kT = 1/σ2 = 280 μm−2 . The partition complex and ParA-ATP dimers are shown as green and red spheres , respectively . Shown here is a representative trajectory in absolute cell coordinate ( 0 µm = old pole; 2 . 5 µm = new pole ) as a function of time in a virtual cell . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 023 The ability of the DNA-relay model to reproduce the experimental results ( Figure 4B , C ) is particularly remarkable given that the model was grounded on experimentally measured data ( Table 2 ) and did not rely on parameter optimization . In this model , the DNA-bound ParA-ATP dimers fluctuate around the equilibrium point of the underlying DNA loci . When the partition complex catches DNA-bound ParA-ATP dimers in a stretched out-of-equilibrium state , it experiences the elastic force , which moves the complex toward the equilibrium point until ATP hydrolysis releases it . Repetition of the process results in a relay of the partition complex from one chromosomal region to another . Directionality arises from the presence of the ParA-ATP dimer gradient . Since there are more DNA-bound ParA-ATP dimers toward the new pole , the partition complex will tend to encounter these DNA-bound ParA-ATP dimers as they stretch toward the partition complex . This results in a net force and hence translocation relay toward the new pole . From our estimate of diffusion coefficient of the partition complex ( DPC = 0 . 0001 μm2 s−1 , see Figure 4—figure supplement 2 ) and its velocity during ParA-dependent phase ( v = 0 . 003 µm/s , Figure 2E ) , one can estimate that the partition complex experiences a force F = v ( kT/DPC ) = 0 . 1 pN ( where k is the Boltzmann constant and T is the absolute temperature ) . This is in good agreement with the characteristic elastic force generated from chromosomal locus dynamics , which can be estimated as F = ksp σlong = 0 . 06 pN ( where ksp is the measured spring constant and σlong is the standard deviation of the DNA locus from equilibrium ) . Thus , one or more interactions with DNA-bound ParA dimers provide sufficient force to account for the observed partitioning velocity , providing further support for the DNA-relay mechanism . The elastic force in the DNA-relay mechanism implies that the partition complex is under tension when it is bound to DNA-associated ParA-ATP dimers ( i . e . , during ParA-dependent translocation ) . We found evidence supporting this notion when we examined the fluorescent GFP-ParB/parS signal . The GFP-ParB/parS signal formed a diffraction-limited focus when located close to the cell poles , that is , before or after segregation when GFP-ParB/parS was not interacting with ParA . However , the fluorescent signal associated with the segregating GFP-ParB/parS region ( i . e . , interacting with DNA-bound ParA dimers ) frequently adopted an extended configuration ( Figure 6A ) , consistent with a force decompacting the parS region . We confirmed that this extended conformation was not due to rapid motions of GFP-ParB/parS during image acquisition as it was readily observed in formaldehyde-fixed cells ( Figure 6B ) . 10 . 7554/eLife . 02758 . 024Figure 6 . The ParB/parS complex transiently adopts an extended conformation during the fast segregation phase . ( A ) A fraction of GFP-ParB/parS complexes ( red arrows ) adopts extended conformation that appears as non-diffraction limited spots in fluorescent images of live CJW4762 cells . ( B ) Same as ( A ) except that cells were fixed with 4% formaldehyde prior imaging . ( C ) The aspect ratio ( AR ) of CFP-ParB/parS complexes in live CJW3367 cells was calculated as the ratio of the longest dimension , l , to the shortest dimension , d , for each partition complex signal . A representative cell is shown with the fluorescent CFP-ParB/parS signals outlined in yellow . ( D ) The propensity of a CFP-ParB/parS partition complex ( PC ) to display a decompacted conformation is shown with respect to cellular position . Individual CFP-ParB/parS complexes were binned according to their relative positions in the cell with old pole = 0 and new pole = 1 ( see ‘Materials and methods’ for pole discrimination ) . The fraction of CFP-ParB/parS complexes with AR >1 . 5 in each bin are shown . ( E ) Top , representative time-lapse sequence ( 4-s intervals ) showing de-compaction and recoiling of a segregating partition complex ( marked by GFP-ParB ) in a CJW4762 cell . Bottom , ARs of the anchored ( red ) and segregating ( blue ) ParB/parS complexes as a function of time is shown . For all the experiments indicated above , the expression of ParB fusion proteins was induced with 0 . 03% xylose in M2G for 60–75 min prior to synchronization and imaging . All scale bars = 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 024 To quantify these observations , we used a filtering-based algorithm to identify fluorescent objects ( CFP-ParB/parS complexes ) in each cell ( Parry et al . , 2014 ) . Analyzing only cells in which the CFP-ParB/parS complex has duplicated and separated , we measured the aspect ratio ( AR ) of each partition complex ( Figure 6C ) , a metric that has also been used to distinguish between compacted and decompacted states of chromatin loci in eukaryotic nuclei ( Verdaasdonk et al . , 2013 ) . A perfectly round CFP-ParB/parS signal is expected to have an AR of 1 while a stretched signal has an AR score greater than 1 ( Figure 6C ) . We visually correlated a subset of the spots with their AR and verified that an AR cutoff of 1 . 5 reliably discriminated between the compacted and decompacted conformations . Using this AR cutoff , we found that the frequency of decompacted conformation ( AR >1 . 5 ) is low near the new and old poles ( Figure 6D ) , where the ParB/parS complexes are expected to either be anchored by PopZ or diffuse locally ( Figure 2A ) . However , the frequency of decompacted CFP-ParB/parS signals increased as the partition complex moved away from the old pole ( Figure 6D ) . Thus , the frequency of decompacted CFP-ParB/parS signals correlated with the concentration of ParA-ATP dimers , suggesting that the stretched conformation adopted by the partition complex is triggered by ParB interacting with ParA-ATP dimers bound to the dynamic DNA . Furthermore , the AR of the segregating partition complex appeared to fluctuate over time ( Figure 6E ) , consistent with the partition complex dynamically responding to fluctuating pulling forces . The DNA-relay model suggests that the ParABS system utilizes intrinsic DNA dynamics inside cells to translocate a cargo . It also suggests that the role of a ParA-ATP dimer is to provide a timed and reversible connection between the partition complex and the dynamic DNA . According to the model , a fluctuating DNA locus is captured out-of-equilibrium by the partition complex ( via ParA-ATP dimers ) . As the DNA locus returns to its equilibrium , it brings the partition complex along . The association time between ParB/parS and the DNA-bound ParA-ATP dimer—which is defined by ParB-stimulated ATP hydrolysis rate ( kcat ) —is critical for this mechanism to work efficiently . Under optimal conditions , the partition complex would be released immediately once equilibrium is reached to catch the next out-of-equilibrium DNA-bound ParA-ATP dimer . This would result in an apparent relay of the partition complex from one DNA-bound ParA-ATP dimer to another . A back-of-the-envelope calculation ( 'Materials and methods' ) estimates that the optimal attachment time would be 36 s . Thus , the optimal rate of ATP hydrolysis , which is the inverse of this optimal attachment time , should be about 0 . 03 s−1 . This value is in remarkably good agreement with the ParB-induced ATP hydrolysis rate ( kcat = 0 . 03 s−1 ) we obtained from our biochemical data ( Figure 1F ) . We expect that if the ATP hydrolysis were too slow , the partition complex would remain bound to a ParA-ATP dimer too long , slowing down progression . Conversely , too high ATP hydrolysis rate would result in premature release of the complex ( before it reached the equilibrium point ) , under-utilizing the potential of the elastic force . To examine how the performance of the DNA-relay model depends on kcat , we varied the kcat values in our simulations . We found that for the estimated DPC = 0 . 0001 μm2 s−1 ( Figure 4—figure supplement 2 ) , decreasing or increasing kcat by 10-fold negatively impacts the speed and robustness of translocation ( Figure 7A , B ) . This suggests that the hydrolysis rate has been optimized during evolution for the proposed DNA-relay mechanism . 10 . 7554/eLife . 02758 . 025Figure 7 . Appropriate ParB-stimulated ParA ATPase rates are important for the robustness of the DNA-relay model . ( A ) Averaged trajectories of the partition complex along the long cell axis during the fast ParA-dependent phase were simulated using varied ParB-stimulated ParA ATPase rates ( kcat ) and a fixed diffusion coefficient for the ParB/parS complex ( DPC ) of 0 . 0001 μm2/s . ( B ) Same data set as ( A ) , except that for each kcat , the fraction of trajectories that completed translocation are shown as a function of time . DOI: http://dx . doi . org/10 . 7554/eLife . 02758 . 025 Central to the mechanism of ParABS-dependent transport is the origin of the translocation force . We did not find any evidence supporting a eukaryotic-like filament-based mechanism . Rather , our experiments and simulations collectively suggest a DNA-relay mechanism , in which the DNA-associated ParA-ATP dimers serve as transient tethers that harness the intrinsic dynamics of the chromosome to relay the partition complex from one DNA region to another . In this model , the translocation force is derived from the elastic property of the chromosome . Elastic dynamics may be inherent to highly compacted and ordered DNA as they have also been observed for chromosomal loci in eukaryotic nuclei ( Verdaasdonk et al . , 2013 ) as well as in nucleoids isolated from lysed E . coli cells ( Cunha et al . , 2005 ) . Furthermore , a modeling study has shown that the E . coli chromosome behaves like a spring filament based on the positional distribution of chromosomal loci across cells ( Wiggins et al . , 2010 ) . A recent theoretical study suggests that a ParA gradient with ParB binding alone may generate a thermodynamic force ( called chemophoresis ) ( Sugawara and Kaneko , 2011 ) . While such force may contribute , it is not clear whether the in vivo conditions ( e . g . , ParA concentration , cargo size , time scale ) can result in a thermodynamic force large enough to account for the translocation kinetics observed inside cells . On the other hand , we show here that the DNA-relay mechanism is sufficient to produce the sub-piconewton force that we estimated for ParA-dependent translocation . The dynamic nature of chromosomes seem to be a universal feature of active cells ( Marshall et al . , 1997; Weber et al . , 2010 , 2012; Javer et al . , 2013; Verdaasdonk et al . , 2013; Zidovska et al . , 2013 ) . Since many ParA-like proteins involved in positioning of various large cytoplasmic cargos ( chromosomal regions , plasmids , chemotaxis clusters ) have a conserved and functionally important DNA-binding surface ( Hester and Lutkenhaus , 2007; Castaing et al . , 2008; Ptacin et al . , 2010; Schofield et al . , 2010; Roberts et al . , 2012 ) , the DNA dynamics may also be exploited in these systems to achieve active partitioning . Another common property of all ParABS systems examined biochemically so far is the relatively low ParA ATPase activity , even after ParB stimulation . Our data suggest that ATPase activity is near-optimal for cargo translocation via the DNA-relay mechanism ( Figure 7 ) . ParA-like proteins have been implicated in other biological functions besides partitioning ( Thanbichler and Shapiro , 2006; Murray and Errington , 2008 ) . For example , B . subtilis Soj controls DNA replication initiation in vegetative cells ( Murray and Errington , 2008 ) . While this distinct function involves interactions with other cellular components ( e . g . , DnaA ) , biochemical evidence suggests that differential regulation of ParA dimeric states by ParB also plays an important role . As shown here , C . crescentus ParA ATPase activity requires high concentration of ParB ( Figure 1F ) , which insulates ParA-ATP dimers from premature turnover by diffusing ParB and restricts stimulation of ATP hydrolysis to the ParB-rich partition complex . In contrast , in B . subtilis , a low concentration of Spo0J ( ParB ) is sufficient to stimulate Soj ATPase activity , such that even diffusing Spo0J promotes ATP hydrolysis ( Scholefield et al . , 2011 ) . As a consequence , in C . crescentus , ParA accumulates as a cloud of DNA-bound ParA-ATP dimers , primed for the transport of the ParB/parS complex , whereas in B . subtilis , Soj mostly exists in a replication-inhibitory monomeric form that cannot bind the DNA ( Scholefield et al . , 2011 ) . Consistent with this notion , a mere overproduction of Soj , which should tip the balance toward the Soj dimer state , results in cloud-like localization of Soj and Spo0J/parS translocation in the receding wave of the Soj cloud ( Marston and Errington , 1999; Quisel et al . , 1999 ) . Thus , a common biochemical framework can be tuned to generate different system behaviors . In conclusion , we present a new physical mechanism for intracellular transport in which the chromosome plays a mechanical function . We anticipate that other facilitating or prohibiting elements ( e . g . , replication , transcription , and entropy-driven polymer separation , chromosome topology and glassy cytoplasm ) ( Jun and Mulder , 2006; Le et al . , 2013; Wang et al . , 2013; Parry et al . , 2014 ) , which were not considered in our DNA-relay model , have an influence on chromosome partitioning . Further quantitative knowledge of these effects will allow us to build upon our DNA-relay model to gain a more comprehensive understanding of chromosome segregation in bacteria . Strains used in this study and methods of strains construction are detailed in Table S1 in Supplementary file 1 . Oligonucleotide primers used in this study are tabulated in Table S2 , which is presented in Supplementary file 2 together with methods of plasmids construction . Synchrony , conjugation , transformation and transduction with the bacteriophage ΦCR30 were performed as previously described ( Ely , 1991 ) . C . crescentus strains were grown at 30°C in the defined minimal M2G medium ( 0 . 87 g/l Na2HPO4 , 0 . 54 g/l KH2PO4 , 0 . 50 g/l NH4Cl , 0 . 2% [wt/vol] glucose , 0 . 5 mM MgSO4 , 0 . 5 mM CaCl2 , 0 . 01 mM FeSO4 ) unless otherwise stated . Since none of the C . crescentus strains used in this study contained replicative plasmids , antibiotics were omitted when cells were grown for the purpose of imaging . For all experiments , cells were harvested from exponentially growing cultures . ParA and ParB are essential for viability in C . crescentus ( Mohl and Gober , 1997 ) . The parA-dendra2 ( this study ) , parA-eyfp ( Schofield et al . , 2010 ) and egfp-parB ( Thanbichler and Shapiro , 2006 ) fusions used in this study support viability when expressed as the only copy , indicating that they are functional . We previously showed that parA-eyfp is functional ( Schofield et al . , 2010 ) . In the M2G medium at 30°C , the doubling times for the strains expressing parA-dendra2 or egfp-parB are 133 ± 2 min and 142 ± 2 min , respectively , as compared to 133 ± 2 min for wild-type CB15N grown under the same condition . For the strains in which the synthesis of ParA and ParB protein fusions was induced from the xylose-inducible promoter ( pXyl ) as a second copy , cells were grown in the presence of 0 . 03% xylose for 60–75 min ( fluorescent ParB fusions ) or 0 . 3% xylose for 1 hr ( ParA-Dendra2 ) . Under these inducing conditions , the localization of these second-copy fusions was visually undistinguishable from that of single copies ( data not shown ) . For the purification of ParA or ParAR195E , BL21 ( DE3 ) cells carrying pET24HT-ParA or pET24HT-ParA ( R195E ) were grown in Luria Broth supplemented with 50 µg/ml kanamycin to an OD600 = 0 . 6 , chilled to 18°C before IPTG ( 1 mM ) was added to induce His6-ParA synthesis overnight at 18°C . Cells were collected by centrifugation at 5000×g for 20 min and stored at −80°C . The cell pellet was resuspended in buffer A1 ( 100 mM Hepes/KOH , 100 mM KCl , 1 mM EDTA and 10% glycerol ) supplemented with EDTA-free Roche protease inhibitor , 1 mM DTT , 0 . 5 mM Mg-ATP , 1 kU DNase I and 0 . 5 mg/ml lysozyme . The cell suspension was incubated on ice for 30 min before 4 M KCl was added to a final concentration of 1 M ( to dissociate ParA from the DNA ) . Following sonication , the sample was clarified by centrifugation at 100 , 000×g for 30 min . One molar imidazole ( pH 7 ) was added to the clarified lysate to a final concentration of 40 µM . The supernatant was then passed through a Ni-NTA column ( Qiagen , Germany ) thrice by gravity flow . The resin was washed with 30 column volumes of buffer A2 ( 25 mM Hepes/KOH pH 7 . 4 , 450 mM KCl , 50 mM potassium glutamate [KGlu] , 1 mM MgSO4 , 40 mM imidazole , 1 mM DTT and 100 µM MgATP ) before elution was carried out with 50 ml of buffer A3 ( 25 mM Hepes/KOH pH 7 . 4 , 450 mM KCl , 50 mM KGlu , 1 mM MgSO4 , 300 mM imidazole , 1 mM DTT and 100 µM MgATP ) . Eluent was collected in 1-ml fractions . Fractions containing high concentrations of protein ( determined by Bradford assay using Bovine IgG as standards ) were pooled . The pooled sample usually had ∼2–2 . 5 mg/ml protein content as determined by Bradford assay . TEV protease was diluted into the sample at ∼1:30 ( wt/wt ) ratio and incubated at room temperature for 2 hr to remove the 6 × His tag . The cleavage efficiency was approximately 90–95% . The sample was then concentrated approximately twofold using Amicon Ultra-14 ( MWCO = 10 kDa ) and subjected to gel filtration fractionation on a Superdex 75 16/60 column ( GE Healthcare Life Sciences , Pittsburgh , PA ) at 1 ml/min flow rate in buffer A4 ( 25 mM HEPES/KOH pH 7 . 5 , 200 mM KGlu , 1 mM MgSO4 , 1 mM DTT , 100 µM Mg-ATP and 5% glycerol ) . Residual His-tagged proteins were removed by passing the sample through a HisTrap HP column ( GE Healthcare Life Sciences ) in buffer A4 supplemented with 40 mM imidazole . A pooled sample was concentrated to ∼1 . 5 mg/ml and subjected to dialysis against buffer A4 supplemented with 20% glycerol overnight at 4°C , and stored in small aliquots at −80°C . Polyhistidine-tagged ParAG16V-YFP was purified using Ni2+ column as described for wild-type protein with the following modifications . To remove the imidazole , we induced ParAG16V-YFP-His6 precipitation by dialyzing the sample into buffer A3 and re-solubilized the precipitated protein ( which we isolated by centrifugation ) in buffer A1 supplemented with 1 M KCl and 1 mM DTT . BL21 ( DE3 ) /pET21b-ParB cells or BL21 ( DE3 ) /pET21b-ParB ( L12A ) cells grown up to OD600 = 0 . 6 were incubated with 1 mM IPTG for 4 hr at 37°C . Cells were collected by centrifugation at 5000×g for 10 min at 4°C and frozen at −80°C . Cells were resuspended in buffer B1 ( 50 mM Hepes/KOH pH 7 . 0 , 25 NaCl , 0 . 1 mM mM EDTA , 5 mM MgCl2 and 1 mM DTT ) supplemented with 1 kU DNase I and Roche EDTA-free protease inhibitor . Cell lysis was induced by passing the sample through a French press twice at 16 , 000 psi . Following 30 min incubation on ice , the sample was clarified by two rounds of centrifugation at 50 , 000×g for 25 min . The clarified lysate was injected into an HiPrep SP 16/10 column ( GE Healthcare Life Sciences ) equilibrated in buffer B1 at 2 ml/ml and washed with four column volumes ( CV ) of buffer B1 before a gradient ( 50–500 mM NaCl ) was developed over 20 CV with buffer B2 ( 50 mM Hepes/KOH pH 7 . 0 , 1 M NaCl , 0 . 1 mM EDTA , 5 mM MgCl2 and 1 mM DTT ) . Fractions containing ParB-His6 were pooled and imidazole was added to ∼15 mM before incubating the sample with NiNTA resin ( Qiagen ) for 1 hr . The mixture was poured into the column to remove unbound proteins . The resin was washed with more than 20 CV of buffer B3 ( 25 mM Hepes/KOH pH 7 . 0 , 300 mM NaCl , 5 mM MgCl2 , 40 mM imidazole and 1 mM DTT ) followed by 2 CV of buffer B3 minus imidazole , and eluted with buffer B4 ( 25 mM Hepes/KOH pH 7 . 0 , 300 mM NaCl , 5 mM MgCl2 , 300 mM imidazole and 1 mM DTT ) . Fractions containing ParB-His6 were pooled and subjected to dialysis in 100 × sample volume of buffer B5 ( 25 Tris/HCl pH 7 . 5 , 50 mM NaCl , 5 mM MgSO4 , 0 . 1 mM EDTA and 1 mM DTT ) overnight at 4°C . The sample was clarified at 30 , 000 × g for 15 min prior to injection into an HiPrep Heparin 16/10 column equilibrated in buffer B5 at 1 . 5 ml/min . The column was washed with 5 CV of buffer B5 before a gradient ( 50–250 mM NaCl ) was developed over 20 CV with buffer B6 ( 25 Tris/HCl pH 7 . 5 , 1000 mM NaCl , 5 mM MgSO4 , 0 . 1 mM EDTA and 1 mM DTT ) . Fractions containing ParB-His6 were concentrated in a Amicon-15 unit and chromatographed over a Superdex 75 16/60 column equilibrated in buffer B7 ( 25 mM Hepes/KOH pH 7 . 5 , 150 mM KGlu , 5 mM MgSO4 , 0 . 1 mM EDTA and 1 mM DTT ) at a flow rate of 0 . 75 ml per min . Pooled samples were subjected to dialysis with two changes of 1 l of buffer B7 + 20% glycerol and stored at −80°C in small aliquots . ParA samples in buffer A4 + 20% glycerol were incubated with 25 mM EDTA for 1 hr at 37°C before their buffer was exchanged with buffer C ( 25 mM Hepes/KOH pH 7 . 5 , 150 mM KGlu , 2 . 5 mM EDTA and 1 mM DTT ) using a spin column to remove ATP . The samples were mixed with 2 . 5 mM Mg-ATP ( for ‘+ATP’ reactions ) or buffer ( for ‘−ATP’ reactions ) , incubated at room temperature for 10 min , and centrifuged at 16 , 000×g for 30 s . 50-microliter samples of ParA at approximately 50 µM were injected into a Superdex 75 10/300 GL column ( GE Healthcare Life Sciences ) equilibrated with 25 mM Hepes/KOH pH 7 . 5 , 150 mM KGlu , 5 mM MgSO4 , 2 mM Mg-ATP , 0 . 1 mM EDTA and 1 mM DTT for the +ATP reactions or 25 mM Hepes/KOH pH 7 . 5 , 150 mM KGlu , 2 . 5 mM EDTA and 1 mM DTT for the −ATP reactions at a flow rate of 0 . 6 ml/min . The elution profiles were monitored at an absorbance of 280 nm . To examine whether ParB forms dimers in solution , we performed a size exclusion chromatography ( SEC ) coupled with UV , on-line laser light scattering ( LS ) and refractive index ( RI ) detectors ( SEC-UV/LS/RI ) . Specifically , purified His6-ParB ( pre-filtered through a 0 . 22 µm filter ) was applied on a Superose 6 , 10/30 , HR SEC column ( GE Healthcare Life Sciences ) equilibrated in 100 mM HEPES , pH 7 . 4 , 150 mM NaCl , 1 mM DTT buffer and chromatographed at a flow rate of 0 . 3 ml/min . Elution from SEC was monitored by a photodiode array ( PDA ) UV/VIS detector ( 996 PDA , Waters Corp . , Milford , MA ) , differential refractometer ( OPTI-Lab , or OPTI-rEx Wyatt Corp . , Santa Barbara , CA ) , and static , multi-angle laser light scattering detector ( DAWN-EOS , Wyatt Corp . ) . The Millennium software ( Waters Corp . ) controlled the HPLC Alliance 2965 ( Waters Corp . ) system , to which the column was connected to , and data collection from the multi-wavelength UV/VIS detector , while the ASTRA software ( Wyatt Corp . ) collected data from the refractive index detector , the light scattering detectors , and recorded the UV trace at 280 nm sent from the PDA detector . Data collection and analyses were carried out at the Keck Foundation Biotechnology Resource Laboratory , Yale University . The ATPase activity of ParA and ParAR195E was measured using an ATP/NADH-linked assay ( De La Cruz et al . , 2000 ) , which we modified for a 96-well plate format . Reactions ( 100 μl ) containing the appropriate concentrations of ParA , ParB and salmon sperm DNA ( Invitrogen ) , 5 × NADH enzyme mix ( 310 µM NADH , 100 U/ml of lactic dehydrogenase , 500 U/ml pyruvate kinase , and 2 . 5 mM phosphoenolpyruvate ) , 25 mM Hepes/KOH ( pH 7 . 4 ) , 10 mM MgSO4 , 150 mM KGlu , 5% glycerol and 1 mM DTT were mixed with 2 . 5 mM ATP unless mentioned otherwise . Absorbance measurements at 340 nm were taken in Corning UV Transparent Flat Bottom 96-well plates in a Synergy 2 plate reader ( BioTek , Winooski , VT ) at 30°C at 1-min intervals . Initial velocities were calculated from a linear regression of each time course and corrected for spontaneous ATP hydrolysis and NADH oxidation . We also ran parallel control experiments to account for residual ATPase in ParB protein preparations . A standard curve with known amounts of NADH was obtained and used to convert the rate of ADP production from absorbance/time to concentration/rate . We first estimated the number of ParB and ParA molecules by quantitative Western blotting of cell lysates . For ParB , swarmer cells of C . crescentus CB15N were isolated by differential centrifugation ( Evinger and Agabian , 1977 ) and resuspended in M2G medium to an OD660 of 0 . 5 . One milliliter of cell culture was pelleted and used to make a cell lysate for Western blotting . The cell lysate was separated on an Any-kD Mini-PROTEAN TGX Precast Gel ( Biorad , Hercules , CA ) , proteins were transferred to a nitrocellulose membrane that was probed with 1:50 , 000 dilution of anti-ParB polyclonal antibody ( Proteintech Group , Chicago , IL ) and 1:10 , 000 dilution of anti-rabbit secondary antibody ( Biorad ) . Signals were developed with an enhanced chemiluminescence reagent ( GE Healthcare Life Sciences ) and detected by exposing the membrane to a Kodak film ( Carestream Health , Rochester , NY ) or using a Typhoon PhosphoImager ( GE Healthcare Life Sciences ) . The intensities of bands were quantified using ImageJ and compared against a standard curve generated from known amounts of ParB-His6 probed on the same blots . Since , at low concentrations , we lost significant amount of ParB-His6 presumably due to nonspecific binding of the protein to the wall of the plastic tubes , the ParB-His6 standard stock was prepared by diluting purified ParB-His6 in SDS buffer denatured cell lysates made from strain MT174 ( which produces GFP-ParB instead of native ParB ) . After normalizing for the fraction of lysate loaded , we calculated the number of ParB molecules in the CB15N lysate by dividing the ParB amount by the molecular weight of ParB and by multiplying by the Avogadro number . In parallel , the number of cells used to prepare the cell lysate used for the Western blots was determined by serial dilution followed by quantification of colony forming unit ( CFU ) on an agar plate containing PYE medium . Using lysates made from four different pellets , we determined that swarmer cells contain 720 ± 80 ( mean ± SD ) ParB molecules ( or 360 ParB dimers ) per cell . The number of ParA molecules per cell was determined using the same protocol as described for ParB with the following modifications . In the absence of high quality anti-ParA antibody , we used JL-8 anti-GFP antibody ( Clontech , Mountain View , CA ) to quantify the amount of ParA-YFP in a C . crescentus strain ( CJW3010 ) in which parA has been cleanly replaced by a parA-yfp fusion at the native chromosomal location ( Schofield et al . , 2010 ) . Varying amount of purified ParA-YFP-His6 ( added to cell lysate prepared from wild-type cells ) was used to create a calibration curve . We used 1 , 000 × dilution of JL-8 as primary antibody and 10 , 000 × dilution of goat anti-mouse secondary antibody ( Biorad ) . Since the dynamics of ParB/parS segregation in the ParA-YFP-producing CJW3010 strain is indistinguishable from that of the wild-type , we assume that wild-type cells contain similar concentration of ParA molecules . To convert the number of molecules per cell to concentration , we divided the number of molecules per cell by the Avogadro number and the estimated volume of a typical swarmer cell . The cytoplasmic volume of a swarmer cell , 0 . 25 fL , was obtained by cryo-tomography from 3D segmentation of a C . crescentus swarmer cell ( Briegel et al . , 2006 ) . The concentrations of ParA and ParB were calculated using the formula = # Molecules/Avogadro Number × Volume M . For all microscopy observations , cells were spotted on 1% agarose pads containing M2G medium , unless specified otherwise . Images were acquired using an Eclipse Ti-U microscope ( Nikon , Tokyo , Japan ) with an Orca-ER camera ( Hamamatsu Photonics , Hamamatsu City , Japan ) and phase-contrast objective Plan Apochromat 100 × /1 . 40 NA ( Carl Zeiss , Oberkochen , Germany ) at room temperature except for time-lapse experiments ( 30°C ) . Images were acquired and processed with either MetaMorph software ( Molecular Devices , Sunnyvale , CA ) or MATLAB ( The MathWorks , Natick , MA ) . Cell mesh creation and fluorescence quantification were done using the open source , MATLAB-based software MicrobeTracker ( Sliusarenko et al . , 2011 ) . Identification of diffraction-limited fluorescence spots was done using SpotFinder , an accessory in MicrobeTracker , unless mentioned otherwise . To measure the fraction of ParB molecules associated with the chromosomal parS region , we used a strain ( MT174 ) in which the parB gene has been substituted by a functional parB-gfp fusion at the native chromosomal location without affecting the operon structure ( Thanbichler and Shapiro , 2006 ) . MT174 cells and wild-type CB15N cells , both from exponential phase cultures , were spotted on the same pad and imaged . Cells were segmented using MicrobeTracker ( Sliusarenko et al . , 2011 ) . The signal intensity ( in the GFP channel ) of each cell normalized by its area was plotted as a histogram , which clearly showed two populations of cells ( Figure 1—figure supplement 3B ) . A bimodal Gaussian model was used to fit the data . Autofluorescence in MT174 cells producing GFP-ParB was subtracted by the normalized signals detected in CB15N cells ( lacking a GFP fusion ) . We used MicrobeTracker to quantify the fluorescence associated with segmented areas along the long cell axis . We assumed that segments with the least fluorescence concentration ( fluorescence within the segment divided by the segment area ) correspond to cell area far away from the GFP-ParB/parS complex and therefore their fluorescence values report the concentration of freely diffusing GFP-ParB . For each cell , we computationally ranked the fluorescence concentration within each cell segment , selected the lowest 10% of the total segments containing the lowest fluorescence concentration , calculated the average , and normalized this value to the total cell area to obtain the fluorescence value associated with the total diffusing GFP-ParB . This value was divided by the total GFP-ParB signal for that particular cell to obtain the fraction of diffusing GFP-ParB signal . The fraction of GFP-ParB in the partition complex = ( 1−fraction of diffusing ParB ) × 100% . To induce the expression of gfp-parB , 0 . 03% xylose was added to an exponential culture of CJW4762 cells for 60–75 min prior to subjecting the culture to synchrony . Isolated swarmer cells were spotted on a 1 . 5% agarose pad and imaged at 30-s intervals . Cell outlines were acquired using MicrobeTracker and the positions of the partition complex spots within each cell were detected and computed using SpotFinder ( Sliusarenko et al . , 2011 ) . The trajectory of each partition complex was built using a custom-written MATLAB script that minimizes the total displacements of the two spots between the current frame and the previous frame ( Supplementary file 3 ) . To analyze segregation , we considered only trajectories that reached the new pole . To identify the ‘fast’ translocation phase in each trajectory , we wrote a MATLAB script that first linearly interpolates and smoothes each trajectory along the long cell axis before calculating the second derivatives of the curve ( Supplementary file 4 ) . The two inflection points covering the longest distance travelled were defined as the start and end of the ‘run’ phase . All traces were inspected by eye and only traces with correctly assigned start and end points were kept ( 141 out of 155 ) . Partition complexes were identified using a previously reported algorithm for detecting fluorescent objects ( both diffraction-limited and non-diffraction limited ) in images ( Parry et al . , 2014 ) . The output was analyzed to compute the major axis length and minor axis length for each partition complex using the regionprops function in MATLAB . The aspect ratio ( or AR , the ratio between the major axis length and the minor axis length ) was calculated for each partition complex . Before plotting AR as a function of cell coordinate , we oriented each cell such that zero corresponded to the old pole using the localization of ParA-YFP as a reference . Glass slides and cover slips used for single-molecule fluorescence microscopy were washed in the following order: 15 min sonication in 1 M KOH , 15 min sonication in milliQ H2O and 15 min sonication in 70% ethanol with 3 × milliQ H2O rinses between solution changes . Cleaned glass slides and cover slips were then dried with pressured air and used within the same day . DNA mass per C . crescentus chromosome=Total bp per chromosome×mass per bp=4×106 bp×650 Da bp−1=2 . 6×109Da×1g6 . 022×1023Da=4 . 31×10−15g Thus , in vivo DNA concentration =DNA massCell volume=4 . 3×10−15 g2 . 5×10−13 ml=17 mg ml−1 To estimate DPC , we analyzed the movement of the ParB/parS complex in the slow , ParA-independent phase ( Shebelut et al . , 2010 ) , which appears to exhibit diffusive behaviors . As shown in Figure 4—figure supplement 2A , we identified three ‘segments’ within the slow phase for each cell: ‘Segment 1’ corresponds to the part of the trajectory when the ParB/parS complex has detached from the old pole but has not yet duplicated . To identify this segment , we computationally looked for the part of trajectory in which there was only one ParB/parS complex in the cell and its position was at least 320 nm from the old cell pole . ‘Segment 2’ corresponds to the part of the trajectory when the ParB/parS complex proximal to the old pole has separated from its sister but has not yet re-attached to the old pole . This segment started 1 min after two ParB/parS complexes became visibly separated ( to exclude possible active splitting mechanism ) and ended when the complex moved within 320 nm from the old pole . ‘Segment 3’ corresponds to the part of the trajectory when the ParB/parS complex distal to the old pole has separated from its sister and has not engaged in the directed ‘fast’ translocation phase . This segment contained the maximum unbiased stretch of the trajectory 1 min after two ParB/parS complexes became visibly separated . We identified this segment by determining the longest part of the trajectory that remains confined within a virtual 500-nm wide box ( i . e . , no net directional motion ) . We combined data from all trajectories for each of these three segments separately and analyzed their respective displacement distributions and mean square displacements ( MSDs ) ( Figure 4—figure supplement 2B , C ) . We estimated DPC using two different approaches . First , we performed a Gaussian fitting ( which is expected for normal diffusion ) to the displacement distributions , which returned best fit DPC values of 0 . 0007 ± 0 . 0001 , 0 . 00012 ± 0 . 0001 and 0 . 00012 ± 0 . 0001 µm2s−1 for ParB/parS motion during segments 1 , 2 and 3 , respectively ( Figure 4—figure supplement 2B ) . Consistent with these estimations , a linear fit to the first four time points of each MSD gave DPC values of 0 . 0004 ± 0 . 0003 , 0 . 0001 ± 0 . 0006 and 0 . 00008 ± 0 . 0001 µm2s−1 for segments 1 , 2 and 3 , respectively ( Figure 4—figure supplement 2C ) . These values are close to the diffusion coefficient ( 1 . 5 × 10−4 µm2s−1 ) of E . coli chromosomal ori region that we estimated from data presented in Kuwada et al . ( 2013 ) . In addition , the good linear MSD fits suggest that the motions of the unattached ParB/parS complex can be approximated as a normal diffusion at a minute scale ( i . e . , on the time-scale of segregation ) . Therefore , unless specified otherwise , we used DPC = 1 × 10−4 µm2s−1 ( the average value for ‘Segment 2’ and ‘Segment 3’ given by the two approaches ) in all simulations . We estimated DA from published results for generalized ( i . e . , time-dependent ) diffusion coefficient of various chromosomal loci in E . coli , Dapp ∼ ( 0 . 25–1 . 5 ) × 10−3 µm2s−0 . 4 ( Weber et al . , 2012; Javer et al . , 2013 ) . This gives an effective diffusion coefficient at 1-ms time scale ( time interval used in our simulations ) : DA = DappΔt −0 . 6 ∼ ( 1 . 5–9 ) × 10−2 µm2s−1 . We used DA value of 0 . 01 µm2s−1 in the simulations of our DNA-relay model . As a control , we simulated the dynamics of DNA-bound ParA dimers in the absence of other factors and found that this estimated DA value closely recapitulated the step-size and position distributions that are experimentally observed for chromosomal loci such as the LacI-labeled groESL locus ( data not shown ) . CJW2966 , CJW5466 and CJW5468 cells were grown at 30°C in M2G supplemented with 5 µg/ml kanamycin . Prior to imaging , synthesis of LacI-CFP was induced with 0 . 03% xylose ( plus 20 µM IPTG to prevent adverse effects on DNA replication and cell division ) . After 60–80 min , cells were washed and spotted on 1 . 2% agarose containing M2G and 20 µM IPTG , and imaged at 2-s time intervals between frames . LacI-CFP-labeled chromosomal loci appeared as diffraction-limited spots . Their positions ( relative to the long and short axes of each cell ) were determined using the SpotFinder ( Sliusarenko et al . , 2011 ) . Only trajectories with a minimal length of five frames were considered for quantitative analysis . To exclude differences in equilibrium position due to cell-to-cell variability , we measured deviations ( Δx ) from the mean position of each individual trajectory . The probability P ( Δx ) of a locus to fluctuate around its equilibrium point is given by: ( 1 ) P ( Δx ) ∼e−E ( Δx ) kTwhere k is the Boltzmann constant , T is the absolute temperature , and E ( Δx ) is the energy associated with the fluctuation . For an elastic force , F = ksp Δx and E ( Δx ) = kspΔx2/2 ( with ksp being the elastic constant ) and the probability of observing a locus at a distance Δx from the equilibrium is a Gaussian distribution ( which is what we observe in experiments ) : ( 2 ) P ( Δx ) ∼e−kspΔx22kT We fitted the experimental distributions separately for deviations along the short and long axes of the cell with one free parameter σ ( with 1/σ2 = ksp/kT ) , obtaining σ = 38 ± 1 nm and 64 ± 2 nm for the short and long axis , respectively . These values were used in the Brownian dynamics simulations . We considered three models in this study . The ‘diffusion’ model contains only the diffusion terms of the Equation 3 below . The ‘diffusion-binding’ model also includes binding and un-binding of the ParB/parS complex to the DNA-bound ParA dimers , but no force terms ( i . e . , immobile DNA-bound ParA dimers ) . The ‘DNA-relay’ model contains all terms described in Equation 3 , which takes into consideration diffusion , binding/unbinding , and elastic forces ( i . e . , dynamic DNA-bound ParA dimers ) . The parameter values used in each model are listed in the Table 2 unless stated otherwise . Two-dimensional Brownian dynamics ( BD ) simulations were performed using a second-order approximation ( Branka and Heyes , 1998 ) . At each time step , the coordinates of the ParB/parS complex and DNA-bound ParA-ATP dimers were calculated as follows: ( 3 ) x ( ti+Δt ) =x ( ti ) +R0+DkTFΔt+D22 ( kT ) 2FF'Δt2+DkTF'R1 The first three terms in Equation 3 correspond to first-order approximations typically used in BD simulations ( Saxton , 2007 ) , where Δt is the time step of simulations , x ( ti ) is the coordinate of ParB/parS or DNA-bound ParA dimers at the previous time point , D is the diffusion coefficient for the component considered ( DPC for the ParB/parS complex or DA for DNA-bound ParA dimers ) , kT is the temperature in energy units , R0 specifies the random displacement due to diffusion ( i . e . , random number from a Gaussian distribution with zero mean <R0> = 0 and variance <R02> = 2DΔt ) , F is the force acting on the particle considered ( partition complex or DNA-bound ParA-ATP dimer ) . The last two terms in Equation 3 correspond to further expansion of the stochastic equations ( Branka and Heyes , 1998 ) , where F’ is the derivative of the force ( F′ = dF/dx ) and R1 is a random number specified by a Gaussian distribution ( with zero mean <R1> = 0 and variance <R12> = ( 2/3 ) DΔt3 ) , which is correlated with R0: <R0 R1> = DΔt2 . The force terms for DNA-bound ParA and ParB/parS are: ( 4 ) FParA=ksp ( xParA−xParA0 ) and FParA'=ksp ( 5 ) FParB=∑ksp ( xParA−xParA0 ) and FParB'=nABksprespectively , where ksp is the spring constant , xParA0 is the equilibrium point of ParA dimers bound to DNA , and nAB is the total number of ParA dimers bound to the partition complex . Note that Equation 5 is the sum over all ParA dimers bound to the partition complex . Also , only the ratio ksp/kT was defined as a model parameter ( determined from experiment , 1/σ2 = ksp/kT ) since all force terms depends only on this combination of ksp and kT . In our models , the ParB/parS complex can be in one of two states: ( 1 ) freely diffusing ( no force terms ) or ( 2 ) in a complex with one or more DNA-bound ParA dimers ( non-zero force terms ) . We modeled ParA dimers and ParB/parS complex as disks with RParA and RPC equal to 2 and 50 nm , respectively . The partition complex binds to ParA every time their disks overlap . At any given time , ParA could be in one of three different states: DNA-associated , in a complex with ParB/parS , or freely diffusing . The motion of DNA-bound ParA was simulated via Equations 3 and 4 . When a ParA dimer was associated with the partition complex , its coordinate increment was equal to the increment of the partition complex . The motion of freely diffusing ParA monomers was not explicitly simulated , since ParA monomers do not interact with the partition complex . The motion of freely diffusing ParA is relatively fast ( DA-free >> DA ) leading to a uniform random distribution across the cytoplasm within seconds . Free ParA molecules bind to DNA with uniform probability with respect to the short cell axis . However , in the long-axis direction , we emulated ParA distribution observed in cells ( Figure 4—figure supplement 1 ) by assuming that the probability of ParA to re-bind to DNA is a linearly increasing function between the current ParB/parS location ( xPC ) and the new pole: ( 6 ) PParA−DNA=2 ( l−xPC ) 2 ( x−xPC ) , for xpc<x<l , where l is a cell length , and the coordinates are defined as x = 0 at the old ole and x = l at the new pole . The transition of ParA between different states was modeled as a stochastic process ( exponential distribution of times ) with average times τdb ( for ParAfree → ParADNA-bound ) and τcat ( for ParAParB/parS bound → ParAfree ) . The rate of ParA dissociation from the DNA upon interaction with the ParB/parS complex is equal to the maximum hydrolysis rate , τcat = 1/kcat ≈ 30 s ( Figure 1F ) , unless mentioned otherwise . Note that because of its slow rate , dissociation of ParA from the DNA due to the basal ATPase activity of DNA-bound ParA was neglected . Since for the plasmid P1 system , the rate of freely diffusing ParA monomers dimerizing and binding to DNA is thought to be comparable to ATP hydrolysis rate ( Vecchiarelli et al . , 2010 ) , we assumed that τdb = τcat = 30 s for all simulations shown here . Note that varying τdb between 1-100 s did not significantly affect the results of our simulations ( data not shown ) . We defined the starting conditions using experimentally measured values . At the beginning of the simulation , the partition complex was placed at x ( 0 ) = 0 . 8 µm ( measured from the old pole , Figure 2B ) , y ( 0 ) = 0 . At the start of each simulation , ParA-ATP dimers were distributed randomly using Equation 6 with xPC = x ( 0 ) . Cell boundaries were modelled as a reflective 2 . 5 µm × 0 . 4 µm rectangular . All simulations were carried out using Δt = 1 ms for a total of 2000 s or until the partition complex crosses xfinish = 2 . 6 μm , which ever happened first . To estimate the optimal attachment time between the partition complex and ParA-ATP dimers , we calculated the average time ( Δt ) required for the partition complex to reach an equilibrium point since its association with a DNA-bound ParA dimer . A particle under elastic force ( F = ksp Δx where ksp is the elastic constant ) moves a distance Δx in a viscous medium during an interval Δt: Δx = v Δt = µ F Δt = ( D/kT ) ksp Δx Δt Here v is the velocity of the particle , µ is the particle mobility linked to the diffusion coefficient DPC as µ = DPC/ ( kT ) , k is the Boltzmann constant and T is the absolute temperature . This allowed us to calculate the optimal time of ATP hydrolysis using σlong2 = kT/ksp: Δt = kT/ ( DPC ksp ) = σlong2/DPC = ( 0 . 06 µm ) 2/10−4 µm2 s−1 = 36 s .	DNA molecules exist in cells as tightly packed structures called chromosomes . During the cell cycle , chromosomes duplicate , and the two copies separate , ready to end up in two separate daughter cells . The process of chromosome separation in cells of higher organisms such as animals and plants is well understood . A protein-based structure called the spindle apparatus guides the separating chromosomes to different ends of the dividing cell . However , how chromosome separation occurs in bacteria is not well understood despite its importance in bacterial multiplication . The nucleoprotein complex ParABS is critical for separating chromosomes in bacteria . The complex is made up of three parts: parS , a stretch of DNA located at the point where chromosome duplication begins; and two proteins called ParB and ParA . While the molecular players are known , how they work together to separate chromosomes is under debate . One popular suggestion is that ParA forms a spindle-like structure . Alternatively , a diffusion-based mechanism has been proposed , where a gradient of ParA molecules bound to the chromosome interacts with a parS/ParB complex , directing the diffusion of the complex . Lim et al . used quantitative microscopy to observe the movement of the parS/ParB complex and the spatial distribution of the ParA proteins in a model bacterium . The results were inconsistent with the presence of a spindle-like structure in the cells . A mathematical model describing chromosome movement—based on the number and activity of ParA and ParB found in live cells—also failed to fit with either the spindle or the diffusion theory . Instead , Lim et al . propose a new model , based on the discovery that chromosomes are elastic . In this ‘DNA-relay model’ , ParA is bound to DNA . When ParB intermittently binds to ParA , it usually catches the ParA-DNA complex when the complex is elastically stretched . The elasticity of the chromosome itself then makes the parS/ParB complex move in the direction where the most ParA molecules are still bound to the chromosome . Lim et al . suggest that similar elastic mechanisms could also be behind more general intracellular transport in bacteria .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "microbiology", "and", "infectious", "disease" ]	2014	Evidence for a DNA-relay mechanism in ParABS-mediated chromosome segregation
Left-right asymmetries in brains are usually minor or cryptic . We report brain asymmetries in the tiny , dorsal tubular nervous system of the ascidian tadpole larva , Ciona intestinalis . Chordate in body plan and development , the larva provides an outstanding example of brain asymmetry . Although early neural development is well studied , detailed cellular organization of the swimming larva’s CNS remains unreported . Using serial-section EM we document the synaptic connectome of the larva’s 177 CNS neurons . These formed 6618 synapses including 1772 neuromuscular junctions , augmented by 1206 gap junctions . Neurons are unipolar with at most a single dendrite , and few synapses . Some synapses are unpolarised , others form reciprocal or serial motifs; 922 were polyadic . Axo-axonal synapses predominate . Most neurons have ciliary organelles , and many features lack structural specialization . Despite equal cell numbers on both sides , neuron identities and pathways differ left/right . Brain vesicle asymmetries include a right ocellus and left coronet cells . Animals exhibit various forms of left-right asymmetry ( Ludwig , 1932; Neville , 1976; Brown and Wolpert , 1990; Palmer , 2009 ) , and anatomical examples in brains are reported both in invertebrates ( Hobert et al . , 2002; Frasnelli et al . , 2012 ) , and vertebrates ( Rogers and Andrew , 2002; Duboc et al . , 2015 ) as differences in cell number ( Blinkov and Glezer , 1968 , their tables 184 and 185 ) and particularly in the lateralization of the mammalian cortex ( Galaburda et al . , 1978 ) . Even predominantly symmetrical bilaterians exhibit various forms of sidedness , in both protostomes ( Grande and Patel , 2009 ) and deuterostomes ( Hamada et al . , 2002; Duboc et al . , 2005 ) . In large , complex brains , these anatomical asymmetries are often structurally cryptic , and gene expression has been used as a proxy to examine the evolution of sidedness . As a sister group to vertebrates ( Cameron et al . , 2000; Satoh et al . , 2014 ) , ascidians have larvae that are chordate in body plan and mode of development , with a dorsal central nervous system ( CNS ) . In the ~1 mm tadpole larva of Ciona intestinalis neurons are distributed rostrocaudally in three main centres , a brain vesicle , motor ganglion and caudal nerve cord ( Katz , 1983; Nicol and Meinertzhagen , 1991; Meinertzhagen et al . , 2004 ) . The CNS forms from a neural tube , yet exhibits left/right differences , and so provides a useful model to study many aspects of brain asymmetry . This issue is important because brain laterality has been associated with increased fitness for animal life ( Duboc et al . , 2015 ) . The most studied tunicate species is Ciona intestinalis ( Satoh , 1994 ) . Not only does its development result from a fixed pattern of cell lineage and result in a mere ~ 2600 cells in the larva of Ciona intestinalis ( Satoh , 1999 ) , but the genome , first in Ciona intestinalis ( Dehal et al . , 2002 ) and now in nine other species ( Brozovic et al . , 2016 ) , has been sequenced . Even though the events of early neural development and the nervous system’s subsequent metamorphosis have been identified , together with many of their underlying causal gene networks ( Satoh , 2003; Sasakura et al . , 2012 ) , the detailed cellular organization of their product , the CNS of the swimming larva , still remains almost entirely unresolved . Ciona releases 5000–10000 eggs per individual ( Petersen and Svane , 1995 ) , and its eggs are released either individually , or in a mucous string ( Svane and Havenhand , 1993 ) . Gametes undergo fertilization , cleavage , development , and then hatch into non-feeding lecithotrophic larvae in the water column . Initially after hatching , larvae swim up toward the surface of the water by negative geotaxis using the otolith cell ( Tsuda et al . , 2003 ) a behaviour retained in ocellus-ablated larvae . Later in larval life , larvae exhibit negative phototaxis , swimming down to find appropriate substrates for settlement ( Tsuda et al . , 2003 ) . The swimming period exhibits three characterized behaviours: tail flicks ( ~10 Hz ) , ‘spontaneous’ swimming ( ~33 Hz ) , and shadow response ( ~32 Hz; Zega et al . , 2006 ) . Larvae swim more frequently and for longer periods earlier in life up to 2 hr post hatching ( hph ) . Of the reported behaviors , the shadow response , in which a dimming of light results in symmetrical swimming , is the best studied , developing at 1 . 5 hph and increasing in tailbeat frequency after 2 hph ( Zega et al . , 2006 ) . In addition to phototactic and geotactic behavior , there is evidence of chemotactic behavior just before settlement ( Svane and Young , 1989 ) and of some mechanosensory responses in swimming larvae ( Bone , 1992 ) . Because larvae do not feed , their main biological imperative is survival and successful settlement to undergo metamorphosis into a sessile adult , in an environment with appropriate food and reproductive resources . Thus , entering the water current and avoiding predation by filter feeders may be the foundation for the larva’s many behavioral networks , especially in early life before settlement . The substrate for these behaviours is the larva’s dorsal central nervous system , which is divided into the anterior sensory brain vesicle ( BV ) , connected by a narrow neck to the motor ganglion ( MG ) within the larval trunk , and a caudal nerve cord ( CNC ) in the tail ( Nicol and Meinertzhagen , 1991 ) . Sensory neurons of the CNS and their interneurons reside in the BV , which has an expanded neural canal and the most complex neuropil . The relay neurons of the posterior brain vesicle extend axons through the neck to the motor ganglion , which overlies the anterior portion of the notochord , and contains neurons of the motor system . At the trunk-tail border , muscle cells of the tail flank the notochord and CNS , and these extend down through the tail alongside the narrow , simple CNC . In addition to the CNS several sensory epidermal neurons ( ENs ) of the peripheral nervous system ( PNS ) populate the dorsal and ventral axes of the larva in a rostrocaudal sequence , with axons running beneath the epidermis ( Imai and Meinertzhagen , 2007b ) . Many asymmetries have been uncovered by the developmental expression of Nodal and its signaling pathways ( Hamada et al . , 2002; Hudson , 2016 ) . As in vertebrates , in ascidians , their sibling group ( Satoh et al . , 2014 ) , Nodal expresses on the left hand side of the developing embryo ( Boorman and Shimeld , 2002a , 2002b; Yoshida and Saiga , 2008 ) . This is true neither of other deuterostomes ( Duboc et al . , 2005 ) nor lophotrochozoans ( Grande and Patel , 2009 ) , while ecdysozoans such as Drosophila and C . elegans lack Nodal ( Schier , 2009 ) , even though the brain in Drosophila is asymmetrical ( Pascual et al . , 2004 ) . The development of brain asymmetry in the ascidian does however depend on the presence of an intact chorion in the embryo ( Shimeld and Levin , 2006; Yoshida and Saiga , 2008; Oonuma et al . , 2016 ) . In contrast to the situation in most chordates , structural brain asymmetries , which include cell numbers , positions , and connections are externally visible in the tadpole larva of ascidians , for example from the pigment spots and right-sided ocellus in the head of Ciona intestinalis ( Eakin and Kuda , 1971; Katz , 1983; Nicol and Meinertzhagen , 1991 ) . Photoreceptor neurons associated with the ocellus are of the ciliary type , with outer segment lamellae orientated parallel to the cilium ( Eakin and Kuda , 1971 ) , in contrast to the perpendicular arrangement found in vertebrate rods ( Lamb and Collin , 2007 ) . The photoreceptor cells of the ocellus on the right of the brain vesicle , which number 20 cells ( Horie et al . , 2008 ) , 17 and 18 of which had been reported by Nicol and Meinertzhagen ( 1991 ) , are twinned with 17 to 19 structurally different coronet cells ( previously claimed hydrostatic pressure receptor cells: Eakin and Kuda , 1971; Nicol and Meinertzhagen , 1991 ) on the left . This sidedness may also correspond to larval behavior because ascidian larvae pursue a helical trajectory when swimming ( McHenry and Strother , 2003; McHenry , 2005 ) and ascidian larvae are thought to use klinotaxis to respond to visual cues by modulating the symmetry of tail kinematics ( McHenry and Strother , 2003 ) . The pattern of helical swimming arises from bilateral contractions of the tiered muscle bands on either side of the notochord . On each side , gap junctions connect all 18 uninucleate muscle cells , arranged in three rows , dorsal medial and ventral ( Bone , 1992 ) . Muscle activity along the tail is thus not segmental , and instead propagates posteriorly in a wave from innervated anterior dorsal and medial muscle cells at the trunk-tail border without the requirement for additional neuronal input along the tail ( Bone , 1992 ) . In the past , the search for brain asymmetries has been frustrated by two features , the presumed relative rarity of such asymmetries , and the lack of structurally identified networks of neurons in which to recognise them . These obstacles are resolved in the CNS of the tadpole larva of Ciona , in which not only is the asymmetrical organization obvious , but the possibility exists to uncover the larva’s complete network of neuronal synaptic connections , or its connectome ( Lichtman and Sanes , 2008 ) . That possibility rests on the tiny size of the brain , the small number of its constituent cells , approximately 330 ( Nicol and Meinertzhagen , 1991 ) , and the morphological simplicity of its neurons ( Imai and Meinertzhagen , 2007a ) . The apparent symmetry of the characteristic swimming pattern of chordate-like tail undulations ( Bone , 1992; Video 1 ) stands in marked contrast to the asymmetries in the CNS that generates them . Here we present the connectome of the CNS of the tadpole larva in Ciona , and reveal its asymmetries in the left and right complements of neurons and their synaptic networks . 10 . 7554/eLife . 16962 . 003Video 1 . Symmetrical undulations of the tail in a swimming Ciona larva . The tail lacks segmentation and in the 2 hr hatchling larva oscillates at 20–30 Hz at the juncture with the rostral trunk ( Bone , 1992 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 003 Synapses were identified based on criteria established in other invertebrate species ( White et al . , 1986; Westfall , 1996; Meinertzhagen , 2016 ) : primarily a cluster of vesicles at a presynaptic membrane . Small electron-lucent vesicles 30 to 60 nm in diameter were found exclusively at some presynaptic sites where they tended to cluster closely to form a tight cumulus ( Figure 2A ) , accompanied at some synapses by larger electron-lucent vesicles ( 70—110 nm ) ( Figure 2B ) . The same vesicle types were found throughout the neuron , but their distribution was not quantified . Some synapses also had dense-core vesicles , large ( 110–140 nm; Figure 2C ) , medium ( 70–80 nm; Figure 2B ) , or small ( 40–60 nm ) ( Figure 2D; Figure 2—source data 1 ) . The size of the internal density within these vesicles varied , some medium sized dense-core vesicles having small cores ( Figure 2E ) . Other synapses had exclusively medium to large dense-core vesicles . Postsynaptic densities ( Gray , 1959; Peters and Palay , 1996 ) were observed at some synapses ( Figure 2C ) but not at all , thus providing an unreliable criterion for a synaptic contact . 10 . 7554/eLife . 16962 . 007Figure 2 . Synapses contain presynaptic vesicles of various sizes and types . ( A ) Tightly packed cumulus of small ( 30–40 nm ) vesicles at a single presynaptic site ( arrow ) . ( B ) Mixed populations of small ( 30–50 nm ) and large ( 70–110 nm ) electron-lucent vesicles ( arrow ) as well as dense-core vesicles of medium size ( arrowhead ) . ( C ) Large ( 100–110 nm ) vesicles with dark cores ( arrowheads ) . ( D ) Synapses containing electron-lucent vesicles ( 30–60 nm; arrow ) as well as small ( 60 nm; small arrowhead ) and medium ( 80 nm; large arrowhead ) dense-core vesicles . See Figure 2—source data 1 for a list of those neurons with synapses having mixed vesicle populations . ( E ) Medium dense-core vesicles with small cores ( arrows ) . ( F ) Membrane apposition ( between arrows ) interpreted as a gap junction , with membrane densities on both sides . Scale bars: 500 nm ( A , B , C and E ) ; 200 nm ( D and F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 00710 . 7554/eLife . 16962 . 008Figure 2—source data 1 . Neurons with synapses having mixed electron-lucent and dense-core vesicle populations ( mixed ) , or exclusively dense-core vesicle ( dcv ) populations , with numbers ( No . ) of synapses of each type , and totals of both dcv and mixed vesicle synapses . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 008 Membrane appositions interpreted as putative gap junctions ( Bennett and Goodenough , 1978 ) were annotated for contacts at which densities were present on the membranes of both sides ( Figure 2F ) , except where such contacts formed directly adjacent to the neural canal , thus excluding junctions provisionally interpreted as desmosomes or adherens junctions . Synapses and gap junctions varied in size , contributing profiles to between 1 and 29 sections ( <66 sections for neuromuscular junctions ) or 55 sections ( gap junctions ) ( Figure 3A ) . 10 . 7554/eLife . 16962 . 009Figure 3 . Synapse numbers ( presynaptic sites ) and sizes for all neurons ( for complete list see Figure 3—source data 1 ) . ( A ) Most synaptic contacts extend over <10 60 nm sections . Those occupying >10 sections are neuromuscular junctions , inputs from relay neurons to MG neurons , and synapses from antenna cells . The frequency curve for chemical synapses reveals more large contacts than for gap junctions . ( B ) Plotted for all neurons , the total depth of presynaptic contact co-varies linearly with the total number of synapses ( R2 = 0 . 91 ) . Removing single-profile synapses eliminates 18% of all synaptic partnerships , and removing all 2-profile synapses would have eliminated a total of 35% of all synaptic partnerships . ( C ) The number of presynaptic sites co-varies with the number of postsynaptic partners according to a power function ( R2 = 0 . 81 ) . The number of synaptic partners is also referred to as the network statistic ‘degree’ , and is mapped to the synaptic network ( Figure 3—figure supplement 2 ) . Five neurons lie well above the curve , having low degrees , with many synapses and few postsynaptic partners . These are: Antenna neuron 1 ( Ant1 ) with many synapses onto seven relay neurons , and the two pairs of anterior-most motor neurons ( MN1 and MN2 ) with many synapses onto muscle . ( D ) Cumulative distribution of the log number of presynaptic sites over the surfaces of neurons of major cell classes . TERM: terminal; AX: axon; CB: cell body; DEN: dendrite . Most synapses are located over axons and terminals ( see Figure 3—figure supplement 1 for ( D ) averages per neuron and ( E ) postsynaptic site distribution ) . ( E ) Proportions of synapses and gap junctions in the connectome formed for particular partnerships . ( F ) Reciprocity of connections in the network given as the proportion of neuron partners that are reciprocally connected and the extent of their reciprocity ( calculated as the cumulative depth of contacts in one direction divided by the sum of the depth of all contacts between the neuron pair ) . The total proportion of reciprocal synaptic connections between neuron pairs is 0 . 39 . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 00910 . 7554/eLife . 16962 . 010Figure 3—source data 1 . Summary of all neurons in the larval CNS of Ciona intestinalis . Neurons listed by ID , with cell type , morphological features , location , presence or absence of cilia , and number of each neuron’s pre- and postsynaptic sites or putative gap junctions ( >0 . 06 µm ) . Ependymal cells excluded . The final column shows left lateral views of individual neuron reconstructions ( whole cells , or terminals for photoreceptors ) as small thumbnails with scale bars: 1 µm ( thick bars: coronet cells , lens cells , photoreceptor terminals and PR-III cells , vacINs ) ; 10 µm ( thin bars: all other cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 01010 . 7554/eLife . 16962 . 011Figure 3—figure supplement 1 . Relationships between the morphology and synaptic output of larval CNS neurons . ( A ) Number of synapses is not correlated with soma volume ( μm3 ) ( r2 = 0 . 4 ) . ( B ) Soma volume does not correlate with the combined surface area of its axon and terminal regions ( μm2 ) ( r2 = 0 . 4 ) . ( C ) Axon and terminal surface area is linearly related to the number of presynaptic sites , but the correlation is not strong ( r2 = 0 . 7 ) . ( D , E ) Distribution of synaptic sites for major neuron types presented as an average number of synapses per neuron per region of the cell for presynaptic sites ( D ) and postsynaptic sites ( E ) . ( F ) Frequency distribution of the geometric means of cumulative synaptic depth in each direction for all neuron pairs . Values >2 follow the tailed distribution up to a maximum geometric mean of 17 . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 01110 . 7554/eLife . 16962 . 012Figure 3—figure supplement 2 . Network graphs with network statistics visualized as attributes . Edge-weighted spring embedded layout applied to both graphs , using EdgeBetweeness . ( A ) Network of chemical synapses . In-degree ( number of presynaptic partners ) mapped to node size , Out-degree ( number of postsynaptic partners ) mapped to node colour , Edge betweenness ( number of shortest paths that go through an edge ) mapped to edge colour , and cumulative synaptic depth mapped to edge width . ( B ) Network of putative gap junctions . Degree ( number of partners , undirected ) mapped to node size and node colour , Edge betweenness mapped to edge colour , and cumulative contact depth mapped to edge width . Scale for In-degree and Out-degree is equivalent to undirected Degree values for the gap junction network . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 012 A total of 301 cells of the CNS were imaged , of which the CNS included 177 neurons with axons and presynaptic sites ( Figure 1—source data 1; Figure 3—source data 1 ) , and constituting the remainder , ependymal cells ( those ciliated cells abutting the canal that lack an axon ) and two cells of the CNS that are ambiguous , having presynaptic sites , but lacking a neuronal form ( Figure 1—source data 1 and Figure 3—source data 1 ) . Cells omitted from our EM series , those rostral to the otolith and caudal to the bipolar tail neurons ( Imai and Meinertzhagen , 2007b; Stolfi et al . , 2015 ) , are presumed to account for the remainder of the >331 cells reported by Nicol and Meinertzhagen ( 1991 ) and thus to number at least 30 . This assumes the constancy of cell number between different sibling batches of larvae . Between the CNS neurons ( and the four bipolar tail neurons ) , we identified 8768 synapses ( 6618 >1 section ) , including 1772 neuromuscular synapses , and 2105 putative gap junctions ( 1206 >1 section ) . Each CNS neuron thus formed on average 49 ( standard deviation , SD 61 ) presynaptic sites with a range of between 1 and 430 synapses and an average of 13 ( SD 23 ) putative gap junctions , with a range between 0 and 166 . Each postsynaptic neuron received an average of 39 ( SD 42 ) synapses in total from all its presynaptic partners , with a range between 0 and 179 . For each neuron , the number of presynaptic sites varies with the number of its postsynaptic partners , plotted for all neurons and their synapses but excluding the neuromuscular junctions ( Figure 3C ) . This relationship indicates that with each additional partner , the number of synapses made by a presynaptic neuron increases , thus reflecting a postsynaptic drive to the total synapse load . There was no overall relationship for each neuron between the volume of its soma and the number of its synapses ( r2 = 0 . 4 ) , nor between soma volume and axon/terminal surface area ( r2 = 0 . 4; Figure 3—figure supplement 1 ) . However , axon/terminal surface area and number of synapses were weakly correlated ( r2 = 0 . 7 ) ( Figure 3—figure supplement 1C ) . Synaptic structure is sometimes unpolarised ( Figure 4A ) , with synaptic vesicles situated on either side of the synaptic cleft; neurons also frequently connected to form both reciprocal ( Figure 4B ) and serial synapses ( Figure 4C; Table 1 ) . Of the 8617 synapses , 922 were polyadic , having multiple postsynaptic elements ( Figure 4C; Figure 4—source data 1; Table 1 ) . The most common of these were dyads , which constituted 93% of all polyadic synapses , and were common especially in antenna neurons ( see below ) . 10 . 7554/eLife . 16962 . 013Figure 4 . Unpolarized , reciprocal , and serial synapses . ( A ) Unpolarized mixed synapse between cell 115 and cell 23 with dense-core ( arrowhead ) and electron-lucent ( arrow ) vesicles on both sides of the synaptic cleft . ( B ) Single section with synapse from cell 102 to cell 120 ( black arrow ) and a reciprocal partner synapse from cell 120 to cell 102 ( red arrow ) . Arrowhead: membrane apposition marking a putative gap junction . ( C ) Serial dyad synapse ( black arrow ) from single neuron onto two postsynaptic targets ( 157 and 74 ) , one of which is presynaptic in the same section ( red arrow ) at a dyad synapse onto two neurites ( AMG1 and AMG2 ) . ( D ) Serial monad synapse ( black arrow ) onto a single postsynaptic target ( 126 ) that is presynaptic at an adjacent synapse ( red arrow ) to a single postsynaptic target ( 116 ) . Scale bars: 1 µm . ( E ) Series of four 60 nm sections through a single synapse . The pre- and postsynaptic cell are labelled in the top image . A clear cumulus of presynaptic vesicles is visible all images , and a clear postsynaptic density in the penultimate image ( arrowhead ) . Scale bar: 500 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 01310 . 7554/eLife . 16962 . 014Figure 4—source data 1 . Comparison of synaptic complements using different parameters and exclusion criteria . Exclusions include synapses onto the basal lamina ( bm ) , synapses onto ependymal cells ( Ep ) , neuromuscular junctions ( Mu ) , synapses onto no apparent postsynaptic neuron ( space ) , and synapses observed in fewer than two sections ( >1 section ) . For values excluding neuromuscular junctions the neuromuscular junction values and percentages are given . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 01410 . 7554/eLife . 16962 . 015Table 1 . Numbers of synapses and gap junctions . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 015Total no . Total no . sectionsMean no . sections/contactNo . synapses>1 sectionMean no . sections/contact >1 section% Unpolarized% Polyad% DcvSynapses8617301633 . 566184 . 35 . 210 . 78Gap junctions320557651 . 812063 . 1 ? 3N/APercentage ( % ) refers to the percentage of all synapses that are unpolarized ( presynaptic vesicles on either side of the cleft between both neuron partners ) ; polyadic ( having >1 postsynaptic neurite ) ; or containing dense-core vesicles ( dcv ) at the presynaptic site . The network forms a single connected component , with all cells ( nodes ) being connected by a synapse ( edge ) to another node in the network ( Figure 3—figure supplement 2 ) . The network statistics ( Table 2 ) reveal that the characteristic path length between two neurons is 2 . 7 ( from one neuron , through one other to its target ) , with neurons having an average of 20 neighbors ( synaptic partners ) and an overall average network clustering coefficient ( existing edges between neighbors of a neuron/possible edges between neighbors of a neuron ) of 0 . 333 . 10 . 7554/eLife . 16962 . 016Table 2 . Network statistics for networks of chemical synapses and putative gap junctions . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 016Statistic Synaptic network Gap junction network Full network CNS neurons only Full network CNS neurons only ( >0 . 06 μm ) Clustering co-efficient0 . 3330 . 3350 . 250 . 305Connected component1171Network diameter9788Radius1414Shortest paths90% [41001] 95% [29759] 85% [31536] 100% [16770] Characteristic path length2 . 6842 . 5413 . 0782 . 775Average number of neighbours20 . 16920 . 6898 . 67410 . 369Number of nodes213177193130Network density000 . 0450 . 08Network heterogeneity--0 . 9350 . 76Number of self-loops1916139Multi-edge node pairs8266993022Network centralisation--0 . 1910 . 257Network statistics calculated using the Cytoscape Network Analyzer for network of chemical synapses ( Synaptic network ) and putative gap junctions ( Gap Junction network ) for both the full network thus including PNS neurons , muscle , ambiguous cells , and synapses onto basal lamina , as well as CNS neurons; and the network for CNS neurons only ( CNS neurons ) . Note that the 'CNS neurons only' network excludes one additional isolated profile of a single branch of one photoreceptor terminal , probably pr10 . Within the CNS the distribution of presynaptic sites over the surface of the neuron varies by cell type . Most neurons are monopolar , <25% only having dendrites , and their axons usually form a clear terminal . Axons fasciculate in bundles but braid their positions within their bundle or sometimes defasciculate . Brain vesicle ( BV ) intrinsic interneurons have approximately equal numbers of presynaptic sites over their axons as their terminals , each constituting approximately 40% of their total . In contrast , axo-axonal synapses are the predominant synapses involving relay and motor ganglion ( MG ) interneurons , comprising >50% for both ascending and descending MG interneurons and >35% for BV relay neurons . The abundance of en passant synaptic contacts in the larval CNS of Ciona is particularly apparent when examining the proportional distribution of postsynaptic sites . Synapses formed by BV relay neurons at their terminals form en passant onto the axons and terminals of both relay and MG neuron targets , which together constitute >20% of all BV relay neurons’ presynaptic contacts . Likewise , >50% of BV intrinsic interneuron synapses form en passant from their axons or terminals onto the axons or terminals of their targets . Synapses from terminal to terminal also constitute the greatest proportion of photoreceptor synapses ( 43% ) , and en passant synapses constitute over 60% of their contacts . Overall , within the CNS and among the axons of the dorsal PNS , 68% of all neuron-neuron synapses terminate on axons or terminals . More synaptic contacts are made upon axons than upon terminals , the latter constituting only 23%–26% of all synapses , whereas those onto axons comprise 42%–44% . These numbers support the impression that presynaptic sites are formed at various places over the cell surface , and that each cell type has a preferred location to form them , but that this location is neither absolute nor exclusive . For further information see Figure 3 and Figure 3—figure supplement 1D The overall cell complement including neurons , ependymal and accessory cells , is closely similar on the two sides ( left: 125; right: 129; midline 46 ) . The brain vesicle has unequal numbers of neurons and pigment cells , however , with more sensory neurons and pigment cells on the right side and more interneurons on the left ( Table 3 ) . Many of these left-side interneurons were previously identified as ciliated ependymal cells ( Nicol and Meinertzhagen , 1991 ) , but have been seen here to possess axons and synapses that were not visible by light microscopy , and are thus to be considered neurons . Excluding ependymal cells , each brain region has approximately equal numbers of cells on both sides ( Table 3 ) . Despite this near equal distribution in their overall numbers , however , examples of asymmetries in the numbers of identified neuron types , or numbers of neurons of each type , were nevertheless also found amongst specific interneuron classes in the brain vesicle . In the ventral motor ganglion most neurons are paired , including five pairs of motor neurons and four pairs of interneurons , whereas caudally among the ascending contralateral interneurons , ACINs , there were two representatives on the left and one on the right , as well as two descending posterior motor ganglion neurons ( PMGNs ) found only on the right ( Table 3 ) . 10 . 7554/eLife . 16962 . 017Table 3 . Numbers of cells in the left , right and centre of the CNS and PNS . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 017Left Centre Right Lens cells3Pigment cells2Total: Pigment and lens cells 5 Coronet1321Photoreceptors 37Antenna neurons11Photoreceptor tract interneurons3Anterior BV neurons291BV peripheral interneurons441Bipolar neurons2Anaxonal arborizing neurons12Posterior BV peripheral interneurons211Photoreceptor relay neurons 6Photoreceptor-peripheral relay neurons 226Photoreceptor-coronet relay neurons 21Antenna-coronet relay neuron1Antenna relay neurons 72Peripheral relay neurons 21Relay interneurons5Total: BV neurons 72 14 57 Neck neurons11Total: Neck neurons 1 1 Ascending MG peripheral interneurons313Descending decussating neurons11MG interneurons33Motor neurons 55Total: MG neurons 12 1 12 Ascending contralateral inhibitory neurons ( ACINs ) 21Posterior MG interneurons2Mid-tail neurons22Total: CNC neurons 4 5 All CNS neurons 88 15 75 Peripheral nervous system Bipolar tail neurons22Peripheral neurons ( RTENa ) 66anterior ATENs22posterior ATENs4DCENs4Total: PNS neurons 8 6 14 Neurons of the left side of the nervous system outnumber those of the right , which in turn outnumber those of the centre . All CNS neurons include known neurons that lack synapses ( ) . **Additional mid-tail neurons which lay beyond the analysed region of the EM series are excluded from the totals . Neurons are distributed unequally with respect to cell type . The most obvious case for sidedness in the CNS has been long recognized from the composition and placement of the ocellus and coronet cells ( Meinertzhagen and Okamura , 2001; Meinertzhagen et al . , 2004 ) . On the left side , we found the 17 enigmatic coronet cells ( Figure 5A–B ) , each structurally distinguished by a single bulbous protrusion and expressing immunoreactivity to dopamine ( Moret et al . , 2005 ) . These have short axons that terminate against the basal lamina where they form synapses with almost exclusively dense-core vesicles 80–110 nm in diameter ( Figure 6A ) . Additional synapses are formed at close range onto other interneurons , including relay neurons ( pr-corRN and ant-corRN ) , or onto neighbouring coronet cells ( Figure 6B ) . Alongside the coronet cells lie ciliated neurons , some of which have cilia that project towards the bulbous protrusions ( Figure 6C ) . 10 . 7554/eLife . 16962 . 018Figure 5 . Sensory neurons and associated cells have sided distributions . Reconstructed coronet cells ( Cor ) with their bulbous protrusions ( BP , one with a black arrow ) and -- in their correct relative position -- six layers of photoreceptor neurons , excluding their terminals , together with otolith ( Ot ) and ocellar ( Oc ) pigment cells . Reconstructions shown from a dorsal ( A ) , or frontal ( B ) view . ( C ) Reconstruction of visual system components ( including photoreceptor tract ( trIN ) and vacuolated sensory ( vacIN ) interneurons , shown from the right side , anterior to the right . ( D ) Sensory neurons ( spheroids ) and their modified cilia reconstructed within the outline of the CNS from left lateral , dorsal , and frontal views . Cells coloured as in panels A-C , PR-I outer segments in yellow , Pr-II outer segments in teal , coronet bulbous protrusions in green . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 01810 . 7554/eLife . 16962 . 019Figure 6 . Synapses of coronet cells . ( A ) Synapse , containing exclusively dense-core vesicles ( arrow ) , from a coronet cell onto the basal lamina ( BL ) . ( B ) Unpolarized synapse between two coronet cells , with dense-core vesicles ( arrows ) on both sides of a synaptic cleft . ( C ) Reconstruction of coronet cells each with a bulbous protrusion ( arrow ) alongside coronet-cell associated somata ( grey ) of ciliated neurons , with cilia reconstructed in black . Scale bars: 1 µm ( in A and B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 019 On the right side , five rows of photoreceptors which , like those of vertebrates , and as previously reported ( Dilly , 1962 , 1969; Eakin and Kuda , 1971 ) , extend a stalk that contains a basal body and expands to form a ciliary outer segment in the ocellus pigment cup . In addition to the Group I photoreceptors of the ocellus , two anterior rows of Group II photoreceptors adjacent to the ocellus have ciliary outer segments that extend into the neural canal , with smaller terminals extending a short distance into the posterior brain vesicle neuropil . The number and arrangement of photoreceptor neurons confirms the presence of the additional class of photoreceptors identified using Ci-opsin immunolabeling , with outer segments outside the ocellus pigment ( Horie et al . , 2005 , 2008 ) . These photoreceptors , along with those of the ocellus , total 30 , as reported previously ( Horie et al . , 2005 , 2008 ) . We identify 23 with outer segments that projected into the ocellus pigment , and 7 into the neural canal ( Video 2; Video 3; Video 4 ) . The 17–18 photoreceptors reported by Nicol and Meinertzhagen ( 1991 ) appear to represent just some of the rows of nuclei from the five fans of ocellus photoreceptors , those that were visible in semithin sections and that lay close to the opaque pigment . In addition to the 30 photoreceptors , three anterior sensory interneurons , unique in having axons within the sensory axon tract , lie on the right side in the anterior brain vesicle ( Figure 5C ) . We also identify a further group of seven Group III ( Horie et al . , 2008 ) right-side photoreceptors which lie posterior and ventral to the 30 ( Figure 5C ) , are vacuolated , and have outer segments that are less well organized . 10 . 7554/eLife . 16962 . 020Video 2 . Rotation of reconstructed sensory structures . Reconstructed pigment cells ( black ) with otolith associated ciliated cells ( yellow ) and vacuoles observed in a variety of cell types ( lime green ) . Outer segments reconstructed as spheres for group I photoreceptors projecting into the ocellus ( darker purple ) and group II photoreceptors projecting into the canal ( lighter purple ) and group III modified outer segments ( blue ) , as well as coronet bulbous protrusions ( orange ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 02010 . 7554/eLife . 16962 . 021Video 3 . Rotation of reconstructed sensory neurons . Reconstruction including transparent cell bodies illustrating pigment cells ( black ) , group I ( dark purple ) , group II ( light purple ) and group III ( blue ) photoreceptors with their outer segments reconstructed as spheres , lens cells ( white ) with their vacuoles reconstructed ( bright green ) ; also shown are vacuolated photoreceptor tract interneurons ( lime ) , antenna cells ( green ) , otolith associated ciliated cells ( yellow ) , and coronet bulbous protrusions . Terminals of antenna and photoreceptor neurons are truncated in this view . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 02110 . 7554/eLife . 16962 . 022Video 4 . Reconstruction of spheroids representing the cell body positions of sensory structures . Pigment , photoreceptor and coronet cells with the bulbous protrusions ( green ) and photoreceptor outer segments ( type I and type II: yellow; and type III: purple ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 022 In addition to receptor neurons , interneurons of the brain vesicle also exhibit sidedness in the position of their somata ( Figure 7 ) . 10 . 7554/eLife . 16962 . 023Figure 7 . Representation and relative sizes of cell bodies and their positions along the neuraxis , with corresponding axon tracts . ( A ) Cell bodies of CNS neurons , dorsal view . Colours denote cell types ( key ) . ( B ) Corresponding axon tracts , shown as skeleton reconstruction , dorsal view , colours as in ( A ) ( for a network graph of synaptic connectome formed by corresponding neurons sorted by connectivity see Figure 7—figure supplement 1 ) . ( C ) Cell bodies of CNS neurons and axon tracts , corresponding to ( A ) and ( B ) , left lateral view . Pigment oc/ot: ocellus and otolith pigment cells; PR ( oc ) : type I photoreceptor; PR ( can ) : type II photoreceptor; PNIN: peripheral interneuron; PNIN ( cilia ) peripheral interneuron with cilium; vac IN: vacuolated photoreceptor-associated interneuron; Antenna: antenna cell; Coronet: coronet cell; aaIN: anaxonal arborizing interneuron; BVlN ( cilia ) : ciliated brain vesicle interneuron: pr-AMG RN: photoreceptor-AMG relay neuron; trIN: photoreceptor tract interneuron; cor-ass BVIN: ciliated coronet associated brain vesicle interneuron; prRN: photoreceptor relay neuron; BVIN ( no cilia ) : brain vesicle interneuron lacking cilium; non-sensory relay neuron ( RN ) ; antRN: antenna relay neuron; ant-corRN: antenna-coronet relay neuron: Eminens: eminens neuron: PNRN: peripheral relay neuron: PBV PNIN: posterior brain vesicle peripheral interneuron; MGINs 1–2: motor ganglion paired interneurons 1 and 2; MGINs 3: motor ganglion paired interneurons 3; ddN: descending decussating neuron pair: AMG: ascending motor ganglion neuron; MN: motor neuron: ACIN: ascending contralateral inhibitory neuron; PMGN: posterior motor ganglion neuron; Midtail neurons: short descending neurons of the caudal nerve cord . Scale bar 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 02310 . 7554/eLife . 16962 . 024Figure 7—figure supplement 1 . Total network of synaptic pathways within the larval CNS of Ciona intestinalis . Network graph of all connections within the larval nervous system generated in Cytoscape . Line width indicates synaptic strength ( key , right ) and arrows indicate direction of synaptic connection . Cells colour-coded by connectivity class ( key ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 024 Right-side interneurons can be identified by various morphological features not resolved in previous light microscopy studies . Two classes of intrinsic interneurons are: Posterior to these intrinsic neurons are three additional classes of interneuron: Most interneurons of the brain vesicle are however located on the left side and many are structurally anonymous . Aside from those mentioned above ( a-c ) , photoreceptor and antenna relay neurons are left-sided . In addition , other relay interneurons that lack direct input from sensory neurons are also left-sided ( Table 3 ) . The only apparent difference in photoreceptor input to left and right subclasses of relay neurons is that Type II canal photoreceptors are presynaptic only to right-side relay neurons of the pr-AMG class ( Figure 9 ) . Unlike their right-side counterparts PR-AMGRN ( R ) ( cells 108 , 116 , 127 , 157 , 123 , and 130 ) , the left-side photoreceptor relay neurons PRRN ( L ) ( cells 74 , 94 , 80 , 86 , 96 , 100 , 121 , and 126 ) receive photoreceptor input exclusively from Type I ocellus photoreceptors ( Figure 9 ) . Two ventral antenna neurons ( Ant1 and Ant2 ) , which are proposed to signal input from otolith position ( Torrence , 1986; Tsuda et al . , 2003 ) , lack obvious sidedness in their cell body positions , and extend collateral axons toward the posterior brain vesicle ( BV ) . The terminals of these antenna cells , however , differ in their synaptic input to sides of the posterior BV: Antenna cell 1 to both sides , and Antenna cell 2 predominantly to right-side relay neurons ( AntRN ) . One right-side antenna relay neuron , located more ventrally ( Ant-corRN; Figure 1—source data 1 ) , is also postsynaptic to coronet cells . 10 . 7554/eLife . 16962 . 026Figure 9 . Asymmetrical sensory input to the two sides of the motor ganglion MG ( L ) and MG ( R ) via relay neurons . Sensory input arises from coronet cells ( Cor ) ; antenna cells Ant1 and Ant2 ( combined as Ant ) ; and photoreceptors ( PR ) of two types: ocelli ( oc: PR I ) and neural canal ( can: PR II ) . Signals are relayed through respective interneuron classes: photoreceptor relay neurons on the left , PRRN ( L ) ; photoreceptor-ascending motor ganglion ( PR-AMGRN ( R ) ) relay neurons on the right ( Figure 8c ) ; and antenna relay neurons ( AntRN ) of the left and right sides . PR-AMG relay neurons of the right side receive input from ascending motor ganglion neurons that is reciprocated . Pathways with weak connections are shown with dashed lines . Details of pathway strength appear in Figure 9—figure supplement 1DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 02610 . 7554/eLife . 16962 . 027Figure 9—figure supplement 1 . Total network of synaptic pathways within the larval CNS of Ciona intestinalis . Network graph of all connections between cells of the larval nervous system grouped by cell type ( cf , Figure 7—figure supplement 1 ) . Line width indicates the total synaptic number ( key , right ) and arrows indicate direction of synaptic connection . Cells colour-coded by connectivity class ( key: Figure 1—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 027 Compared with the brain vesicle , the motor ganglion has paired neurons and thus appears more bilaterally symmetrical ( Figure 7; Figure 10—figure supplement 1 ) , but nevertheless receives asymmetrical inputs from the relay neurons of the brain vesicle ( Figure 10 ) . These project mostly to the motor ganglion’s interneurons , however , although some sparse connections are made directly to the motor neurons themselves ( Figure 10—figure supplement 1 ) . The shortest sensory pathway to any motor neuron connects via a brain vesicle interneuron , and is thus disynaptic , although most direct pathways involve two interneurons and are thus trisynaptic ( Figure 11 ) . However , these shortest paths fail to depict the complexity of integration revealed in the total network ( Figure 7—figure supplement 1; Figure 9—figure supplement 1 ) . 10 . 7554/eLife . 16962 . 028Figure 10 . Classes of relay neurons ( presynaptic ) in the CNS of Ciona and the inputs these provide to cells on the left and right sides of the motor ganglion ( for details of relay inputs see Figure 10—figure supplement 1 and for antenna pathway see Figure 10—figure supplement 2 ) . For relay neuron class names see the key in Figure 1—source data 1 ) . Each circle represents the input synapses to the first ( column 1 ) , second ( column 2 ) or both ( column 3 ) paired MG interneurons . Inputs to the left ( blue ) or right ( red ) partners are shown as an angular subtense of a circle the area of which represents the overall synaptic strength . Most sensory inputs are predominantly one-sided , some entirely so . Total synaptic input varies widely ( see Figure 10—source data 1 for actual values and proportions ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 02810 . 7554/eLife . 16962 . 029Figure 10—source data 1 . Relay neuron inputs to the left and right motor ganglion . Values refer to the total number of synapses and their proportions of the whole population . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 02910 . 7554/eLife . 16962 . 030Figure 10—figure supplement 1 . Reconstructions of motor ganglion neurons populated with photoreceptor and antenna relay neuron synaptic input sites , colour-coded by relay neuron type ( key ) . Synaptic sites are marked by 4 nm spheres regardless of their actual size . prRN: photoreceptor relay neuron; pr/trINRN: relay neurons with input from photoreceptors and photoreceptor tract interneuron; pr/AMGRN: relay neurons with inputs from photoreceptor and peripheral pathway; pr/cor/antRN: relay neuron with inputs from photoreceptor , coronet cells , and antenna 2 neuron; pr/ant2RN: relay neurons with inputs from photoreceptors and antenna 2 neuron; ant2RN: relay neurons with inputs from antenna 2 neuron; ant1/2RN: relay neurons with input from both antenna neurons . Anterior to the left . ( A ) Dorsal view of all target neurons in the MG . ( B ) Target motor ganglion neuron pairs in ( A ) populated with postsynaptic sites received from relay neurons . Each neuron is rotated to show synaptic sites along its inside edge . Scale bars: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 03010 . 7554/eLife . 16962 . 031Figure 10—figure supplement 2 . Antenna cell relay neuron input to the motor ganglion . ( A ) Network of antenna cell relay neuron synaptic connections to components of the motor ganglion . Cells are colour-coded , and synapse strength is denoted by line thicknesses ( key , right ) and varies over a >50 fold range in cumulative synaptic depth . ( B ) Summarizing ( A ) : Antenna cells ( 1 and 2 ) provide input to three classes of relay neurons ( a , b and c ) represented in Figure ( A ) that provide input in turn to left- and right-side motor ganglion cells . Both antenna cell 1 and 2 networks form feedforward triplet motifs onto relay neuron classes ( a to c ) which themselves form symmetrical feedforward motifs . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 03110 . 7554/eLife . 16962 . 032Figure 11 . The shortest CNS pathways between sensory neurons and motor neurons for different sensory modalities are three-synapse arcs . Four modalities are indicated , from top to bottom: light , gravity , coronet cells ( possibly hydrostatic pressure ) and PNS mechano/chemosensory . Members of the same cell types are assigned the same colour . Each pathway originates in the particular class of sensory neuron ( photoreceptor: PR; antenna neuron: Ant; coronet cell , Cor; and PNS sensory neuron , pns ) connecting via relay neurons ( RN ) and interneurons ( IN ) of the brain vesicle , to interneurons ( IN ) of the motor ganglion , to motor neurons ( MNs ) . AMG: addition relay neurons of the motor ganglion receive input from the PNS . These pathways are all interconnected , overlapping networks for all sensory modalities thus underlying the complex reality of the actual behavioral network ( Figure 9—figure supplement 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 032 The relay neurons of the photoreceptor and antennal pathways project asymmetrically ( Figure 10—source data 1; Figure 10—figure supplement 2 ) , photoreceptor interneurons connecting 70% to left-side interneurons , and antenna interneurons connecting 70% to right-side interneurons , of the motor ganglion ( Figure 10; Figure 10—source data 1 ) . The asymmetry is greatest among relay interneurons that receive additional input from coronet cells . Those that receive photoreceptor combined with coronet cell input project entirely to the left , and those that receive antennal and coronet cell input project entirely to the right . The asymmetry in pathways to motor ganglion interneurons parallels that of the direct input to motor neurons ( photoreceptor pathways connecting to the left and antenna to the right ) ( Figure 10; Figure 10—source data 1 ) . The five pairs of motor neurons ( MN1-MN5 ) form morphologically distinct neuromuscular junctions . Two anterior primary motor neurons ( MN1 and MN2 ) each form 200–360 large synapses with many vesicles and postsynaptic densities on the muscle ( Figure 12A ) , whereas the three more posterior motor neurons ( MN3-MN5 ) each form 15–50 smaller neuromuscular junctions , often with fewer vesicles and less obvious muscle postsynaptic specializations ( Figure 12B ) . The input to left and right muscle at these junctions is asymmetrical from all motor neuron pairs except the MN2 neuron pair , which provides symmetrical synaptic input in both number ( 46% to left and 54% to right ) and total size ( 49% to left and 51% to right ) . Asymmetries in synapses observed between left and right muscle are not great , however , falling within a left:right ratio range between 60:40 and 40:60 , MN1 having greater input to the right , and MN3-MN5 greater to the left ( Table 4 ) . 10 . 7554/eLife . 16962 . 033Figure 12 . Ciona intestinalis larval motor neuron terminals and neuromuscular junctions . ( A ) Neuromuscular junction ( arrow ) of MN1 , with a postsynaptic specialization on the muscle ( arrowheads ) . A basal lamina ( red arrow ) extends in the cleft between neuron and muscle . ( B ) Two adjacent neuromuscular synapses ( arrows ) with postsynaptic cisternae ( arrowheads ) , but lacking postsynaptic membrane densities . ( C , D ) Enlarged views of anterior tail with reconstructed puncta representing colour-coded neuromuscular junctions of each motor neuron pair . ( D ) Top: Dorsal view . Bottom: Left lateral view . Scale bars: 1 µm ( A , B ) ; 10 µm ( C , D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 03310 . 7554/eLife . 16962 . 034Table 4 . Input to left and right dorsal and medial muscle bands from motor neuron pairs at their neuromuscular junctions . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 034Motor neuron pair Left muscle band Right muscle band Ratios Dorsal Medial Dorsal Medial Left: Right No . syn No . sec No . syn No . sec No . syn No . sec No . syn No . sec No . syn No . sec MN1 19296947145230118113055840: 6039: 61MN2 2241583258163646: 5449: 51MN3 421562810160: 4061: 39MN4 451893011660: 4062: 38MN5 21128155558: 4270: 30Number of synapses ( No . syn ) and number of synaptic profiles ( No . sec ) provided for each motor neuron and left:right ratios expressed as percentages of neuromuscular junction input from left and right partners for each motor neuron pair . Motor neurons also form synapses with each other , with some further asymmetries between connections on the left and right sides ( Figure 13A ) . On the right side only , MN1 is presynaptic to MN4 and postsynaptic to MN3 . MN2 also shows asymmetries , receiving input from MN3 on the right , but providing input to MN3 of the left , and receiving feedback input from MN5 just on the left side . These synaptic asymmetries contrast with the obvious symmetry of the network of gap junctions between the motor neurons ( Figure 13A–D ) . 10 . 7554/eLife . 16962 . 035Figure 13 . The networks of motor neurons MN1-MN5 and descending ipsilateral neurons ( MG1-MG3 ) of the left and right side of the motor ganglion ( MG ) . ( A ) Synaptic network of motor neurons 1–5 on the left ( MN1-5L ) and right ( MN1-5R ) sides . ( B ) Network of putative gap junctions between motor neurons of the MG . ( C ) Summary diagrams of motor neuron synaptic networks of left and right sides . ( D ) Summary diagrams of putative gap junction network of motor neurons of the left and right sides . Dotted lines represent tentative connections , dashed lines minimal contacts , and solid line connections with many large contact sites . ( E ) Synaptic network of descending ipsilateral interneurons ( MGINs ) of the motor ganglion . ( F ) Network of putative gap junctions between descending ipsilateral interneurons of the motor ganglion . For a , b , e and f: Arrows illustrate polarity of synapse , line thickness show the cumulative depth of synaptic contact in µm ( see Materials and methods ) . Red lines illustrate synaptic contacts that differ between left and right sides of the MG . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 035 The synaptic strengths ( see Materials and methods ) of the interneuron network on left and right sides are also asymmetrical ( Figure 13E ) . In particular , the strength of the synaptic input from the first interneuron ( cell MGIN1 ) to the second ( cell MGIN2 ) is far greater on the left side . This asymmetry is paralleled by the gap junction network , in which the second interneuron is connected to the first only on the left side . In addition , this second left interneuron ( MGIN2L ) is also connected to the contralateral first right interneuron ( cell MGIN1R ) by gap junctions ( Figure 13F ) . Sidedness in the neural network of the motor ganglion is also evident in the synaptic connections between interneurons and motor neurons ( Figure 14 ) . The first motor neuron receives more inputs on the right side from both the first and third interneuron ( cells MGIN1 and MGIN3 ) than it does on the left ( synaptic depths 1 . 08 and 1 . 2 µm versus 0 . 42 and 0 . 72 µm ) . The second motor neuron ( MN2 ) likewise receives more inputs from the second interneuron on the right ( MGIN2R ) than the left ( MGIN2L ) side ( synaptic depths 4 . 69 versus 1 . 24 µm ) . Other asymmetries are apparent in connections between cells that are reciprocal on only one side: MN1L to MGIN3L , MN3L to MGIN2L , on the left , and MN5R to MGIN1R , MN1R to MGIN2R , and MGIN2R to MN4R on the right ( Figure 14A ) . These pathways refer to chemical synapses ( Figure 14A ) , relative to which those exhibited by gap junctions ( Figure 14B ) are more left/right symmetrical and involve more connected partners across the midline , especially for the anterior components . 10 . 7554/eLife . 16962 . 036Figure 14 . Left-right asymmetries in the overall synaptic pathways of the motor ganglion . Pathways shown are between motor neurons ( MNs ) , descending ipsilateral interneurons ( MGINs ) and descending mid-tail neurons . ( A ) Synaptic network with arrows indicating polarity of synaptic contacts . ( B ) Summary network of gap junctions for descending motor ganglion neurons illustrated in ( A ) . Blue lines represent gap junction inputs from motor neurons , green represent gap junction inputs from interneurons , and pale blue are gap junction inputs from mid-tail neurons . Pathway strength varies over a wide ( >25 times ) range and is more left-right symmetrical than the synaptic network in ( A ) . In both ( A ) and ( B ) pathways shown by orange lines are left/right asymmetrical , and those in pink are present only on one side ( key ) . Thickness of lines indicates cumulative depth of synaptic contacts ( see Materials and methods ) ( scale ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 036 Aside from the obvious asymmetry in their number , ACINs are also asymmetric in their projections and connections between left and right sides of the posterior motor ganglion ( Video 5 ) . On the right , although the ACIN crosses the midline , it does not pass the ependymal cell to extend into the right neuropil , and so forms no contralateral synapses . On the left , however , both ACINs decussate fully , crossing to the contralateral neuropil and forming synapses there . These synapses are as a result not formed directly onto contralateral motor neurons , as previously proposed ( Horie et al . , 2010 ) , but instead onto interneurons ( cells MGIN1R , MGIN2R and MGIN3R ) of the right side ( Figure 15A ) . Both left and right ACINs are presynaptic to their ipsilateral motor neurons ( Figure 15B ) and first two pairs of interneurons . All three ACINs are also presynaptic to a right-side bipolar tail neuron ( cell BTN2 , reported in Stolfi et al . , 2015 ) , and additionally form multiple synaptic contacts directly onto the ventral basal lamina ( Figure 15C ) . That input resembles the synaptic input from the motor neuron terminals , which form onto the basal lamina that ensheathes the muscle ( Sanes et al . , 1978 ) . 10 . 7554/eLife . 16962 . 037Figure 15 . ACIN synapses and network . ( A ) Presynaptic site ( arrow ) from the left ACIN onto contralateral MGIN interneurons at a dyad synapse . BTN2: bipolar tail neuron profile . ( B ) Dyad synapse ( arrow ) onto ipsilateral motor neuron MN3R and an unpaired tail interneuron ( PMGN2 ) on the right side . ( C ) Synapse ( arrow ) from ACIN1L onto the ventral basal lamina ( BL ) opposite the notochord . Scale bar ( a-c ) : 1 µm . ( D ) Network diagram of ACIN pathways . Layout plotted as an edge-weighted spring embedded network ( Cytoscape 3 . 1 . 0: NRNB . org ) based on synapse pathway strengths ( see Materials and methods ) . Right and left neuropiles are each enclosed in a dashed line . Pathway strengths are shown as the line thickness sorted by the cumulative depth of synaptic profiles ( key ) . The right side includes two sided PMGN interneurons and their partners . Note reciprocity of connections for ipsilateral but not contralateral partners . Thus ACINs are presynaptic to contralateral partners but not postsynaptic . Cell types abbreviated as in Figure 1—source data 1 ( E ) Dorsal view of reconstructed ACINs . Scale bar 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 03710 . 7554/eLife . 16962 . 038Video 5 . Rotations of reconstructed ACINs decorated with their presynaptic sites . Reconstructed ACINs with presynaptic sites colour-coded by postsynaptic cell type: basal lamina ( black ) ; motor neuron ( blue ) , descending MG interneuron ( green ) , bipolar tail neuron ( red ) , and posterior MG descending neuron ( brown ) DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 038 Presynaptic input to ACINs is not only asymmetric , but differs between each of the three ACIN cells . All are postsynaptic , however , to at least one ipsilateral interneuron and one ipsilateral motor neuron ( Figure 15D ) . Both left ACINs receive input from MN3L and MGIN2L , but ACIN1L receives additional input from MN1L and a descending decussating neuron ( cell ddR ) , while ACIN2L receives input from MGIN1L and the third a contralateral interneuron ( MGIN3R ) . On the right , the only symmetrical input partner is the ipsilateral first interneuron , MGIN1 ( R ) , with asymmetrical input from MN2R as well as right-side bipolar tail neurons . Summarizing: the connectivity matrix of all neurons reveals the following features ( Figure 16 ) : ( a ) Most cell types form synapses among members of the same class of neuron . ( b ) Many synapses also form on the basal lamina and from motor neurons onto muscle cells . ( c ) A high degree of reciprocity is manifest between members of different neuron classes , especially those that are interneurons . ( d ) Neurons are both pre- and postsynaptic , those that have many presynaptic sites generally also have many that are postsynaptic , except for sensory neurons in which presynaptic sites predominate . ( e ) The pathway strength is in general greater in the motor ganglion but also high for the relay neurons and among neurons of the pathway from the peripheral neurons . ( f ) Cell classes with fewer representative neurons tend to have greater pathway strengths . Features of the sidedness of the matrix are reported in Figure 16—figure supplement 110 . 7554/eLife . 16962 . 039Figure 16 . Entire connectivity matrix for the complete brain of a larva of Ciona intestinalis . Shown for all synapses are the pre- ( rows ) and post- ( columns ) synaptic cells , colour-coded by cell type ( see Figure 1—source data 1 ) and arranged in their rostro-caudal sequence along the longitudinal axis ( presented in Figure 16—figure supplement 1 is the same matrix with cells sorted into left and right sides ) . Each intercept is colour-coded for the cumulative depth of presynaptic contacts made by that neuron upon its postsynaptic partner ( key , bottom ) . In the case of dyads or triads , all connections are plotted . Also included are muscle cells , and the basal lamina of the CNS , both of which are exclusively postsynaptic . Other cell types , particularly ependymal cells lacking axons , are excluded . Muscle cells of the dorsal and medial bands are pooled on each side , because these are connected via gap junctions ( Bone , 1992 ) . The matrix is bounded by nested boxes between specific cell types . The smaller boxes enclosed by dashed lines indicate the connections between neurons of the same subtype . These are enclosed within boxes bounded by coloured lines , which indicate connections between neurons of the same brain region . Neurons of the brain vesicle are segregated into sensory neurons ( orange lines , Sensory ) , intrinsic interneurons ( pink lines , BVIN ) and relay neurons ( green lines , RN ) . Remaining boxes are as follows , neurons of the: PNS ( PNS ) ; motor ganglion ( MG ) ; and tail neurons of the caudal nerve cord ( CNC ) . For matrix file see ( Figure 16—source data 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 03910 . 7554/eLife . 16962 . 040Figure 16—source data 1 . Matrix in Figure 16 1 as excel file . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 04010 . 7554/eLife . 16962 . 041Figure 16—source data 2 . Matrix in Figure 16—figure supplement 2 as excel file . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 04110 . 7554/eLife . 16962 . 042Figure 16—figure supplement 1 . Matrix of connections from Figure 16 sorted by left and right sides . Cell types and connection strength are both coded by colour as in Figure 16 Rows and columns are sorted starting with cells connecting from left ( L ) to right ( R ) , left to left , right to left and right to right . Midline ( M ) cells separate these quadrants . Remaining cell types ( notochord and basal lamina , bm ) bound the four quadrants . Muscle cells presented on their respective sides and pooled as in Figure 16DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 04210 . 7554/eLife . 16962 . 043Figure 16—figure supplement 2 . Entire matrix of putative gap junctions for the complete brain of a larva of Ciona intestinalis . Shown for all neuron partners with a cumulative membrane contact depth of >0 . 12 µm , colour-coded by cell type and arranged in their rostro-caudal sequence along the longitudinal axis . Each intercept is colour-coded for the cumulative membrane contact depth of contacts made by that neuron upon its partner ( key , bottom ) . The matrix is bounded by nested boxes between specific cell types . The smaller boxes enclosed by dashed lines indicate the connections between neurons of the same subtype . These are enclosed within boxes bounded by coloured lines , which indicate connections between neurons of the same brain region . Neurons of the brain vesicle are segregated into sensory neurons ( orange lines , Sensory ) , intrinsic interneurons ( pink lines , BVIN ) and relay neurons ( green lines , RN ) . Remaining boxes are as follows , neurons of the: PNS ( PNS ) ; motor ganglion ( MG ) ; and tail neurons of the caudal nerve cord ( CNC ) . Matrix file provided in Figure 16—source data 2DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 043 We report the full synaptic connectome of a single tadpole larva of a model chordate species , the ascidian Ciona intestinalis , and use this to identify the complete inventory of the many asymmetrical features in its CNS . Some connectomic analyses rely on symmetry of the connections between left and right sides to validate the synaptic connections of neurons that are paired ( Durbin , 1987; Randel et al . , 2014; Ohyama et al . , 2015 ) . This approach has not been possible in Ciona , because many cells are not bilaterally paired in the brain vesicle , while in the motor ganglion where cells are paired , presynaptic inputs from the brain vesicle are similarly asymmetrical . Additionally , we cannot exclude the possibility that minor left/right asymmetries might be the product of developmental noise or imprecision , insofar as only a small number of larvae need to survive and eventually become adults . Our findings reveal not only those features with possible counterparts in the vertebrate CNS , but also the many features of all nervous systems . These include the wide range and combination of cells that share synaptic contacts , the poorly segregated distribution of synapses over the neuron surface; synapses onto non-neuronal cells and basal lamina; unpolarized and mixed vesicle synapses resembling those in cnidarians ( Westfall , 1996 ) ; and the apparent general redundancy in the connectome . Unpolarized synapses have been previously reported between coelenterate ( Horridge and Mackay , 1962 ) and pulmonate mollusc ( McCarrager and Chase , 1985 ) neurons , while synapses onto the basal lamina have been reported in muscle after removing the underlying myofibres ( Sanes et al . , 1978 ) . Synaptic sites that occur onto the basal lamina , and that therefore lack postsynaptic partners , resemble other presynaptic sites in the CNS , and resemble neuromuscular junctions , where neurotransmitter must cross the basal lamina to act on adjacent muscle cells . Some presumed sites of synaptic release across the basal lamina lie opposite non-neuronal cells , including epidermal cells or cells of the notochord . Neurotransmitter receptor expression studies are critical to our understanding of the roles these special sites of synaptic vesicle release may play . Sites that lack adjacent cells may be sites of neuromodulator release . This is particularly likely in the case of coronet cells , which express reporters for dopamine and have exclusively dense-core vesicles at their synapses onto the basal lamina . Synaptic reciprocity and serial networks are commonplace in many nervous systems ( e . g . Dowling and Boycott , 1966 ) . Aided by the numerical simplicity of the Ciona CNS , we are also able to detect features such as cilia that may lie undetected in highly populated vertebrate brains , and confirm that many neurons in Ciona are ciliated ( Figure 3—figure supplement 1 ) . An obvious point of comparison is set by the nervous system of C . elegans , the sole precedent for a completely reported connectome , with which the Ciona larva is dimensionally comparable and with which it has a numerically similar network -- 302 identified neurons in the nervous system of a single hermaphrodite C . elegans ( White et al . , 1986; Varshney et al . , 2011 ) compared with 177 neurons in the larval CNS of Ciona . Neurons form comparable numbers of synapses in both these model nervous systems -- an average of 37 presynaptic sites and 7 putative gap junctions ( >1 section ) compared with 28 synapses and 3 . 2 gap junctions per non-pharyngeal neuron in C . elegans ( calculated from data at http://www . wormatlas . org/hermaphrodite/nervous/Neuroframeset . html ) . These numbers are smaller than in Drosophila , in which optic lobe neurons may have in excess of 100 presynaptic sites ( Meinertzhagen and Sorra , 2001; Takemura et al . , 2008 , 2015 ) and clearly less than typical vertebrate neurons ( for example a mouse somatosensory cortex neuron has about 8200 synapses: Schüz and Palm , 1989 ) . Further discussion of this topic is provided elsewhere ( Meinertzhagen , 2010 ) . For brain asymmetry to appear in the hatched larva , the expression of Nodal and ciliary action are both required during embryonic development ( Nishide et al . , 2012; Thompson et al . , 2012 ) . These events are perturbed during the process of dechorionation that has been used for many experimental interventions , especially electroporation of genetic reagents ( Shimeld and Levin , 2006 ) , so that many asymmetries may have escaped detection in previous reports that are revealed in our larva reared with an intact chorion . Furthermore , within the chorion the developing embryo invariably curls around itself along the left side of the trunk ( Katsumoto et al . , 2013 ) , and this may further influence sidedness in the larval brain , in ways that are lost after dechorionation . In addition to pigment cell displacement , arrestin expression indicating photoreceptor cell fate , is significantly altered , often expanding to the left brain in dechorionated embryos ( Oonuma et al . , 2016 ) . The overall asymmetry of the larval ascidian CNS finds deep parallels with asymmetries in other chordates , including vertebrates ( Boorman and Shimeld , 2002b ) , and clear differences from those of other deuterostomes , such as echinoderms ( Duboc et al . , 2005 ) . Most difficulties in comparing between ascidian larval and vertebrate nervous systems come however from differences in their respective cell numbers . The extreme miniaturization of the former exposes sidedness in the ascidian larval brain that can actually exist in both chordate clades , such as in the vertebrate epithalamus ( Concha and Wilson , 2001; Hamada et al . , 2002 ) . Structurally most obvious in the ascidian larva is sidedness in the position of the ocellus , which is driven by the same left-side action of Nodal ( Yoshida and Saiga , 2011 ) as drives asymmetry in all chordate brains , including those of vertebrates ( Halpern , 2003; Carl et al . , 2007 ) . Migration of the ocellus pigment driven in Ciona by Nodal ( Yoshida and Saiga , 2011 ) calls to mind the vertebrate pineal , which is also lateralized based on Nodal’s action ( Carl et al . , 2007 ) . Nodal acts on the left side of the CNS in both , but in Ciona the pigment cells migrate to the right , whereas vertebrate parapineal cells migrate to the left . Moreover , based on the projection of their outer segments , which is inward , not toward the outside world , Ciona’s larval eyes are more akin to lateral eyes than to a single pineal ( Lamb , 2013 ) . The photoreceptor cells of the ocellus likely share a common lineage , blastomeres A9 . 14 and A9 . 16 on the right side ( Oonuma et al . , 2016 ) , with the coronet cells of the left side ( Cole and Meinertzhagen , 2004 ) . Ascidian coronet cells are likewise components of a morphologically homologous structure that is bilateral in vertebrates , the saccus vasculosus ( Smeets et al . , 1983 ) . In vertebrates , the saccus vasculosus comprises coronet cells in addition to neurons contacting the cerebrospinal fluid . In Ciona we report new ciliated coronet-associated neurons , with cilia that project into the neural canal toward the coronet cells’ bulbous protrusions . In both vertebrate and ascidian cases , it is these neurons rather than the coronet cells that give rise to an axonal pathway , forming the axon tract emanating from the saccus vasculosus in vertebrates ( Rodríguez-Moldes and Anadón , 1988 ) and the coronet complex in Ciona , further strengthening the similarities between coronet complex and saccus vasculosus . Both ocellus and coronet complex structures are lateralized in Ciona , unlike the single medial structures reported for cyclopic mutants ( Belloni et al . , 1996; Chiang et al . , 1996 ) or fused , unpaired saccus vasculosus phenotypes ( Nieuwenhuys , 1998 ) . Given that ascidian larvae swim in a helical pattern ( McHenry , 2005 ) and have a single-sided ocellus , their phototactic behaviour follows a helical , not visual pattern , as defined by Randel and Jékely ( 2016 ) . Unlike simple helical phototaxis , however , the complement of cell types and complexity of component connections involved in Ciona’s visual circuit ( Video 6 ) compare much more closely to circuits responsible for visual phototaxis than those for simple helical phototaxis ( Randel and Jékely , 2016: their Figure 2f , g ) . Helical swimming of the ascidian larva provides a mechanism by which a preexisting bilateral visual phototaxis circuit could have been co-opted into a complex hybrid helical phototaxis circuit , still allowing mechanisms such as delay , sensory integration , and modulation to take place , and unlike the more direct helical phototaxis mechanisms in ciliated forms , such as protists and trochophore larvae ( Randel and Jékely , 2016 ) . 10 . 7554/eLife . 16962 . 044Video 6 . Animated reconstruction of the photoreceptor pathway . Cell bodies , shown as spheroids , from photoreceptor through relay neurons to the motor ganglion . Cells are colour-coded as in Figure 1—source data 1DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 044 The presence of unilateral systems in the larval CNS of Ciona that show homology with bilateral structures in the vertebrate brain can be interpreted as a reduction in one side , and as an outcome of the small cell numbers in ascidian larvae . A defining feature of larval Ciona and its CNS is indeed the miniaturization of both . The larva’s small cell numbers can be seen as a direct outcome of the few embryonic cleavage generations , no more than 14 for the entire CNS ( Cole and Meinertzhagen , 2004 ) , the lack of metamerism ( Garstang , 1928; Crowther and Whittaker , 1994 ) , the lack of feeding in this lecithotrophic larvae , and consequently its short life . The small number of its component neurons is reflected in turn in the relatively small numbers of synapses formed by each , conforming to an overall relationship between neuron number and synapse number per neuron seen in nervous systems generally . In contrast to both is the richness of cell types . Among the 177 larval neurons we can distinguish at least 25 different types and 52 subtypes identified on the basis of morphological and connectivity differences . C . elegans has a numerically comparable richness , with 118 cell types among its 302 neurons ( White et al . , 1986 ) . The asymmetries we observe start to appear with the failure of cells to pair left and right , at around 75% of embryonic development ( Cole and Meinertzhagen , 2004 ) . After this the brain vesicle pushes to the right between 75% and 85% of embryonic development , and cell positions begin to shift , followed by the loss of strict bilateral symmetry among the cells of the motor ganglion . The pattern of cell lineage is left/right symmetrical until the 11th cleavage , with ventral divisions becoming desynchronized after the 10th cleavage ( Cole and Meinertzhagen , 2004 ) . The photoreceptor pigment cell begins its migration at the 11th generation . The ACINs appear after the 11th embryonic cleavage ( Nishitsuji et al . , 2012 ) and fate choices select cells as either neurons or ependymal cells at around this stage too ( Cole and Meinertzhagen , 2004 ) . The asymmetry in position within the motor ganglion and between the cell numbers of the two sides of the caudal nerve cord is apparent by the final stages of embryonic development . Total cell numbers are mostly left-right symmetrical throughout the CNS , and sidedness is mostly a question of cell fate . Thus the fate choice , for example between neuron and ependymal cell type , generates the sidedness in the numbers of cells comprising the same class . A striking example is provided by the ACIN neurons in the rostral caudal nerve cord . Thus A11 . 116 undergoes two divisions , first to yield A12 . 231 and A12 . 232 ( Nishitsuji et al . , 2012: their Figure 5B ) , and a further round of divisions , yielding a total of four 13th-generation cells that become two ACIN and two ependymal cells . The fate decision to generate ependymal and ACIN cells is then proposed to occur after this final division , based on the expression of extracellular signals and transcription factors . The total number of ACINs varies between larvae ( Nishino et al . , 2010 ) , and in the larva we report , this variation manifests itself as an asymmetry , in which we observe two progeny of the left side as ACINs , with only a single ACIN present on the right , the other site being occupied by ependymal cell tail 4 . Thus variation in cell fate decision is , we propose , the basis for the left/right asymmetry in ACIN neurons . We also find neurons on the right side ( PMGN1 and PMGN2 ) of a hitherto unreported type that are located anterior to the single ACIN ( ACIN2L ) of that side , but posterior to motor neuron pair MN5 . Their asymmetrical location suggests they may represent similar late choices between neuronal and ependymal cell fate . Together with these asymmetries in cell types , there are clear asymmetries in connectivity . These are most obvious in inputs to , and connections within , the motor ganglion . Similar asymmetries may exist in other motor systems but are revealed only from comparisons between the left and right sides of a complete network , and so have rarely been revealed . In the mouse , however , an imbalance index for the motor innervation of interscutularis muscles reveals an asymmetry in the morphological features of motor innervation ( Lu et al . , 2009 ) , while in the polychaete Platynereis there is also an asymmetric connectivity pattern for one class of motor neuron , both for the inputs it provides to the muscle it innervates and the inputs it provides to contralateral neuron partners ( Randel et al . , 2014 , 2015 ) . These instances compare to both the asymmetrical input of , for example , MN1 to muscle and other network asymmetries we find in Ciona in which , like Platynereis , an input from two ACINs on the left side of the brain to interneurons on the right side is not matched by a contralateral input from the ACIN on the right side . Our conclusions are based on a single larva , although four unrelated sibling larvae have been reported with closely similar overall cell complements ( Nicol and Meinertzhagen , 1991 ) . Data are still lacking on the constancy of neuron cell types and their connections between individuals . The consistency of asymmetrical differences we see in this larva will only be resolved when its sibling larvae are examined , as we now undertake . Mechanosensory drive of symmetrical swimming was initially dismissed because tails can retain swimming independent of the head , and no known input from mechanosensory cells to motor neurons was known . However , we find multiple inputs from mechanosensory tail neurons’ interneurons to motor neurons , both from AMGs in the dorsal MG , and as reported from the bipolar tail neurons ( Stolfi et al . , 2015 ) . It is not yet clear whether the cilia of DCENs or VCENs may alternatively have thermoreceptive , electroreceptive , or chemoreceptive properties , although there is evidence for chemotactic behavior during larval settlement , mediated by epidermal sensory neurons . A proposed central pattern generator ( CPG ) for Ciona has been compared ( Horie et al . , 2010 ) to the vertebrate swimming CPG of the lamprey ( Grillner and Wallén , 1999 ) but in Ciona omits the role of excitatory interneurons in general and ipsilateral connections within the motor network in particular . Moreover the depicted network is left-right symmetrical and with proposed direct contralateral input to motor neurons , features we now show to be lacking in Ciona’s CNS connectome . This difference highlights the power of a complete connectome to reveal the actual connections between identified neurons . We find the left-right difference of the ACIN connections , in particular , to be most surprising , and endorsed by the failure of the right ACIN to send a neurite into the left neuropile even though it crosses the midline . Additional features of the network’s right side also support this asymmetry: the lack of ACIN1 on the right side , and the two additional right-side interneurons with neurites that remain ipsilateral , and the lack of inhibitory feedback to the first interneuron on the right from the ipsilateral ACIN . An additional left-right asymmetry is provided by the input from the first interneurons MG1 L and R to the second interneurons MG2 L and R , which is far greater on the left than on the right side , and supported by an asymmetrical distribution of gap junctions . The latter is endorsing , and unusual because putative pathways formed by gap junctions are in general more symmetrical than those formed by chemical synapses . Larvae swim in a helical pattern ( McHenry , 2005 ) from simple bilateral flexions of the tail ( Bone , 1992; Nishino et al . , 2010 ) . The helical pattern may reflect the asymmetry of motor pathways , but varying the direction of incident light relative to the ocellus could also generate helical klinotaxis , enablng phototaxis in response to cyclical changes in light intensity as the shadow of the ocellus pigment sweeps across the outer segments ( McHenry and Strother , 2003; McHenry , 2005; Randel and Jékely , 2016 ) . Based on gravitaxis in young larvae ( Tsuda et al . , 2003 ) we also predict that the antenna neurons’ circuits underlie a directional response to gravity . It is interesting that the visual pathway originating in the right-sided ocellus and the gravity pathway from the antennal cells , both converge asymmetrically in the motor ganglion , the visual pathway stronger to the left of the ganglion and the gravity pathway to the right . These asymmetries in sensory input to the MG , particularly for the antenna pathway , suggest that a sensory input ought to generate a sided swimming movement , such as a large unilateral contraction ( Video 7 ) . We predict from the network that this should predominantly be on a fixed side of each animal , as video recordings indeed suggest ( Video 8; Ryan , unpublished observations ) . 10 . 7554/eLife . 16962 . 045Video 7 . Unilateral tail flick . A larva exhibits a unilateral tail flick . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 04510 . 7554/eLife . 16962 . 046Video 8 . Asymmetrical tail flicks . A larva exhibits repeated tail flicks to the same side of the trunk . DOI: http://dx . doi . org/10 . 7554/eLife . 16962 . 046 Despite its shared evolutionary ancestry with chordates , the synaptic network we report incorporates many pathways that are manifestly left-right asymmetrical but that compare to pathways in the vertebrate brain so far reported to be bilaterally symmetrical in form and function . We interpret these asymmetries as an ascidian specialization brought about by the small cell numbers and rapid development of the larval stage , and compatible with the larva’s sensorimotor behaviour and helical swimming pattern . Adult sea squirts , Ciona intestinalis ( L . ) , were collected by Mr . Peter Darnell from Mahone Bay , Nova Scotia . Adults were kept in tanks under constant illumination at the Aquatron facility of Dalhousie University , with flowing sea-water ( 5–6 L/min ) at ~18°C . Adults kept for 1–5 days were removed from the tank and dissected to expose the oviduct and sperm duct . Eggs were collected from the oviduct using a pipette and placed in Petri dishes containing sea-water filtered through a Nalgene 0 . 2 µm syringe filter . The animals were then washed with sea-water and the sperm duct pierced with a pipette and sperm sucked directly into the pipette and placed in a microcentrifuge tube . Sperm from one adult was added two drops at a time into a Petri dish containing eggs from a different adult and the dish gently swirled to distribute the sperm . Eggs and sperm were left for 15 min for fertilization to occur , then eggs were rinsed several times with filtered sea-water and placed in a Petri dish with filtered sea-water , wrapped in aluminium foil , and placed in an incubator at 18°C . Dark-reared larvae were removed from the 18°C incubator after 20 hr , and larvae ( 21 hr post fertilization/2 hr post hatching ) were fixed at 4°C for 2 hr in 1% OsO4 in 0 . 2M Na2PO4 ( phosphate buffer ) adjusted to pH 7 . 2 with HCl . The animals were then post-fixed in 0 . 2M phosphate buffer containing 2% glutaraldehyde for 1 hr at 4°C . Prior to fixation newly hatched larvae received light from a dissecting microscope fibre optic illuminator only briefly to check that they were swimming , and to pipette them into the fixative . Fixed specimens were dehydrated in an ethanol series ( 30% , 50% , 70% , for 10 min each , followed by 90% 95% , 100% then propylene oxide for 5 min each ) . After dehydration , specimens were placed overnight at 23°C in a dish containing propylene oxide and Poly/Bed 812 resin 1:1 . Sibling larvae , the products of a single cross , were next transferred to 100% resin for 3 hr and then placed in 100% resin and polymerized at 60°C for 48 hr . Larvae were sectioned and checked for acceptable fixation . Nervous tissue from marine animals presents special problems for good EM fixation ( Cobb and Pentreath , 1978 ) and a single larva selected for its clear synaptic vesicle profiles and intact cell membranes was cut in a series at 60 , 70 or 100 nm , as reported in Results ( Figure 1A ) . All sections were post-stained for 5–6 min in freshly prepared aqueous uranyl acetate followed by 2–3 min in lead citrate . Sections were viewed using an FEI Tecnai 12 electron microscope operated at 80kV and images captured initially using a Kodak Megaview II camera with software ( AnalySIS: SIS GmbH , Münster , Germany ) , or a later Gatan 832 Orius SC1000 CCD camera using Gatan DigitalMicrograph software ( Gatan Inc . , Pleasanton , CA ) . High magnification 3 . 85 nm per pixel images were collected for the neuropil region of each section . The profile area ranged from five 5 × 5 montages per section in the posterior brain vesicle to single 2 × 2 montages in the tail . Independent lower magnification 13 . 9 nm per pixel images of the entire CNS and overlying epidermis were collected for every section in the anterior BV , and for every fourth section through the posterior brain vesicle , neck , motor ganglion and anterior tail . The montages were compiled automatically with the Gatan DigitalMicrograph software or with AnalySIS . Given a limitation of the Gatan software-generated montages , final montages were also manually montaged in Adobe Photoshop . Images were imported into either a high magnification or low magnification series in Reconstruct ( Fiala , 2005 ) , and sections manually aligned using this software . All profiles in every third section , somata in the low-magnification series and all profiles in the high-magnification series , were then traced completely . Fewer than five neurites were candidate orphans that lacked synapses , and thus did not contribute to the CNS connectome . Skeletonization of neurite projections was accomplished using the function Z-Trace in Reconstruct , which connects the mid-point of each traced profile . In the high magnification series , traces were hidden and all sections then blindly annotated for synapses , putative gap junctions , dense-core vesicles and cilia . After each block of sections was annotated , the traces were then made visible and annotated elements assigned to specific pre- and postsynaptic elements . Blind annotation was duplicated in 100 section blocks by an independent annotator ( Ms Carlie Langille ) . Most annotations of each viewer duplicated existing synaptic contacts seen by the other , but neither viewer annotated a synapse between two partners that did not replicate a synapse formed elsewhere by the same two cells . Of the differences observed , 95% were simply those between the numbers of sections in which a synapse was observed . Each neuron was classified from structural criteria , mostly from the identity of its presynaptic partner ( s ) ( see key in Figure 1—source data 1 ) . Synapses were identified based on the criteria established in C . elegans ( White et al . , 1986 ) of a cluster of vesicles at a presynaptic membrane . Although postsynaptic densities were observed at some synapses ( Peters and Palay , 1996 ) , these were either not clear or not present at all sites with a presynaptic vesicle cluster . Putative gap junctions were annotated at sites with juxtaposed membranes and densities on the membranes of both sides , except where such contacts were directly adjacent to the neural canal , which are candidate desmosomes and were not studied further . The numbers of synapses and the numbers of sections in which a synaptic profile , a single imaged cross-section of a synapse , was observed from each were both measures used to quantify synaptic strength , as was their product , the total depth of synaptic profiles . The cumulative depth of synaptic contact was calculated by multiplying the number of sections in which a synapse was observed by the section thickness . These values for each presynaptic neuron were linearly proportional ( Figure 3B ) ; the relationship remained unchanged for the numbers of synapses per neuron when small synapses – those containing only a single section -- were removed . This depth in µm was used for network diagrams as a putative proxy for synaptic strength . Nucleus location was used to determine neuron sidedness . A line through the midline of the CNS drawn from the ventral to dorsal surfaces through the neural canal was used to classify nuclei as left or right . Those with nuclei intersected by the midline were classified as midline neurons . Nuclear x/y positions are reported in Figure 1—source data 1 Partitioning the parts of each neuron ( cell body/soma , axon , terminal , or dendrite ) was determined by examining the reconstructed neuron and identifying the sections at the boundaries between these parts . Axons were distinguished from their somata of origin by the reduction in profile diameter , and terminals were distinguished from axons where we observed branching or expansions with presynaptic sites near their final termination . Many terminals had but small swellings , and their synapses occurred onto collateral terminals or axons , so were considered en passant . Dendrites were those regions proximal to the cell body in which short neurites extended , with or without branching . Some relay neurons contain expanded branched regions near their axon hillock , which we classified as ‘BVterm’ and these were included in the analysis of dendritic synapse . Exclusively postsynaptic dendrites were lacking in the CNS . Despite our attempts to partition them in this way , neurons were characterized by a lack of segregation of pre- and postsynaptic sites along their length and a general absence of defining features for each region . For Video 1 and Video 7 , larval swimming was recorded using a Motionscope CCD camera by Redlake Imaging ( Model PCI 2000 S #1108–0009 ) mounted on a Leica MZFLIII dissecting microscope equipped with a Plan Apo 1/0x objective and recording these on Sterling computer using MIDAS software . Clips were selected and converted to . mov format using iMovie . For Video 8 , hatched larvae were recorded using a Leica MZFLIII dissecting microscope equipped with a Plan Apo 1/0x objective , capturing their images with a high resolution CCD camera ( Elmo TSM 41OH ) on VHS video tapes , and then transforming these to digitized image sequences at up to 60 fps by software ( QuickTime Pro: Apple Inc . ) .	Brains are made up of a network of nerve cells ( neurons ) that are connected to each other by junctions called synapses . The neurons on the left and right sides of the brain form different patterns of connections , but this asymmetry can be difficult to spot because the brain is large and complex . Understanding how the whole network operates is key to understanding how the brain works . However , a full map of all the connections between neurons – known as a connectome – has only been described for one species so far , a nematode worm called C . elegans . The tadpole larva of the common sea squirt has a fairly simple brain distantly related to our own but made up of only about 330 cells . Ryan et al . used a technique called electron microscopy to study thin sections from the brains of sea squirt larvae to reveal this animal’s connectome and investigate left-right asymmetry in the brain . The analysis revealed 177 neurons in this larval brain , just over half of its brain cells . These can be split into at least 25 types and each neuron has a simple , mostly unbranched shape with , on average , 49 synapses with other cells . This means that , even though it has such a small number of neurons , the neuron network is still relatively complex . The shortest sensory pathway to any muscle connects via three synapses , although most pathways involve more . The left and right sides of the brain differ in the types of neurons they contain and the connections these form , even though both sides have the same number of cells . The findings of Ryan et al . reveal the second animal connectome and lay the groundwork for future studies on how each neuron in the network influences the behaviour of the sea squirt’s larva . Further work is also required to find out how the patterns of synapses in the brain change as the larva ages , and whether the connectome differs between siblings .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	The CNS connectome of a tadpole larva of Ciona intestinalis (L.) highlights sidedness in the brain of a chordate sibling
UNC93B1 , a multipass transmembrane protein required for TLR3 , TLR7 , TLR9 , TLR11 , TLR12 , and TLR13 function , controls trafficking of TLRs from the endoplasmic reticulum ( ER ) to endolysosomes . The mechanisms by which UNC93B1 mediates these regulatory effects remain unclear . Here , we demonstrate that UNC93B1 enters the secretory pathway and directly controls the packaging of TLRs into COPII vesicles that bud from the ER . Unlike other COPII loading factors , UNC93B1 remains associated with the TLRs through post-Golgi sorting steps . Unexpectedly , these steps are different among endosomal TLRs . TLR9 requires UNC93B1-mediated recruitment of adaptor protein complex 2 ( AP-2 ) for delivery to endolysosomes while TLR7 , TLR11 , TLR12 , and TLR13 utilize alternative trafficking pathways . Thus , our study describes a mechanism for differential sorting of endosomal TLRs by UNC93B1 , which may explain the distinct roles played by these receptors in certain autoimmune diseases . Toll-like receptors ( TLRs ) recognize conserved microbial features and initiate signals critical for induction of immune responses to infection . A subset of TLRs ( TLR3 , TLR7 , TLR8 , and TLR9 ) recognizes forms of nucleic acids , including double-stranded RNA , single-stranded RNA , and DNA ( Barbalat et al . , 2011 ) . This specificity facilitates recognition of a broad array of microbes but introduces the potential for recognition of self-nucleic acids . TLR7 and TLR9 recognition of self-RNA and self-DNA , respectively , contributes to autoimmune diseases such as systemic lupus erythematosus ( SLE ) ( Marshak-Rothstein , 2006; Christensen and Shlomchik , 2007 ) . Discrimination between self and microbial nucleic acids cannot be achieved solely through recognition of distinct features but instead relies on differential delivery of these potential ligands to TLRs ( Barton and Kagan , 2009 ) . All of the TLRs capable of nucleic acid recognition localize within endosomal compartments which sequesters these receptors away from self nucleic acids in the extracellular space ( Barton and Kagan , 2009 ) . Our previous studies as well as work from other groups indicate that a requirement for ectodomain cleavage of intracellular TLRs further restricts receptor activation to protease-rich acidic compartments ( Ewald et al . , 2008 , 2011; Park et al . , 2008; Garcia-Cattaneo et al . , 2012 ) . Bypassing this requirement enables responses to extracellular self nucleic acid and leads to fatal inflammatory disease in mice ( Mouchess et al . , 2011 ) . Moreover , the system appears carefully balanced as simply overexpressing TLR7 in mice causes responses to self-RNA and development of an SLE-like disease ( Pisitkun et al . , 2006; Subramanian et al . , 2006; Deane et al . , 2007 ) . Thus , defining the regulatory steps that control TLR localization and influence the threshold of receptor activation has important implications for self/non-self discrimination . TLR9 and other intracellular TLRs must traffic from the endoplasmic reticulum ( ER ) to endolysosomes before responding to ligands . UNC93B1 , a multi-pass transmembrane protein localized to the ER , appears to facilitate this trafficking ( Brinkmann et al . , 2007; Kim et al . , 2008 ) . Mice homozygous for a nonfunctional Unc93b1 ( H412R ) allele ( Unc93b13d/3d ) fail to respond to TLR3 , TLR7 , or TLR9 ligands , and mice and humans deficient in UNC93B1 are highly susceptible to viral infection ( Casrouge et al . , 2006; Tabeta et al . , 2006; Lafaille et al . , 2012 ) . More recently , UNC93B1 has been implicated in the function of TLR11 , TLR12 , and TLR13 ( Pifer et al . , 2011; Shi et al . , 2011; Koblansky et al . , 2012; Oldenburg et al . , 2012 ) . UNC93B1 is not required for responses by surface localized TLRs such as TLR2 and TLR4 ( Tabeta et al . , 2006 ) . UNC93B1 associates with endosomal TLRs , and in cells with defective UNC93B1 , TLR9 and TLR7 fail to leave the ER ( Brinkmann et al . , 2007; Kim et al . , 2008 ) . However , the mechanism by which UNC93B1 facilitates TLR trafficking to endosomal compartments remains enigmatic , especially considering its reported direct translocation from the ER to endolysosomes ( Kim et al . , 2008 ) . This pathway is inconsistent with our findings that TLR9 and TLR7 traffic through the general secretory pathway en route to endosomes ( Ewald et al . , 2008 , 2011 ) . In addition , mice expressing an aspartic acid to alanine mutation at amino acid position 34 in UNC93B1 ( Unc93b1D34A/D34A ) were recently shown to develop spontaneous autoimmunity due to enhanced TLR7 responses and diminished TLR9 responses ( Fukui et al . , 2009 , 2011 ) . These findings suggest that regulation of TLRs by UNC93B1 can influence the relative thresholds of receptor activation . For these reasons , we have sought to define the molecular basis by which UNC93B1 controls endosomal TLR trafficking and function . Beyond the implied role for UNC93B1 discussed above , little is known about the molecular mechanisms that mediate proper localization of endosomal TLRs and no other factor required specifically for endosomal TLR trafficking has been identified . Nevertheless , several reports suggest that endosomal TLR trafficking may be influenced at both ER and post-Golgi trafficking steps . Gp96 functions as an ER folding chaperone for many TLRs , including TLR9 , and PRAT4A has been implicated in TLR trafficking from the ER ( Takahashi et al . , 2007; Yang et al . , 2007; Lee et al . , 2012 ) . Additionally , the HRS/ESCRT pathway is involved in post-Golgi trafficking by sorting ubiquitinated TLR7 and TLR9 to endosomal compartments ( Chiang et al . , 2012 ) , and the adaptor protein-3 ( AP-3 ) has been reported to target TLR9 and TLR7 to lysosome related organelles specialized for type I IFN induction ( Honda et al . , 2005; Blasius et al . , 2010; Sasai et al . , 2010 ) . Interestingly , UNC93B1 trafficking to these compartments is also impaired in AP-3 deficient cells ( Sasai et al . , 2010 ) . Whether UNC93B1 interacts with other components implicated in trafficking of endosomal TLRs remains to be determined . In this study , we report that UNC93B1 is required for multiple steps of TLR trafficking . UNC93B1 plays a direct role in facilitating exit of TLRs from the ER as well as a later role in recruitment of adaptor protein-2 ( AP-2 ) to facilitate endocytosis of TLR9 from the plasma membrane . Surprisingly , TLR7 does not have the same requirements for UNC93B1 and utilizes distinct trafficking machinery to reach endolysosomes . Thus , our results describe how UNC93B1 controls endosomal TLR trafficking and provide the first mechanistic basis for differential regulation of these receptors . UNC93B1 has been described as an ER-resident trafficking chaperone that translocates TLRs directly from the ER to endolysosomes upon TLR activation ( Kim et al . , 2008 ) . This model is based in part on the observation that UNC93B1 never acquires Endoglycosidase H ( EndoH ) -resistant glycans ( Brinkmann et al . , 2007 ) , which are acquired only when proteins traffic through the medial Golgi . Because this proposed function for UNC93B1 conflicts with our model of TLR9 trafficking ( Ewald et al . , 2008 , 2011 ) , we first examined whether UNC93B1 is present in endolysosomal compartments in unstimulated cells . Wildtype ( WT ) UNC93B1 but not the nonfunctional ( H412R ) mutant was detectable in phagosomes purified from unstimulated RAW264 cells ( Figure 1A ) . Moreover , a portion of UNC93B1-WT gained EndoH-resistance in multiple cell types , while UNC93B1-H412R was entirely EndoH-sensitive ( Figure 1B–F ) . These results agree with a previous report that the H412R mutant fails to leave the ER ( Kim et al . , 2008 ) . To formally demonstrate that the increased molecular weight of UNC93B1-WT is due to N-linked glycans , we mutated Asn-251 , which is within a consensus N-glycosylation site , and this mutant failed to acquire EndoH-resistant glycans ( Figure 1G ) . Based on the acquisition of EndoH-resistant glycans by UNC93B1 , we examined whether UNC93B1 is detectable within COPII vesicles , which mediate transport of cargo between the ER and Golgi ( Zanetti et al . , 2012 ) . Using an in vitro COPII budding assay ( Kim et al . , 2005; Merte et al . , 2010 ) , we compared levels of UNC93B1-WT and UNC93B1-H412R in purified vesicles . UNC93B1-WT , but not H412R , was clearly present within the vesicles , further supporting a model in which UNC93B1 exits the ER through the general secretory pathway ( Figure 1H ) . Altogether , these data indicate that a pool of UNC93B1 protein exits the ER and traffics through the Golgi in unstimulated cells . Moreover , transit from the ER to the Golgi may be important for UNC93B1 function , as the nonfunctional UNC93B1-H412R mutant fails to enter COPII vesicles and does not reach the medial Golgi . 10 . 7554/eLife . 00291 . 003Figure 1 . UNC93B1 traffics to the Golgi en route to endolysosomes . ( A ) UNC93B1 is present in phagolysosomes of unstimulated cells . Phagosomes ( PHG ) isolated by flotation from RAW264 cells only ( Ø ) or expressing GFP tagged UNC93B1-WT or UNC93B1-H412R and cells prior to isolation ( Pre ) were separated by SDS-PAGE , and immunoblotted with anti-GFP , anti-LAMP1 ( lysosome marker ) , and anti-calnexin ( ER marker ) . ( B ) A portion of UNC93B1 protein traffics to the Golgi apparatus . Wildtype UNC93B1 ( WT ) or H412R , each with a C-terminal 3× FLAG tag , were expressed in HEK293Ts by transient transfection . The immunoprecipitated proteins were treated with EndoH ( E ) , PNGaseF ( P ) or left untreated ( − ) , separated by SDS-PAGE , and visualized by immunoblot with anti-FLAG antibody . Bands representing EndoH-sensitive ( white arrow ) and resistant ( black arrow ) forms of UNC93B1 are indicated . ( C ) – ( F ) UNC93B1 acquires EndoH-resistant modifications . UNC93B1 tagged with GFP ( C ) or myc-His ( D ) from transiently transfected HEK293Ts , and FLAG tagged UNC93B1 expressed in MEFs ( E ) or 3d iMac cells ( F ) were analyzed for the presence of EndoH-resistant glycans . Lysates were separated by SDS-PAGE and immunoblotted with the indicated antibodies . EndoH-sensitive ( white arrow ) and EndoH-resistant ( black arrow ) forms are indicated . ( G ) Mutation of UNC93B1 glycosylation sites abolishes EndoH resistant forms . Lysates from HEK293Ts transiently transfected with FLAG tagged UNC93B1-WT , -N251A or -N251A/N272A were separated by SDS-PAGE and immunoblotted with anti-FLAG antibody . ( H ) UNC93B1 is loaded into COPII vesicles . Digitonin-permeabilized COS7 cells expressing 3× FLAG-tagged UNC93B1-WT or UNC93B1-H412R , or no cells ( Ø ) were incubated with ATP regenerating system , GTP , and rat liver cytosol , as indicated , in an in vitro COPII budding assay . Vesicles purified by ultracentrifugation were analyzed by SDS-PAGE and immunoblot using the indicated antibodies . 20% of the COS7 cells prior to the budding reaction serves as a loading control ( 20% donor ) . ERGIC/p58 serves as a positive control for the formation of COPII vesicles . Results are representative of at least three experiments ( A–G ) or two experiments ( H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00291 . 003 Our previous work reported that three species of TLR9 can be detected within macrophages , representing distinct maturation stages: an initial 150-kDa species with EndoH-sensitive glycans corresponding to the ER-resident protein ( TLR9-ER ) , a larger species with EndoH-resistant glycans corresponding to full-length receptor that has passed through the Golgi ( TLR9-Precursor ) , and a 80-kDa band with EndoH-resistant glycans corresponding to the mature , cleaved receptor within endolysosomes ( TLR9-Cleaved ) ( Figure 2A , lane 1 ) ( Ewald et al . , 2008 , 2011 ) . To examine how UNC93B1 function impacts TLR9 localization , we compared these three forms of TLR9 in immortalized macrophages derived from Unc93b13d/3d mice and complemented with UNC93B1-WT or UNC93B1-H412R . While all three bands were present in macrophages with functional UNC93B1 , only the ER-resident form was detectable in cells expressing the UNC93B1-H412R mutant ( Figure 2A ) , consistent with our previous analysis of TLR9 in UNC93B1 shRNA knockdown cells ( Ewald et al . , 2008 ) . These data indicate that TLR9 does not reach the medial Golgi in the absence of functional UNC93B1 . Because a pool of UNC93B1 can traffic from ER to Golgi by entering COPII vesicles ( Figure 1B ) , we considered whether UNC93B1 regulates this aspect of TLR9 trafficking . Indeed , analysis of TLR9 loading into COPII vesicles revealed that TLR9 was only detectable in the presence of functional UNC93B1 whereas a control traffic protein , ERGIC/p58 , was packaged independently of UNC93B1 ( Figure 2B ) . 10 . 7554/eLife . 00291 . 004Figure 2 . UNC93B1 controls ER exit of TLRs 3 , 7 , 9 , 11 , and 13 . ( A ) TLR9 fails to exit the ER in cells lacking functional UNC93B1 . Lysates from 3d iMac cells complemented with either UNC93B1-WT or UNC93B1-H412R and expressing TLR9-HA were analyzed by SDS-PAGE and immunoblotted with the indicated antibodies . The precursor ( black arrow ) , ER ( white arrow ) and cleaved ( grey arrow ) forms of TLR9-HA are indicated . ( B ) UNC93B1 is required for TLR9 loading into COPII vesicles . RAW264 macrophages stably transduced with retroviruses encoding control or Unc93b1-directed shRNA and expressing TLR9-HA were used in an in vitro COPII budding assay as described in ( Figure 1H ) . Lysates of purified vesicles or donor membranes were probed with the indicated antibodies . ( C ) The transmembrane and cytosolic domain of TLR9 is sufficient to confer UNC93B1-dependence . ( Left ) schematic of TLR9 and the CD4-TLR9 chimera . Transmembrane ( TM ) , ectodomain ( Ecto ) and cytosolic domain ( Cyto ) are indicated . ( Right ) CD4-TLR9 was expressed in HEK293Ts together with FLAG-tagged UNC93B1-WT or UNC93B1-H412R . Total lysates were analyzed by SDS-PAGE and immunoblotted with anti-CD4 and anti-FLAG antibodies . EndoH-sensitive ( white arrow ) and resistant ( black arrow ) forms are indicated . ( D ) CD4 trafficking to the cell surface is normal in Unc93b13d/3d cells . Splenocytes from C57BL/6 ( blue line ) or Unc93b13d/3d ( red line ) mice were stained with anti-CD4 and analyzed by flow cytometry . ( E ) CD4-TLR chimeric proteins for each of the indicated TLRs were expressed in HEK293Ts together with FLAG-tagged UNC93B1-WT or UNC93B1-H412R . Lysates were separated by SDS-PAGE and visualized by immunoblot with anti-HA and anti-FLAG antibodies . EndoH-sensitive ( white arrows ) and resistant ( black arrows ) forms are indicated . The chimeras were constructed as shown in Figure 1E , except with the addition of a C-terminal HA tag . Results are representative of at least three experiments ( A , C , and E ) or two experiments ( B and D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00291 . 004 Quality control mechanisms ensure that only properly folded proteins can exit the ER , and one mechanism by which UNC93B1 could regulate ER exit of TLR9 is through regulation of TLR9 folding , as has been reported for gp96 ( Yang et al . , 2007 ) . To address whether UNC93B1 serves as a folding chaperone , we tested whether a chimeric CD4-TLR9 protein , consisting of the ectodomain of CD4 fused to the transmembrane and cytosolic regions of TLR9 ( Figure 2C , left ) , required UNC93B1 function . Because trafficking of CD4 is not UNC93B1-dependent ( Figure 2D ) , this chimera can be used to test whether TLR9 requires UNC93B1 to ensure correct folding of the TLR9 ectodomain . CD4-TLR9 acquired EndoH-resistant glycans when expressed with UNC93B1-WT but not when expressed with mutant UNC93B1-H412R ( Figure 2C , right ) . Thus , the requirement for UNC93B1 is not based on TLR9 ectodomain folding . Furthermore , the transmembrane domain and cytosolic regions of TLR9 are sufficient to mediate UNC93B1-dependent trafficking . Taken together , these data indicate that UNC93B1 regulates ER to Golgi transport of TLR9 . While we cannot rule out that a pool of UNC93B1 bypasses the Golgi en route to endosomes as suggested by others ( Brinkmann et al . , 2007; Kim et al . , 2008 ) , this route does not seem relevant for UNC93B1-dependent TLR9 trafficking . Instead , UNC93B1 appears to control TLR9 entry into COPII vesicles . We next sought to determine whether the dependence on UNC93B1 for ER exit is a general property of endosomal TLRs . We generated a panel of CD4-TLR fusion proteins and tested whether they required UNC93B1 to exit the ER , based on acquisition of EndoH-resistant glycans . As expected , ER exit of both CD4-TLR3 and CD4-TLR7 required UNC93B1 , which is consistent with defective TLR3 and TLR7 signaling in Unc93b13d/3d mice ( Tabeta et al . , 2006 ) , whereas CD4-TLR4 trafficked independently of UNC93B1 ( Figure 2E ) . CD4-TLR11 and CD4-TLR13 also required UNC93B1 for ER exit ( Figure 2E ) . Less is known about the localization or trafficking of these TLRs , although biochemical studies have suggested that these TLRs can associate with UNC93B1 ( Brinkmann et al . , 2007; Melo et al . , 2010; Pifer et al . , 2011 ) . Our results are the first to show that UNC93B1 regulates the trafficking of TLR11 and TLR13 by controlling ER exit . Altogether , our data indicate that the role for UNC93B1 in controlling ER exit extends to all known endosomal TLRs ( TLR3 , TLR7 , TLR9 ) as well as TLR11 and TLR13 . To define regions within UNC93B1 necessary for ER export of TLRs , we generated a series of N-and C-terminal truncation mutants . Based on the predicted topology of UNC93B1 ( Brinkmann et al . , 2007 ) , the N- and C-termini face the cytosol , so we reasoned that these regions would be most likely to interact with putative trafficking factors ( Figure 3A ) . Interestingly , N-terminal truncations resulted in a strong defect in TLR9 trafficking from the ER as evidenced by significantly reduced or absent precursor and cleaved forms of TLR9 in cells expressing the Δ50 and Δ57 UNC93B1 truncations ( Figure 3B ) . This defective trafficking resulted in impaired responses to TLR9 ligands in macrophages , while TLR2 responses remained intact ( Figure 3C ) . Taken together , our data indicate that the N-terminal region of UNC93B1 contains residues important for trafficking of TLR9 and UNC93B1 from the ER to endolysosomes . These findings are in agreement with a previous study showing that TLR9 responsiveness is strongly dependent on the N-terminus of UNC93B1 , although this study did not directly examine TLR9 trafficking ( Fukui et al . , 2009 ) . Furthermore , an UNC93B1 point mutant ( D34A ) , which reduced TLR9 signaling but enhanced TLR7 signaling ( Fukui et al . , 2009 , 2011 ) , reduced TLR9 transport and cleavage ( Figure 3D ) . 10 . 7554/eLife . 00291 . 005Figure 3 . UNC93B1 mutants reveal two distinct roles in TLR9 trafficking . ( A ) Schematic of predicted UNC93B1 topology . UNC93B1 is a 12 pass transmembrane protein with terminal regions facing cytosol . Two putative N-linked glycosylation ( N-Glyc ) sites are indicated ( top ) . Schematic of UNC93B1 ( 1–598 a . a . ) . The black numbered boxes represent predicted transmembrane domains , white boxes represent regions predicted to face the cytosol , and grey boxes represent regions predicted to face the lumen . Truncation and point mutations are indicated with arrows ( bottom ) . ( B ) N- and C-terminal mutants of UNC93B1 have distinct TLR9 trafficking outcomes . Lysates from 3d iMac cells expressing TLR9-HA and complemented with mutant forms of GFP-tagged UNC93B1 were subjected to SDS-PAGE and immunoblotted with anti-HA and anti-GFP antibodies . The precursor ( black arrow ) , ER ( white arrow ) and cleaved ( grey arrow ) forms of TLR9-HA are indicated . n . s . indicates a non-specific band . ( C ) N- and C-terminal mutants of UNC93B1 have diminished TLR9 signaling . 3d iMac cells complemented with WT or indicated mutant alleles of UNC93B1 were harvested for intracellular TNFα staining 5 hr after stimulation with 3 μM CpG , 1 μg/ml Pam3CSK4 or left unstimulated . Percentages of TNF-producing cells after gating on UNC93B1-GFP positive cells are plotted . ( D ) TLR9 trafficking to endolysosomes is impaired in UNC93B1-D34A cells . 3d iMac cells were complemented with GFP-tagged UNC93B1-WT , -H412R or -D34A . Lysates were separated by SDS-PAGE and visualized by immunoblot with anti-HA and anti-GFP antibodies . The precursor ( black arrow ) , ER ( white arrow ) and cleaved ( grey arrow ) forms of TLR9-HA are indicated . ( E ) UNC93B1 interacts with the cleaved form of TLR9 . Immunoprecipitation of FLAG tagged UNC93B1 ( WT or H412R ) in 3d iMac cells expressing TLR9-HA was performed in 1% digitonin with anti-FLAG matrix . UNC93B1 associated proteins were analyzed by SDS-PAGE and immunoblot with anti-FLAG and anti-HA antibodies . ER ( white arrows ) and cleaved ( grey arrows ) forms of TLR9-HA are indicated . All results are representative of at least three experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00291 . 005 Our results thus far suggest that UNC93B1 controls the function of multiple TLRs by regulating their exit from the ER . However , UNC93B1 itself also exits the ER and is present in endolysosomes ( Figure 1A , H ) . Moreover , we could detect both the precursor and cleaved forms of TLR9 associated with immunoprecipitated UNC93B1 , suggesting that UNC93B1 and TLR9 remain associated after leaving the ER ( Figure 3E ) . To test whether this post-ER interaction may have functional consequences , we screened UNC93B1 mutants for any role in TLR9 function beyond ER export . Strikingly , truncation of the UNC93B1 C-terminus ( Δ523 and Δ538 ) resulted in an accumulation of the Golgi-modified precursor form of TLR9 and a marked reduction of the cleaved receptor ( Figure 3B ) . This pattern suggests that TLR9 trafficking is blocked at some point after the medial Golgi , in contrast to the trafficking defect observed with N-terminal deletion mutants . Consistent with this interpretation , we observed reduced TLR9 signaling in cells expressing the Δ523 and Δ538 UNC93B1 mutants ( Figure 3C ) . Thus , in addition to its role in ER export , UNC93B1 appears to regulate post-ER trafficking of TLR9 . We next sought to define the mechanism underlying the requirement for UNC93B1 in post-ER TLR9 trafficking . We focused on residues 539–542 , YRYL , because they are evolutionarily conserved and fit the consensus for a YxxΦ motif ( where Φ represents a hydrophobic residue ) , which may mediate protein interactions or serve as a site for phosphorylation ( Songyang et al . , 1993; Ohno et al . , 1995 , 1996; Crump et al . , 1998; Bonifacino and Traub , 2003 ) ( Figure 4A ) . Mutation of Tyr539 to an Ala ( Y539A ) resulted in accumulation of the Golgi-modified precursor form of TLR9 and reduced TLR9 cleavage , similar to the block in TLR9 trafficking observed with the Δ538 truncation mutant ( Figure 4B ) . Mutation of Leu-542 to Ala ( L542A ) and Tyr-539 to Phe ( Y539F ) gave similar results , although the block was not as complete as Y539A ( Figure 4C ) . Responsiveness to TLR9 ligands was also reduced in UNC93B1-Y539A expressing cells , consistent with the block of TLR9 processing ( Figure 4D ) . 10 . 7554/eLife . 00291 . 006Figure 4 . UNC93B1 controls post-Golgi trafficking of TLR9 by recruiting AP-2 . ( A ) Multi-species protein sequence alignment of residues in the UNC93B1 C-terminal tail . The YxxΦ motif , YRYL , is boxed . ( B ) Precursor TLR9 accumulates in UNC93B1-Y539A expressing cells . Lysates from 3d iMac cells expressing TLR9-HA were complemented with the indicated GFP tagged UNC93B1 , analyzed by SDS-PAGE , and immunoblotted with anti-HA and anti-GFP antibodies . The precursor ( black arrow ) , ER ( white arrow ) and cleaved ( grey arrow ) forms of TLR9-HA are indicated . n . s . indicates a non-specific band . ( C ) Mutations of YxxΦ in UNC93B1 confer partial phenotypes in TLR9 trafficking when compared with Y539A . Lysates from 3d iMac cells expressing TLR9-HA and GFP tagged UNC93B1-WT , H412R , Δ538 , L542A , Y539F , Y539A were analyzed by SDS-PAGE and immunoblotted with anti-HA and anti-GFP antibodies . Presence of precursor ( black arrow ) , ER ( white arrow ) and cleaved ( grey arrow ) forms of TLR9 are indicated . ( D ) TLR9 signaling is impaired in UNC93B1-Y539A cells . 3d iMac cells , complemented with GFP tagged UNC93B1-WT , -H412R , or -Y539A , were harvested for intracellular TNFα staining 5 hr after stimulation with 3 μM CpG , 1 μg/ml Pam3CSK4 ( Pam ) or left unstimulated . Percentages of TNF-producing cells after gating on UNC93B1-GFP positive cells are plotted . ( E ) The C-terminal tail of UNC93B1 interacts with AP-2μ . Results from a yeast two-hybrid assay testing for interaction between the AP ( -1A , -2 , -3A , -4 ) µ subunits and the N- or C-terminal cytosolic regions of UNC93B1 ( N- or C-tail UNC ) , are shown . Growth on –His–Trp–Leu plates ( –His ) or –Ade–Trp–Leu ( –Ade ) indicates interaction . Growth on –Trp–Leu plates ( +His+Ade ) serves as a control . ( F ) – ( G ) Tyr-539 on UNC93B1 mediates interaction with AP-2 complex . ( F ) Results from a yeast two-hybrid assay testing for interaction between the AP-2µ subunit and the C-terminal cytosolic region of UNC93B1 ( UNC93B1 C-tail ) from WT , the Y539A mutant , or the Δ538 mutant are shown . Growth on –His–Trp–Leu plates ( –His ) indicates interaction . Growth on –Trp–Leu plates ( +His ) serves as a control . ( G ) HEK293T were transiently transfected with AP-2µ-HA and FLAG tagged UNC93B1-WT , -Y539A . Cell lysates were incubated with anti-FLAG matrix , and UNC93B1-associated proteins were eluted with FLAG peptide , separated by SDS-PAGE and visualized by immunoblot with anti-HA or anti-FLAG antibodies . UNC93B1 associated AP-2µ is indicated with black arrow . Results are representative of at least three experiments ( B , D–F ) or two experiments ( C and G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00291 . 006 YxxΦ motifs can serve as binding sites for clathrin adaptor protein ( AP ) complexes , so we tested whether the C-terminal tail of UNC93B1 could interact with any of the four mammalian AP complexes . AP complexes consist of four subunits , of which the µ subunit typically determines cargo specificity ( Bonifacino and Traub , 2003; Ohno , 2006 ) . Because these interactions are often weak , we used a yeast two-hybrid ( Y2H ) assay in which the μ subunits of AP-1 , AP-2 , AP-3 , and AP-4 were fused to the Gal4 activation domain and the UNC93B1 N- and C-terminal tails were fused to the Gal4 DNA binding domain . The C-terminal cytosolic region of UNC93B1 interacted strongly with AP-2μ , but not with the µ subunits of any of the other AP complexes ( Figure 4E ) . Importantly , interaction with AP-2μ was abolished by the Δ538 truncation and the Y539A mutation ( Figure 4F ) . Additionally , we were able to co-immunoprecipitate AP-2μ with UNC93B1-WT , but not with UNC93B1-Y539A ( Figure 4G ) . Taken together , these data suggest that UNC93B1 directly recruits AP-2 via a YxxΦ motif present in its C-terminal cytosolic tail . AP-2 complexes direct clathrin-mediated endocytosis of cargo from the plasma membrane . Therefore , our results suggest that UNC93B1 and TLR9 traffic together to the surface and require AP-2-mediated internalization to reach endocytic compartments . To test this possibility , we examined whether TLR9 required UNC93B1-mediated recruitment of AP-2 to gain access to endolysosomal compartments . Using immunofluorescence microscopy , we detected reduced colocalization of TLR9 and the lysosomal marker Lamp-1 in UNC93B1-Y539A-expressing cells ( Figure 5A ) . We also compared surface expression of N-terminally tagged TLR9 in cells expressing different UNC93B1 alleles . While surface TLR9 was only weakly detectable in cells expressing UNC93B1-WT , expression of UNC93B1-Y539A increased the levels of FLAG-TLR9 at the cell surface ( Figure 5B ) . Furthermore , as we found for TLR9-HA , the FLAG-TLR9 Golgi-modified precursor form accumulated in UNC93B1-Y539A expressing cells ( Figure 5C ) . This result suggests that disruption of the interaction between UNC93B1 and AP-2 leads to defective endocytosis of TLR9 from the cell surface . In support of this model , siRNA knockdown of AP-2µ1 resulted in a similar increase in surface expression of TLR9 ( Figure 5D ) and accumulation of the TLR9 precursor form ( Figure 5E ) . As a positive control for AP-2µ1 knockdown , we observed accumulation of CD71 ( transferrin receptor ) at the cell surface ( Figure 5D ) . Taken together , these data support a model in which UNC93B1 interacts with AP-2 to traffic TLR9 from the cell surface to endolysosomes . 10 . 7554/eLife . 00291 . 007Figure 5 . Failure to recruit AP-2 by UNC93B1 results in cell surface accumulation of TLR9 . ( A ) TLR9 fails to traffic to endolysosomes in UNC93B1-Y539A expressing cells . Localization of TLR9-HA and Lamp-1 in 3d iMac cells complemented with UNC93B1-WT , -H412R , or -Y539A was determined by immunofluorescence microscopy . Representative images of Lamp-1 ( left ) , TLR9 ( middle ) , and a pseudo-colored merged image for each UNC93B1 allele are shown . Arrowheads indicate areas of Lamp-1/TLR9 colocalization . ( B ) TLR9 accumulates at the cell surface in UNC93B1-Y539A expressing cells . HEK293Ts transfected with N-terminally tagged 3× FLAG-TLR9 and the indicated UNC93B1 alleles were stained with anti-FLAG and goat anti-mouse IgG secondary antibodies . FLAG staining was measured by flow cytometry . ( C ) The TLR9 precursor form accumulates in HEK293Ts expressing N-terminally tagged FLAG-TLR9 and UNC93B1-Y539A . Lysates from HEK293Ts stably expressing N-terminally tagged 3× FLAG TLR9 and GFP tagged UNC93B1-WT , -H412R , or -Y539A were harvested , treated with EndoH , separated by SDS-PAGE , and visualized by immunoblot with anti-FLAG and anti-GFP antibodies . ( D ) TLR9 accumulates at the cell surface in cells lacking AP-2 . HEK293Ts stably expressing 3× FLAG-TLR9 with or without WT UNC93B1 were treated with AP-2μ ( AP2M1 ) or control siRNA for 96 hr . FLAG staining was measured on intact ( surface ) or permeabilized ( Perm ) cells as described in ( G ) . Anti-CD71 ( transferrin receptor ) staining serves as a control for AP-2 knockdown . ( E ) Absence of AP-2 causes accumulation of TLR9 precursor form . HEK293Ts were treated with control or AP-2µ siRNA . After 48 hr , siRNA treated cells were transiently transfected with TLR9-HA and GFP tagged UNC93B1-WT . Lysates were harvested 24 hr later for EndoH assay as described in ( A ) . The TLR9 precursor form ( black arrow ) is indicated in each panel . Results are representative of at least three experiments ( B and D ) Or two ( A , C , and E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00291 . 007 Having established the mechanisms by which UNC93B1 facilitates TLR9 delivery to endolysosomes , we next examined whether TLR7 is under similar regulation . However , for reasons that remain unclear , detection of endogenous or epitope-tagged exogenous TLR7 protein is quite challenging . Consequently , the evidence that TLR7 undergoes ectodomain processing is limited ( Ewald et al . , 2011 ) , and one group has reported that TLR7 is not cleaved ( Park et al . , 2008 ) . We synthesized a TLR7 coding sequence optimized for efficient translation ( see ‘Materials and methods’ ) and found that our ability to detect TLR7 protein was significantly improved . Cleaved TLR7 was detected in macrophages expressing UNC93B1-WT but not in macrophages expressing UNC93B1-H412R ( Figure 6A ) . Both the full-length and cleaved forms of TLR7 possessed EndoH-resistant glycans , although fewer of the TLR7 N-linked glycans acquired these modifications than for TLR9 ( Figure 6B , top ) . These results confirm our previous findings that TLR7 undergoes ectodomain proteolysis and traffics through the Golgi en route to endolysosomes . Importantly , our results confirm that TLR7 trafficking is dependent on UNC93B1 . 10 . 7554/eLife . 00291 . 008Figure 6 . Differential trafficking of TLR7 and TLR9 . ( A ) TLR7 trafficking can be monitored biochemically and is UNC93B1-dependent . Lysates from 3d iMac cells expressing TLR7-HA and complemented with GFP-tagged UNC93B1-WT , -H412R or -Y539A were separated by SDS-PAGE and immunoblotted with anti-HA and anti-GFP antibodies . ER ( white arrows ) and cleaved ( grey arrows ) forms of TLR7-HA are indicated . ( B ) TLR7 acquires EndoH resistance . TLR7-HA was immunoprecipitated from 3d iMac cells expressing UNC93B1-WT , -H412R , or -Y539A , treated with EndoH ( E ) , PNGaseF ( P ) or left untreated ( − ) , and visualized by anti-HA immunoblot . Asterisk indicates precursor form of the receptor . The bracket indicates the migration difference between EndoH-treated and PNGaseF-treated TLR7 . ( C ) UNC93B1 interacts with ER and cleaved forms of TLR7 . GFP-tagged UNC93B1-WT or -H412R were immunoprecipitated from 3d iMac cells expressing TLR7-HA . UNC93B1 associated proteins were analyzed by SDS-PAGE and immunoblotted with anti-HA and anti-GFP antibodies . ER ( white arrows ) and cleaved ( grey arrows ) forms of TLR7-HA are indicated . ( D ) TLR7 signaling is normal in UNC93B1-Y539A cells . 3d iMac cells complemented with GFP tagged UNC93B1-WT , -H412R , or -Y539A were stimulated with 100 to 25 ng/ml R848 for 5 hr and intracellular TNFα stain was performed . Percentages of TNF-producing cells after gating on UNC93B1-GFP positive cells are plotted . ( E ) TLR7 interacts with AP-4μ . Results from a yeast two-hybrid assay testing for interaction between the AP-1 , AP-2 , AP-3 , and AP-4 µ subunits and the C-terminal cytosolic region of TLR7 . Growth on –His–Trp–Leu plates ( –His ) indicates interaction . Growth on –Trp–Leu plates ( +His ) serves as a control . ( F ) TLR7-Y892A is unable to interact with AP-4μ . Results from a yeast two-hybrid assay testing for interaction between the AP-4μ and TLR7 YxxΦ mutants ( Y882A , Y892A , or double ) . Conditions are as described in ( E ) . ( G ) TLR7-Y892A does not respond to TLR7 ligands . HEK293T cells were transiently transfected with an NF-κB luciferase reporter as well as expression plasmids encoding TLR7 , TLR7-Y892A , or empty vector . Luciferase production was assayed 16 hr after stimulation with 10 µg/ml R848 . ( H ) TLR7-Y892A trafficking is impaired . TLR7−/− bone marrow derived macrophages ( BMMs ) were transduced with HA-tagged TLR7-WT or TLR7-Y892A . Cell lysates were analyzed by SDS-PAGE and immunoblotted with anti-HA antibodies . ER ( white arrows ) and cleaved ( grey arrows ) forms of TLR7-HA are indicated . Results are representative of at least three experiments ( A , D , and E ) or two experiments ( B , C , and F–H ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00291 . 008 We next examined whether UNC93B1 controls post-Golgi sorting of TLR7 as it does for TLR9 . Similar to TLR9 , both the full-length and cleaved forms of TLR7 associated with UNC93B1 suggesting that UNC93B1 traffics with TLR7 to endolysosomes ( Figure 6C ) . However , unlike TLR9 , the response to TLR7 ligands was normal in UNC93B1-Y539A expressing cells ( Figure 6D ) . Moreover , the amount of cleaved TLR7 was unaffected , and the TLR7 Golgi-modified precursor form did not accumulate in these cells ( Figure 6A , B ) . Thus , delivery of TLR7 to endolysosomes appears to be independent of the UNC93B1/AP-2 pathway . If TLR7 does not require the UNC93B1-mediated recruitment of AP-2 for proper trafficking , then how does the receptor reach endolysosomes ? We considered the possibility that TLR7 may recruit AP-2 directly , thereby obviating the need for UNC93B1 . However , when we screened by Y2H for interactions between the cytosolic region of TLR7 and AP-1 , AP-2 , AP-3 , and AP-4 , we found that TLR7 interacted specifically with AP-4μ but not AP-2μ ( Figure 6E ) . We identified three potential YxxΦ motifs within TLR7 , and mutating one of these ( Tyr-892 ) disrupted the interaction with AP-4 ( Figure 6F ) . In addition , a TLR7-Y892A mutant no longer responded to TLR7 ligands ( Figure 6G ) and displayed reduced ectodomain processing when expressed in bone marrow derived macrophages , suggesting that AP-4 is required for TLR7 delivery to endosomes ( Figure 6H ) . AP-4 has been implicated in vesicular trafficking between the trans Golgi network ( TGN ) and endosomes ( Dell'Angelica et al . , 1999; Aguilar et al . , 2001; Simmen et al . , 2002; Barois and Bakke , 2005; Burgos et al . , 2010 ) . Therefore , the association between AP-4 and TLR7 suggests that TLR7 is diverted from the secretory pathway at the TGN and delivered directly to endosomes while continually associating with UNC93B1 . Thus , the mechanisms controlling trafficking of TLR7 and TLR9 to the compartments from which they signal are distinct . We next sought to determine whether the post-Golgi trafficking of other UNC93B1-dependent TLRs is similar to either of the pathways we have described for TLR7 and TLR9 . First , we tested whether TLR3 , TLR11 , and TLR13 responses were affected in UNC93B1-Y539A expressing cells . Unfortunately , both wildtype and UNC93B1-Y539A cells were unresponsive to Poly I:C , profilin , and flagellin ( data not shown ) , so we could not evaluate TLR3 or TLR11 signaling ( Alexopoulou et al . , 2001; Yarovinsky et al . , 2005; Mathur et al . , 2012 ) . However , cells expressing wildtype UNC93B1 responded robustly to the oligoribonucleotide sequence derived from Staphylococcus aureus ( hereafter referred to as Sa ORN ) that was recently shown to stimulate TLR13 ( Li and Chen , 2012; Oldenburg et al . , 2012 ) . The response of UNC93B1-Y539A cells stimulated with Sa ORN was comparable to that of wildtype cells , suggesting that TLR13 does not require the UNC93B1/AP-2 pathway to access endosomal signaling compartments ( Figure 7A ) . 10 . 7554/eLife . 00291 . 009Figure 7 . TLR11 , TLR12 , and TLR13 traffic independently of the UNC93B1/AP-2 pathway . ( A ) TLR13 signaling is unaffected by UNC93B1-Y539A . 3d iMac cells expressing UNC93B1-WT , -H412R , or -Y539A were stimulated with 1 μg/ml Pam3CSK4 ( black ) or 1 nM Sa ORN complexed with DOTAP ( grey ) and harvested for intracellular TNFα staining 5 hr after stimulation . Percentages of TNF-producing cells are plotted . ( B ) TLR13 trafficking is unaffected by UNC93B1-Y539A . TLR13-HA was immunoprecipitated from 3d iMac cells expressing UNC93B1-WT , -H412R , or -Y539A , treated with EndoH ( E ) , PNGaseF ( P ) or left untreated ( N ) , and visualized by anti-HA immunoblot . Asterisk indicates full length EndoH resistant form of the receptor . ( C ) and ( D ) TLR11 ( C ) and TLR12 ( D ) trafficking is unaffected by UNC93B1-Y539A . TLR11 and TLR12 was treated as in ( B ) . Asterisk indicates full length EndoH resistant form of the receptor . The bracket indicates the migration difference between EndoH-treated and PNGaseF-treated cleaved form of the receptors . Results are representative of at least three experiments ( A ) or two experiments ( B–D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00291 . 009 We also examined trafficking of other UNC93B1-dependent TLRs biochemically in cells expressing UNC93B1-WT , UNC93B1-H412R , or UNC93B1-Y539A . TLR13 acquired EndoH-resistant glycans , and the glycosylation was absent in UNC93B1-H412R cells but unaffected in UNC93B1-Y539A expressing cells . Moreover , the full-length EndoH-resistant form did not accumulate in these cells , as we observed for TLR9 . Together with the responses to TLR13 ligands , these results indicate that TLR13 traffics independently of the UNC93B1/AP-2 pathway . We performed similar experiments with TLR11 and TLR12 , which both recognize profilin ( Yarovinsky et al . , 2005; Koblansky et al . , 2012 ) . Both of these receptors appear to undergo ectodomain cleavage , although the sizes of the cleaved receptors are quite distinct: ∼45 kDa for TLR11 and ∼75 kDa for TLR12 ( Figure 7C , D ) . Importantly , both receptors acquired EndoH-resistant glycans , and the glycosylation was unaffected in UNC93B1-Y539A expressing cells . Thus , TLR11 and TLR12 also do not require the UNC93B1/AP-2 pathway for proper trafficking . Altogether , these results suggest that TLR9 trafficking may be unusual among the endosomal TLRs by trafficking via the plasma membrane en route to endosomes . It is important to note , though , that we have not examined TLR3 trafficking . Thus , it remains possible that TLR3 may also utilize this route , especially considering the several reports that describe TLR3 surface expression ( Qi et al . , 2010 , 2012; Pohar et al . , 2012; Weber et al . , 2012 ) . We have shown that UNC93B1 associates with both the full-length and cleaved forms of TLR9 and TLR7 ( Figures 3E , 6C ) ; however , the differential post-Golgi trafficking routes taken by TLR7 and TLR9 suggest that UNC93B1 must associate with TLR7 and TLR9 in separate complexes . To test this possibility directly , we examined interactions between TLR7 , TLR9 , and UNC93B1 by immunoprecipitation . While both TLR7 and TLR9 could precipitate UNC93B1 , we could never detect interactions between the two receptors ( Figure 8 ) . Thus , UNC93B1 binding to TLR7 appears to preclude binding to TLR9 and vice versa . These data support the idea that endosomal TLRs compete for binding to UNC93B1 in order to exit from the ER ( Wang et al . , 2006; Fukui et al . , 2009 , 2011 ) . 10 . 7554/eLife . 00291 . 010Figure 8 . TLR7 and TLR9 association with UNC93B1 is mutually exclusive . TLR9 and TLR7 associate with UNC93B1 in distinct complexes . 3d iMac cells expressing TLR9-HA , TLR7-V5-His , and WT UNC93B1-GFP ( WT ) or no UNC93B1 ( − ) were lysed in TNT buffer conditions then incubated with anti-HA matrix or anti-V5 protein A/G beads . Immunoprecipitated proteins were analyzed by SDS-PAGE and immunoblotted with HA , GFP and V5 antibodies . Asterisk indicates remaining UNC93B1-GFP signal . Results are representative of two experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00291 . 010 The studies presented here address several poorly understood aspects of TLR cell biology . First , we describe mechanisms by which UNC93B1 can regulate multiple aspects of TLR trafficking . UNC93B1 controls exit of at least six TLRs ( TLR3 , TLR7 , TLR9 , TLR11 , TLR12 and TLR13 ) from the ER by regulating their loading into COPII vesicles . In addition , we show that UNC93B1 acts at the cell surface by recruiting clathrin AP-2 to internalize TLR9 from the plasma membrane into endosomes . Perhaps most surprisingly , our results indicate that TLR7 and TLR9 have different requirements for UNC93B1 and demonstrate that the localization of these receptors is controlled through distinct post-Golgi trafficking mechanisms . This last point may provide an explanation for the contrasting roles played by TLR7 and TLR9 in SLE . Previous work has demonstrated that TLR9 and TLR7 fail to reach endosomes in the absence of functional UNC93B1 ( Kim et al . , 2008 ) . Interactions between UNC93B1 and TLRs appear necessary for proper trafficking ( Brinkmann et al . , 2007 ) , but until our work it was unclear how UNC93B1 mediated delivery of TLRs to endosomal compartments . The most direct examination of UNC93B1 function concluded that the protein translocates with TLRs from the ER to endolysosomes without passing through the Golgi apparatus ( Brinkmann et al . , 2007; Kim et al . , 2008 ) . In contrast , we propose that UNC93B1 is required for loading of TLR9 into COPII vesicles , which direct transport from the ER to the cis-Golgi ( Zanetti et al . , 2012 ) . Multiple lines of evidence support the conclusion that UNC93B1 regulates this essential aspect of ER exit . First , UNC93B1 is detected within COPII vesicles , but the nonfunctional UNC93B1-H412R mutant is not . Second , a fraction of UNC93B1 protein acquires EndoH-resistant glycans , indicative of trafficking through the medial-Golgi . Again , the UNC93B1-H412R mutant does not acquire EndoH resistance . Third , in cells lacking UNC93B1 , TLR9 is not detected within COPII vesicles . Fourth , the EndoH-resistant precursor and cleaved forms of TLR9 are absent in cells not expressing functional UNC93B1 . Importantly , we ruled out the possibility that UNC93B1 functions as a folding chaperone by showing that CD4-TLR fusion proteins require UNC93B1 for ER exit . Altogether , these results argue that UNC93B1 regulates TLR passage through the general secretory pathway , instead of controlling direct translocation between the ER and endosomes . We cannot formally rule out that a pool of UNC93B1 uses this non-canonical route , but it would seem to have little if any relevance for TLR trafficking . The recent identification of Sec22b as a factor required for ER to endosome translocation may allow for further investigation of this possibility ( Cebrian et al . , 2011 ) . Our work suggests that at least six , and possibly seven , TLRs are regulated by UNC93B1 in mice . TLR3 , TLR7 , TLR9 , TLR11 , TLR12 and TLR13 require UNC93B1 for exit from the ER . It is likely that TLR8 also requires UNC93B1 ( Itoh et al . , 2011 ) , based on its similarity to TLR7 , but we have not yet tested this possibility . Regulation of ER exit of TLR11 , TLR12 and TLR13 by UNC93B1 is not completely unexpected , as these TLRs were previously linked to UNC93B1 ( Brinkmann et al . , 2007; Melo et al . , 2010; Pifer et al . , 2011; Koblansky et al . , 2012; Oldenburg et al . , 2012 ) . The determinants of UNC93B1 binding to each of these TLRs remain poorly defined . Residues within the first 50 amino acids of UNC93B1 appear necessary for export of TLR9 but not for export of TLR7 . This result suggests that distinct regions of UNC93B1 are required for association with TLR9 and TLR7 , although measuring how these mutations impact TLR/UNC93B1 interactions has been challenging for us and for other groups ( Fukui et al . , 2009 ) . The fact that UNC93B1-D34A-expressing mice exhibit enhanced TLR7 responses and develop SLE suggests that UNC93B1 is limiting in the ER , at least for TLR7 ( Fukui et al . , 2011 ) . Presumably , the inability of UNC93B1-D34A to interact with TLR9 results in greater TLR7 export and enhanced signaling . These results and our finding that TLR7 and TLR9 association with UNC93B1 is mutually exclusive suggest that each UNC93B1 molecule may interact with a single TLR . This specificity is particularly important in light of our findings that trafficking routes of TLR7 and TLR9 are distinct . Whether UNC93B1 is similarly selective for other TLRs and identification of the domains that mediate any selectivity are important topics for future studies . Selective export of individual TLRs may provide a mechanism for differential regulation of endosomal TLR responses between cell types or in response to external signals . Through analysis of UNC93B1 mutants we identified an additional role for UNC93B1 in directing post-Golgi trafficking of TLR9 . Unlike typical COPII loading factors , UNC93B1 remains associated with its cargo , traffics with TLR9 to the cell surface , and associates with AP-2 via a YXXΦ motif ( Figure 9 ) . Recruitment of AP-2 is necessary for internalization of TLR9 and subsequent trafficking to endolysosomes . This AP-2-dependent trafficking route has been described for several proteins localized to endolysosomes , including LAMP-1 , LAMP-2 , and MHC Class II ( Gough et al . , 1999; Janvier and Bonifacino , 2005; McCormick et al . , 2005 ) . In some cases , the same protein can access endocytic compartments through multiple routes . For example , LAMP-1/2 can also traffic directly from the TGN to endosomes ( Karlsson and Carlsson , 1998 ) . While we cannot rule out that TLR9 may reach endolysosomes through multiple routes , AP-2-mediated internalization appears to be the main pathway of delivery , at least in the cells types we have examined . It is interesting that TLR9 and UNC93B1 would have evolved such dependency on this route , especially considering the potential for self-DNA recognition associated with surface localization of TLR9 ( Barton et al . , 2006; Mouchess et al . , 2011 ) . Indeed , our results appear to underscore why the requirement for ectodomain processing is a critical mechanism ensuring that receptors at the plasma membrane remain nonfunctional ( Ewald et al . , 2008; Mouchess et al . , 2011 ) . 10 . 7554/eLife . 00291 . 011Figure 9 . Trafficking pathways controlling localization of endosomal TLRs . UNC93B1 interacts with several TLRs in the ER and facilitates loading into COPII vesicles . Unlike typical COPII loading factors , UNC93B1 remains associated with TLR9 and TLR7 after exit from the ER . Through its recruitment of AP-2 , UNC93B1 is necessary for endocytosis of TLR9 from the plasma membrane into endosomes . TLR7 does not rely on this trafficking route . Instead , TLR7 utilizes AP-4 to bypass the cell surface and traffic directly to endosomes . This difference in trafficking may result in TLR9 and TLR7 accessing distinct compartments with unique functional properties related to the function of each receptor . DOI: http://dx . doi . org/10 . 7554/eLife . 00291 . 011 One of the most exciting aspects of our study is the observation that trafficking of TLR7 and TLR9 are distinct . We find no evidence that TLR7 requires UNC93B1 for recruitment of AP-2 . Instead , TLR7 appears to employ AP-4 in a direct route of traffic from the TGN to the endosome . These results provide the first evidence that different pathways control TLR7 and TLR9 trafficking and localization . One potential implication of this result is that these pathways may deliver TLR9 and TLR7 to distinct compartments with different access to ligands or distinct signaling properties . We find that TLR11 , TLR12 and TLR13 , like TLR7 , do not require the UNC93B1/AP-2 pathway . However , it remains to be determined whether these TLRs utilize the AP-4 pathway or another pathway to traffic from TGN to endosomes . Further compartmental specialization is generated by AP-3 , which interacts with TLR9 and directs the receptor to endosomal compartments dedicated to type I interferon signaling ( Honda et al . , 2005; Blasius et al . , 2010; Sasai et al . , 2010 ) . TLR7 may also utilize AP-3 to reach this specialized compartment , although we could not detect interaction between AP-3 and TLR7 ( Figure 6E ) . Whether there is transport of TLRs between each of the compartments serviced by AP-2 , AP-3 , and AP-4 remains an open question . The use of distinct molecular pathways to regulate endosomal TLRs may allow for differential regulation of trafficking , either in response to external cues or between different cell types . In addition , TLR7- and TLR9-containing compartments may have differing abilities to access internalized ligands , influencing responses to microbial or self ligands . These possibilities are particularly intriguing when considering the contrasting roles played by TLR7 and TLR9 in SLE , where loss of TLR7 protects against disease while loss of TLR9 exacerbates disease ( Christensen et al . , 2006 ) . Our findings raise the possibility that distinct cell biological regulation may underlie the different roles played by these receptors in autoimmune disease . Defining the mechanisms underlying this regulation may help explain the etiology of certain autoimmune diseases as well as provide opportunities to selectively manipulate distinct aspects of TLR activation . The following antibodies were used for immunoblots , immunoprecipitations , or flow cytometry: anti-HA as purified antibody or matrix ( 3F10; Roche , Indianapolis , IN ) , anti-FLAG as purified antibody or matrix ( M2 and M5; Sigma-Aldrich , St . Louis , MO ) , anti-GFP ( JL-8; Clontech Laboratories , Inc , Mountain View , CA ) , anti-GFP as purified or matrix ( RQ2; MBL International Corporation , Woburn , MA ) , anti-myc ( purified mouse; Invitrogen , Grand Island , NY ) , anti-V5 ( mouse monoclonal; Invitrogen ) anti-CD4 ( RM4-5; BD Biosciences , San Jose , CA ) , anti-Lamp-1 ( 1D4B; BD Biosciences ) , anti-calnexin ( rabbit polyclonal; Enzo Life Sciences , Farmingdale , NY ) , anti-ERGIC/p58 ( rabbit ) has been previously described ( Merte et al . , 2010 ) , anti-TNFα-PE or -APC ( MP6-XT22; eBiosciences , San Diego , CA ) , anti-CD71-APC ( OKT9; eBiosciences ) , goat anti-mouse IgG-AlexaFluor647 ( Invitrogen ) , goat anti-rat-HRP , sheep anti-mouse-HRP , and donkey anti-rabbit-HRP ( GE Healthcare , Waukesha , WI ) . For immunofluorescence: rabbit anti-HA ( Y11; Santa Cruz Biotechnology , Dallas , TX ) , goat anti-mouse Cy3 ( Jackson Immunoresearch , West Grove , PA ) , goat anti-rabbit-AlexaFluor647 ( Invitrogen ) . The following TLR ligands were used to stimulate cells: CpG ODN ( TCCATGACGTTCCTGACGTT , all phosphorothioate linkages ) and Sa ORN ( ‘Sa17’; Oldenburg et al . , 2012 ) ( GACGGAAAGACCCCGUG RNA sequence purchased from Integrated DNA Technologies , San Diego , CA ) , R848 ( InvivoGen , San Diego , CA ) , and Pam3CSK4 ( InvivoGen ) . 3× FLAG peptide was purchased from Sigma-Aldrich . Digitonin was purchased from Wako Pure Chemical Industries , Ltd . ( Richmond , VA ) or Calbiochem ( Billerica , MA ) . Lipofectamine-LTX reagent ( Invitrogen ) was used for transient transfection of plasmid DNA . Lipofectamine RNAiMAX reagent ( Invitrogen ) was used for siRNA delivery . DOTAP liposomal transfection reagent ( Roche ) was used for transfection of Sa ORN in PBS . OptiMEM-I ( Invitrogen ) was used as media to form nucleic acid complexes for transient transfections . Unc93b13d/3d mice ( Tabeta et al . , 2006 ) were obtained from the MMRRC at University of California , Davis . C57Bl/6 were purchased from The Jackson Laboratory ( Bar Harbor , ME ) . All mice were housed in the animal facilities at the University of California , Berkeley according to guidelines of the Institutional Animal Care and Use Committee . Pfu Turbo polymerase ( Agilent Technologies , Santa Clara , CA ) was used according to manufacturer's instructions for site directed mutagenesis . The following mouse stem cell virus ( MSCV ) -based retroviral vectors were used to express UNC93B1 , TLR9 , and TLR7 in cell lines: MSCV2 . 2 ( IRES-GFP ) , MSCV-Thy1 . 1 ( IRES Thy1 . 1 ) , MIGR2 ( IRES-hCD2 ) . The following epitope tags were fused to the C-terminus of UNC93B1: 3× FLAG ( DYKDHDGDYKDHDIDYKDDDDK ) , Myc ( EQKLISEEDL ) , HA ( YPYDVPDYA ) and entire eGFP cDNA derived from pIRES-eGFP plasmid ( Clontech ) . TLR9 was fused to HA at the C-terminal end or with 3× FLAG at the N-terminal end as previously described ( Ewald et al . , 2008; Mouchess et al . , 2011 ) . TLR7 sequence was synthesized after codon optimization by Invitrogen's GeneArt Gene Synthesis service and cloned into same vectors as TLR9 and tagged with C-terminal HA or C-terminal V5-His ( GKPIPNPLLGLDST-HHHHHH ) . TLR11 , TLR12 , TLR13 was C-terminally tagged with HA and cloned into MSCV2 . 2 . UNC93B1 shRNA and control were generated in MSCV-Lmp , as previously described ( Ewald et al . , 2008 ) . CD4-TLR chimeras were generated in pCDNA3 . 1 ( Invitrogen ) . CD4 extracellular domain ( mouse 1–390 a . a . ) was fused to transmembrane domain and cytosolic regions of the following TLRs and C-terminally tagged with HA: TLR4 ( 620–835 a . a . ) , TLR9 ( mouse 803–1032 a . a . ) , TLR3 ( human 691–904 a . a . ) , TLR7 ( human 825–1049 ) , TLR11 ( mouse 703–926 a . a . ) , TLR13 ( mouse 770–991 a . a . ) . Rat AP-2μ-HA containing an internal HA tag in pCDNA3 was provided by A . Sorkin ( while at the University of California , San Diego , CA , now at the University of Pittsburgh , PA ) ( Nesterov et al . , 1999 ) . For yeast-two-hybrid assays , N-terminal ( 1–59 a . a . ) and C-terminal ( 515–598 a . a . ) cytosolic regions of UNC93B1 were fused to Gal4 DNA binding domain ( DBD ) by cloning into pGBT9 ( Clontech ) . TLR7 cytosolic region ( mouse 862–1061 a . a . ) , TLR9 cytosolic region ( mouse 838–1032 a . a . ) , fused to Gal4-DBD by cloning into pGBT9 . AP-1Aμ , AP-2μ , AP-3Aμ , AP-3Bμ , AP-4μ were cloned into pACT2 ( Clontech ) were provided by J . Bonifacino ( National Institutes of Health , Bethesda , MD ) . HEK293T cells were obtained from American Type Culture Collection ( ATCC , Manassas , VA ) . GP2-293 packaging cell lines were obtained from Clontech . Phoenix-Eco ( ØNX-E ) cells were provided by G . Nolan ( Stanford University , Palo Alto , CA ) . Mouse embryonic fibroblasts ( MEFs ) are TLR2/TLR4 double knockout genotype immortalized with SV40 large T-antigen . COS7 were obtained from the Berkeley cell culture facility . The above cell lines were cultured in DMEM supplemented with 10% ( vol/vol ) FCS , L-glutamine , penicillin-streptomycin , sodium pyruvate , and HEPES ( pH 7 . 2 ) ( Invitrogen ) . RAW264 macrophage cell lines ( ATCC ) and immortalized macrophages ( generated as described below ) were cultured in RPMI 1640 ( same supplements as above ) . To generate immortalized macrophage cell lines , bone marrow from Unc93b13d/3d mice was cultured in RPMI 1640 media supplemented with supernatant containing M-CSF , as previously described ( Arpaia et al . , 2011 ) , as well as virus encoding both v-raf and v-myc ( Blasi et al . , 1985 ) . After 8 days , macrophages were removed from M-CSF-containing media and cultured in RPMI 1640 media with added supplements as described above . For retroviral transduction of immortalized macrophages , VSV-G-psuedotyped retrovirus was made in GP2-293 packaging cells ( Clontech ) . GP2-293 cells were transfected with retroviral vectors and pVSV-G using Lipofectamine LTX reagent . 24 hr post-transfection , cells were incubated at 32°C . 48 hr post-transfection viral supernatant ( with polybrene at final 5 μg/ml ) and was used to infect target cells overnight at 32°C and protein expression was checked 48 hr later . Target cells were sorted on MoFlo Beckman Coulter Sorter to match expression . For retroviral transduction of bone marrow derived macrophages , retrovirus was produced with the ØNX-E packaging line . Bone marrow cells were transduced with viral supernatant on two successive days while cultured in M-CSF containing RPMI media until harvested on day 8 . Activation of NF-κB in HEK293T cells was performed as previously described ( Ewald et al . , 2008 ) . Briefly , transfections were performed in OptiMEM-I ( Invitrogen ) with LTX transfection reagent ( Invitrogen ) according to manufacturer's guidelines . Cells were stimulated with 1–10 µg/ml R848 after 24 hr and lysed by passive lysis after an additional 12–16 hr . Luciferase activity was measured on a LMaxII-384 luminometer ( Molecular Devices , Sunnyvale , CA ) . Cell lysates were prepared with TNT buffer ( 20 mM Tris [pH 8 . 0] , 200 mM NaCl , 1% Triton X-100 , 4 mM EDTA and supplemented with EDTA-free complete protease inhibitor cocktail; Roche ) unless otherwise noted . Digitonin lysis buffer ( 50 mM Tris [pH 7 . 4] , 150 mM NaCl , 5 mM EDTA [pH 8 . 0] , 1% Digitonin added fresh and supplemented with EDTA-free complete protease inhibitor cocktail ) was used for co-immunoprecipitations . Lysates were cleared of insoluble material by centrifugation . For immunoprecipitations , lysates were incubated with anti-HA matrix , anti-FLAG matrix , anti-GFP matrix or with purified antibody conjugated to Protein A/G beads ( ThermoFisher Pierce , Rockford , IL ) and precipitated proteins were denatured in SDS-PAGE buffer separated by SDS-PAGE ( Tris–HCl self cast gels or Bio-Rad TGX precast gels [Bio-Rad , Hercules , CA] ) , and probed by the indicated antibodies . For anti-FLAG matrix immunoprecipitations , 3× FLAG peptide ( Sigma-Aldrich ) was used to elute . Immunoprecpitated proteins or total lysate were denatured and treated with Endoglycosidase H or PNGase F according to manufacturer's instructions . All enzymes and buffers were purchased from New England Biolabs ( Ipswich , MA ) . To measure TNFα production , we added brefeldinA to cells 30 min after stimulation , and cells were collected after an additional 4 hr , and cells were stained for intracellular cytokines with a Fixation & Permeabilization kit according to manufacturer's instructions ( eBioscience ) . For FLAG-TLR surface expression , HEK293T cells stably expressing N-terminally tagged 3× FLAG-TLR9 were stained with anti-FLAG ( M5; Sigma-Aldrich ) antibody followed by Alexa 647 goat anti-mouse IgG secondary antibody ( Invitrogen ) . All data were collected on LSR II ( Becton Dickinson , Franklin Lakes , NJ ) or FC-500 ( Beckman Coulter , Indianapolis , IN ) flow cytometers and were analyzed with FloJo software ( TreeStar , Inc . Ashland , OR ) . Co-localization studies were performed on Leica TCS confocal microscope . The images were taken with a 40× oil immersion objective and treated with 2× digital zoom . All images were processed by Adobe Photoshop . Cells were allowed to settle overnight on coverslips . Coverslips were washed with PBS , fixed with 4% paraformaldehyde/PBS , and permeabilized with 0 . 5% saponin/PBS . Slides were incubated in freshly made 0 . 1% sodium borohydride 0 . 1% saponin before being stained in 1% Bovine Serum Albumin ( Fisher Scientific , Pittsburgh , PA ) /0 . 1% saponin in PBS with rabbit anti-HA ( Santa Cruz ) , rat anti-Lamp-1 ( BD Biosciences ) then with secondary antibodies , goat anti-Rabbit 647 ( Invitrogen ) and goat anti-Rat Cy3 ( Jackson Immunoresearch ) . RAW264 cells were used to isolate phagosomes as previously described ( Ewald et al . , 2008 ) . Briefly , cells were incubated with 2 μM latex beads ( Polysciences , Inc . , Warrington , PA ) for 1 hr . Cells were disrupted by dounce homogenization to release intact phagosomes . Following centrifugation in sucrose step gradient , phagosomes were harvested from the 20–10% sucrose interface . Lysates were analyzed by immunoblot . 3 . 5 × 105 HEK293Ts were plated in 2 ml antibiotic free media per well in six-well plates and reverse transfected with 5 μl 20 μM siRNA in 500 μl OptiMEM-I and 5 μl of Lipofectamine RNAiMAX for 48–96 hr until harvest for flow cytometry or SDS-PAGE and immunoblot analysis . siRNA duplexes against human AP-2μ were purchased from ThermoFisher Dharmacon RNAi Technologies ( Waltham , MA ) with the following sequence: 5′-GGAGAACAGUUCUUGCGGC-3′ and with the following conditions: ON-TARGET - Enhanced Antisense Loading , Standard ( A4 ) , UU added to 3′ end . Control siRNA ( ON-TARGETplus Non-targeting siRNA #1 ) was purchased from Dharmacon . Cells of an overnight culture ( 2 . 5 ml ) of yeast strain PJ69-4a in YPD media grown at 30°C were washed in H2O and mixed sequentially with the following: 50% PEG-3500 ( Sigma-Aldrich ) , 10 mM lithium acetate in Tris-EDTA ( TE ) ( pH 7 . 5 ) , denatured salmon sperm DNA ( Invitrogen ) and 100 ng of plasmid constructs , then mixed . Yeast were incubated for 30 min at 30°C and heat-shocked for 15 min at 42°C . Cells were centrifuged , resuspended in H2O , and spread on YNB plates with –Trp –Leu dropout mix . Liquid cultures in Trp-Leu broth were grown at 30°C overnight . Cells were normalized to 1 . 0 OD and 1:10 dilutions made . 4 μl of each dilution was plated on –Trp–Leu–His , and –Trp–Leu–Ade at 30°C . YPD , YNB and dropout mixtures purchased from Sunrise Science . Growth on –Trp–Leu–His or –Trp–Leu–Ade plates was recorded at day 3 . COPII vesicle formation was performed as described previously ( Kim et al . , 2005; Merte et al . , 2010 ) . In brief , RAW264 cells grown in 10 × 10-mm plates or COS7 cells grown in 6 × 100-mm plates were washed in PBS , removed from plates with trypsin , and washed again in PBS containing 10 μg/ml soybean trypsin inhibitor . Cells were permeablized with 40 μg/ml digitonin for 5 min in ice-cold KHM buffer ( 110 mM KOAc , 20 mM Hepes pH 7 . 2 and 2 mM Mg ( OAC ) 2 ) and washed and resuspended in 100 μl KHM . Each reaction contained KHM and where indicated an ATP regenerating system ( 40 mm creatine phosphate , 0 . 2 mg/ml creatine phosphokinase , and 1 mm ATP ) , 0 . 2 mm GTP , and rat liver cytosol ( prepared as described previously; Kim et al . , 2005 ) . Reactions were incubated at 30°C for 60 min . A 75-μl aliquot of the vesicle fraction was separated from the donor microsomal fraction by centrifugation at 14 , 000×g for 20 min at 4°C . Donor fraction was lysed in 75 μl of Buffer C ( 10 mm Tris–HCl [pH 7 . 6] , 100 mm NaCl , 10% [w/v] SDS plus protease inhibitor mixture ) . The vesicles were collected by centrifugation at 50 , 000 rpm at 4°C in a Beckman TLA100 rotor for 30 min . Isolated vesicles were lysed in 20 μl of Buffer C . Donor membrane ( 20% total ) and isolated vesicles ( 75% of total ) were separated by SDS-PAGE and analyzed by immunoblotting .	Toll-like receptors ( TLRs ) are proteins that are responsible for recognizing specific molecules associated with invading pathogens , known as pathogen-associated molecular patterns . Upon detecting these signals , TLRs activate the body's immune response , which fights the infection . A subset of TLRs recognizes nucleic acids , including DNA and RNA , enabling the immune system to respond to foreign material from a diverse range of bacteria and viruses . However , some of the body's own DNA and RNA is also found outside cells ( e . g . , in the bloodstream ) and TLRs must be able to discriminate between these nucleic acids and those belonging to pathogens , because failure to tell the difference between the two could result in autoimmune disease . To reduce this risk , TLRs are sequestered inside the cell within membrane-bound compartments known as endosomes . UNC93B1 is a transmembrane protein that is known to control the movement of TLRs from the endoplasmic reticulum—where TLRs are assembled—to endosomes . However , the exact mechanisms by which this protein controls TLR trafficking were unclear . Now Lee et al . reveal that it directly controls the packaging of at least six TLRs at the endoplasmic reticulum: it helps to load these TLRs into vesicles , which are in turn processed by the Golgi apparatus—the organelle wherein proteins are sorted and packaged en route to their final destinations . Surprisingly , UNC93B1 remains associated with the TLRs even after Golgi processing . Lee et al . also reveal that specific endosomal TLRs are subject to distinct post-Golgi trafficking mechanisms . In order for TLR9 to be delivered to the endosome , UNC93B1 must recruit an adaptor protein called AP-2 , whereas other TLRs appear to require different actions by UNC93B1 . By defining the mechanisms that underlie the differential trafficking of endosomal TLRs , Lee et al . suggest that we may learn how to manipulate distinct aspects of TLR activation , and also gain insights into the causes of certain autoimmune diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "immunology", "and", "inflammation" ]	2013	UNC93B1 mediates differential trafficking of endosomal TLRs
Understanding how fungi specialize on their plant host is crucial for developing sustainable disease control . A traditional , centuries-old rice agro-system of the Yuanyang terraces was used as a model to show that virulence effectors of the rice blast fungus Magnaporthe oryzaeh play a key role in its specialization on locally grown indica or japonica local rice subspecies . Our results have indicated that major differences in several components of basal immunity and effector-triggered immunity of the japonica and indica rice varieties are associated with specialization of M . oryzae . These differences thus play a key role in determining M . oryzae host specificity and may limit the spread of the pathogen within the Yuanyang agro-system . Specifically , the AVR-Pia effector has been identified as a possible determinant of the specialization of M . oryzae to local japonica rice . Understanding the mechanisms determining host range of plant pathogens is crucial for disease management strategies , phytosanitary regulations and policies . The recurrent emergence of new pathogen lineages specialized to novel plant species or newly bred resistant varieties is a major limitation to agricultural production , and there is tremendous interest in developing sustainable strategies to prevent pathogen emergence and spread ( McDonald , 2010; Giraud et al . , 2010 ) . Finding durable methods of controlling the host range of pathogens requires the understanding of the molecular and physiological determinants of pathogen variation in fitness across space and hosts ( Williams , 2010; Barrett et al . , 2008 ) . Before the advent of molecular genetic methods , classical studies in plant pathology have documented patterns of pathogen fitness on different hosts , including pathogenicity ( the capacity to infect ) and virulence ( the quantity of symptoms ) ( Johnson , 1961; Nadler , 1995; Brown , 1994 ) . Variations in pathogen fitness have been repeatedly investigated for numerous agricultural pathosystems using controlled cross-inoculation experiments or inoculation on series of differential hosts . Numerous studies have reported evidence for pathogen local adaptation , where local pathogens have a greater average fitness on their local hosts than immigrants ( Kaltz and Shykoff , 1998; Laine and Barrès , 2013 ) . Higher pathogen fitness on hosts living in the same habitat is consistent with evolutionary theory , which predicts that parasites should be ahead of their hosts in the co-evolutionary race due to their higher mutation rates , shorter generation times and huge populations sizes ( Gandon and Michalakis , 2002 ) . Trade-offs among pathogen fitness traits ( e . g . between pathogenicity and transmission success rate ) are also frequently invoked in theoretical models to explain the maintenance of variation in pathogenicity and resistance ( Brown and Tellier , 2011; Laine and Tellier , 2008 ) . However , although it is important to elucidate the origin and maintenance of variations in pathogen fitness on different hosts for developing durable means of controlling disease , most current understanding is still largely based on theoretical predictions ( Brown and Tellier , 2011; Laine and Tellier , 2008; Schulze-Lefert and Panstruga , 2011; Thrall et al . , 2015 ) . Thus there is a lack of studies investigating the molecular or physiological bases of variation in pathogen fitness across pathogen populations , especially in fungi ( Poppe et al . , 2015 ) . Our current knowledge about the genetic basis of fungal pathogen specialization determining host range is mainly based on comparative genomics and functional analyses of candidate genes . These studies revealed the pivotal role of effector proteins that are secreted during infection and target cellular processes of the host to promote infection . In plant pathogenic fungi , the most prominent class of effectors are small secreted proteins . They are believed to be mostly involved in the suppression of host immunity and in particular so-called pattern-triggered immunity activated by conserved microbial molecular patterns , such as fungal cell wall components ( Lo Presti et al . , 2015 ) . Comparative genomics have revealed distinct repertoires of effectors between related pathogens specialized on different hosts . This suggests that variation in the composition of pathogen effector repertoires contributes to variation in pathogen fitness on different hosts ( reviewed in [Schulze-Lefert and Panstruga , 2011] ) . In the case of Magnaporthe oryzae for instance , pathogenicity toward rice was correlated with the presence of certain effectors ( Chiapello et al . , 2015 ) . The role of variation in pathogen effector repertoires in pathogen specialization is supported by the fact that dispensable , lineage-specific chromosomes containing effectors appear to control adaptation to hosts in a number of fungal plant pathogens ( e . g . [Ma et al . , 2010] ) . In the rice blast fungus M . oryzae , the role of effectors in specialization is supported by circumstantial evidence stemming from the comparisons of isolates specialized to rice and Setaria millet . While the effector-coding gene AVR1-CO39 was absent from rice-infecting isolates , transgenic expression of AVR1-CO39 rendered rice isolates non-pathogenic onto rice carrying the appropriate immune receptor ( Couch et al . , 2005 ) . Similar situations have been reported for the PWL2 gene that prevents pathogenicity on weeping lovegrass ( Sweigard et al . , 1995 ) and the Pwt3 and Pwt4 genes that prevent pathogenicity on wheat ( Takabayashi et al . , 2002 ) . Thus , whereas the role of effectors in pathogenicity has been demonstrated in several cases ( for review [Asai and Shirasu , 2015] ) , their role in variations in host range among pathogen populations remains largely unknown . In some cases , certain effectors can be recognized by plant immune receptors ( often called resistance proteins ) , leading to the activation of the so-called effector-triggered immunity . The effectors revealed by their activity rendering some isolates non-pathogenic ( ‘avirulent’ ) on some hosts represent a sub-category later called Avr-effectors ( Jones and Dangl , 2006 ) . The strong and specific resistance level conferred by effector-triggered immunity contrasts with basal immunity which is weak and not specific and relies on a combination of different mechanisms like constitutive expression of defense genes and pattern-triggered immunity ( Lo Presti et al . , 2015; Vergne et al . , 2010 ) . The respective roles of effector-triggered and basal immune responses in pathogen host range variations have yet to be investigated . Rice blast caused by M . oryzae is currently the most damaging rice disease worldwide , occurring on all cultivated subspecies and varietal types of rice . Four major lineages of M . oryzae causing rice blast can be distinguished on a worldwide scale ( Saleh et al . , 2014 ) . The rice - M . oryzae pathosystem is particularly well-suited for studying specialization to the host since a large number of effectors and resistance ( Pi ) genes coding for immune receptors have been cloned and basal immunity is now well-understood ( Azizi et al . , 2016; Liu et al . , 2014 ) . Moreover , large-scale cross-inoculation experiments of a collection of rice varieties with a collection of rice blast samples representing the worldwide diversity revealed patterns of pathogen fitness that suggest the existence of specialization to hosts in this pathosystem ( Gallet et al . , 2016 ) . Strains originating from japonica rice infected most japonica varieties but could not infect indica varieties whereas strains derived from indica rice infected indica and japonica varieties in controlled conditions . Inoculation onto varieties containing different Pi resistance genes suggested an important role of these genes in the observed patterns of pathogenicity . This is due to the fact that strains originating from japonica hosts were able to overcome less resistance genes than strains originating from indica hosts . However , conclusions regarding the determinants of host range and pathogen specialization were hindered by the fact that the plants and fungal isolates tested had not been collected at the same sites , and were not actively involved in co-evolutionary interactions . In this study , we investigated the molecular basis of M . oryzae specialization to its hosts: rice subspecies japonica and indica . To investigate the mechanisms of specialization in M . oryzae populations actively co-evolving with their hosts , a traditional agro-system from the Yuanyang terraces ( Yunnan , China; http://whc . unesco . org/en/list/1111/ ) where indica and japonica rice varieties have been grown side-by-side for several centuries ( He , 2011 ) was used . First , we showed that pathogen populations are specialized to indica and japonica rice varieties . Next , we investigated the role of plant immunity in shaping variations in pathogen fitness and the contribution of effectors to this pattern . We discovered that specialization of M . oryzae isolates to japonica and indica varieties grown in Yuanyang is correlated with , respectively , the deployment of a large number of Avr-effectors in japonica-borne isolates ( i . e . effectors triggering complete resistance in some plant genotypes ) and a large depletion of Avr-effectors in indica-borne isolates . These contrasting effector repertoires mirror the significant immunity differences between japonica and indica local varieties . We provide further evidence that the AVR-Pia effector is possibly a key player in the pattern of specialization to indica or japonica rice varieties observed . A total of 214 M . oryzae isolates were collected from rice plants ( Oryza sativa ) between 2009 and 2013 in the Yuanyang terraces where the majority of cultivated rice belongs to the indica sub-species ( 98% of the ~1000 ha; data from local station ) . Isolates were sampled on both indica and japonica sub-species ( n = 177 and n = 37 , respectively ) and genotyped using 13 microsatellites . Over this period , the two indica Acuce and Xiao Gu and the two japonica Huang Pi Nuo and Nuo Gu represented the most commonly grown varieties in Yuanyang . Neighbor-joining analysis of genetic distances ( Figure 1A ) combined with DAPC ( Figure 1—figure supplement 1 ) circumscribed a single group representing the vast majority ( 92% ) of isolates collected on japonica rice ( half on Huang Pi Nuo and one quarter on Nuo Gu varieties ) and therefore referred to as the ‘japonica-borne’ ( JB ) group . Conversely , 94% of isolates collected on indica rice were not assigned to the JB group , forming four main clusters collectively referred to as the ‘indica-borne’ ( IB ) group ( Figure 1A; Figure 1—figure supplement 1 ) . Population genetic analysis based on linkage disequilibrium did not support the existence of regular sexual reproduction ( Figure 1—figure supplement 2 ) . No clear pattern of association was found between the making up of pathogen clusters and other possible structuring factors such as the year of sampling or altitude ( Figure 1—figure supplement 3 ) . Thus the rice subspecies appear to be the most potent factor structuring pathogen populations in this agro-system . 10 . 7554/eLife . 19377 . 003Figure 1 . Variability in microsatellite genotype and pathogenicity phenotype of M . oryzae isolates harvested on indica and japonica rice grown in Yuanyang terraces . ( A ) Midpoint rooted neighbor-joining dendrogram representing the proportion of shared microsatellite alleles among multilocus genotypes . Two hundred fourteen isolates ( the prefix ‘CH’ visible in C was removed from A for clarity ) were genotyped using 13 microsatellites . Only one representative of multilocus genotypes repeated multiple times was kept , and for each repeated multilocus genotype the corresponding isolates are listed at the tip of a branch ( 74 unique multilocus genotypes in total ) . Bootstrap supports are indicated by a black dot when >40% ( 1000 resamplings ) . Isolates harvested on japonica and indica rice are indicated in black and grey respectively . Six isolates ( i . e . CH1180 , CH1189 , CH1195 , and CH1317 , collected on indica; CH1208 and CH1301 , collected on japonica ) that show in the dendrogram an intermediate position , were assigned to the IB group because of their pathogenicity phenotypes and their separation from the other JB genotypes in the DAPC ( Figure 1—figure supplement 1 ) . The isolates selected for ( C ) and Figure 3 are marked by a ‘’ and a ‘+’ respectively . The cluster of Japonica-borne isolates ( abbreviated JB; within the square with dashed lines ) was defined following Discriminant Analysis of Principal Components; the remaining samples represent the cluster of Indica-borne ( IB ) isolates . ( B ) Distribution of the two clusters identified based on microsatellite variation ( japonica borne ‘JB’ and indica borne ‘IB’ ) on japonica and indica hosts in the Yuanyang terraces . The fact that the distributions are largely non-overlapping ( the JB and IB clusters are mostly found on Japonica and Indica hosts , respectively ) suggests local adaptation of the pathogens to their respective hosts . The numbers of isolates are indicated above the bars . ( C ) Pathogenicity profiles of 30 isolates on two japonica and five indica varieties ( ‘R’ and ‘S’ stand for resistance and susceptibility , respectively ) . The 30 representative isolates were selected from the analysis presented in ( A ) and inoculated onto seven major rice varieties grown in Yuanyang ( HPN: Huang Pi Nuo; NG: Nuo Guo; ZN: Zi Nuo; XG: Xiao Gu; LJG: Li Jiao Gu; AZG: Ai Zhe Gu; Acuce ) . HPN , NG and ZN are all glutinous rice varieties . The isolates marked with a ‘+’ in ( A ) and ( C ) are those used in Figure 3; all isolates included in ( C ) are marked with a ‘’ in ( A ) . The qualitative analysis of symptoms presented here suggests that japonica-borne ( JB ) isolates cannot attack indica rice whereas indica-borne ( IB ) isolates can attack japonica . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 00310 . 7554/eLife . 19377 . 004Figure 1—source data 1 . The data relates to Figure 1 . For each of the 215 Magnaporthe oryzae isolate , the name of the variety on which it was collected is indicated and the rice subspecies is indicated ( indica or japonica ) . The ‘pyrm’ columns represent microsatellites names from Saleh et al . ( 2014 ) . Other informations like GPS position , altitude and town where the isolates were collected are also indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 00410 . 7554/eLife . 19377 . 005Figure 1—figure supplement 1 . Neighbor-joining tree representing the genetic distance ( in terms of proportion of shared alleles ) between the 74 unique microsatellite genotypes characterized in Yuanyang ( left ) and patterns of memberships in K = 2 to K = 10 clusters as inferred using DAPC ( right ) . The isolates selected for Figure 1C and Figure 3 are marked by a ‘’ and a ‘+’ respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 00510 . 7554/eLife . 19377 . 006Figure 1—figure supplement 2 . Summary statistics of genetic variability in the four clusters of Magnaporthe oryzae identified using Discriminant Analysis of Principal Components and neighbor-joining analysis of genetic distance ( see Figure 1—figure supplement 1 ) . G:N is the number of unique genotypes , divided by the total number of genotypes . rd is the standardized index of association a measure of multilocus linkage disequilibrium , computed on clone corrected datasets ( i . e . keeping a single representative for each multilocus genotypes represented multiple times ) ; significance was determined using 1000 randomizations to simulate random mating . H is the unbiased gene diversity , averaged across loci . *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 00610 . 7554/eLife . 19377 . 007Figure 1—figure supplement 3 . Midpoint rooted neighbor-joining dendrogram representing the proportion of shared microsatellite alleles among the 214 multilocus genotypes originating from Yuanyang terraces . The topology is different from Figure 1A , which was based on unique multilocus genotype only , while this analysis was based on the full set of isolates . Isolates were colored according to their collection year ( left-hand side panel ) or the collection altitude ( right-hand side panel ) , showing no clear pattern of clustering of isolates according to these factors . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 00710 . 7554/eLife . 19377 . 008Figure 1—figure supplement 4 . Scale used for scoring global incompatibility/compatibility ( Resistance/Susceptibility ) . The small brown lesions are indicative of infection sites were the fungus was stopped ( Hypersensitive-response ) whereas large and greyish lesions represent successful penetration and multiplication of the pathogen . This notation system is the same than in Gallet et al ( Gallet et al . , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 008 We used cross-inoculation assays in controlled conditions to evaluate how far pathogen specialization to host subspecies contributed to the observed pattern of pathogen population subdivision between JB and IB groups . Thirty representative isolates were selected ( marked with a ‘’ on Figure 1 ) and inoculated onto the varieties most frequently cultivated in the Yuanyang terraces to evaluate qualitatively cross-pathogenicity using a classification of either resistant or susceptible plant ( Figure 1—figure supplement 4 ) . Japonica rice were susceptible to all isolates from the JB group but only few instances of susceptibility of indica varieties were found with these isolates ( Figure 1C ) . By contrast , both indica and japonica varieties were susceptible to most IB isolates ( see below ) . Altogether , these results suggest that JB and IB groups of M . oryzae are differentially adapted to japonica and indica varieties grown in Yuanyang . We investigated the factors preventing infection of indica varieties by JB isolates in both field ( Figure 1B ) and controlled conditions ( Figure 1C ) . We first evaluated the role of effector-triggered immunity by assessing the content in terms of major resistance ( Pi ) of Yuanyang rice varieties . We used whole-genome sequencing of four Yuanyang varieties ( Figure 2—figure supplement 1 ) and pathogenicity assays ( Figure 2—figure supplement 2 ) to allow detection of 19 cloned Pi genes . These analyses showed that the two major indica rice varieties Xiao Gu and Acuce harbored 8 and 7 Pi genes compared to 2 and 3 in the Huang Pi Nuo and Nuo Gu japonica varieties ( Figure 2A ) . In addition , the indica varieties from Yuanyang had a higher content of Pi genes compared to indica varieties grown in a non-traditional agro-system nearby ( Figure 2—figure supplement 3 ) . Interestingly , one of the most frequent Pi genes in the indica varieties from Yuanyang was the Pia gene that had a low frequency in varieties from the non-traditional agro-system . 10 . 7554/eLife . 19377 . 009Figure 2 . Evaluation of rice Pi and blast Avr-effectors genes in Yuanyang rice varieties and M . oryzae isolates . ( A ) Presence ( 1 ) /absence ( 0 ) of 19 cloned Pi gene based on sequence analysis . Pair-end reads from the four varieties were produced ( 53 to 58 million reads ) by whole-genome sequencing . The reads ( ~150 nucleotides ) were mapped on the corresponding Pi sequences using SOAPaligner ( http://soap . genomics . org . cn/soapaligner . html ) and two mismatches were allowed . Some genes ( noted 0* ) were present but contained a premature stop codon or were not functional according to inoculation tests ( Figure 2—figure supplement 2 ) . The sequences used are listed in Figure 2—figure supplement 1 . Note that when two genes are required for resistance ( Pia , Pi5 , Pik , Pik-s and Pik-m ) , the sequences of both genes were analyzed . ( B ) The number of Avr function of 14 effectors was measured ( Figure 2—figure supplement 4 ) in the JB and IB groups of 30 isolates defined in Figure 1C . Each dot represents an isolate and the median ( black bar ) is indicated . The average values between JB and IB isolates are significantly different ( p<10−7; t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 00910 . 7554/eLife . 19377 . 010Figure 2—figure supplement 1 . Estimation of the presence of Pi genes by mapping reads from whole-genome sequencing on the corresponding Pi gene sequence . The pair-end reads from four species were aligned to the reference library using SOAPaligner ( http://soap . genomics . org . cn/soapaligner . html ) , respectively , and just two mismatches were allowed . The parameters were “soap -a HH-8_AC_left_reads . fq -b HH-8_AC_right_reads . fq -D ref_library . fasta -o HH-8_AC . PE_align −2 HH-8_AC . SE_align -v 2 m 250 -x 450 r 2 p 20” ( for HH-8_AC , other were same ) . The reads which uniquely mapped to a gene were used to calculate coverage . The list below provides the reads’ statistics of each genome . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 01010 . 7554/eLife . 19377 . 011Figure 2—figure supplement 2 . Estimation of the presence of the indicated Pi genes by inoculation with GUY11 transformed with the cognate Avr-Effector . “1” is presence ( plant is resistant ) , “0” is absence of detection ( plant is susceptible ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 01110 . 7554/eLife . 19377 . 012Figure 2—figure supplement 3 . Estimation of the frequency of Pi genes in 18 ( other than those listed in Figure 2A ) indica , traditional varieties from Yuanyang terraces ( YYT ) and in 15 modern , improved indica varieties from the Shipping county ( SP; Yunnan , China ) used as a local geographical control . The presence of the Pi gene was estimated using PCR primers listed . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 01210 . 7554/eLife . 19377 . 013Figure 2—figure supplement 4 . Estimation of Avr-Effector complement using rice differential lines in 30 selected isolates from Yuanyang ( these isolates are indicated by a ‘’ on Figure 1A showing neutral diversity ) . The Resistant ( R ) and susceptible ( S ) classes were established using the scale described in Figure 1—figure supplement 4 . The isolates indicated with a ‘+’ were used for pathogenicity tests in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 013 Secondly , we tested whether JB isolates were excluded from indica hosts due to multiple Avr-effectors matching Pi genes and thus conferring ‘avirulence’ . Thirty representative isolates were selected based on the results of population structure analyses ( Figure 1C ) and inoculated on a set of rice lines diagnostic for 14 major Pi genes ( [Berruyer et al . , 2003] and references therein ) , thus allowing the identification of the 14 corresponding Avr-effectors ( Figure 2—figure supplement 4 ) . On average , isolates from the JB group had 11 . 9 Avr-effectors ( Figure 2B ) , which is significantly different from ( p<10−7 ) and almost twice as much as in the IB group with 6 . 5 Avr-effectors . These results suggest that the high content of Avr-effectors in JB isolates accounts for their lack of pathogenicity on indica varieties that contain many Pi genes . Thus multiple gene-for-gene interactions , such as the Pia/AVR-Pia interaction , could act individually or in combination to prevent disease . The type of lesions caused by IB and JB isolates on japonica rice from Yuanyang were strikingly different ( Figure 3—figure supplement 1 ) . Lesions caused by JB isolates had a drastically reduced brown halo , a phenotype that is associated with resistance and correlated with the production of reactive oxygen species ( Hayashi et al . , 2016 ) . This indicated that JB isolates may have the capacity to inhibit this important component of basal immunity ( Jones and Dangl , 2006 ) . This also suggested that IB isolates may face high basal immunity in Yuanyang japonica varieties . To test this hypothesis , we first evaluated basal immunity conferring partial protection in the absence of major resistance genes and relying on a combination of several different molecular processes such as preformed defense ( Vergne et al . , 2010; Delteil et al . , 2012 ) and pattern-triggered immunity ( Lo Presti et al . , 2015 ) . We inoculated four broadly infecting isolates that harbor few Avr-effectors ( Gallet et al . , 2016 ) to evaluate resistance under conditions that minimize effector-triggered immunity ( Vergne et al . , 2010 ) . This analysis showed that japonica varieties from Yuanyang are more resistant than indica varieties ( Figure 3—figure supplement 2 ) , comparable to the level observed in Nipponbare , which is renowned for its elevated basal immunity ( Vergne et al . , 2010 ) . To analyze preformed defense in Yuanyang japonica varieties , the constitutive expression of defense-related genes which is a hallmark of basal immunity towards the rice blast fungus ( Vergne et al . , 2010; Delteil et al . , 2012 ) , was determined . When measuring the expression of 16 PR genes and 12 genes involved in resistance signaling that are frequently co-regulated with defense genes ( see list in Figure 3—figure supplement 3 ) , more than half of the genes ( 16/28 ) and more than two thirds of the PR genes ( 11/16 ) showed higher constitutive expression in Yuanyang varieties compared to Nipponbare renowned for its elevated constitutive defense ( Vergne et al . , 2010 ) ( Figure 3A ) , with the Huang Pi Nuo variety showing the strongest constitutive expression of defense-related markers ( Figure 3—figure supplement 4 ) . In the japonica varieties , in particular Nuo Gu , as opposed to the indica varieties Acuce and Xiao Gu , the constitutive expression of six out of 16 analyzed PR genes was also significantly higher ( see Mock treatment in Figure 3B; Figure 3—figure supplement 5A ) . Thus , Yuanyang varieties displayed elevated constitutive expression of a large set of defense-related genes , with the japonica varieties Huang Pi Nuo and Nuo Gu showing the strongest levels of expression . 10 . 7554/eLife . 19377 . 014Figure 3 . Constitutive and inducible defense in Yuanyang rice varieties . The expression of defense-related genes ( see Figure 3—figure supplement 3 ) in indica ( Acuce , Xiao Gu ) and japonica ( Huang Pi Nuo and Nuo Gu ) varieties grown in Yuanyang as well as the japonica Nipponbare variety was measured by RT-qPCR . In order to make genes with different expression levels comparable , the different values obtained for each gene were normalized by the average value of the considered gene across all measures . ( A ) The constitutive expression of 16 Pathogenesis-related genes and 11 genes involved in basal immunity signaling was measured by RT-qPCR . The mean values from four biological replicates was normalized and used for hierarchical clustering using hcluster algorithm ( www . omicshare . com/tools ) . The corresponding mean , SD and statistical tests can be found in Figure 3—figure supplement 4 . ( B ) The two major indica ( Acuce and Xiao Gu ) and japonica varieties found in Yuanyang ( Huang Pi Nuo and Nuo Gu ) were inoculated with the virulent isolate Guy11 or mock treated . The expression of 16 Pathogenesis Related-genes was measured and the mean from four biological replicates was calculated . Each dot represents the average expression value from 16 defense genes . The black bar represents the median value . The corresponding mean , SD and statistical tests can be found in Figure 3—figure supplement 5 . ( C ) The expression of 16 Pathogenesis-related genes was measured 24 hr after chitin ( 100 μg/mL ) treatment . The mean value from four biological replicates was calculated and each dot represents this value for one gene . The corresponding mean , SD and statistical tests can be found in Figure 3—figure supplement 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 01410 . 7554/eLife . 19377 . 015Figure 3—source data 1 . The data relates to Figure 3 . The expression values of defense-related genes ( columns ) normalized by the Actin gene are given . All values were also normalized using the mean of each gene in order to make all genes comparable with each other . In the first columns , the first element represents the treatment ( chi=chitin ) , the second the variety , the third the time after treatment and the last the replicate number . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 01510 . 7554/eLife . 19377 . 016Figure 3—source data 2 . The data relates to Figure 3—figure supplement 2 . Columns 2-5 represent susceptibility to the corresponding M . oryzae isolate ( CD203 , CM28 , CL26 , and GY11 ) . The data are then normalized by the value in the Maratelli ( used as universal susceptible control ) variety to allow comparisons between isolates . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 01610 . 7554/eLife . 19377 . 017Figure 3—source data 3 . The data relates to Figure 3—figure supplement 4 . The expression of defense-related genes was evaluated by RT-qPCR and the data normalized by the actin gene is given . Each condition was replicated 3–4 times ( column C ) . AC=Acuce , XG=Xiao Gu , HPN=Huang Pi Nuo , NG=Nuo Gu are rice varities . CL26 ( A ) and CM28 ( B ) are isolates of Magnaporthe oryzae isolates . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 01710 . 7554/eLife . 19377 . 018Figure 3—figure supplement 1 . Examples of symptoms on Huang Pi Nuo . The red and blue arrows indicate typical susceptible and resistance symptoms respectively . The yellow arrow shows a susceptible lesion with surrounding browning , a phenomenon associated with a local resistance response ( Hayashi et al . , 2016 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 01810 . 7554/eLife . 19377 . 019Figure 3—figure supplement 2 . Average susceptibility of Yuanyang terraces varieties . Four multi-virulent isolates ( CD203 , GUY11 , CL26 and CM28 [Vergne et al . , 2010] ) were used to evaluate susceptibility in the absence of major Avr/Pi interactions . The number of susceptible lesions/unit surface was measured for each isolate X variety combination in four replicates of 6 leaves . The median value of four measures is indicated . Basal immunity can be extrapolated considering that it is inversely correlated to global susceptibility . Acuce , Xiao Gu are indica varieties . Huang Pi Nuo , Nuo Gu and Nipponbare are japonica varieties . Letters above indicate significantly different groups of values based on T-test ( p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 01910 . 7554/eLife . 19377 . 020Figure 3—figure supplement 3 . Accessions and names of genes used for expression analysis . The primers can be found in ( Vergne et al . , 2010 ) and ( Delteil et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 02010 . 7554/eLife . 19377 . 021Figure 3—figure supplement 4 . Constitutive expression of defense-related genes in Yuanyang japonica and Nipponbare varieties . The japonica variety Nipponbare with high basal immunity was compared to the Yuanyang japonica varieties Huang Pi Nuo and Nuo Gu . The constitutive expression of 16 pathogenesis-related genes and 11 genes involved in basal immunity signaling was measured by RT-qPCR . In order to compare all genes together , all values for one gene were normalized with the average value of the considered gene for all measures . The values are the means and SD from four biological replicates . The ‘’ denote statistical differences ( t-test; p<0 . 05 ) between Nipponbare and the considered variety . A red and green dot indicates that Nipponbare expression is lower and higher than either Huang Pi Nuo or Nuo Gu respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 02110 . 7554/eLife . 19377 . 022Figure 3—figure supplement 5 . Constitutive and fungal-induced expression of pathogenesis-related genes in Yuanyang japonica and indica varieties . The expression of 16 pathogenesis-related genes was measured by RT-qPCR in the absence of treatment . ( A ) or 24 after inoculation by M . oryzae ( isolate Guy11 ) . In order to compare all genes together , all values for one gene were normalized with the average value of the considered gene for all measures . The values are the means and SD from four biological replicates . The japonica variety Nuo Gu was compared to the Yuanyang indica varieties Acuce and Xiao Gu . The ‘’ denotes statistical differences ( t-test; p<0 . 05 ) between Nuo Gu and the considered variety . A red and green dot indicates that Nuo Gu expression is higher and lower than either Acuce or Xiao Gu . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 02210 . 7554/eLife . 19377 . 023Figure 3—figure supplement 6 . Constitutive and fungal-induced expression of pathogenesis-related genes in Yuanyang japonica and indica varieties . The two major indica ( Acuce and Xiao Gu ) and japonica ( Huang Pi Nuo and Nuo Gu ) varieties found in Yuanyang were mock treated or inoculated with the virulent isolates CL26 ( A ) and CM28 ( B ) . The expression of Pathogenesis-Related genes was measured by RT-qPCR and the mean from four biological replicates was calculated . Each dot represents the average expression value of 16 Pathogenesis Related-genes . The black bar represents the median value . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 02310 . 7554/eLife . 19377 . 024Figure 3—figure supplement 7 . Chitin-induced expression of defense-related genes in Yuanyang japonica and Nipponbare varieties . The japonica variety Nipponbare with high basal immunity was compared to the Yuanyang japonica varieties Huang Pi Nuo and Nuo Gu . The expression of 16 pathogenesis-related genes and 11 genes involved in basal immunity signaling was measured by RT-qPCR 24 hr after chitin treatment ( 100 ug/mL ) . In order to compare all genes together , all values for one gene were normalized with the average value of the considered gene for all measures . The values are the means and SD from four biological replicates . The ‘’ denote statistical differences ( t-test; p<0 . 05 ) between Nipponbare and the considered variety . A red and green dot indicates that Nipponbare expression is lower and higher than either Huang Pi Nuo or Nuo Gu respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 024 The comparison of PR gene expression after infection with the virulent isolate Guy11 in two japonica and two indica typical varieties from Yuanyang also showed a higher induction of defense in the japonica varieties than in the indica varieties ( Figure 3B; Figure 3—figure supplement 5B ) . This trend was also visible for two other broadly infecting isolates ( Figure 3—figure supplement 6 ) . To evaluate the contribution of pattern-triggered immunity to the elevated basal immunity of Yuanyang japonica varieties , we measured the responsiveness to exogenous chitin , a fungal cell wall component inducing immune responses in rice ( Azizi et al . , 2016; Liu et al . , 2014 ) and other plants ( Lo Presti et al . , 2015 ) . We compared Huang Pi Nuo and Nuo Gu to Nipponbare which displays high basal immunity ( Vergne et al . , 2010 ) . One day after chitin treatment , the Yuanyang japonica varieties and Nipponbare showed a strong induction of chitin-responsive genes ( Figure 3C; Figure 3—figure supplement 7 ) with a stronger induction of the PR genes in the Yuanyang varieties , particularly in Nuo Gu . Altogether these data indicate that the Yuanyang japonica varieties display hallmarks of an elevated basal immunity compared to Yuanyang indica varieties . Since Yuanyang japonica varieties display elevated basal immunity and are infected by JB isolates having a large Avr-effector complement , we hypothesized that the large set of Avr-effectors in JB isolates was required to counter the basal immunity of japonica varieties . Under this hypothesis , a first expectation was that the number of Avr-effectors should be positively correlated with virulence on Yuanyang japonica varieties . Seven isolates from the JB and IB groups with a number of Avr-effectors ranging from 5 to 13 ( Figure 1C ) were inoculated on the two major japonica varieties in Yuanyang , Huang Pi Nuo and Nuo Gu . This showed that two components of virulence , aggressiveness ( lesion surface ) and infectivity ( percentage of susceptible lesions per leaf ) were strongly correlated with the number of Avr-effectors ( Figure 4 ) . 10 . 7554/eLife . 19377 . 025Figure 4 . Effector complement and virulence of Yuanyang isolates . Seven isolates with Avr-Effector number ranging from 5 to 13 ( see isolates marked with ‘+’ in Figure 1 ) were selected and inoculated onto japonica varieties Huang Pi Nuo and Nuo Gu . The two quantitative components of virulence ( lesion surface and percentage of susceptible lesions over total lesion number ) are indicated as means and standard deviation from three biological replicates ( each replicate included six independent plants ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 02510 . 7554/eLife . 19377 . 026Figure 4—source data 1 . The data relates to Figure 4 . The file presents the data of inoculation of the Huang Pi Nuo ( HPN ) and Nuo Gu ( NG ) by different isolates of M . oryzae ( strains ) for each of which the number of AVR-effector is indicated . The first two spreadsheets are surfaces of susceptible lesions with HPN and NG . The last two spreadsheets are the data for calculating the percentage of susceptible lesions . Data of resistant ( R ) and susceptible ( S ) lesions are provided and used for calculating the total number of lesions . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 02610 . 7554/eLife . 19377 . 027Figure 4—source data 2 . The data relates to Figure 4—figure supplement 1 . The values represent the number of lesions per unit surface for each of the replicates . The first lane represents the isolate used for inoculating the Nipponbare ( Nip ) or CEBiP ( cebip ) mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 02710 . 7554/eLife . 19377 . 028Figure 4—figure supplement 1 . A large complement of effector is no required on immune-deficient plants . The requirement for Avr-Effectors depends on plant pattern-triggered immunity . Yuanyang isolates with variable number of Avr-Effectors were inoculated onto the cebip mutant and the corresponding Nipponbare ( Nip ) control . The number beside the isolate name corresponds to the number of Avr-Effector genes as evaluated in Figure 2—figure supplement 3 . The number of susceptible lesions per surface unit was measured seven days after inoculation . The black bar represents the median of all data from three replicates . For each isolate , the significance of the difference between cebip and Nipponbare was tested using a T-test ( p<0 . 05; p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 028 The second expectation was that the effect of mutations negatively impacting the plant immune system could be reduced when tested with isolates carrying a large effector complement . Indeed a large effector complement may have the capacity to dampen basal immunity down to the low level found in immune-deficient plants . We thus tested the capacity of Yuanyang isolates with different numbers of Avr-effectors to infect the cebip mutant , defective in mounting chitin pattern-triggered immunity and showing reduced basal immunity ( Delteil et al . , 2012 ) . As previously described , the immuno-depressed plants were less resistant than wild type plants . However , this reduced resistance was only observed with isolates containing a small Avr-effector complement but not with JB isolates containing a large Avr-effector complement ( Figure 4—figure supplement 1 ) . These results indicate that basal immunity has a strong negative impact on the virulence of isolates harboring a relatively limited set of Avr-effectors but not of isolates with a large Avr-effector arsenal . Overall , our findings hence validate the hypothesis that the Avr-effector complement in JB isolates is critical for dampening the basal immunity of Yuanyang japonica rice . The best candidates for individual Avr-effectors with a substantial impact on virulence on Yuanyang japonica rice were AVR-Pia and AVR-Pii since both are present in almost all JB isolates and completely absent from IB isolates ( Figure 2—figure supplement 4 ) . Since numerous effectors in JB isolates could have redundant functions , we reasoned that deleting AVR-Pia or AVR-Pii from these isolates may not strongly affect virulence . We therefore decided to test the role of these effectors by transferring them into the AVR depleted strain Guy11 ( 26 ) . Only AVR-Pia showed significant effects ( Figure 5—figure supplement 1 ) that were further confirmed by comparing three independent transgenic strains expressing AVR-Pia and three independent strains containing the empty vector on several japonica rice varieties ( Figure 5 ) . Two key parameters of virulence , the percentage of susceptible lesions and the surface of individual lesions , were significantly higher for the AVR-Pia transgenic strain and the effect of AVR-Pia was most significant on the Yuanyang variety Huang Pi Nuo . This indicates that AVR-Pia makes a significant contribution to the virulence of M . oryzae on this particular japonica variety , potentially by interfering with cellular host processes important for infection . However , AVR-Pia alone was not sufficient to increase virulence on the Nuo Gu japonica variety , as expected due to the strong inducibility of pattern-triggered immunity in this variety ( Figure 3C ) . The specialization of the rice blast fungus to indica or japonica varieties has been occasionally suggested in the literature but has never been documented convincingly in a real agro-system where both types of hosts occur ( Gallet et al . , 2016; Shang et al . , 2016 ) . Most notably , previous studies suffered from the use of fungal isolates and plant genotypes that were un-paired , i . e . not obtained from the same plant or area . This therefore did not allow the clear detection of the differential adaptation of pathogens to their local hosts . In our study , we used a large collection of isolates ( >200 ) and their paired rice hosts to investigate the specialization of M . oryzae to the host genotype ( Figure 1 ) . The population of M . oryzae in the traditional agro-system of the Yuanyang terraces is highly diverse and structured ( Figure 1 and Figure 1—figure supplements 1–3 ) . In particular , there is strong genetic differentiation between isolates from rice host plants belonging to the indica or the japonica subspecies ( IB and JB isolates respectively ) . Interestingly , IB isolates that were rarely sampled on japonica rice in the field were pathogenic on japonica varieties from the Yuanyang terraces or other origins in controlled conditions ( Figure 1B ) . However , they showed reduced virulence on japonica rice that was only detected when quantitative differences were recorded in infection experiments under controlled conditions ( Figure 4 ) . This reduced virulence of the IB isolates may explain their exclusion from japonica rice in the Yuanyang agro-system . Recently , it has been shown that isolates of Pyrenophora teres f . teres cause ~9% more lesions on their local barley hosts than immigrants ( Rau et al . , 2015 ) , a value similar to what we observe ( Figure 5 ) . Contrasting with the pattern uncovered for IB isolates , JB isolates were non-pathogenic on indica varieties under controlled conditions and are largely excluded from such hosts in the field . Therefore , we found clear evidence that , in the Yuanyang terraces , M . oryzae isolates are specialized to indica or japonica hosts and that fitness differences on these hosts correlates with genetic differentiation , indicating that there is local , host-driven adaptation in the rice-M . oryzae pathosystem . The situation observed in the Yuanyang agro-system is reminiscent of what has been described for several natural ecosystems ( for review [Greischar and Koskella , 2007] ) and represents one of the very few examples known thus far of specialization of a pathogen to its host in an agro-system . 10 . 7554/eLife . 19377 . 029Figure 5 . Impact of the Avr-Pia gene on the virulence of M . oryzae on Yuanyang japonica rice varieties . Three independent transgenic isolates expressing the Avr-Pia gene under its native promoter in the Guy11 background ( +Avr-Pia ) or three transgenic containing an empty vector ( +EV ) were inoculated onto the two major japonica varieties grown in Yuanyang ( Huang Pi Nuo and Nuo Gu ) and two other japonica variety Nipponbare . On Huang Pi Nuo , Avr-Pia isolates are significantly different from isolates containing an empty vector ( ANOVA followed by T-test; p<0 . 05; ***p<0 . 001 ) for two parameters of virulence , the percentage of susceptible lesions ( lower panel ) and the surface of individual lesions ( upper panel ) . Bars represent average values based on three biological replicates for each of the three Avr-Pia and control strains . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 02910 . 7554/eLife . 19377 . 030Figure 5—source data 1 . The data relates to Figure 5 . Strains 1 to 3 are independent GY11 transformants with either AVR-Pia transgene or empty vector ( EV ) . Each strain was replicated three times . HPN= Hunag Pi Nuo , NG= Nu Gu , NB= Nipponbare Spreadsheet 5A and 5B are data for surface of susceptible lesions and susceptible lesions over total lesions . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 03010 . 7554/eLife . 19377 . 031Figure 5—figure supplement 1 . The Avr-Pia but not Avr-Pii gene affects virulence . ( A ) Evaluation of the effect of Avr-Effectors on the virulence on the japonica rice Huang Pi Nuo ( not containing the corresponding Pi genes ) and ( B ) on indica rice Xiao Gu ( containing the corresponding Pi genes ) . For Xiao Gu , the number of susceptible lesions is given since this genotype was resistant to the isolates containing any of the three AVR gene tested . For Huang Pi Nuo , the GUY11+AVR isolates were virulent and the surface of individual lesions is provided as an indication of virulence . DOI: http://dx . doi . org/10 . 7554/eLife . 19377 . 031 Our data enables the development of a model that can explain the specialization of M . oryzae to japonica or indica rice varieties grown in Yuanyang . We demonstrated that the Japonica varieties from the Yuanyang terraces harbor an elevated basal immunity ( Figure 3 and Figure 3—figure supplements 3–7 ) but a low content in major resistance ( Pi ) genes ( Figure 2A ) . The larger effector repertoire of Japonica-infecting isolates ( Figure 2B ) could enable them to overcome the elevated basal immunity in Yuanyang japonica varieties ( Figure 3 ) . In parallel , the relatively larger effector complement of JB isolates would lead to a strong fitness cost on Yuanyang indica varieties ( e . g . Figure 5—figure supplement 1 ) that contain a large set of Pi genes ( Figure 2A ) , with many effectors acting as Avr-effectors ( i . e . effectors triggering complete resistance in plant genotypes carrying a Pi genes that recognize them ) . By contrast , indica-infecting isolates lack multiple effectors , including Avr-effectors , and are therefore able to escape detection by matching Pi genes in indica varieties . However , the lack of these effectors induces a strong fitness cost on japonica varieties ( Figure 4 ) that have a higher level of basal immunity ( Figure 3 ) . This model explains why japonica rice varieties , which represent only 2% of the rice grown in Yuanyang , are almost exclusively infected by the particular type of JB isolates and not by the IB isolates that are very much dominating in terms of population size and are able , at least under controlled conditions , to infect japonica rice . In their description of a unifying concept for non-host resistance , host range and pathogen speciation , Schulze-Lefert and Panstruga ( Schulze-Lefert and Panstruga , 2011 ) proposed that pattern-triggered immunity and effector-triggered immunity both contribute to non-host resistance , while effector-triggered immunity was proposed to mainly drive host range , i . e . the range of host genotypes that can be infected within a given host species . Our results suggest that several components of basal immunity ( including pattern-triggered immunity and constitutive expression of defense ) are also important determinants of host range in an agronomical context . Basal immunity in japonica leads to the accumulation of effectors that are detrimental on indica since varieties of this latter rice subspecies display strong effector-triggered immunity thanks to their extended repertoires of immune receptors for Avr-effectors . Effectors , fitness cost and differential adaptation are linked in different theoretical models Giraud et al . , 2010; Laine and Barrès , 2013; Barrett et al . , 2009] and references therein ) . On the one hand , positive and negative fitness costs of effectors were demonstrated in rare field studies involving the fungus Leptosphaeria maculans ( Huang et al . , 2006 , 2010 ) or the bacterium Xanthomonas oryzae pv oryzae ( Vera Cruz et al . , 2000 ) . Under controlled conditions , a reduced number of ‘avirulence’ activities had a significant negative impact on fitness as in the case of Phytophthora infestans infecting potato ( Montarry et al . , 2010 ) . This is similar to what we observe with IB isolates that lost many Avr-effectors ( Figure 2B ) , probably to become infectious on indica varieties . At the same time , this loss of Avr-effectors may negatively impact fitness on japonica rice on which larger repertoires of Avr-effectors are required ( Figure 4 ) . On the other hand , fitness cost and specialization have been observed in many instances ( reviewed in [Laine and Barrès , 2013] ) . By contrast , there are yet only few cases demonstrating the relationship between effector suites and specialization in plants or in animals . In plants , a major support for the relationship between effector content and specialization to host plants comes from experiments where the transfer of an entire mobile chromosome harboring effectors could convert a non-pathogenic strain of Fusarium oxysporum into a pathogen of tomato ( Ma et al . , 2010 ) . In animals , a convincing example was provided in the case of the transfer of an effector from Coxiella burnetii that could extend host cell range of Legionella pneumophila ( Lührmann et al . , 2010 ) . Our data suggest that the JB isolates have a characteristic effector suite ( Figure 2B ) that is required for specialization to japonica rice in Yuanyang ( Figure 1 ) . Our data also suggests that , amongst them , AVR-Pia is a major determinant of specialization to either indica or japonica rice subspecies found in the Yuanyang terraces . Indeed we show that ( i ) the sole presence of AVR-Pia increases several fitness parameters ( Figure 5 ) on the japonica rice Huang Pi Nuo that has elevated basal immunity ( Figure 3 ) and ( ii ) that AVR-Pia is present in almost all JB isolates and absent from IB isolates ( Figure 2—figure supplement 4 ) . Additional experiments , such as mutating AVR-Pia in JB isolates , could provide further insights into the role of this gene in differential adaptation to indica and japonica varieties . To our knowledge this is also the first report in M . oryzae of an Avr-effector demonstrated to confer increased fitness . Interestingly , despite the gain of fitness provided by AVR-Pia on some japonica variety , this gene was lost at a high frequency in M . oryzae strains at the worldwide level ( Yoshida et al . , 2009; Cesari et al . , 2013 ) . This suggests that the Pia resistance gene was in the past widely distributed and therefore counter-selected strains harboring AVR-Pia . The findings reported here allowed us to propose a model describing the molecular underpinnings of the specialization of M . oryzae to japonica and indica varieties in Yuanyang . Our work will form the basis of testable hypotheses to determine whether the molecular mechanisms described by our model represent fundamental features of the specialization of M . oryzae to indica and japonica rice subspecies and in other pathosystems . This work suggests that the appropriate deployment of contrasting immune systems in the field can dramatically impact pathogen populations . In their pioneer work showing that mixtures can produce resistance , Zhu et al . ( Zhu et al . , 2000 ) reported a field situation where inter-cropping rice varieties dramatically reduced blast severity levels . Quite interestingly , this work involved a japonica variety ( including Huang Pi Nuo , a . k . a Huang Ke Nuo ) and two modern hybrid indica varieties ( Zhu et al . , 2004 ) . We propose that part of the observed reduction of disease in this seminal work could be due to mechanisms similar to the ones uncovered in the Yuanyang terraces . The different types of resistance factors deployed may have been exposed to specialized pathogen populations , whose reduced virulence on their non-native alternate hosts would have reduced the global disease burden . Rice blast lesions identified in the field were put under 100% humidity . The resulting fungal colonies were transferred to sterile medium and single spores were isolated , DNA was extracted and analyzed using microsatellite markers according to Saleh et al ( Saleh et al . , 2014 ) . Plants and Magnaporthe oryzae were grown as described in Berruyer et al ( Berruyer et al . , 2003 ) . Fungal spores ( 50 , 000 spores/mL ) were inoculated by spray after three weeks ( fourth leaf stage ) and symptoms measured seven days after inoculation . Resistance ( R ) or susceptibility ( S ) scores were established as in Gallet et al ( Gallet et al . , 2016 ) . The cebip mutant used is in the Nipponbare background ( Delteil et al . , 2012 ) . For Avr-Effector diagnostics ( Figure 2B ) , we used the rice lines described in Berruyer et al ( Berruyer et al . , 2003 ) that allow the identification of virulence and to some extent AVR functions . For chitin treatment , three week-old plants were sprayed with 0 . 02% tween 20 ( mock ) or 100 μg/mL of chitin solubilized in 0 . 02% tween 20 . The experiment was repeated four times . This chitin contains 2 to 8-mers of oligosaccharide ( YSK , Yaizu Suisankagaku Industry , Japan ) . The third last fully expanded leaves were harvested 24 hr after treatment for gene expression analysis . We used DAPC and neighbor-joining analysis of genetic distances to infer population subdivision . For both DAPC , we retained the first 20 principal components , and the first six discriminant functions . Genetic distances were computed as the proportion of shared alleles using a custom-made script , and the neighbor-joining was computed using the neighbor program from the Phylip package ( http://evolution . genetics . washington . edu/phylip/progs . data . dist . html ) . All analyses were carried out on clone corrected datasets ( i . e . on datasets for which a single representative was kept for multilocus microsatellite genotypes represented multiple times ) . The index of association rd is a measure of multilocus linkage disequilibrium ranges from 0 ( complete panmixia ) to 1 ( strict clonality ) ( Agapow and Burt , 2001 ) . The rd statistic and other summary statistics of genetic variability were computed for all clusters using a custom-made script . Significance of rd values was established by comparing the observed values with the distributions obtained by 1000 randomizations . The multi-virulent isolate Guy11 was transformed with plasmids carrying AVR-Pia or Avr-Pii ( Yoshida et al . , 2009 ) . For each Avr , the corresponding empty vector was used to build control strains . Single spores from transgenic events selected on the adapted antibiotic were isolated and amplified . The functionality of each Avr gene was tested with an inoculation of the rice differential line carrying the corresponding Pi gene . RNA was extracted from either healthy or inoculated leaves and cDNA produced as in Delteil et al ( Delteil et al . , 2012 ) . The primers for the rice marker genes used were previously shown to work in indica and japonica background ( Vergne et al . , 2010 ) . Each sample consisted of at least eight plants randomly chosen and for each condition , three to four independent samples were analyzed to build the mean expression . All expression data were normalized using the expression of the constitutive Actin gene .	Microbes that cause diseases in plants are a threat to food security . For example , the rice blast fungus Magnaporthe oryzae causes the loss of enough rice to feed 60 million people each year . Disease-causing microbes must overcome the plant’s first line of defense , which includes preformed barriers and antimicrobial responses that are triggered by characteristic molecules found in many different microbes . The microbes that can overcome this first line of defense typically do so with an arsenal of proteins called effectors that interfere with specific biological processes in the plant . To counteract this interference , some plants have evolved genes that encode proteins that detect these effectors and trigger stronger antimicrobial responses . For centuries , farmers and plant breeders have selected for these resistance genes when trying to breed crops that are more resistant to disease . However , over time , disease-causing microbes have lost effectors , which means that several resistance genes have rapidly become ineffective . Some researchers predicted that growing a mixture of varieties of a given crop together might be a better way of protecting crop yields . Over 16 years ago , this idea was proved successful against the rice blast fungus for rice plants grown in China . However , the exact reasons why this strategy worked and its effects on the fungus were not clear . Now Liao , Huang et al . have taken another look at rice varieties grown via the traditional method of terraces of rice paddies in Yuanyang . Some of these varieties had a strong first line of defense and few resistance genes , while others relied much more on resistance genes to protect themselves again the rice blast fungus . Liao , Huang et al . found that growing rice varieties with such different immune systems forces some of the rice blast fungi to accumulate effector proteins to combat the first line of defense , whereas other fungi had to get rid of these effectors to avoid being recognized by the major resistance genes . These two forces led to the evolution of two specialized populations of fungi that can infect specific rice varieties but not others . This means that the fungi cannot spread in the landscape , and so the fields of rice become resistant as a whole . These new findings demonstrate the importance of diversity in rice for sustainable crop protection . The next challenge will be to demonstrate if a similar approach can also protect other major crops grown in different agricultural settings .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "plant", "biology" ]	2016	Pathogen effectors and plant immunity determine specialization of the blast fungus to rice subspecies
Social interactions involve multi-modal signaling . Here , we study interacting rats to investigate audio-haptic coordination and multisensory integration in the auditory cortex . We find that facial touch is associated with an increased rate of ultrasonic vocalizations , which are emitted at the whisking rate ( ∼8 Hz ) and preferentially initiated in the retraction phase of whisking . In a small subset of auditory cortex regular-spiking neurons , we observed excitatory and heterogeneous responses to ultrasonic vocalizations . Most fast-spiking neurons showed a stronger response to calls . Interestingly , facial touch-induced inhibition in the primary auditory cortex and off-responses after termination of touch were twofold stronger than responses to vocalizations . Further , touch modulated the responsiveness of auditory cortex neurons to ultrasonic vocalizations . In summary , facial touch during social interactions involves precisely orchestrated calling-whisking patterns . While ultrasonic vocalizations elicited a rather weak population response from the regular spikers , the modulation of neuronal responses by facial touch was remarkably strong . Rats are highly social animals that display complex behaviors ( aggression , dominance , mating , parental care , and play [Barnett , 1958] ) , which involve the use of multi-modal signaling and sensing ( Brecht and Freiwald , 2012 ) . These social interactions are initially characterized by anogenital sniffing followed by facial contacts which occur at a constant rate ( Wolfe et al . , 2011 ) . Facial contacts have also been shown to be involved in dominance-related behaviors in naturalistic settings ( Blanchard et al . , 2001 ) . Facial contacts consist of extensive whisker-to-whisker and snout-to-snout touch episodes , largely mediated by macrovibrissae ( Wolfe et al . , 2011 ) . The use of macrovibrissae results in a strong activation of the barrel cortex and , importantly , the neurons appear to represent the social context of these interactions by distinct firing rates ( Bobrov et al . , 2014 ) . Ultrasonic vocalizations ( USVs ) form another important component of rodent social interactions ( Brudzynski and Pniak , 2002; Wright et al . , 2010 ) . Rats produce two distinct classes of USVs , 22 kHz alarm calls and 50 kHz appetitive vocalizations ( McGinnis and Vakulenko , 2003; Brudzynski , 2009 ) , which serve as indicators of negative and positive affective states , respectively ( Knutson et al . , 2002 ) . While the alarm calls are elicited by a range of threats ( dominant or aggressive conspecifics , predators , and aversive stimuli [Brudzynski , 2009] ) , the 50-kHz calls are produced in anticipation of/in response to direct social contact ( Blanchard et al . , 1993; Bialy et al . , 2000; Brudzynski and Pniak , 2002 ) , mating ( Bialy et al . , 2000 ) , and homo-specific ( rough-and-tumble ) or hetero-specific ( tickling ) play behaviors ( Burgdorf et al . , 2008 ) . Further , playback experiments have demonstrated that 50-kHz vocalizations can induce approach behavior ( Wöhr and Schwarting , 2007 ) . Studies on the neuronal representation of USVs have largely relied on playback experiments , for example , to demonstrate the induction of c-fos expression ( Sadananda et al . , 2008 ) . Reliable and preferential responses to USVs have been reported upon playback in anaesthetized ( Kim and Bao , 2013 ) or awake rats ( Carruthers et al . , 2013 ) . However in the awake auditory cortex , a sparse representation of playback sounds has been demonstrated ( Hromádka et al . , 2008 ) . Little is known about the representation of the whole repertoire of conspecific calls in awake , behaving animals . Likewise , there is little information about the integration of multisensory social signals in the auditory cortex , with the notable exception of experience-dependent modulation of auditory responses by natural odors in mice ( Cohen et al . , 2011 ) . To address these issues , we employed the gap paradigm ( Wolfe et al . , 2011; von Heimendahl et al . , 2012; Bobrov et al . , 2014 ) , wherein a combination of USVs and facial touch enables the study of multisensory coordination and integration of social signals in the auditory cortex . Specifically , we pose the following questions: ( i ) How are calls and whisking related on coarse and fine time scales ? ( ii ) How does auditory cortex respond to calls in interacting animals ? ( iii ) How does auditory cortex respond to facial touch ? ( iv ) Are responses to calls in the auditory cortex modulated by touch ? To study the relationship between calling and social interactions , we aligned the USVs to the onsets of the facial touch episodes ( Figure 1D ) . We analyzed three scenarios: ( i ) when the subject rat was present alone on the setup ( ‘Subject alone’ ) , ( ii ) after the introduction of stimulus rats during which interactions took place ( ‘Social setting’ ) , and ( iii ) specifically during the facial touch episodes ( ‘Facial touch’ ) . In all these scenarios , a vast majority of USVs ( ∼80% ) were composed of just three call categories , that is , trill , complex , and flat ( Figure 1E ) . Interestingly , the proportion of trills was greater in social contexts whereas the proportion of flats was greater when the animals were alone . A similar trend was observed in some of the minor call categories ( Figure 1—figure supplement 1C ) . While the baseline calling rate was low when the subject rat was alone ( 0 . 13 Hz ) , a substantial increase ( 0 . 80 Hz ) occurred in the social setting . Within the facial touch episodes , a further increase in vocalization ( 1 . 50 Hz ) was observed . Indeed , the calling rate for all call categories increased during facial touch compared to periods outside of it ( Figure 1F , Figure 1—figure supplement 1D ) , with trills showing the highest increase . Analysis of peri-stimulus time histograms ( PSTHs ) also revealed an increase in calling associated with the onset ( Figure 1G ) and a sharp decrease with the offset ( Figure 1H ) of facial touch . For interactions with non-social stimuli , the calling rate was lower ( 0 . 36 Hz ) and there was no increase in calling during touch ( Figure 1I ) . On the whole , this increase in calling rate indicates a role for USVs in social communication in our paradigm . Since facial touch was associated with whisker contacts and extensive vocalization , we wondered if whisking and calling were temporally related . To assess this , we tracked precise whisker positions in high-speed videos and mapped the call times onto this information . As shown in the representative example ( Figure 2A ) , calls were typically emitted in the retraction phase of the whisking cycle of the call emitter ( top trace ) , whereas there was no systematic relationship of calling to the whisking of the interacting partner ( bottom trace ) . Hence , when all whisking traces were aligned and averaged relative to calls , the whisking rhythmicity was preserved in the resulting call-triggered whisking average ( Figure 2B ) . We observed that the relationship of call onsets to whisking phase was distributed significantly non-uniformly relative to whisking for call emitter ( n = 664 , p = 0 . 0014 , Figure 2C , top , Hodges–Anje test ) but not for interacting partner ( n = 705 , p = 0 . 06 , Figure 2C , bottom ) . There also appears to be a substantial bias for calls during the retraction phase of the emitter's whisking ( 381 in retraction vs 283 in protraction , Figure 2C , top ) unlike for calls during the interacting partner's whisking ( 352 in retraction vs 353 in protraction , Figure 2C , bottom ) . 10 . 7554/eLife . 03185 . 007Figure 2 . Whisking and vocalization are coordinated during facial touch . ( A ) Sample whisker traces of a call emitter ( top ) when aligned to its own vocalizations ( gray bars ) indicate a correlation with the retraction phase of the whisker . This was not the case with the whisker trace of the interacting partner ( bottom ) . ( B ) Call-triggered whisking average wherein the average of all whisking traces is aligned with respect to the emitter's call onset . ( C ) Distribution of call onsets were significantly non-uniform ( Hodges–Anje test ) relative to whisking cycle of emitter itself ( top ) but not for the interacting partner ( bottom ) . There appears to be a substantial bias for calls during the retraction phase of the emitter's whisking ( top ) unlike for calls during the interacting partner's whisking ( bottom ) . ( D ) Predominant whisking rates determined by power spectral density of whisker traces during social touch is 8 . 1 Hz ( left ) . Distribution of time intervals between whisking cycles within one animal ( auto ) correspondingly shows a peak at ca . 112 ms ( middle ) . Points of maximum protraction were used for binning . Inter-animal ( cross ) whisking interval distribution suggests that the interacting animals do not whisk with a fixed phase relationship ( right ) . ( E ) Power spectral density analysis to determine the predominant calling rates revealed a 7 . 6 Hz predominant frequency component ( left ) . The distribution of time intervals between start of two subsequent vocalizations peaks around 144 ms ( ‘auto’ , middle ) . The effect is not as pronounced when triggering to calls that were assigned to a different animal ( ‘cross’ , right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03185 . 007 We next analyzed the predominant whisking and vocalization rates and observed a peak at ∼8 Hz for both by power spectral density analysis ( Figure 2D , E , left panels ) . Analysis of the distribution of time intervals between whisking cycles revealed a peak at 112 ms within animals ( ‘auto’ , Figure 2D , middle ) but not across animals ( ‘cross’ , Figure 2D , right ) . This was consistent with previous observations that whisking is not coordinated across animals ( Wolfe et al . , 2011 ) . Similarly , in the analysis of the distribution of time intervals between the start of two subsequent vocalizations of the putative call emitter , a prominent peak was observed at 144 ms ( ‘auto’ , Figure 2E , middle ) . Such rhythmicity was not obvious when the call onsets across interaction partners were analyzed ( ‘cross’ , Figure 2E , right ) , suggesting that the coordination of whisking and vocalization is restricted to individuals . This analysis is constrained by the fact that ∼1 in 5 calls is incorrectly assigned to a particular call source . Sessions with a single source of USVs ( where subject rats interact with plastinated rats/objects ) would ideally provide unambiguous evidence as there is only one source of USVs . However , since very few calls are produced in such sessions , we presented subject rats with anaesthetized stimulus rats to elicit extensive vocalization . Analysis of tracked whiskers in the high-speed videos relative to call onset shows that the temporal coordination of calling and whisking is indeed observable ( Video 1 ) . To study the neuronal representation of USVs , we performed extracellular tetrode recordings in the auditory cortex ( Figure 3—figure supplement 1A–C ) of awake , behaving rats ( four females , four males ) while they interacted with conspecifics . Single units were identified based on separation and stability criteria ( see ‘Materials and methods’ ) and classified as putative regular-spiking ( RS ) or fast-spiking ( FS ) neurons ( Figure 3—figure supplement 1D–H ) . Analysis of PSTHs triggered to the call onsets revealed the presence of several neurons in the auditory cortex that responded strongly to the calls vocalized within that session ( Figure 3A , C; Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 03185 . 008Figure 3 . Responses to USVs in auditory cortex are excitatory and cell-type dependent . ( A ) Representative PSTHs showing the response of a RS neuron to all USVs ( top left ) and to individual call categories that is , trill ( top right ) , complex ( bottom left ) , and flat ( bottom right ) ( bin size: 10 ms ) . ( B ) Population data showing responses of RS neurons plotted as firing rate in response to USVs vs baseline ( left , Wilcoxon signed rank test ) . The population showed a significant modulation in firing rate in response to USVs . The mean response indices ± SEM to either all calls or the major call categories also reveal little or no overall modulation ( right ) . ( C ) Response of a representative FS neuron showing a stronger modulation to all calls and also to complex and trill calls ( bin size: 10 ms ) . ( D ) Almost all FS neurons were also significantly upregulated by USVs ( left , Wilcoxon signed rank test ) and this was evident in their mean response indices ± SEM to all calls and to the various call categories ( right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03185 . 00810 . 7554/eLife . 03185 . 009Figure 3—figure supplement 1 . Locations of recording sites in the auditory cortex and cell-type classification . ( A ) Photomicrograph of cytochrome oxidase stained brain section showing electrolytic lesions ( arrowheads ) along two tetrode tracts ( solid lines ) passing through the auditory cortex ( bregma: −3 . 72 mm; magnification: 2×; scale bar: 1 mm ) . Dotted lines demarcate the sub-regions: auditory cortex dorsal ( AuD ) , primary auditory cortex ( Au1 ) , and auditory cortex ventral ( AuV ) . ( B ) Locations of recording sites across the auditory cortices of four female and four male subject rats . Bregma positions are as indicated in the rat brain atlas ( reproduced with permission from Paxinos and Watson , 2007 , The Rat Brain in Stereotaxic Coordinates , sixth Edition , copyright Elsevier 2007 , All Rights Reserved ) . ( C ) Location of recording sites as in ( B ) . ( D ) Comparison of firing rates with full spike width indicated a bi-modal distribution of auditory cortex units . ( E ) A similar comparison with the second half width also resulted in a bi-modal distribution . ( F ) Using these features , k-means clustering of units with two clusters resulted in well separated populations ( indicated by dotted line ) . ( G ) Spikes from these two populations resulted in well-defined average spike shapes , with the longer waveforms ( orange ) being classified as putative regular-spiking ( RS ) neurons and the thinner waveforms ( green ) classified as putative fast-spiking ( FS ) neurons . ( H ) Overall firing rates of these two classes were also highly differentiated ( median firing rates: RS: 4 . 4 Hz vs FS: 11 . 1 Hz ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03185 . 00910 . 7554/eLife . 03185 . 010Figure 3—figure supplement 2 . Responses of auditory cortex neurons to ultrasonic vocalizations are heterogeneous . ( A–E ) Representative PSTHs showing the response of RS neurons to all USVs and to individual call categories that is , trill , complex , and flat ( bin size: 10 ms ) . ( F–H ) Representative PSTHs showing the response of FS neurons to all USVs and to individual call categories that is , trill , complex , and flat ( bin size: 10 ms ) . Neurons that appear to be more strongly modulated by one call category than others are indicated by an asterisk ( * ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03185 . 01010 . 7554/eLife . 03185 . 011Figure 3—figure supplement 3 . Population responses of auditory cortex neurons to different USV call categories . ( A ) Population response of RS neurons ( top ) to various call categories compared to baseline firing rates showed significant modulation by trill , complex , and flat categories ( Wilcoxon signed rank test ) . FS neurons ( bottom ) were also significantly modulated by trill category . ( B ) Scatter plot showing significant response of RS neurons to the ‘best call’ , that is , maximal excitatory response when compared to the baseline firing rate ( Wilcoxon signed rank test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03185 . 01110 . 7554/eLife . 03185 . 012Figure 3—figure supplement 4 . Population responses to own vs stimulus calls and in various sub-regions of the auditory cortex . ( A ) RS neurons showed significant modulation to USVs of stimulus but not own calls ( top ) whereas FS neurons are modulated by own but not stimulus calls ( bottom , Wilcoxon signed rank test ) . ( B ) Sub-region-wise comparison showed that AuD and Au1 RS neurons were significantly modulated by USVs ( top ) whereas Au1 FS neurons alone were modulated by USVs ( bottom , Wilcoxon signed rank test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03185 . 01210 . 7554/eLife . 03185 . 013Figure 3—figure supplement 5 . Auditory cortex neurons do not show any locking to the phase of whisking . ( A ) Polar plots showing the preferred phase of auditory cortex neurons triggered to own ( left ) and stimulus whisking ( right , Rayleigh vector lengths 0 . 1–0 . 5 with 0 . 2 indicated in red ) . Neurons that passed the shuffling test criteria are indicated in green . At the individual level neurons that have a Rayleigh vector length >0 . 2 and have passed the shuffling test are considered to be locked to whisking . At the population level , there appears to be no significant locking of spiking with either the retraction or protraction phase of whisking . DOI: http://dx . doi . org/10 . 7554/eLife . 03185 . 013 Strikingly , most auditory cortex neurons showed little or no response to calls while ∼10% of RS neurons showed heterogeneous and excitatory responses ( Figure 3A , Figure 3—figure supplement 2A–E ) . Some RS neurons had a fast onset-associated response ( response latency <25 ms of call onset; Figure 3A , Figure 3—figure supplement 2A , B ) , while the others showed a more sustained response ( Figure 3—figure supplement 2C–E ) . In addition , a few late responders were also observed ( response latency >50 ms after call onset; Figure 3—figure supplement 2E ) . USVs elicited a small ( ∼5% ) but significant increase in the population of RS neurons sampled ( Figure 3B , left ) , as reflected in their firing rates ( mean response to all calls ± SEM: 9 . 3 ± 0 . 82 Hz vs baseline: 8 . 8 ± 0 . 77 Hz; n = 172 , p = 0 . 0006 , Wilcoxon signed rank test ) . The median firing rates however were lower ( response to calls: 6 . 06 Hz vs baseline: 5 . 42 Hz ) . This would suggest that the overall increase in the mean firing rate was contributed to by a small fraction of neurons within the population . This is in line with the observation that most neurons had little or no response to USVs despite extensive sampling of the auditory cortex across several animals ( Figure 3—figure supplement 1B , C ) . FS neurons on the other hand showed a robust excitatory modulation by USVs ( Figure 3C , Figure 3—figure supplement 2F–H ) . Most FS neurons had a sustained response to calls ( Figure 3C , Figure 3—figure supplement 2F ) . In addition , few cells with sharp short latency ( Figure 3—figure supplement 2G ) or delayed ( Figure 3—figure supplement 2H ) responses were also observed . The mean firing rates showed a consistent upregulation ( response to calls: 19 . 11 ± 3 . 94 Hz vs baseline: 16 . 92 ± 4 . 07 Hz , n = 23 , p = 0 . 006 , Wilcoxon signed rank test ) , which is seen in the scatter plots ( Figure 3D , left ) . Analysis of call category specificity revealed that many neurons ( RS and FS ) responded to more than one of the major call categories ( Figure 3A , C , Figure 3—figure supplement 2A , C , D , F–H ) . Neurons more strongly modulated by one call category than others were also observed ( indicated by asterisks; Figure 3—figure supplement 2B ) . Population level analysis showed that there is a significant modulation of RS neurons by all the major call categories ( Figure 3—figure supplement 3A , top row ) . FS neurons however appear to be significantly modulated by trills alone ( Figure 3—figure supplement 3A , bottom row ) . The lack of significant modulation by complex and flat calls is probably an artifact caused by the small sample size as most FS neurons show activation in the scatter plots ( Figure 3—figure supplement 3A , bottom row ) . We next wanted to check if the population as a whole has a greater preference for one of the call categories . For this , we computed responses to the ‘best call’ that is , the major call type that elicits the highest modulation . While this scenario biased towards larger effects shows a highly significant modulation by the ‘best call’ ( p < 0 . 0001 , Figure 3—figure supplement 2B ) , there appears to be no overwhelming population of trill-/complex-/flat-preferring neurons . To compare the responses across cell types and call categories , we computed response indices for each neuron ( see ‘Materials and methods’ ) . The mean response indices of RS neurons to all calls and the major call categories showed a very small positive modulation ( Figure 3B , right ) , with no preference to any specific call category . Mean response indices of FS neurons , on the other hand , showed a robust positive modulation to all calls and to individual call categories ( Figure 3D , right ) . Also , the response indices of FS neurons were significantly different from those of RS neurons ( p < 0 . 0001 , unpaired Mann–Whitney test ) . We also determined that while RS neurons significantly responded to calls from the stimulus animals , this was not the case with the subject animal's own calls ( Figure 3—figure supplement 4A , top row ) . However , a pair-wise comparison of each RS neuron's response to own vs stimulus calls was not significantly different ( p = 0 . 85 , Wilcoxon signed rank test ) , suggesting that this was a population level effect . Interestingly , the reverse was true for the FS neurons , which showed a clear preference for own calls but not for the stimulus calls ( Figure 3—figure supplement 4A , bottom row ) . Pairwise comparisons of FS neuronal responses showed a significant difference ( p = 0 . 03 ) , suggesting that they could indeed discriminate between own and stimulus calls . We also compared the responses across various auditory cortex sub-fields , that is , auditory cortex dorsal ( AuD ) , primary auditory cortex ( Au1 ) , and auditory cortex ventral ( AuV ) . AuD and Au1 RS neurons as well as Au1 FS neurons were significantly modulated by USVs . However , the apparent lack of modulation in AuV RS , AuD , and AuV FS neurons is confounded by the small sample sizes ( Figure 3—figure supplement 4B ) . Previously , having observed a correlation between whisking and calling , we next tested if the neuronal firing rate is correlated with the whisker position . Towards this , whisker tracking was performed on high-speed videos to determine whisker angle and phase . Spike time stamps aligned to this information was used to generate polar plots and compute Rayleigh vector lengths for each neuron which showed >10 spikes each in touch episodes . The phase locking was also tested with a shuffling test that is insensitive to the spike count ( see ‘Materials and methods’ ) . At the population level , there was no strong preference to the whisking phase ( Figure 3—figure supplement 5A ) . A few cells that appeared to have a strong locking to the retraction phase were also strong responders to USVs , suggesting that this phase locking is an artifact from locking of whisking and vocalization . Unexpectedly , we observed that facial touch resulted in an inhibition of several RS ( Figure 4A–D ) and FS ( Figure 4E–G ) neurons . Exemplary spike raster ( Figure 4A ) and PSTHs of RS ( Figure 4B ) and FS ( Figure 4E ) neurons triggered to the onset ( left ) and end ( right ) of facial touch show the inhibition elicited due to touch and an off-response , which usually occurred within a 200-ms window at the end of touch . A substantial fraction of neurons in our dataset showed this distinct off-response ( ∼19% ) . This phenomenon appears to occur in Au1 ( Figure 4C ) but not in the other auditory cortex sub-regions ( Figure 4—figure supplement 1A , B ) . 10 . 7554/eLife . 03185 . 014Figure 4 . Facial touch evokes inhibition and off-responses in primary auditory cortex ( Au1 ) . ( A ) Schematic showing the spiking activity ( rasters ) of a representative Au1 RS neuron aligned to a facial touch episode ( gray bar ) . Facial touch onset results in an inhibition in firing rate and is associated with an increase upon the end of touch . ( B ) Representative PSTHs of a RS neuron triggered to onset ( left ) and offset ( right ) of facial touch demonstrating the inhibitory effect during touch and an off-response at the end of touch ( bin size: 77 ms ) . ( C ) Population response of RS neurons plotted as firing rate after touch onset vs baseline ( left ) shows a significant inhibition due to facial touch . However , the off-response ( in a 200 ms window after the end of facial touch ) led to an increase in firing rate which was significantly higher than in touch ( right , Wilcoxon signed rank test ) . ( D ) Mean response indices ± SEM of RS neurons during facial touch and off-response were significantly different ( Wilcoxon signed rank test ) . ( E ) Representative PSTHs of a FS neuron triggered to facial touch onset ( left ) and offset ( right ) demonstrating a similar inhibitory effect due to touch and the off-response at the end of touch ( bin size: 77 ms ) . ( F ) Population response of FS neurons showing a significant inhibition ( Wilcoxon signed rank test ) due to touch ( left ) that was released at the end of touch ( right ) . ( G ) Mean response indices ± SEM of FS neurons were also significantly different ( Wilcoxon signed rank test ) between facial touch and off-response windows . DOI: http://dx . doi . org/10 . 7554/eLife . 03185 . 01410 . 7554/eLife . 03185 . 015Figure 4—figure supplement 1 . Facial touch does not elicit inhibition in AuD and AuV; whisker/snout touch does not lead to measurable changes in sound intensities . ( A ) Facial touch onset-associated inhibition ( left ) and offset associated increase in firing rate ( right ) do not occur in AuD RS neurons . ( B ) Similarly , these phenomena were also not observed in AuV RS neurons . ( C ) Plots of change in relative power measured during social interactions revealed no change when triggered to whisker touch ( left ) and snout touch ( middle ) , whereas when triggered to USVs ( right ) a large increase was observed . Dotted lines denote the pre- and post-triggered averages . DOI: http://dx . doi . org/10 . 7554/eLife . 03185 . 015 In Au1 RS neurons , the firing rates reduced from 8 . 18 ± 1 . 03 Hz to 7 . 82 ± 1 . 04 Hz during the touch episode ( n = 105 , p = 0 . 0067 , Wilcoxon signed rank test , Figure 4C , left ) . Compared to touch , the firing rate in the off-response window was significantly higher at 8 . 49 ± 1 . 1 Hz ( n = 105 , p = 0 . 0019 , Figure 4C , right ) . Unexpectedly , this ∼10% increase in the population response was about twice as large as the excitatory responses elicited by USVs in RS cells . Similarly , for Au1 FS neurons , the baseline firing rates decreased from 14 . 6 ± 4 . 18 Hz to 13 . 82 ± 4 . 03 Hz due to touch ( n = 15 , p = 0 . 0302 , Figure 4F , left ) . The firing rate during the off-response window was higher at 17 . 03 ± 4 . 71 Hz when compared to the touch episode ( p = 0 . 0043 , Figure 4F , right ) . The off-response to facial touch is remarkable as it was the largest population response we observed in the auditory cortex . Mean response indices for Au1 RS neurons ( Figure 4D ) showed that the overall inhibition due to touch was reversed after the touch episode ( p = 0 . 0329 ) . A similar inhibition at touch onset and reversal at the end of touch were also evident in Au1 FS neurons ( p = 0 . 0026 , Figure 4G ) . It could be hypothesized that these responses in Au1 are brought about not by the touch itself but by the sounds produced due to touch . To test this , we measured the sound intensities triggered to whisker and snout touch and did not observe any increase in intensity . USVs lead to a large increase as expected ( Figure 4—figure supplement 1C ) . Given the robust inhibition observed in Au1 , we next asked if facial touch modulates the responses of Au1 neurons to USVs . Indeed , we observed that in RS neurons , the response to calls was modulated by touch ( Figure 5A ) . The population response is also evident by an increased modulation by USVs in touch ( greater scatter , Figure 5B , right ) as compared to neurons responding to calls out of touch ( Figure 5B left ) . Pairwise comparison of each neuron's response to USVs in ( right ) and out ( left ) of touch revealed a highly significant differential modulation ( p = 0 . 0038 , Wilcoxon signed rank test , Figure 5B ) . This was also evident in the frequency distribution histograms of the response indices . While response indices of RS neurons to calls out of touch were largely centered around zero indicating little or no modulation ( Figure 5C , left ) , they were more spread out during touch ( Figure 5C , right ) indicating significantly greater modulation ( p = 0 . 0072 , Kolmogorov–Smirnov test ) . However , in FS neurons , responsiveness to calls appeared to be less strongly modulated by touch at the individual ( Figure 5D ) and population levels ( p = 0 . 3303 , Figure 5E ) . Correspondingly , the spread of activation indices did not differ in and out of touch ( p = 0 . 3752 , Figure 5F ) . 10 . 7554/eLife . 03185 . 016Figure 5 . Stronger and variable modulation of responses to vocalizations by facial touch . ( A ) Representative PSTHs of a RS neuron showing a stronger response to vocalizations during facial touch ( right ) compared to out of touch ( left , bin size: 10 ms ) . ( B ) Population response of RS neurons also demonstrated an increased modulation to vocalizations ( both excitation and inhibition ) during touch ( right ) compared to outside of it ( left ) , which were significantly different when subjected to a pair-wise comparison ( Wilcoxon signed rank test ) . ( C ) Distribution of response indices showed that a large fraction of RS neurons were not modulated by vocalizations when out of touch ( left ) while a significant amount of modulation occurred during touch ( right , Kolmogorov–Smirnov test ) . ( D ) Representative PSTHs of a FS neuron showing a higher response to vocalizations during touch ( right ) compared to the response when out of touch ( left , bin size: 10 ms ) . ( E ) Population response of FS neurons however did not show any significant difference ( Wilcoxon signed rank test ) in modulation to vocalizations either in ( right ) or out ( left ) of touch . ( F ) Distribution of response indices in FS neurons also did not show any significant differences in modulation ( Kolmogorov–Smirnov test ) either in ( right ) or out ( left ) of touch . DOI: http://dx . doi . org/10 . 7554/eLife . 03185 . 01610 . 7554/eLife . 03185 . 017Figure 5—figure supplement 1 . Sampling bias does not account for modulation of responsiveness to calls during touch; calls both from own and stimulus animals lead to increased modulation in primary auditory cortex during touch . ( A ) Distribution of response indices from bootstrapped data out of touch ( left ) shows that a large fraction of RS neurons are not modulated by vocalizations . This was significantly different ( Kolmogorov–Smirnov test ) from the response during facial touch ( right ) where many cells showed increased modulation . ( B ) Similar analysis of the distribution of response indices of FS neurons revealed that there was no significant differences ( Kolmogorov–Smirnov test ) between bootstrapped out of touch ( left ) and in touch ( right ) data . ( C ) Distribution of response indices shows that many RS neurons exhibit higher modulation to own calls during touch ( right ) , which was significantly different ( Kolmogorov–Smirnov test ) from the modulation to calls out of touch ( left ) . ( D ) A similar distribution of responses indices to stimulus calls was also observed in RS neurons out of ( left ) and in ( right ) touch ( Kolmogorov–Smirnov test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03185 . 017 The apparent increased modulation in responsiveness to calls during touch could be due to a sampling artifact as only a fraction of calls ( 19 . 8% ) occur during touch . To rule out this potential confound , we applied bootstrapping analysis ( see ‘Materials and methods’ ) to compute the mean response indices for our data set and plotted the frequency distribution histograms as before . These bootstrapped data show that neuronal responses in RS ( but not FS ) neurons were strongly modulated during touch , independent of the number of samples ( Figure 5—figure supplement 1A , B ) . Another possible reason for this greater modulation could be the higher effective intensity of vocalizations in touch when the animals were closely juxtaposed . This would suggest that the response of RS neurons would be stronger to stimulus calls in touch while the response to the subject's own calls would not show much modulation ( under the assumption that the subject rat vocalizes and perceives its own calls at roughly the same intensities ) . There , however , does not appear to be any greater responsiveness to stimulus calls during touch . This would suggest that the intensity differences are probably not responsible for the higher modulation during touch ( Figure 5—figure supplement 1C , D ) . Taken together , these results suggest that the responsiveness to USVs in the auditory cortex is modulated by facial touch . Social interactions consist of complex behaviors which employ a range of multi-modal signaling and sensing ( Brecht and Freiwald , 2012 ) . Social transmission of food preference which relies on the combined use of olfactory and gustatory cues ( Galef and Wigmore , 1983 ) is probably the most extensively studied example of this . In addition , the use of unimodal signals including vision ( ear wiggling [Erskine , 1989] ) , smell ( cheek gland pheromones , [Kannan and Archunan , 2001] , and somatosensation [Blanchard et al . , 2001] ) has also been reported . However , little is known about the multisensory integration of these signals , which would play a critical role in facilitating social interactions . In this inquiry , we study interacting rats to investigate audio-haptic coordination and multisensory integration in the auditory cortex . We demonstrate that facial touch during social interactions is associated with an increased production of USVs . We observe a temporal coordination of vocalization and whisking , with calls being associated with the retraction of whiskers . USVs elicited excitatory responses in a small fraction of RS neurons in the auditory cortex , whereas almost all FS neurons showed a strong activation . Facial touch however resulted in a robust inhibition of the primary auditory cortex . Moreover , we observed a remarkable off-response at the end of the touch episode , possibly reflecting a release of touch-induced inhibition , which was surprisingly the largest response modulation observed in our study . Finally , it appears that the response of auditory cortex neurons to USVs is modulated by facial touch . Previous studies from our laboratory have used the gap paradigm to study social interactions and have demonstrated that spontaneously interacting rats use their whiskers for extensive facial interactions ( Wolfe et al . , 2011; von Heimendahl et al . , 2012 ) . Interestingly , the social context ( sex and sexual status ) of these interactions appears to be represented in the barrel cortex neurons ( Bobrov et al . , 2014 ) . A little explored component of these interactions has been USVs , which are known to contribute to species-specific social signaling ( Brudzynski and Pniak , 2002; Brudzynski , 2013 ) . A reduction in social interaction and ultrasonic communication has also been reported in a mouse model of monogenic heritable autism ( Jamain et al . , 2008 ) . Indeed , using the gap paradigm , we observed an increase in the rate of vocalizations during the presentation of social ( but not non-social ) stimuli . Interestingly , the calling rate also showed a sharp decrease at the end of interactions . We classified the USVs into specific call types largely based on an earlier description ( Wright et al . , 2010 ) . We observe that a vast majority of calls in our paradigm belong to one of the three categories that is , trill , complex , and flat . Further , the proportion of trills and flats appear to be modulated by the presence of conspecifics . These results are in line with the finding that there is a greater prevalence of flats in singly tested rats whereas trill vocalization is increased in pair-tested rats ( Wright et al . , 2010 ) . A role for trills in social contexts has also been reported during play behavior ( Schwarting et al . , 2007 ) . Taken together , these results would suggest that specific call categories could play a role during specific components of social interactions . This idea is supported by a recent study which reports that male mice emit distinct vocalizations when females leave the environment ( Yang et al . , 2013 ) . Since facial touch involves the extensive use of whiskers and USVs , we employed high-speed videography to track the position of the whiskers with high temporal resolution and mapped the vocalization data onto this information . We observed a locking of calling to whisking , with calls being produced during the retraction phase of the whisking cycle . The retraction phase has been correlated to the exhalation phase of breathing ( Moore et al . , 2013 ) , which would correspond to call production . Previous studies have shown that in addition to a correlation of respiration and USV production ( Roberts , 1972; Riede , 2011 ) , sniffing and whisking are also tightly correlated ( Welker , 1964; Ranade et al . , 2013 ) . The neural substrate of such coordination is thought to lie in the joint architecture of pattern generators for whisking and breathing ( Moore et al . , 2013 ) . Further , our results on the ∼8 Hz vocalization rate are in line with an earlier report that there is a selective increase in the representation of sounds repeated at an ethological rate ( Kim and Bao , 2009 ) . Our findings on the temporally precise audio-haptic coordination in social signaling are reminiscent of much earlier findings on the multisensory orchestration of sensory acquisition ( Welker , 1971 ) . RS neurons in the auditory cortex largely had little or no response to USVs . A very small fraction ( ∼10% ) however showed robust responses to calls . When we did not observe more numerous responses to USVs , we initially wondered if the freely interacting animals simply did not hear many of the USVs . However , this explanation can be ruled out as: ( i ) responses to both own and stimulus were the same even though the subject rats were obviously always close to their own calls and ( ii ) FS cells responded robustly and significantly to calls . To our knowledge , this is the first report of neuronal responses to conspecific calls in awake behaving rats . Previous reports on the neuronal encoding of USVs in the rat auditory cortex have employed playback experiments in anesthetized preparations ( Kim and Bao , 2013 ) or awake rats ( Carruthers et al . , 2013 ) . While these studies report reliable population responses to vocalizations , our data are very similar to another study that shows that there is a sparse representation of sounds in the unanaesthetized auditory cortex ( Hromádka et al . , 2008 ) . Using a method that is not biased to neuronal activity ( cell-attached recordings ) , the authors report that <5% of neurons responded robustly to sounds at any instant . Further , they also report that narrow-spiking inter-neurons are highly responsive . Of the strongly modulated neurons , we observe a range of responses: most cells displayed early ( <25 ms after call onset ) increases in firing rates , while a few late responders ( >50 ms after call onset ) were also observed . Responses tended to be sharp or sustained , with a few cells showing robust modulation by one or more call categories . At the population level , the major call categories evoked strong activation in the RS cell population whereas FS neurons appear to be modulated only by trills . Also , there does not appear to be any population level preference for one of these call types with relation to their best response . Interestingly , RS neurons seem to prefer the stimulus calls whereas FS neurons were strongly modulated by the animal's own calls . However , a cell-wise analysis of response to own vs stimulus calls does not show any difference , suggesting that this phenomenon is brought out at the population level . The role of FS neurons in blanking out the animal's own calls and facilitating increased response to stimulus calls would be a tempting speculation . There also appears to be auditory cortex sub-region-wise differences , in that AuD and Au1 respond significantly to calls . Despite the temporal coordination of whisking and calling , we do not observe any strong preference of the auditory cortex to the whisking phase at the population level . Rare examples of neurons strongly locked to retraction could be artifacts from locking of whisking and vocalization . We next analyzed the response of auditory cortex neurons to facial touch and observed an unexpected inhibition . This was particularly evident as an off-response; wherein at the end of the touch episode , a burst of firing was observed in the PSTHs indicating a possible release from inhibition . This was observed both in RS and FS neurons , but only in those from Au1 . These observations are potentially confounded by the fact that whisker/snout contacts could in themselves produce sounds . Indeed , high sensitivity to broadband stimuli has been reported in the cat auditory cortex ( Bar-Yosef and Nelken , 2007 ) . Sound intensity measurements were performed but no change was detected due to whisker movement and contacts , at least at the microphones . Both social and non-social touch induce inhibition , suggesting that the modulation we observe is likely to be tactile in nature . An anatomical basis for these observations would lie in the direct connectivity between somatosensory and auditory cortices ( Budinger et al . , 2006 ) . Even prior to the cortical processing , multisensory integration of sound and touch has been shown to occur in the dorsal cochlear nucleus ( Young et al . , 1995; Kanold and Young , 2001; Shore , 2005 ) . It has also been suggested that a possible role for touch in the auditory areas is to enhance responses to non-self vocalizations while at the same time suppressing responses to self-generated sounds such as calls or respiration ( Shore and Zhou , 2006 ) . Multisensory integration of visual ( Bizley et al . , 2007 ) and olfactory information ( Cohen et al . , 2011 ) in the auditory cortex has been demonstrated in ferrets and rats , respectively . Evidence for multisensory integration in the primary auditory cortex also comes from studies on the modulation by touch ( Fu et al . , 2003; Kayser et al . , 2005 ) and vision ( Kayser et al . , 2007 ) in primates . The extensive modulation of primary auditory cortex by touch implies a possible role for touch-induced inhibition during social interactions where there is an increased vocalization . To test this , we analyzed the responsiveness of Au1 neurons to USVs in and out of touch . Interestingly , RS neurons exhibited a greater modulation to calls during touch as compared to out of touch . However , this was not the case with FS neurons . This large modulation of call responsiveness by touch could be construed as a sampling issue as calls during touch are fewer than those out of touch . However , bootstrapping analysis rules out this interpretation . Similarly , a higher intensity of stimulus calls during touch ( due to close proximity ) is less likely to be a contributing factor . Interestingly , there are several lines of evidence to suggest that inhibition in the auditory cortex could actually be responsible for an increased responsiveness to auditory stimuli . It has been reported that balanced inhibition underlies tuning and sharpens spike timing in auditory cortex ( Wehr and Zador , 2003 ) and that inhibition might contribute to auditory processing ( Hamilton et al . , 2013; Shamma , 2013 ) . In the light of these findings , it would be interesting to understand how multisensory response modulation of auditory cortex neurons affects the behavior of interacting animals . Wistar rats ( 45- to 60-day old , female and male ) were commercially procured ( Harlan , Eystrup , Germany ) and housed with a 12:12 hr inverted light/dark cycle and ad libitum access to food and water . While implanted ( ‘subject’ ) rats were housed individually after surgery , ‘stimulus’ rats were housed in groups of 2–3 . After a 1-week post shipment recovery , rats were handled for 2–3 days , following which they were habituated to the behavioral setup for 3–4 days . All experimental procedures were performed in accordance to German regulations on animal welfare ( Permit no . G0259/09 ) . The gap paradigm ( Wolfe et al . , 2011; Bobrov et al . , 2014 ) was used to study social interactions between conspecifics , wherein facial touch episodes occur freely across a gap between rats placed on two platforms ( 30 cm × 25 cm , Figure 1A ) . The platforms ( enclosed by 35 cm high walls on three sides ) were elevated ( 20 cm ) and placed in a Faraday cage to reduce electrical noise . The gap between the platforms was set to 20 ± 2 cm , depending on the size of the interacting partners . The entire setup was enclosed with black curtains and the room was darkened during experiments . An overhead low-speed camera ( 30 Hz ) was used for continuous videography under infrared illumination ( ABUS , Wetter , Germany ) . In most experiments , each recording session consisted of a 5 min baseline during which the subject rat was alone in the setup , a 5 min interaction time when stimulus animals/objects were presented across the gap , and another 5 min baseline at the end of interactions . On each recording day , 2–7 such recording sessions were conducted with various combinations of stimuli presented in a pseudo-random order . Stimuli presented included female and male conspecifics ( 60- to 120-day old ) , an object ( Styrofoam block ) and plastinated rats . These stimuli were presented either individually or in pairs ( Figure 1A ) . Foam mats on the stimulus platform were changed between recording sessions to minimize olfactory cues . Offline analyses of videos were used to identify episodes of social facial interactions , and the following behavioral events were scored for: whisker overlap onset ( Figure 1B , left ) , snout touch onset ( Figure 1B , right ) , snout touch offset , and whisker overlap offset . The time of placement/removal of stimuli into/out of the setup was also scored for . Ultrasonic vocalizations produced by the rats were recorded using four microphones ( condenser ultrasound CM16/CMPA , frequency range 10–200 kHz , Avisoft Bioacoustics , Berlin , Germany ) placed under the elevated platforms ( Figure 1A ) . Data were acquired using UltraSoundGate 416H at a sampling rate of 250 kHz and 16-bit resolution using Avisoft-RECORDER USGH software ( Avisoft Bioacoustics , Berlin , Germany ) . Acoustic analysis was performed using Avisoft SASLab Pro ( Avisoft Bioacoustics , Berlin , Germany ) . Spectrograms were generated using the following fast Fourier transform ( FFT ) parameters: length of 1024 points and an overlap of 93 . 75% ( FlatTop window , 100% frame size ) . The spectrograms had a frequency resolution of 244 Hz and a time resolution of 0 . 256 ms . Custom-written MATLAB codes ( Source code 1 ) were also used to generate spectrograms using the Blackman–Harris window which resulted in a clearer appearance of frequency-modulated calls with steeper decays and a higher resolution ( MathWorks , Natick , MA , USA ) . FFT parameters were: 500 points length and 80% overlap , resulting in 58 Hz frequency resolution and 0 . 4 ms time resolution . Start and end times of calls were manually set , and call category assignment was done as per an earlier classification ( Wright et al . , 2010 ) . A total of 56 , 525 USVs were identified and classified by two different experimenters while being blind to the behavioral events . A small fraction of calls ( 5 . 8% ) could not be unambiguously assigned into any one of these categories and were classified as ‘unspecified’ . In addition to call duration , mean frequency and band widths were computed . A subset ( 48 , 170 vocalizations ) was acquired alongside recordings from the auditory cortex and used to compute neuronal responses to USVs . To assign call source , low-speed videos of periods with only one animal present were superimposed with ultrasound recordings from all four channels . Analysis of these videos demonstrated a high degree of vocalization directionality . Hence , a comparison of call intensities on each of the channels was used to assign the source . A majority of USVs ( 80% ) were assigned to either the subject or stimulus rats while 20% remained unassigned ( and subsequently omitted from all emitter-dependent analyses ) . To check if the fraction of unassigned calls was greater during touch episodes where close facial proximity could obscure the exact source , we analyzed 11 , 194 calls that occurred during touch . Only 17 . 8% of these were unassigned , which is similar to the overall unassigned fraction , ruling out any obfuscation of source assignment due to touch . Noise ( generally not directional ) or simultaneous calls ( rare ) perturb this algorithm in principle , but visual inspection of spectrograms showed that such interferences were the exception . To estimate the accuracy of this method , we analyzed the false positive rates ( i . e . , calls wrongly assigned to stimuli ) in two conditions: ( i ) recordings of subject rats interacting with plastinated rats and objects: in 13 recordings with 1833 calls , 1286 calls ( 70% ) were correctly assigned to the subject , 284 calls ( 15% ) were wrongly assigned to stimuli , and 263 calls ( 14% ) were unassigned , ( ii ) recordings when subject rats were present alone on the setup: out of 4299 calls , 3406 calls ( 79% ) were correctly assigned , 196 calls ( 5% ) were wrongly assigned to stimuli , and 697 calls ( 16% ) were unassigned . Having excluded the unassigned calls from emitter-dependant analyses , we hence estimate that >4 out of 5 calls were assigned to the correct source . To compare intensity changes due to whisker touch , snout touch , and USVs , power spectral densities ( PSD , as a function of time and frequency ) were computed around these triggers . Integration over time verified that in the analyzed time window , power spectrum was indifferent to the trigger and frequency distribution was identical before and after it . Power was then cumulated over frequencies , which yields relative intensity as a function of time . Since the intensity values span several orders of magnitude , the geometric mean was used for comparison . Same number of triggers was randomly chosen to circumvent sample size issues . Interactions with vocalizations occurring in the time frame of interest as well as vocalizations that occurred simultaneously or in succession were excluded . All remaining episodes were averaged and traces plotted relative to the average during the pre-trigger phase . The pre-trigger average PSD was comparable across triggers ( in the order of 10−8 V2 Hz−1 ) . Whiskers of both subject and stimulus rats were tagged under 2–4% isoflurane anesthesia with small spherical drops of high-viscosity epoxy glue ( Dymax 3021 UV-adhesive , Dymax Europe , Wiesbaden , Germany ) that was hardened using ultraviolet light ( Bluewave 50 , Dymax Europe , Wiesbaden , Germany ) . The tag was covered with silver paint and fixed with superglue . Stimulus rats also received a black dot on the head to facilitate head tracking . Social interactions were recorded using a high-speed camera ( A504k , Basler AG , Ahrensburg , Germany ) at 250 Hz with 1280 × 1024 pixels . Video frames were streamed directly to a PCIe 1429 express card ( National Instruments Corporation , Austin , TX , USA ) . Acquisition was controlled by custom-written Labview ( National Instruments Corporation , Austin , TX , USA ) programs . Whisker ( one per animal ) tracking was performed for a subset of 58 interactions ( 194 s of video ) where the whiskers were clearly visible . Tracking was done as has been described earlier ( Wolfe et al . , 2011; Bobrov et al . , 2014 ) using a custom-written code ( Source code 2 ) . The process consisted of manual setting of head center and nose positions , automated contour detection and setting of whisker base and tag positions . Head axis ( posterior to anterior ) and whisker direction ( outwards from whisker base ) were acquired with custom-written MATLAB software . The head–whisker-angle was defined as zero in the orthogonal position , with protraction taking positive values up to a theoretical maximum of 180° and retraction being negative analogously . Whisker traces , that is , time series of whisker angle , were processed for signal cleansing , quality criteria , and signal decomposition as described elsewhere ( Hill et al . , 2011 ) . This includes up-sampling to 1000 Hz and 1–25 Hz bandpass filtering ( fourth order Butterworth filter ) . Whisker cycles were disregarded when not between 50 and 250 ms length or when amplitude was <7 . 5° . Hilbert transform yielded the phase within the whisking cycle starting at maximum protraction ( 0° ) , via maximum retraction ( back-directed movement up to 180° ) and return ( ≤360° ) . Whisking cycle phases at the onset of calls were analyzed for both the calling animal and the one towards which the calls were putatively directed . To compute the locking of neuronal firing to whisker positions , spike time stamps for each cell were aligned to whisker traces and binned in a circular fashion before being normalized for phase occupancy . To the resulting polar plots , circular Rayleigh statistics were applied to calculate the Rayleigh vector ( the summation of unit vectors in all individual angle observations to find a tendency in overall direction ) . The following criteria was set to identify putative phase locked cells: ( i ) minimum number of spikes ( 10 ) in a tracked episode , ( ii ) Rayleigh vector greater than 0 . 2 , and ( iii ) shuffling test ( see below ) showing significance at a confidence interval of 5% . The shuffling test is necessary because regular tests , such as Rayleigh or Hodges–Ajne test for non-uniformity of circular data , are sensitive to very low- or high-spike numbers because of the extreme ( high or low ) variance among bins . Also , the number of bins can affect the test result . To test independently of these , the spike train was randomly time shifted relative to the whisker phase trace . The Rayleigh vector was computed again and this procedure repeated 10 , 000 times to simulate a random set of Rayleigh vectors from the spike series . If the original Rayleigh vector was in the 95% quantile of that distribution , it was presumed to be phase locked . To analyze periodicity of calling , the intervals between successive calls were computed . For whisking cycles , the point of maximum protraction served as a marker . The spectral power density of whisking and calling was determined using Welch's method ( ‘pwelch’ , MATLAB signal processing toolbox ) in the range of 4–25 Hz . This was applied to the approximately continuous whisker trace and to the outline of the calling interval histogram , the latter with lower resolution and less power due to the rough discretization . Neuronal activity was recorded using a chronic microdrive ( Harlan 8-drive , Neuralynx , Bozeman , MT , USA ) consisting of eight independently movable tetrodes ( arranged in a 4 × 2 array ) . The tetrodes were fashioned out of 12 . 5-µm diameter nichrome wire ( California Fine Wire Company , Grover Beach , CA , USA ) and gold-plated to a resistance of 250–300 kΩ ( nanoZ , Neuralynx , Bozeman , MT , USA ) . The microdrives were implanted on 8 ‘subject’ rats ( four male , four female ) spanning the following Bregma locations: −3 . 12 to −6 . 00 mm AP; 6 . 5 to 7 . 00 mm ML ( Paxinos and Watson , 2006 ) . For implanting the drive , rats were subjected to ketamine ( 100 mg/kg body wt ) /xylazine ( 7 . 5 mg/kg body wt ) anesthesia . Booster doses of anaesthetics were administered as required . Body temperature was maintained with a heating pad and continuously monitored by a rectal probe ( Stoelting , Wood Dale , IL , USA ) . After securing the animal's head onto a stereotactic apparatus ( Narashige Scientific Instrument Lab . , Tokyo , Japan ) , lidocaine was injected subcutaneously under the scalp . After retraction of the temporal muscle , the cleaned skull surface was treated with a UV-activated etchant-cum-glue ( Optibond All-In-One , Kerr Italia , Salerno , Italy ) . Gold-plated screws were fixed away from the craniotomy site to anchor the drive and provide for grounding . After craniotomy and durectomy , the microdrive was positioned on the brain and the area was covered with 1% agarose . The microdrive was secured with dental cement ( Paladur , Heraeus Kulzer , Hanau , Germany ) . Tetrodes were lowered into the brain and recordings typically began 1–2 days after surgery . Tetrodes were advanced by a minimum of 80 µm between recording days to ensure that new high quality units were sampled during the course of the experiment . After passing through a unity-gain headstage , signals were transmitted via a tether cable to an amplifier ( Digital Lynx , Neuralynx , Bozeman , MT , USA ) . Spike signals were amplified ( 10× ) , digitized at 32 kHz , and bandpass-filtered between 0 . 6 and 6 kHz . Events that crossed a user-set threshold were recorded for 1 ms ( 250 µs before and 750 µs after voltage peak ) . At the end of the experiment , subject rats were deeply anaesthetized using ketamine/xylazine and electrolytic lesions ( Figure 3—figure supplement 1A ) were performed by injecting 10 µA negative current through the tetrodes for 10 s ( nanoZ , Neuralynx , Bozeman , MT , USA ) . Transcardiac perfusion was performed with cold phosphate buffer and 4% paraformaldehyde . The brains were dissected out and post-fixed in 4% paraformaldehyde ( overnight ) . Coronal sections ( 150 µm ) were stained for cytochrome oxidase and visualized using light microscopy to identify the lesions . Recording depths ( determined by number of microdrive turns ) were used to identify exact recording sites and cortical layers ( Paxinos and Watson , 2006 ) relative to lesions after accounting for shrinkage during tissue processing using Neurolucida ( MBF Bioscience , Williston , VT , USA ) . Amplitude and principal components were used for offline spike sorting using the semiautomatic clustering algorithm KlustaKwik ( KD Harris , Rutgers University , Newark , NJ , USA ) . Manual correction and refinement were applied with MClust ( AD Redish , University of Minnesota , Minneapolis , MN , USA ) using MATLAB ( MathWorks , Natick , MA , USA ) . Spike features ( energy and first derivative of energy ) were used for separation . Inclusion criteria for single units were determined by refractory period , separation quality , and stability . Inter-spike interval histograms with minimal or no contamination in the first 2 ms were indicative of a single unit . Separation quality was determined by L-ratio ( <0 . 2 ) and isolation distance ( >15 ) ( Schmitzer-Torbert et al . , 2005 ) . Stability of the units was quantified as follows: for each recording session , time periods outside interactions ( as many in number and length as interactions on that day ) were selected . This was performed on randomly distributed periods for 1000 permutations , and a linear correlation between time and firing rate was calculated . Average Pearson's R value was used as a measure of stability ( higher R means stronger drift ) and units with a value >0 . 4 were excluded from analysis . Spike shapes were used to classify units as putative regular-spiking ( RS ) or fast-spiking ( FS ) neurons . RS neurons have been shown to have wider action potentials while FS neurons on the other hand have narrower action potentials ( Atencio and Schreiner , 2008 ) . Auditory cortex single unit waveforms normalized by peak voltage were used to compute a whole host of features ( height , trough , width parameters ) . Of these , two closely related features , full spike width and second half width showed bi-modal distributions ( Figure 3—figure supplement 1D , E ) . For this analysis , the waveforms were compared to the widest spike in the dataset ( which was designated the value of 1 ) . Consequently , the spike widths of thinner spikes get assigned with negative values ( Figure 3—figure supplement 1D , E ) . k-means clustering of units with two clusters resulted in well separated populations ( indicated by dotted line , [Figure 3—figure supplement 1F] ) , albeit indicating a high correlation between these two features ( arising possibly due to the spike waveforms being narrower than measured , which in turn is determined by filter settings ) . Indeed , spikes from these two populations resulted in well-defined average spike shapes ( Figure 3—figure supplement 1G ) and were well differentiated with respect to firing rates ( Figure 3—figure supplement 1H ) . To quantify the response of each neuron , we computed the average firing rate in the following response windows: in call duration alone and in a call duration + 25 ms window . For RS neurons , pair-wise comparison of basal firing rates were significantly different to the firing rates in the call duration alone ( p = 0 . 0049 ) and call duration + 25 ms ( p = 0 . 0006 , Wilcoxon matched pairs signed rank test ) response windows . However for FS neurons , the response was significantly different in the call duration + 25 ms window ( p = 0 . 006 ) , while it was not the case during the call duration alone ( p = 0 . 0605 ) . We also computed the onset responses in various time windows ( 0–25 , 26–50 , 51–75 , 76–100 ms after call onset ) as reported earlier ( Hromádka et al . , 2008 ) . For RS neurons , we found significant increase in all four time windows while for FS it was significant only for the 26–50 and 51–75 ms time windows . It must be noted that since the stimuli are not all of the same duration ( for e . g . , trills: 50 . 9 ± 26 . 2 ms , mean ± SD , Table 1 ) , the ideal response windows should span the entire call durations . Thus it appears that the call duration + 25 ms response window sufficiently well describes the responses of both RS and FS neurons in our study and this was used to compute the neuronal responses to USVs . For facial touch , firing rate was defined as average firing rate during all interactions with an interaction partner . The off-response due to facial touch was calculated as the mean firing rate in a 200 ms window after the end of the touch episode . These firing rates were compared with a matched baseline period , which was as long as the USVs/interactions and shifted −10 , 000 ms relative to these on the spike train ( jumping over any periods which contained a USV or interaction ) . For each neuron , a response index was calculated as follows: Response Index = ( in − out ) / ( in + out ) , where in and out are the firing rates during a call/interaction and baseline firing rates , respectively . Data were analyzed using Prism 6 ( GraphPad Software Inc . , La Jolla , CA , USA ) or MATLAB ( MathWorks , Natick , MA , USA ) and are presented as mean ± SEM unless stated otherwise . Whisking phase preference of vocalizations was tested by Hodges–Ajne test . Since most of the data were not normally distributed ( as determined by D'Agostino & Pearson omnibus normality test ) , differences between groups were tested with Wilcoxon signed rank test for paired data and Mann–Whitney U test for unpaired data . Comparison of distribution widths was performed by Kolmogorov–Smirnov test .	Rats are highly social creatures , preferring to live in large groups within an established hierarchy . Social interactions range from play , mating , and parental care to displays of aggression and dominance and involve the use of odors , touch , and vocal calls . Touch typically takes the form of snout-to-snout contact , while most vocalizations are ultrasonic , with calls of different frequencies used to signal alarm or pleasure . To date , most studies of rat vocalizations have involved playback of recorded calls to anaesthetized animals , and relatively little is known about how freely moving rats respond to calls . Rao et al . have now addressed this question by recording video footage of rats interacting with other animals or with objects and then using electrodes to record signals in the brains of these rats . The video footage revealed that rats produce more vocal calls during social interactions than they do during non-social interactions . Moreover , bursts of calls appear to signal the beginning and end of bouts of snout-to-snout contact , suggesting that rodent communication involves the coordinated use of both tactile and vocal cues . Surprisingly , electrode recordings from the part of the brain that responds to sound—the auditory cortex—revealed that most neurons in this region did not respond to ultrasonic calls . However , a type of neuron called a fast-spiking neuron did respond strongly to these calls . The work of Rao et al . shows that information from multiple senses is directly combined early in the processing of sensory information . Exactly why tactile stimuli should inhibit the auditory cortex is not clear , but there is some evidence that this may increase the rat's sensitivity to sounds . Further experiments are required to test this possibility and to determine how integrating information from multiple senses affects rodent behavior . This will help us to understand how the brain generates coherent social behaviour from signals arriving through distinct sensory channels .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Vocalization–whisking coordination and multisensory integration of social signals in rat auditory cortex
Storing temporal sequences of events ( i . e . , sequence memory ) is fundamental to many cognitive functions . However , it is unknown how the sequence order information is maintained and represented in working memory and its behavioral significance , particularly in human subjects . We recorded electroencephalography ( EEG ) in combination with a temporal response function ( TRF ) method to dissociate item-specific neuronal reactivations . We demonstrate that serially remembered items are successively reactivated during memory retention . The sequential replay displays two interesting properties compared to the actual sequence . First , the item-by-item reactivation is compressed within a 200 – 400 ms window , suggesting that external events are associated within a plasticity-relevant window to facilitate memory consolidation . Second , the replay is in a temporally reversed order and is strongly related to the recency effect in behavior . This fast-backward replay , previously revealed in rat hippocampus and demonstrated here in human cortical activities , might constitute a general neural mechanism for sequence memory and learning . Storing and retrieving temporal sequences of events ( i . e . , sequence memory ) , a capacity shared across species , is crucial to many cognitive functions , including speech recognition , movement planning , and episodic memory ( Doyon et al . , 2003; Giraud and Poeppel , 2012 ) . For example , we retrieve a stream of serially ordered events when recalling past personal experience , and we memorize numbers in order when calling a friend on the phone . Swinging the racket to hit an incoming ball in a tennis match similarly involves the planning and controlling of a chain of movement elements over time . To accomplish sequence memory , two core components – the content ( items ) and the ordinal information ( temporal order ) – of a sequence are vital to be encoded and maintained in working memory . There exists ample evidence that neural responses during the retention interval show a sustained load-dependent enhancement ( Buschman et al . , 2011; Jensen et al . , 2002; Klimesch et al . , 1999; Sauseng et al . , 2009; Vogel and Machizawa , 2004; Xu and Chun , 2006 ) , suggesting that the maintenance of mnemonic contents is implemented by recurrent feedback loops ( Luck and Vogel , 2013 ) , as well as suppression of irrelevant information ( Sauseng et al . , 2009; Bonnefond and Jensen , 2012 ) . That being said , retention of the sequence order information in memory cannot solely rely on an overall enhancement of neural activity and presumably requires the temporally segregated representations of individual items . Theoretical models postulate that sequence memory is mediated by a theta-gamma coupled neuronal oscillatory mechanism ( Jensen and Lisman , 2005; Lisman and Idiart , 1995 ) , such that individual items of the list/sequence , encoded in gamma-band activities , occur at the different phases of a theta-band rhythm . A recent MEG study provides important evidence supporting this theta-gamma coupling model during the memory encoding period ( Heusser et al . , 2016 ) . By asking subjects to mentally replay short sound or video clips , the temporal phase patterns of the neural activities during memory retrieval are found to be similar to that during memory encoding , further advocating preservation of temporal information of the stimulus in working memory ( Michelmann et al . , 2016 ) . Working memory contents have also been suggested to be maintained in rapid transition in dynamic hidden states ( Stokes , 2015; Wolff et al . , 2017 ) . However , it remains unknown how the human brain represents and maintains temporal order information during the memory retention period , when the stimulus sequence is not present . Despite this gap , it is well-established in animal studies that place cells in rodent hippocampus fire in the same or reverse order in which they appeared during previous spatial navigation , in both awake and asleep status ( Diba and Buzsáki , 2007; Foster and Wilson , 2006; Louie and Wilson , 2001; Skaggs and McNaughton , 1996 ) . In contrast to rodent studies in which sequential trajectory of place cell firing could be explicitly assessed to explore reactivation profiles during memory maintenance , human studies have provided more limited access to the item-specific reactivation profiles at the neuronal ensemble level . To overcome this difficulty , in the present study , we employed sequence memory paradigms in combination with a temporal response function ( TRF ) method ( Crosse et al . , 2016; Jia et al . , 2017; Lalor et al . , 2006 ) to dissociate the item-specific neuronal response during the memory retention interval . Our results demonstrate that each item in a memory list , characterized by an alpha-band neural profile , is successively reactivated to mediate the representation and preservation of ordinal information in memory . Importantly , this item-by-item neuronal profile does not simply echo the sequence in the external world but instead displays two intriguing properties compared to the actual stimulus sequence: reversal in order and compression in time . Specifically , the item that occupies a late position in the list is reactivated earlier ( e . g . , for sequence 1 – 2 , the reactivation profile is 2 – 1 in order ) , and the whole list encompassing all items is compacted within a 200 – 400 ms temporal window . Most importantly , this backward reactivation profile is directly related to the subsequent recency effect in memory behavioral performance . Our study provides novel neural evidence in humans that the sequence order information is encoded and preserved in a fast and reverse reactivation sequence . Previous studies have provided evidence supporting the object-based nature of working memory such that all features of an object , even the task-irrelevant ones , will be stored automatically in working memory ( Gao et al . , 2011; Hollingworth and Luck , 2009; Hyun et al . , 2009; Luck and Vogel , 1997 ) . Most importantly , color features have been found to be the type of feature having the strongest conjunction with other features within an object ( Gao et al . , 2011; Johnson et al . , 2008; Wheeler and Treisman , 2002 ) . Based on these findings , in Experiment 1 , we used task-irrelevant color probes that were either memory-related or non-memory-related to examine whether we could tag the memory reactivations during the retention period when the to-be-remembered features ( i . e . , orientation features ) were not present . We recorded 64-channel EEG signals from human participants performing an adapted working memory task ( Sawaki and Luck , 2011 ) . Each trial consisted of three periods: encoding , maintaining , and recalling ( Figure 1A ) . In short , subjects were instructed to memorize the orientation of the cued bar ( ‘encoding’ ) and then held the information in working memory ( ‘maintaining’ ) , and finally performed a memory test on the orientation feature ( ‘recalling’ ) . During the maintaining period , participants performed a central fixation task ( to control eye movements ) , with two task-irrelevant color discs presented simultaneously . Crucially , the two probes were either memory-related ( i . e . , with memory-matching color , WM ) or non-memory-related ( i . e . , with memory-nonmatching color , NWM ) , and neither contained the to-be-memorized orientation features . They were displayed at either the left or right side of the fixation cross to exclude spatial memory effects . Next , a TRF approach ( Jia et al . , 2017; Lalor and Foxe , 2010; Ding and Simon , 2012 ) , which has been used to assess brain response that tracks ongoing changes in sound envelope ( Lalor and Foxe , 2010; Ding and Simon , 2012 ) , visual luminance ( Lalor et al . , 2006; VanRullen and Macdonald , 2012 ) , and even high-level properties ( Liu et al . , 2017; Broderick et al . , 2018 ) , was employed to extract and isolate neuronal responses for the ongoing luminance change of the two probes ( WM and NWM ) , respectively , throughout the maintaining period . Specifically , the luminance of the two discs was modulated continuously according to two 5 s random temporal sequences that were generated anew in each trial ( Figure 1B ) . The TRF responses ( Figure 1C ) for the WM disc probe ( WM-TRF ) and NWM disc probe ( NWM-TRF ) were then calculated and dissociated from the same 5 s EEG recordings , based on their corresponding stimulus temporal sequences ( Figure 1B ) . Note that the TRF response is defined as the brain response to a unit luminance in the stimulus sequence , as a function of temporal lag after each transient . As a result , the two calculated TRF responses ( i . e . , WM-TRF and NWM-TRF ) would represent the impulse response for the luminance sequence of the WM and NWM probes , respectively , throughout the memory retention interval ( see details of TRF approach in Materials and methods ) . If the color feature is not specifically bound to the associated orientation features held in memory , we would expect no difference between the WM-TRF and NWM-TRF responses . On the other hand , if the two probes show distinct responses , it suggests that the task-irrelevant color probes could be used to tag the memory-related activations during retention . Eighteen participants participated in Experiment 1 , and their memory performance was well above chance ( N = 18 , mean accuracy = 0 . 888 , s . e . = 0 . 029; One-sample t-test , df = 17 , t = 26 . 34 , p < 0 . 001 , CI: [0 . 357 , 0 . 419] , Cohen’s d = 6 . 208 ) . First , the overall TRF response for the color probes showed clear alpha-band patterns and was robust at single-subject level ( Figure 2—figure supplement 1 ) , consistent with previous findings ( Jia et al . , 2017; VanRullen and Macdonald , 2012 ) . We next performed a spectrotemporal analysis on the WM-TRF and NWM-TRF responses ( see the corresponding TRF waveforms in Figure 2—figure supplement 2A ) to examine their fine dynamic structures as a function of frequency ( 0 – 30 Hz ) and time ( 0 – 0 . 6 s ) , and this was done on each channel and in each participant separately . As shown in Figure 2A , both WM-TRF ( left panel ) and NWM-TRF ( right panel ) demonstrated clear sustained alpha-band activations ( 8 – 11 Hz; permutation test , p < 0 . 05 , corrected , see Figure 2—figure supplement 2C ) , consistent with previous findings ( VanRullen and Macdonald , 2012 ) . Given the prominent alpha-band activations in the TRF responses , we then first selected channels that showed overall significant larger WM + NWM alpha-band responses ( one-sample t-test , p < 0 . 05; compared to the average of all channels ) . Among the resulting channels , three electrodes ( red dots in Figure 2B ) passed the statistical test on the WM-NWM alpha-band difference ( bootstrap test , p < 0 . 06 , see details in Materials and methods; FDR corrected across channels and time ) and were defined as a ‘channel-of-interest’ for allsubsequent analyses , including those in Experiment 2 . As shown in Figure 2C , a ‘channel-of-interest’ showed enhanced alpha-band power for WM-TRF compared to NWM-TRF within the latency of around 200 – 500 ms . We then extracted the alpha-band time courses within the channel-of-interest . As shown in Figure 2D , WM ( red ) showed significant alpha-band enhancement over NWM ( black ) from 310 ms to 370 ms ( bootstrap test , p = 0 . 046 , FDR corrected across time ) , supporting the alpha-band memory effect . The WM and NWM probes were presented in two visual fields , and therefore the observed alpha memory effect could be a result of spatial attention ( Worden et al . , 2000 ) . We performed two additional analyses to address this issue . First , we examined the alpha-band memory effects ( WM – NWM ) for the left and right discs separately and observed a similar spatial distribution without statistically significant lateralization effects ( paired t-test , N = 18; see Figure 2—figure supplement 3A ) . Second , we compared the alpha power between channels that are contralateral to the WM probe and channels contralateral to the NWM probe and did not find any significant difference ( paired t-test , N = 18; see Figure 2—figure supplement 3B ) . Thus , the alpha-mediated memory effects during maintenance could not be interpreted by spatial attention . This is also consistent with recent findings supporting memory-related hidden states in the absence of attention ( Wolff et al . , 2017 ) . It is noteworthy that during the central fixation task ( see eye movement profile in Figure 2—figure supplement 4B ) , WM and NWM probes were both task-irrelevant , and the only difference between them regarded color ( i . e . , memory-related or non-memory-related ) . Therefore , the observed difference between the WM-TRF and NWM-TRF responses indicates that the task-irrelevant color probes did carry memory-related information , consistent with previous findings advocating the object-based nature of working memory ( Gao et al . , 2011; Hollingworth and Luck , 2009; Hyun et al . , 2009 ) . The results also suggest that we could use task-irrelevant color probes to tag the memory-associated reactivations during maintenance , using the alpha-band temporal profile as a neural signature ( no effects in other frequency bands , Figure 2—figure supplement 3C ) . See further analysis on the alpha-band in TRF responses in Figure 2—figure supplement 4A . We next examined how a temporal sequence and the associated order information is maintained in working memory . Nineteen participants participated in Experiment 2 . As shown in Figure 3A , participants were first presented with a sample array containing three bars with different orientations and different colors . Next , two of the three bars were randomly selected and serially cued , and importantly , participants were asked to memorize not only the orientation of the two cued bars but also their temporal order ( i . e . , the 1st and the 2nd orientation ) . During the following 5 s ‘maintaining period’ , participants performed a central fixation task and held the orientation sequence in memory . In the final ‘recalling period’ , participants compared the orientation of the bar in the test array with the memorized orientations ( e . g . is it more similar to the orientation of the 1st or the 2nd cued bars ? ) . Therefore , subjects should be memorizing the temporal order of the orientation sequence . Crucially , the probe array during the maintaining period consisted of three disc probes – one 1st-memory-related probe ( i . e . , matching color with the 1st cued bar ) , one 2nd-memory-related probe ( i . e . , matching color with the 2nd cued bar ) , and one non-memory-related probe ( i . e . , matching color with the non-cued bar ) . The three discs were presented at three random spatial locations on a ring to exclude spatial memory effects . Again , the luminance of the three disc probes was modulated continuously according to three random sequences respectively . The corresponding TRF responses ( 1st-WM-TRF , 2nd-WM-TRF , and NWM-TRF ) were then computed from the same 5 s EEG recordings , and a spectrotemporal analysis was performed on them for each channel and in each participant separately ( see TRF waveforms in Figure 2—figure supplement 2B ) . Notably , all the analysis was performed in the channel-of-interest , which was independently defined in Experiment 1 ( see Figure 2B ) . First , significant recency effect ( i . e . , 2nd item better than 1st item ) in behavioral performance was found ( N = 19 , 1st item: mean = 0 . 794 , s . e . = 0 . 020; 2nd item: mean = 0 . 840 , s . e = 0 . 014; recency effect: mean = 0 . 047 , s . e . = 0 . 020; paired t-test , df = 18 , t = 2 . 378 , p = 0 . 029 , CI: [0 . 005 , 0 . 089] , Cohen’s d = 0 . 532 ) . Moreover , the alpha-band memory effects ( i . e . , WM – NWM difference averaged within the whole 600 ms time range , bootstrap test , p = 0 . 047 ) revealed in Experiment 1 were still present ( Figure 3B ) . Most interestingly , the 1st-WM-TRF and the 2nd-WM-TRF displayed distinct alpha-band temporal courses ( Figure 3C ) : the 1st-WM-TRF alpha-band response ( red line ) was temporally delayed by approximately 200 ms relative to that of the 2nd-WM-TRF ( blue line ) response ( alpha-band peak latency , Wilcoxon signed-rank test , p = 0 . 036 ) . We next normalized the alpha-band power temporal profiles by subtracting the averaged alpha-band time courses across all the three TRFs from that for 1st-WM-TRF , 2nd-WM-TRF , and NWM-TRF , respectively . Figure 3D illustrates the normalized alpha-band temporal profile , showing a reverse sequential activation profile ( i . e . , alpha-band activation for the 1st item followed by that for the 2nd item ) . Specifically , the 2nd-WM-TRF showed a response at around 80–100 ms , whereas the 1st-WM-TRF showed a response at around 260–300 ms ( p < 0 . 05 , one-sided t-test ) . The backward sequential response profiles were consistent across electrodes and items ( see control analysis for each channel and for different item combinations in Figure 3—figure supplement 1CD ) . To examine the statistical significance of the sequential activation , we calculated the cross-correlation coefficient between the normalized alpha-band profile for the 1st-WM-TRF and the 2nd-WM-TRF responses and then performed a permutation test by shuffling condition labels . As shown in Figure 3E ( top panel ) , the 1st-to-2nd cross-correlation coefficient at a temporal lag of around –170 ms was significant ( permutation test , p < 0 . 05 , corrected ) , supporting that the 2nd item alpha-band activations essentially preceded those of the 1st item ( i . e . , negative temporal lag ) by around 170 ms . Notably , the sequential activation pattern was still present for direct comparisons between the 1st-WM-TRF and 2nd -WM-TRF responses ( Figure 3—figure supplement 1A; Figure 3—figure supplement 1B ) , supporting that it was not involvement of the NWM item ( i . e . , normalization ) that caused the results . Furthermore , we observed a similar but weaker trend of reverse serial reactivation profile in a three-item sequence memory task ( Figure 3—figure supplement 2 ) . Finally , in addition to the item-by-item sequential responses , we also observed a trend of repeated activations for the same item ( Figure 3D ) . We thus calculated the within-item autocorrelation coefficient to examine the temporal period of the recurrent activations . As shown in Figure 3E ( bottom panel ) , the temporal lag of ~390 ms was statistically significant ( permutation test , shuffling between condition labels , p < 0 . 05 , corrected ) . This suggests that the two items in the memory list are further organized into a recurring temporal chunk of approximately 400 ms , within which sequentially memorized items reactivate one after another in a reversed order . We further evaluated whether the alpha-mediated memory reactivation during the retention period might have behavioral relevance . We first examined the relationship between the WM – NWM alpha-band enhancement and the one-item memory performance in Experiment 1 . Participants were grouped based on their memory accuracy into ‘high-performance’ ( N = 9 , mean = 0 . 907 , s . e . = 0 . 018 ) and ‘low-performance’ ( N = 9 , mean = 0 . 796 , s . e . = 0 . 053 ) ( Figure 4A ) . We then used the same analysis to assess the alpha-band courses for the two groups separately . Similar to the grand average results ( Figure 2D ) , the high-performance group showed enhanced WM – NWM alpha-band power difference within 300–400 ms ( Figure 4B ) , whereas the low-performance group showed weaker enhancement ( Figure 4C ) . As shown in Figure 4D , the high- and low-performance groups displayed non-overlapping confidence intervals in the alpha-band memory effects ( WM – NWM alpha power difference averaged from 310 to 370 ms , a time range selected based on group results , see Figure 2D ) . Thus , the stronger the alpha-band memory effects ( i . e . , Alpha WM > Alpha NWM ) , the better the corresponding memory behavioral performances . Participants in Experiment 2 ( 2-item sequence memory ) were divided into two groups based on their memory recency effects ( memory performance 2nd item – memory performance 1st item ) : the ‘high-recency’ group ( N = 9 , mean = 0 . 116 , s . e . = 0 . 020 ) and the ‘low-recency’ group ( N = 9 , mean = –0 . 021 , s . e . = 0 . 017 ) ( Figure 4E ) . First , both groups ( Figure 4FG ) showed similar reverse sequential responses profiles , similar to those in the group results ( see Figure 3D ) . Notably , the ‘high-recency’ group displayed a strong and significant backward-sequential alpha profile ( Figure 4F; permutation test , 1 – 2 cross-correlation coefficient at a lag of −170 ms , p < 0 . 05 , Figure 4—figure supplement 1A ) , whereas the ‘low-recency’ group displayed weak and nonsignificant sequential responses ( Figure 4G; permutation test , 1 – 2 cross-correlation coefficient at lag of −170 ms , p = 0 . 28; Figure 4—figure supplement 1 ) . To further compare the difference in reactivation profiles between the two groups , we examined the 2nd-1st alpha power difference for the two groups at three time ranges ( phase 1: 140 – 180 ms; phase 2: 280 – 320 ms; phase 3: 470 – 510 ms; shaded regions in Figure 4FG ) , which were selected based on group results ( Figure 3D ) . As shown in Figure 4H , interestingly , the main difference between the high- and low-recency groups was on the 2nd TRF response ( Figure 4FGH ) . Specifically , the high-recency group elicited a stronger 2nd item response than the low-recency group ( phase 1 and phase 3 , non-overlapping confidence intervals , Figure 4H; permutation test by shuffling between high- and low-recency groups , p = 0 . 08 ) , whereas both groups elicited similar 1st item responses ( phase 2 , Figure 4H; permutation test by shuffling between high- and low-recency groups , p = 0 . 48 ) . Thus , our results support an essential association between the backward sequential reactivations and recency effect in sequence memory behavior . We also performed an additional analysis by dividing subjects based on their memory accuracy rather than the recency , and the two groups elicited quite similar backward sequential responses ( Figure 3—figure supplement 1E ) , excluding the possibility that it is memory strength rather than the recency effect that leads to the observed backward sequential responses . Taken together , the alpha-mediated reactivation profile during the memory retention period is related to subsequent memory behavioral performance . Importantly , there exists an essential relationship between the backward item-by-item reactivation during maintenance and the subsequent recency effect , a behavioral index for sequence memory . In Experiment 2 , subjects were instructed to decide whether the orientation of the bar in the test array was similar to the orientation of the 1st or the 2nd memorized orientations , and the behavioral performance showed a significant recency effect ( i . e . , 2nd item better than 1st item ) . Meanwhile , although the accuracy for both items was high , response bias might still be a confounding factor for the observed recency effect ( e . g . , biased to report the probe to better resemble the 2nd item ) . To address the issue , we ran a control study ( N = 11 ) employing a new behavioral procedure that would not be contaminated by response bias . Specifically , the experiment paradigm was the same as that for Experiment 2 except that rather than making a binary response in the recalling period , subjects were instructed to rotate a bar to the memorized orientation . By calculating the angular deviation between the reported orientation and the true orientation of the memorized bar , in combination with a probabilistic model ( Bays et al . , 2009 ) , we could estimate the target response probability for the 1st and the 2nd WM items separately , independent of response bias . As shown in Figure 5A , the target response probability was significantly larger for the 2nd item than for the 1st item ( df = 10 , one-tail paired t-test , t = 2 . 052 , p = 0 . 034 , CI: [−0 . 003 , 0 . 083] , Cohen’s d = 0 . 618 ) , confirming the recency effect and excluding the response bias interpretation . Furthermore , we combined the neural recordings in the control study ( N = 11 ) with those from Experiment 2 ( N = 19 ) , resulting in a 30-subject dataset . As shown in Figure 5B , the normalized alpha-band reactivation profile ( N = 30 ) showed a similar reverse sequential pattern . The 1st-to-2nd cross-correlation coefficient at a temporal lag of around –170 ms was significant ( permutation test , p < 0 . 05 , corrected; Figure 5C ) , again supporting that the 2nd item alpha-band activations essentially preceded those for the 1st item ( i . e . , negative temporal lag ) by around 170 ms . We recorded EEG responses in human participants performing sequence memory tasks and employed a TRF approach to probe the neuronal response that specifically tracks each item of the temporal sequence during memory retention . Our results consistently demonstrate that individual items are successively reactivated , characterized by a sequence of alpha-band activities . Compared to the actual stimulus sequence , the serial neural replay is temporally compacted within a 200 –400 ms chunk and is reversed in temporal order . Crucially , the backward reactivation is associated with the recency effect , a behavioral index signifying backward memory priority . Taken together , our results constitute novel neural evidence in humans that the sequence order information is encoded and maintained in short-term memory by fast and backward serial reactivations . Temporally sequenced activations have also been reported in several recent human studies . For example , after being trained in a reasoning task that involves selecting a statistically related object path , the brain spontaneously displays a fast successive representation of states ( Kurth-Nelson et al . , 2016 ) . Recently , an fMRI study demonstrated that flashing only the starting point of a learned sequence would trigger a prediction wave of responses in primary visual cortices ( Ekman et al . , 2017 ) . Finally , a MEG study examined the theta-gamma coupling strength as participants were presented with picture sequences , and found that the gamma power shifts along the phase of a theta rhythm as more items are added to memory , thus supporting the ‘phase coding’ model for episodic memory formation ( Heusser et al . , 2016 ) . Our results are different from this important work and reveal distinct mechanisms . First , we examined the memory retention period when items were not presented , whereas the previous work studied the encoding period when items appeared serially . Second , we demonstrate backward reactivations whereas the previous study supports a forward-playing profile , further suggesting a fundamental distinction between memory encoding and retention . We demonstrate that the item-by-item reactivation during memory retention period is temporally reversed in order . Previous neurophysiological recordings in monkeys have revealed forward serial activations in prefrontal population activity ( Siegel et al . , 2009 ) . Sequences within the hippocampus in rats have been observed in both forward and reverse order ( Diba and Buzsáki , 2007; Foster and Wilson , 2006; Louie and Wilson , 2001; Skaggs and McNaughton , 1996 ) , and interestingly , the backward sequence has been shown during awake periods immediately after spatial experience ( Foster and Wilson , 2006 ) . A recent MEG work found a backward trajectory of representations of states ( Kurth-Nelson et al . , 2016 ) during a reasoning task . Why is the sequential reactivation reversed in temporal order during memory maintenance ? One interpretation comes from a reinforcement learning model , which proposes that the recent item would be the starting and anchoring point for propagating information backward along incoming trajectories to build a recency-weighted running average of the rewards received for each action taken ( Foster and Wilson , 2006; Bornstein et al . , 2017; Gershman and Daw , 2017 ) . Another interesting possibility is derived from the ‘activity-silent’ model ( Stokes , 2015 ) , which proposes that the contents of working memory are mediated by dynamic hidden states . These memory-related hidden states could be pinged and inferred out during maintenance in the absence of attention and lingering delay activity ( Wolff et al . , 2017 ) . In our experiments , the TRF response could be regarded as the emerging item-specific activation after each pinging ( i . e . , each unit luminance transient in the probe stimulus ) . Consequently , the item associated with stronger excitability ( i . e . , the recent item ) would appear with shorter latency compared to the item associated with weaker excitability ( i . e . , the early item ) , resulting in reversed sequential reactivation as we observed here . Furthermore , instead of presenting each item sequentially ( Siegel et al . , 2009 ) , here all the items in the list were displayed simultaneously followed by serial cuing ( to exclude the possibility that serial presentation might result in different encoding strengths ) . How the sequence is presented ( sequential or simultaneous ) could be another factor accounting for the order of reactivations . Memory reactivations during the rehearsal period are compressed in time , in agreement with previous neurophysiological recordings in animals ( Foster and Wilson , 2006; Euston et al . , 2007; Ji and Wilson , 2007; Skaggs et al . , 1996; Xu et al . , 2012 ) . Reorganization of events with regard to the internal temporal framework , presumably mediated by neuronal oscillations of various rhythms ( e . g . , theta , alpha , gamma , etc . ) , may contribute to memory consolidation . Discrete epochs of events that are temporally separated in the external world can be successfully associated in the brain within an appropriate time frame , which could be vital for operation of the neuronal plasticity mechanism ( Jensen and Lisman , 2005; Nabavi et al . , 2014 ) , and a similar time compression mechanism could mediate sequence maintenance in working memory . A recent EEG study revealed that attention samples multiple visual objects in a temporally sequential manner even when subjects maintained sustained attention on one object ( Jia et al . , 2017 ) . We would argue that sequential attentional sampling could not account for the present observations . First , a previous study showed that attention would first sample the item associated with highest attentional priority and then would shift to an item associated with less attentional priority . However , here , all items in the sequence should be attended to with the same priority and thus would not be sequentially sampled by attention . Second , a previous sequential pattern ( attended before unattended ) could not reasonably account for the backward reactivation . Furthermore , the backward sequential pattern and its association with the recency effect could be explained by a ‘memory strength’ interpretation . For example , the better the memory strength is of the item ( e . g . , better memory strength for the 2nd item over the 1st item ) , the earlier would be the response latency ( e . g . , earlier response for the 2nd item compared to the 1st item ) . To test the account , we performed a control analysis by grouping subjects in Experiment 2 into two groups based on their memory accuracy rather than the recency , and examined the respective response profiles . If the alternative interpretation were correct , we would expect to see a temporal shift in the response pattern between the two groups such that the high-accuracy group would show earlier response compared to the low-accuracy group . However , the two groups elicited quite similar backward sequential responses ( Figure 3—figure supplement 1E ) , excluding the possibility that it is memory strength rather than a recency effect that leads to the observed backward sequential responses Finally , our results exhibit alpha-mediated responses during memory retention , consistent with previous findings on working memory ( Jensen et al . , 2002 ) . It is well acknowledged that alpha-band neuronal activities have a key function in working memory by efficiently suppressing distracting information ( Bonnefond and Jensen , 2012 ) . For instance , increased memory load is associated with enhanced alpha-band activities ( Jensen et al . , 2002; Klimesch et al . , 1999; Tuladhar et al . , 2007 ) , and repeated transcranial magnetic stimulation ( TMS ) at alpha frequency modulates the short-term memory capacity ( Sauseng et al . , 2009 ) . Memory reinstatement has been found to be accompanied by alpha-band activities , which carry a temporal phase signature of the replayed stimulus ( Michelmann et al . , 2016 ) . In addition to the traditionally posited inhibitory function in many cognitive processes ( Sauseng et al . , 2009; Bonnefond and Jensen , 2012; Klimesch , 2012; Händel et al . , 2011 ) , alpha-band neuronal oscillations have recently been found to implement an ‘‘echo’’ or reverberation of the inputs . These alpha echoes are indeed enhanced by visual attention , supporting an additional role of the alpha-band rhythm in the maintenance of sensory representations over time ( VanRullen and Macdonald , 2012 ) . Taken together , we reason that the memory-related alpha response in our results reflects memory reactivations rather than inhibition , based on several facts . First , previous studies using a TRF approach have revealed enhancement in alpha-echoes under attended conditions ( Jia et al . , 2017; VanRullen and Macdonald , 2012 ) , which could not be interpreted by inhibitory account . Second , the memory-related alpha responses did not show spatial attention effects ( see Figure 2—figure supplement 3A ) , which are known to be mediated by inhibitory alpha activities ( Worden et al . , 2000 ) . Finally , WM showed stronger alpha responses than NWM here and were strongly associated with memory behavior , further excluding the inhibitory interpretation . We propose that the alpha-band activity in TRF responses might represent the stimulus-induced phase resetting of the intrinsic alpha-band oscillation in EEG activities ( Makeig et al . , 2002 ) , given their close associations ( Figure 2—figure supplement 4A ) . In conclusion , our results provide converging evidence that sequence order information is represented and consolidated in short-term memory by an item-by-item serial reactivation mechanism . Crucially , the serial reactivation , temporally compressed ( i . e . , within a 200 – 400 ms window ) and reversed in temporal order ( i . e . , backward reactivation ) , is essentially associated with the subsequent memory behavioral performance . This fast-backward replay during maintenance , previously revealed in rat hippocampus for spatial navigation and shown here in humans for a non-spatial task , might therefore constitute a general neural mechanism for sequence memory and learning . Twenty subjects participated in Experiment 1 . Twenty subjects participated in Experiment 2 . Eleven subjects participated in the control study . During data collection and preprocessing , two subjects in Experiment 1 , and one subject in Experiment 2 were excluded because of overall low behavioral performance , excessive eye movements , or not finishing the whole experiment . All participants had normal or corrected-to-normal vision and had no history of psychiatric or neurological disorders . All experiments were carried out in accordance with the Declaration of Helsinki . All participants provided written informed consent prior to the start of the experiment , which was approved by the Research Ethics Committee at Peking University ( 2015-03-05c2 ) . Participants sat in a dark room in front of a CRT monitor with 100 Hz refresh rate , and their heads were stabilized using a chin rest . In each trial of all experiments , there were three periods: the encoding period , maintaining period , and recalling period ( Figure 1A ) . Participants were asked to only memorize the orientation of the cued bars ( 0 . 38°×1 . 15° visual angle ) . During each trial , all participants were instructed to keep the number of eye blinks to a minimum . Eye movements were monitored using an EyeLink 1000 eye tracker ( SR Research ) . The experimental trial was initiated only when fixation was located within 1° visual angle of the fixation cross . Results showed that the participants maintained good fixation on the central cross ( within 1° ) throughout each trial ( Figure 2—figure supplement 4B ) . EEG was recorded continuously using a 64-channel EasyCap and two BrainAmp amplifiers ( BrainProducts ) . Vertical and horizontal electrooculograms were recorded by two additional electrodes around the participants’ eyes . EEG data were offline band-pass filtered between 2 and 50 Hz . Independent component analysis was performed independently for each subject to remove eye-movement and artifact components , and the remaining components were back-projected onto the EEG electrode space . All channels were then referenced to the average value of all channels . The EEG was downsampled to 100 Hz before TRF calculation to be matched with the temporal resolution of the luminance sequences . To avoid the influence of the onset EEG response in each trial , which may bias the estimated TRF results , we extracted the middle part of the 5 s EEG trial responses ( 0 . 5 – 4 . 5 s ) for further TRF calculation .	Have you ever played the ‘Memory Maze Challenge’ game , or its predecessor from the 1980s , ‘Simon’ ? Players must memorize a sequence of colored lights , and then reproduce the sequence by tapping the colors on a pad . The sequence becomes longer with each trial , making the task more and more difficult . One wrong response and the game is over . Storing and retrieving sequences is key to many cognitive processes , from following speech to hitting a tennis ball to recalling what you did last week . Such tasks require memorizing the order in which items occur as well as the items themselves . But how do we hold this information in memory ? Huang et al . reveal the answer by using scalp electrodes to record the brain activity of healthy volunteers as they memorize and then recall a sequence . Memorizing , or encoding , each of the items in the sequence triggered a distinct pattern of brain activity . As the volunteers held the sequence in memory , their brains replayed these activity patterns one after the other . But this replay showed two non-intuitive features . First , it was speeded up relative to the original encoding . In fact , the brain compressed the entire sequence into about 200 to 400 milliseconds . Second , the brain replayed the sequence backwards . The activity pattern corresponding to the last item was replayed first , while that corresponding to the first item was replayed last . This ‘fast-backward’ replay may explain why we tend to recall items at the end of a list better than those in the middle , a phenomenon known as the recency effect . The results of Huang et al . suggest that when we hold a list of items in memory , the brain does not replay the list in its original form , like an echo . Instead , the brain restructures and reorganizes the list , compressing and reversing it . This process , which is also seen in rodents , helps the brain to incorporate the list of items into existing neuronal networks for memory storage .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Fast-backward replay of sequentially memorized items in humans
Human-associated microbial communities vary across individuals: possible contributing factors include ( genetic ) relatedness , diet , and age . However , our surroundings , including individuals with whom we interact , also likely shape our microbial communities . To quantify this microbial exchange , we surveyed fecal , oral , and skin microbiota from 60 families ( spousal units with children , dogs , both , or neither ) . Household members , particularly couples , shared more of their microbiota than individuals from different households , with stronger effects of co-habitation on skin than oral or fecal microbiota . Dog ownership significantly increased the shared skin microbiota in cohabiting adults , and dog-owning adults shared more ‘skin’ microbiota with their own dogs than with other dogs . Although the degree to which these shared microbes have a true niche on the human body , vs transient detection after direct contact , is unknown , these results suggest that direct and frequent contact with our cohabitants may significantly shape the composition of our microbial communities . Recent studies of the human microbiota have focused on multiple body sites in unrelated adults ( Costello et al . , 2009; Human Microbiome Project Consortium , 2012 ) or on a single body site , such as the gut , in single individuals over time ( Koenig et al . , 2011 ) or family units , including those with mono- and dizygotic twin pairs ( Turnbaugh et al . , 2008; Yatsunenko et al . , 2012 ) . Genetically related individuals , regardless of whether they cohabitate or not at the time of sampling , tend to share more of their gut ( fecal ) microbes than unrelated individuals ( Caugant et al . , 1984; Zoetendal et al . , 2001; Stewart et al . , 2005; Rajilić-Stojanović et al . , 2007; Turnbaugh et al . , 2008 ) . However , monozygotic twins are not significantly more similar than dizygotic twins ( Turnbaugh et al . , 2008 ) , indicating that this effect may be influenced by more than genetic similarity . In a study of US teenage mono- and dizygotic twins and their biological parents , we observed that the composition of the fecal microbiota of teens was more similar to that of their parents than unrelated adults , and as similar to that of their fathers as their mothers ( Yatsunenko et al . , 2012 ) . Moreover , mothers and fathers shared more similar bacterial communities in their guts compared to unrelated individuals living in other households ( Yatsunenko et al . , 2012 ) , indicating that a shared environment or lifestyle ( e . g . , contact with same microbial sources or diet ) affects the similarity of the fecal microbiota . Family members may also share intestinal bacteria with their household pets ( Caugant et al . , 1984 ) . Because we leave microbes from our bodies on the surfaces we touch ( Fierer et al . , 2010; Flores et al . , 2011 ) and at least a moderate level of microbial exchange is facilitated by direct contact , it is conceivable that our body site-associated microbial communities are shaped in part by our surroundings and those we contact on a daily basis . Whether similar patterns to those mentioned exist within non-gut body sites , whether body sites respond differently to factors such as cohabitation and family structure , and how these patterns change with host age remain unknown . To test the hypothesis that more microbes are shared between individuals who share a greater number of potential microbial sources , we examined the extent to which microbiota are shared among members of households composed of cohabiting heterosexual adults with and without children ( offspring ) , and with and without dogs . If our hypothesis were supported , then cohabiting family members would have microbiota more similar to each other than to members of different households . Furthermore , cohabiting couples with either children or dogs would share more microbial taxa in one or more of their body habitats than those without either , because such households contain , in addition to a shared environment , additional shared sources of potentially unique microbes with which couples are in close contact . We sampled 159 individuals comprising 17 families with cohabiting children aged 6 months to 18 years , 17 families with one or more dogs but no children , 8 families with both children and dogs , and 18 families with neither children nor dogs . Each family consisted of at least two cohabiting adults ( which we define as ‘partners’ or ‘couples’ ) between the ages of 26 and 87 years , and all children included in this study were biologically related to and cohabited with the focal couple . Sampling was performed as described in Costello et al . ( 2009 ) . For humans , fecal , oral ( dorsal tongue ) , forehead , and right and left palm communities were sampled ( n = 5 samples per individual all taken at a single time point ) . Dogs were sampled similarly ( n = 36 ) , except that all four paws were swabbed ( n = 7 samples per dog taken at the same time that humans were sampled ) . The age and gender of humans surveyed in each family plus the number and breed of dogs in families with pets are summarized in Table 1 . All samples were initially frozen at −20°C before they were transferred to the laboratory where they were stored at −80°C until they were subjected to DNA extraction , PCR of the variable region 2 ( V2 ) of bacterial 16S rRNA genes and subsequent multiplex sequencing with an Illumina GAIIx instrument ( Illumina , Inc . , San Diego , CA; n = 969 samples used for the analyses reported , 74 , 855 , 127 total reads; average read length , 105 ± 19 nt ) . The resulting 16S rRNA dataset was analyzed using UniFrac ( Lozupone and Knight , 2005 ) , a phylogeny-based measure of the degree of similarity between microbial communities , to assess patterns of similarity within and between families across body sites . 10 . 7554/eLife . 00458 . 003Table 1 . Summary of the number , age classification , and gender of humans surveyed in each family and the number and types of animals in families with petsDOI: http://dx . doi . org/10 . 7554/eLife . 00458 . 003Adults ( Sex ) Infants ( Sex ) Adolescents ( Sex ) Seniors ( Sex ) Dogs ( breed ) Other pets2 ( M , F ) 1 ( F ) 1 ( F ) Cat2 ( M , F ) 1 ( Unknown ) 2 ( M , F ) Cats2 ( M , F ) 1 ( F ) 2 ( M , F ) 2 ( Unknown ) 2 ( M , F ) 1 ( F ) 2 ( Unknown ) 2 ( M , F ) 1 ( F ) 1 ( M ) 2 ( M , F ) Cats , guinea pigs2 ( M , F ) 2 ( M , F ) 2 ( Unknown ) 2 ( M , F ) 3 ( Unknown ) 2 ( M , F ) 1 ( Unknown ) 2 ( M , F ) 2 ( M , F ) 1 ( F ) Cat , fish2 ( M , F ) 1 ( M ) Cat2 ( M , F ) 2 ( M , F ) 2 ( M , F ) Fish2 ( M , F ) 1 ( F ) 1 ( Jack Russell Terrier ) Cat2 ( M , F ) 1 ( Australian Cattle mix ) 2 ( M , F ) 2 ( M , F ) Cat , tarantula2 ( M , F ) 1 ( Labrador/Golden mix ) Cat2 ( M , F ) 1 ( Springer Spaniel ) Cat2 ( M , F ) 1 ( M ) 1 ( M ) Cat2 ( M , F ) Cats2 ( M , F ) Reptiles , amphibians2 ( M , F ) Cat2 ( M , F ) Cats2 ( M , F ) Fish2 ( M , F ) 1 ( F ) 2 ( M , M ) 2 ( M , F ) 2 ( M , F ) Cat2 ( M , F ) 1 ( F ) 3 ( M , F , M ) 1 ( M ) Cat2 ( M , F ) 1 ( Border Collie ) 2 ( M , F ) 2 ( Kelpie , Standard Poodle ) 2 ( M , F ) 1 ( F ) 1 ( Border Collie mix ) 2 ( M , F ) 1 ( M ) 2 ( Boxer , Boxer ) 2 ( M , F ) 2 ( Labrador , Labrador ) 2 ( M , F ) 2 ( M , F ) 2 ( English Setter , Labrador ) 2 ( M , F ) 2 ( Labrador , Australian Shepherd/Spaniel mix ) Chickens2 ( M , F ) 2 ( M , M ) Cats , rabbit , reptiles2 ( M , F ) 1 ( F ) 2 ( M , F ) 1 ( F ) 1 ( German Shepherd mix ) 2 ( M , F ) 1 ( Bernese Mountain ) 2 ( M , F ) 2 ( M , F ) 2 ( M , F ) 1 ( M ) 1 ( M ) Cat2 ( M , F ) 3 ( F , M , F ) 1 ( German Shepherd/Malamute mix ) 2 ( M , F ) 1 ( F ) 1 ( F ) 2 ( M , F ) 2 ( Australian Shepherd , Australian Shepherd ) 2 ( M , F ) 1 ( M ) 2 ( Border Collie/German Shepherd mix , Labrador mix ) 2 ( M , F ) 1 ( M ) Cat1 ( F ) 1 ( M ) Cat2 ( M , F ) 1 ( F ) 2 ( Siberian Husky , Greater Swiss Mountain ) 2 ( M , F ) 2 ( M , F ) Cat2 ( M , F ) 1 ( Unknown ) 2 ( M , F ) 2 ( M , F ) 1 ( Unknown ) 2 ( M , F ) 10612261536*These dogs were not sampled and thus not included in the total number of dogs . Each row is one family and the last row contains the total for each column . Infants were considered to be individuals aged 0–12 months , children/adolescents as 1–17 years , adults as 18–59 years and seniors as ≥60 years Our results revealed that family unit had a strong effect on human microbial community composition across all body sites: at each site , family membership explained a large proportion of the variability in bacterial diversity as measured using Faith's phylogenetic diversity ( PD ) ( Faith , 1992 ) , suggesting that family members tend to harbor similar levels of bacterial diversity ( Table 2 ) . Composition was also significantly affected by family membership across all body sites such that communities were more similar within families than between them ( Table 3 and Figure 1A–D , dogs , if present , are shown together with family members ) . This pattern was strongest for skin , a body habitat in constant contact with the external environment . An analysis of similarity ( ANOSIM ) showed that R ranged from 0 . 21 to 0 . 62 for humans; the value was higher for dogs with R = 0 . 71 for forehead and R = 0 . 83 for paws . On the forehead and palms , all except the father-to-infant within-family distances were significantly smaller than between-family distances ( Figure 1A , B and Table 4 ) . 10 . 7554/eLife . 00458 . 004Table 2 . Summary of the optimal linear mixed model explaining microbial phylogenetic diversity of body sites in relation to the main factorsDOI: http://dx . doi . org/10 . 7554/eLife . 00458 . 004Age groupBody siteTermTypeEstimateSE%VariabilityAllPalms ( L&R ) NoDogFixed−0 . 190 . 06FamilyRandom38 . 98AgeRandom17 . 14PlateRandom4 . 34FamSizeRandom0 . 38LaneRandom0BSRandom0Residual39 . 16ForeheadFamilyRandom24 . 85AgeRandom15 . 41LaneRandom8 . 50FamSizeRandom7 . 81PlateRandom6 . 33Residual37 . 10FecalAgeRandom45 . 64PlateRandom7 . 15FamSizeRandom3 . 08FamilyRandom2 . 24LaneRandom2 . 97 × 10-9Residual41 . 88OralAgeRandom35 . 92FamilyRandom14 . 54LaneRandom3 . 64PlateRandom4 . 60 × 10-10FamSizeRandom0Residual45 . 90AdultsPalms ( L&R ) NoDogsFixed−0 . 220 . 068MaleFixed−0 . 110 . 031FamilyRandom46 . 90PlateRandom2 . 77FamSizRandom5 . 17 × 10-11LaneRandom2 . 16 × 10-12BSRandom0Residual50 . 33ForeheadNoDogsFixed−0 . 270 . 087FamilyRandom34 . 91FamSizeRandom9 . 08PlateRandom0 . 80LaneRandom0Residual55 . 21FecalFamilyRandom20 . 20LaneRandom9 . 77FamSizeRandom0PlateRandom0Residual70 . 03OralFamilyRandom26 . 16FamSizeRandom14 . 59LaneRandom3 . 43PlateRandom3 . 42Residual52 . 40The model takes into account the variability between age groups ( Age ) , families ( Family ) , family sizes ( FamSize ) , sequencing lanes ( Lane ) , and primer plates ( Plate ) . Variability between left and right palms ( BS ) is also controlled for in the palm model . The table gives parameter estimates and standard errors for the significant terms in the model and the percentage of explained variability for each of the random effects ordered from highest to lowest . 10 . 7554/eLife . 00458 . 005Table 3 . Summary of a permutational analysis of variance ( PERMANOVA ) assessing the effect of family membership on unweighted UniFrac distances between familiesDOI: http://dx . doi . org/10 . 7554/eLife . 00458 . 005Body siteSource of variationDfMSPseudo-FPVPalms ( L&R ) Family membership690 . 32742 . 07020 . 0010 . 21Residual2100 . 158150 . 40ForeheadFamily membership700 . 237581 . 46030 . 0010 . 18Residual910 . 16270 . 40FecalFamily membership660 . 2121 . 27590 . 0010 . 14Residual920 . 166150 . 41OralFamily membership680 . 125441 . 68030 . 0010 . 15Residual960 . 074650 . 27V are the estimates of the components of variance for the factors . Statistically significant effects are in bold . 10 . 7554/eLife . 00458 . 006Figure 1 . Community similarity within and between families across body sites , and taxa contributing to these differences . Panels ( A–D ) show average unweighted UniFrac distances between family members ( blue ) and between members of different families ( red ) . ‘Child’ refers to all offspring aged 3–18 years who cohabit with the parents . ‘Infants’ were considered to be individuals aged 0–12 months . Palm/Paw refers to the right palm in the human comparisons and the back left paw in the dog comparison . Although there are distinguishable differences between the left and right palm communities within and across individuals ( Fierer et al . , 2008 ) , the same analysis using the left palms showed a similar pattern ( Table 2 ) and neither composition nor diversity were different enough between palms or among the four dog paws to affect the overall patterns . Mean ± 95% CI and R values ( ANOSIM ) are shown . p<0 . 05 and *p<0 . 001 based on 10 , 000 permutations . Panel ( E ) shows the families of bacteria that exhibit the greatest differences in the number of phylotypes ( OTUs ) shared within and between adult partners on the right palm . Bars represent the average number of shared phylotypes for a given bacterial family within partners from the same family ( blue ) and between partners of different families ( red ) . Mean ± 95% CI shown . p<0 . 05 after Bonferroni correction ( Wilcoxon test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00458 . 00610 . 7554/eLife . 00458 . 007Table 4 . Summary of ANOSIM analyses of the differences between within-family and between-family community comparisons using unweighted ( unwtd ) and weighted ( wtd ) UniFracDOI: http://dx . doi . org/10 . 7554/eLife . 00458 . 007ComparisonBody siteFam ( N ) Ind ( N ) R ( unwtd ) pR ( wtd ) pFamiliesForehead601510 . 49<0 . 00010 . 0100 . 39Right palm601410 . 62<0 . 00010 . 180 . 0008Left palm561220 . 61<0 . 00010 . 30<0 . 0001Fecal591510 . 19<0 . 00010 . 0680 . 056Oral591550 . 28<0 . 00010 . 32<0 . 0001SpousesForehead601140 . 28<0 . 00010 . 0910 . 0048Right palm591050 . 35<0 . 00010 . 110 . 0042Left palm55910 . 31<0 . 00010 . 130 . 002Fecal591140 . 21<0 . 00010 . 0770 . 0032Oral591170 . 18<0 . 00010 . 18<0 . 0001Father–ChildForehead14310 . 170 . 0250 . 0260 . 36Right palm14310 . 210 . 00850 . 110 . 096Fecal14330 . 170 . 0530 . 0660 . 20Oral14330 . 070 . 0690 . 230 . 0051Mother–ChildForehead14310 . 250 . 00240 . 0180 . 41Right palm14300 . 200 . 0130 . 00190 . 51Fecal14330 . 100 . 110 . 100 . 095Oral14330 . 210 . 00470 . 260 . 0012Father-InfantForehead12240 . 250 . 00340 . 0300 . 99Right Palm12210 . 160 . 0160 . 00630 . 96Fecal12230 . 0570 . 96−0 . 100 . 97Oral12240 . 0610 . 61−0 . 00860 . 70Mother–InfantForehead12230 . 230 . 0025−0 . 0250 . 99Right palm12210 . 34<0 . 00010 . 0120 . 87Fecal12230 . 0660 . 94−0 . 0360 . 86Oral12240 . 0920 . 230 . 0740 . 42DogsForehead12220 . 710 . 00020 . 56<0 . 0001Back left paw12230 . 83<0 . 00010 . 68<0 . 0001Fecal12250 . 250 . 00790 . 200 . 039Oral12250 . 250 . 0170 . 300 . 023‘Child’ refers to all offspring aged 3–18 years who cohabit with the parents . ‘Infant’ is considered to be an individual aged 0–12 months . The number of families and individuals used in each analysis is shown . Statistically significant comparisons ( p<0 . 05 ) are bolded . Those both statistically significant and of relatively large magnitude ( R>0 . 25 ) are bolded and italicized . For tongue and feces , we observed effects at the level of family and partner , although generally of smaller magnitude than for skin sites ( R < 0 . 30 ) . The comparison between partners , which is nested within the comparison between family members , is the strongest for all body sites and likely drives similarities by family at these sites . In contrast , a weaker effect , or no effect , was observed for parent–offspring pairs ( Figure 1C , D , R < 0 . 15 ) . This effect seemed to depend primarily on the age of the child . Although parents may share significantly more similar tongue and gut communities with their own children than with other children at older ages ( 3–18 years ) , the same is not the case for parents and their infants . These findings agree with previous studies that found the fecal microbiota of teens to be more similar to that of their parents than to unrelated adults ( Yatsunenko et al . , 2012 ) . However , our results also suggest that effects of cohabitation are insufficiently strong to overcome differences due to age ( which are discussed in more detail in the section ‘Effect of age on the human microbiota’ ) , particularly between infants and adults , whose microbiota differ substantially ( Palmer et al . , 2007; Koenig et al . , 2011 ) . We concluded that a shared environment may homogenize skin communities through contact with common surfaces ( including each other ) . Likewise , it may be easier to exchange skin microbes via exposure to home surfaces or indoor air ( both of which are typically dominated by skin-associated microbes ; Fierer et al . , 2010 ) , than it is to exchange gut or mouth bacteria , potentially because skin surfaces may be less ‘selective’ environments compared to the gut or mouth environments . Our results suggest that observed microbiota developmental dynamics depend on the body site under consideration ( Table 2 and Figure 2 ) . Here , we define ‘development’ as the rate and pattern with which new hosts ( i . e . , infants and children ) acquire adult-like microbiota over time . As noted previously ( e . g . , Palmer et al . , 2007; Koenig et al . , 2011; Yatsunenko et al . , 2012 ) , the development of the gut microbiota involves profound alterations in diversity and composition that take place over a relatively protracted timeframe ( nominally , 0–3 years in age ) ( Table 2 and Figure 2 ) . Our study enables us to ask further whether similar dynamics are observed in the contemporaneously sampled oral and skin communities of the same individuals . For oral communities , diversity changed substantially with age ( Table 2 ) , with a notable increase between the age of 0 and 3 ( Figure 2B ) , while compositional development , though significant , involved more subtle shifts than those observed over the same age range in the gut ( Table 5 ) . On the skin , diversity and composition ( i . e . , here , strictly membership ) changed relatively little with age ( Table 2 and Figure 2 ) . Interestingly , however , using a distance metric that emphasizes abundance ( weighted UniFrac ) reveals a strong developmental shift in skin microbiota on the forehead ( Figure 2A ) , a trend driven in part by changes in the relative abundance of dominant taxa rather than the acquisition/loss of unique taxa with age . This result is consistent with earlier studies ( Somerville , 1969; Leyden et al . , 1975 ) . For example , we see the relative abundance of Propionibacteria significantly increase on the forehead with age ( Table 6 ) , which has been shown to be associated with increasing levels of sebum production ( Leyden et al . , 1975; McGinley et al . , 1980 ) . The lack of a significant effect of age with unweighted UniFrac and the small amount of the variance in microbial diversity explained by age suggests developmental dynamics affected more by environmental exposures ( ostensibly , to adult skin microbiota ) than by age-associated shifts in the selective landscape ( e . g . , via introduction of solid food , emergence of teeth , etc . ) . 10 . 7554/eLife . 00458 . 008Figure 2 . Approach towards or departure from the ‘adult’ state in each body site with age . ( A ) Each point represents the average distance ( unweighted UniFrac in red; weighted UniFrac in blue ) between each participant and all other participants in the ‘adult’ age bracket . Here we define baseline ‘adult’ as 30–45 years in age ( the results are not sensitive to this threshold ) . R2 values ( linear regression model ) are shown . p<0 . 01 , p<0 . 001 . ( B ) Phylogenetic diversity ( PD ) of the communities on each body site is plotted for all of the offspring in the study ( aged 0–18 years ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00458 . 00810 . 7554/eLife . 00458 . 009Table 5 . Summary of a permutational distance-based linear model testing the effect of age ( Age ) , gender ( Sex ) , pet ownership ( Dog or Cat ) , and family size ( FamSize ) on unweighted and weighted UniFrac distance ( measures of community dissimilarity ) DOI: http://dx . doi . org/10 . 7554/eLife . 00458 . 009AgeGroupBody siteBest modelAICcPsuedo-Fpr2UnweightedAllPalms ( L&R ) ( + ) Age−453 . 495 . 91820 . 0012 . 2 ( + ) Dog−455 . 944 . 48370 . 0011 . 19 ( + ) FamSize−454 . 952 . 03780 . 0010 . 62Forehead ( + ) Age−264 . 312 . 75790 . 0011 . 69Fecal ( + ) Age−269 . 344 . 37750 . 0012 . 71Oral ( + ) Age−386 . 582 . 47740 . 0021 . 5AdultsPalms ( L&R ) ( + ) Dog−284 . 034 . 3030 . 0012 . 43Forehead ( + ) Dog−166 . 12 . 38780 . 0012 . 33FecalNoneNANANANAOralNoneNANANANAInfantsPalms ( L&R ) NoneNANANANAForeheadNoneNANANANAFecalNoneNANANANAOralNoneNANANANASeniorsPalms ( L&R ) NoneNANANANAForeheadNoneNANANANAFecalNoneNANANANAOralNoneNANANANAWeightedAllPalms ( L&R ) ( + ) Age−727 . 6623 . 5970 . 0019 . 05 ( + ) Sex−730 . 845 . 2180 . 0011 . 57 ( + ) FamSize−733 . 394 . 57950 . 0020 . 39Forehead ( + ) Age−418 . 219 . 4630 . 00115 . 11 ( + ) Sex−423 . 787 . 69390 . 0011 . 29 ( + ) FamSize−424 . 893 . 16750 . 0280 . 22Fecal ( + ) Age−353 . 082 . 40440 . 0611 . 51Oral ( + ) FamSize−416 . 323 . 40730 . 0192 . 05AdultsPalms ( L&R ) ( + ) Sex−462 . 244 . 50580 . 0025 . 36 ( + ) FamSize−464 . 113 . 91490 . 0031 . 09 ( + ) Dog−465 . 653 . 58580 . 0030 . 22Forehead ( + ) Sex−295 . 944 . 36280 . 0084 . 18FecalNoneNANANANAOralNoneNANANANAThis analysis was performed for the entire data set ( All ) as well as separately for age groups . The terms from the best model for each body site are shown , along with the percent of total variation explained ( r2 ) . Those terms of statistically significant and largest effect are bolded for each body site . 10 . 7554/eLife . 00458 . 010Table 6 . Summary of taxon abundances ( % ) present on the forehead for each age groupDOI: http://dx . doi . org/10 . 7554/eLife . 00458 . 010TaxonInfantsChildren/AdolescentsAdultsSeniorsActinobacteria Propionibacteriaceae ( 8 . 78 × 10-13 ) 6 . 3125131 Corynebacteriaceae0 . 72 . 04 . 29 . 3 Micrococcaceae* ( 0 . 0072 ) 2 . 02 . 71 . 22 . 5Bacteroidetes Prevotellaceae* ( 7 . 02 × 10-9 ) 7 . 75 . 61 . 41 . 5 Porphyromonadaceae* ( 8 . 53 × 10-12 ) 2 . 83 . 10 . 40 . 8 Flavobacteriaceae1 . 21 . 41 . 12 . 0 Bacteroidaceae0 . 51 . 41 . 11 . 9Firmicutes Streptococcaceae* ( 5 . 32 × 10-37 ) 47275 . 58 . 2 Staphylococcaceae* ( 0 . 0088 ) 2 . 12 . 6124 . 3 Carnobacteriaceae* ( 5 . 99 × 10-24 ) 4 . 83 . 40 . 50 . 7 Veillonellaceae* ( 1 . 25 × 10-13 ) 4 . 61 . 90 . 71 . 3Alphaproteobacteria Sphingomonadaceae0 . 91 . 41 . 30 . 9Betaproteobacteria Neisseriaceae2 . 44 . 12 . 36 . 9 Comamonadaceae0 . 41 . 21 . 52 . 9Gammaproteobacteria Pasteurellaceae* ( 1 . 64 × 10-6 ) 5 . 37 . 01 . 32 . 4 Moraxellaceae0 . 72 . 21 . 71 . 8A significant effect of age ( p<0 . 05 after Bonferroni correction; exact p-values are shown in parentheses ) . Infants were considered to be individuals aged 0–12 months , children/adolescents as 1–17 years , adults as 18–59 years and seniors as ≥60 years . Family level abundances of >1% were subjected to ANOVA analysis in QIIME . We next examined which groups of taxa are shared more between cohabiting partners than by adults from different families . Figure 1E shows an example of the specific taxa that are shared within and between adult partners on the right palm . The taxa driving these differences on the palm are lineages commonly reported in surveys of the human skin microbiota such as Propionibacteria ( Costello et al . , 2009; Grice et al . , 2009 ) . Two of these taxa , Prevotella and Veillonella , are primarily associated with the human oral community ( Nasidze et al . , 2009 ) , suggesting that at least for cohabiting couples , oral-skin transfer may be moderately frequent . Because these taxa are also found in the gut , we tried to determine the level at which the palm communities contained taxa derived from either oral or fecal sources . Using SourceTracker ( Knights et al . , 2011 ) , we estimated that on average , ∼11% of the palm community is likely from oral sources , as opposed to <2% from fecal sources . Given that oral bacteria can persist on skin for at least 8 hr ( Costello et al . , 2009 ) , we do not know whether these patterns are due to repeated inoculation from oral-skin contact or a true establishment of oral microbes on skin habitats . However , these results do suggest that close physical contact ( such as that between cohabiting couples ) can affect the taxonomic composition of the skin and may explain why these communities are more similar . Interestingly , the similarity in the microbiota of cohabiting individuals extends beyond human-to-human relationships to pet-to-pet and even human-to-pet relationships . The patterns of similarity between cohabiting dogs mimic that of cohabiting people , with skin ( fur ) sites showing the greatest degree of similarity ( Figure 1A–D ) . Moreover , from a microbial perspective , the skin communities of adults are on average more similar to those of their own dog ( s ) than to other dogs ( Figure 3 ) . Thus , we further explored the effect of dogs on the overall bacterial diversity and composition of their cohabiting owners in more detail . A principal components analysis ( PCoA ) of the human skin communities did not show strong clustering of the sites by dog-owning status in the three main axes , suggesting that dogs do not have a large effect . However , once age was accounted for , we found that dog ownership also affects the skin communities of adults , such that dog owners share more similar communities than expected by chance ( Table 5 ) . This effect is not seen when the weighted UniFrac measure is used , suggesting that dog-owners share similar communities mainly due to the addition of rare rather than abundant taxa . Such effects were detected in adults , but not infants or seniors , and may be due to behavioral differences between age groups that were not measured , such as variation in levels of contact with dogs . Alternatively , the presence of strong age affects as indicated earlier combined with small sample sizes of children with and without dogs may have obscured our ability to detect a significant effect of dogs . 10 . 7554/eLife . 00458 . 011Figure 3 . Community similarity and phylotype sharing between dogs-owners and their dogs . The left panel shows the average unweighted UniFrac distance between adult dog-owners and their dogs ( blue ) , between dog-owners and other ( not their own ) dogs ( red ) , and between adults who do not own dogs and dogs ( green ) . The right panel shows the number of phylotypes shared for the same categories . Comparisons are labeled on the y-axis such that the first body site listed corresponds to the dog and the second site corresponds to the human . Mean ± 95% CI shown . The presence of asterisks lacking brackets indicates that all pairwise comparisons within that group are significant . Generally , dog-owners tend to share more similar communities and more phylotypes with their own dogs than with other dogs . p<0 . 05 , *p<0 . 001 after Bonferroni correction ( Wilcoxon test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00458 . 011 Because dogs appear to have the largest effect on the skin communities of their cohabiting adult owners , we then explored the differences in the number of shared phylotypes ( OTUs ) , and the overall bacterial diversity on the skin of adults with dogs as well as adults with children . Adults who have dogs share more bacterial phylotypes with each other than they do with adults who do not have dogs ( Figure 4 , top right ) . Having a dog , then , has an effect of similar size on the number of taxa shared in human skin communities as the effect of living together ( i . e . , two people who have dogs but do not live together share , on average , about as many phylotypes as two people who live together but who do not have a dog ) . Adults who have a dog and live together share the greatest number of skin phylotypes while adults who neither have a dog nor live together share the least . We tested the effects of gender , pet ownership , and cohabitation of children using a linear mixed effects model , taking into account the variability in diversity due to family membership , age , and technical differences ( e . g . , sequencing lane ) . Of these factors , only dog ownership and gender significantly affected diversity ( Table 7 ) . Adults who own dogs tended to have a higher diversity of bacteria on their skin ( hands and forehead ) than those without dogs ( Figure 4 , top left , p<0 . 001 , Student's t-test with 10 , 000 Monte Carlo simulations; Table 2 ) . It should also be noted that consistent with previous studies ( Fierer et al . , 2008 ) , we found that adult females tend to have a higher diversity of bacteria on their hands than adult males ( Table 2 ) . 10 . 7554/eLife . 00458 . 012Figure 4 . Alpha diversity and shared phylotypes in couples with and without dogs and children . The left panels show rarefaction curves for skin communities of couples ( including seniors ) who have dogs ( top , in red ) , those without dogs ( top , in blue ) , couples ( excluding seniors ) with infants/children ( bottom , in red ) , and those without infants/children ( bottom , in blue ) . Mean ± 95% CI shown . The right panels show the average number of phylotypes shared among individuals from the same categories shown in the left panels . Mean ± 95% CI shown . p<0 . 05 , *p<0 . 001 after Bonferroni correction ( Wilcoxon test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00458 . 01210 . 7554/eLife . 00458 . 013Table 7 . Summary of a linear mixed effects analysis on the response of bacterial diversity across the body sites using the full data set and then filtered to include just adultsDOI: http://dx . doi . org/10 . 7554/eLife . 00458 . 013Age groupBody siteFixed effectsRandom effectsModelΔAICPr ( Chi ) AllPalms ( L&R ) Dog + Cat + Sex + KidBS + Ag + Fa + FS + Pl + LaFull0NA ( − ) Dog2 . 910 . 027 ( − ) Cat−5 . 601 ( − ) Sex1 . 730 . 053 ( − ) Kid−5 . 621ForeheadDog + Cat + Sex + KidAg + Fa + FS + Pl + LaFull0NA ( − ) Dog0 . 570 . 11 ( − ) Cat−5 . 251 ( − ) Sex0 . 970 . 085 ( − ) Kid−4 . 221FecalDog + Cat + Sex + KidAg + Fa + FS + Pl + LaFull0NA ( − ) Dog−0 . 0490 . 16 ( − ) Cat−0 . 150 . 17 ( − ) Sex0 . 330 . 13 ( − ) Kid−0 . 130 . 17OralDog + Cat + Sex + KidAg + Fa + FS + Pl + LaFull0NA ( − ) Dog1 . 220 . 072 ( − ) Cat−1 . 950 . 82 ( − ) Sex−1 . 880 . 73 ( − ) Kid−1 . 780 . 64AdultsPalms ( L&R ) Dog + Cat + Sex + KidBS + Fa + FS + Pl + LaFull0NA ( − ) Dog4 . 090 . 014 ( − ) Cat−5 . 291 ( − ) Sex4 . 110 . 013 ( − ) Kid−5 . 691ForeheadDog + Cat + Sex + KidFa + FS + Pl + LaFull0NA ( − ) Dog2 . 520 . 033 ( − ) Cat−4 . 991 ( − ) Sex−0 . 370 . 20 ( − ) Kid−4 . 381FecalDog + Cat + Sex + KidFa + FS + Pl + LaFull0NA ( − ) Dog0 . 300 . 13 ( − ) Cat0 . 240 . 13 ( − ) Sex−1 . 440 . 45 ( − ) Kid−0 . 420 . 21OralDog + Cat + Sex + KidFa + FS + Pl + LaFull0NA ( − ) Dog−0 . 290 . 19 ( − ) Cat−6 . 501 ( − ) Sex−7 . 221 ( − ) Kid−5 . 411The factors tested are co-habitation of dogs ( Dog ) , cats ( Cat ) , children ( Kid ) and host gender ( Sex ) . The base model takes into account the variability between age groups ( Ag ) , families ( Fa ) , family sizes ( FS ) , sequencing lanes ( La ) , and primer plates ( Pl ) . Variability between left and right palms ( BS ) is also controlled for in the palm model . The change in model fit resulting from exclusion of each fixed effect based on Akaike's Information Criterion ( AIC ) is shown . Statistically significant values are in bold text . In contrast to the skin communities , effects of gender or dogs were not detected in the gut or oral communities ( Table 7 ) . In fact , none of the tested factors were identified as important in these communities . Curiously , owning other types of indoor pets ( i . e . , cats [which were not sampled for this study] ) did not have a significant effect on the diversity ( Table 7 ) , overall similarity ( Table 5 ) , or amount of taxa shared between the skin communities of adult partners ( p=0 . 92 , Wilcoxon test ) . Although family size ( i . e . , having a child or children ) did seem to have a significant effect on whether individuals shared more similar skin communities , having a child in our study cohort also did not have a comparable effect to age or dog ownership on community similarity or diversity ( Figure 4 , bottom , p=0 . 05 for diversity; Tables 5 and 7 ) . For example , family size only explained a small proportion of the variability in diversity across all body sites ( oral: 0% , palms: <1% , fecal: 3% , forehead: 8% ) compared to age ( oral: 36% , palms: 17% , fecal: 46% , forehead: 15% ) . Although the mean difference in the number of phylotypes shared between couples paired from different families with and without children was significant , the size of the difference was minor ( <2 phylotypes ) and is likely due to the large number of observations in the within and between categories ( 2757 and 3071 respectively ) , which gave us the power to detect very small effect sizes ( d = 0 . 085 based on a power analysis given these sample sizes ) . One possible explanation for the large effect of dogs in comparison to children and other pets may be that individuals with dogs harbor taxa different from those without dogs , largely due to the presence of dog-derived bacterial taxa on their skin and presumably from frequent direct contact . One of the main taxa driving the pattern of similarity between dog owners is a family of Betaproteobacteria ( Methylophilaceae ) , a group that was also highly abundant ( 4 . 6% ) in the mouths of the dogs in this study , consistent with a common occurrence of oral–skin transfer between dogs and their owners . Other taxa include several families of Actinobacteria and a family of Acidobacteria commonly associated with soil ( Lauber et al . , 2009 ) , all of which were present on the paws and forehead of dogs , although in relatively low abundance ( <1% ) . In addition , characterization of the dog oral and ‘skin’ ( fur and paws ) microbiota revealed a greater diversity of taxa than described in humans ( Table 8 ) . Whereas human skin tends to be dominated by a few taxa at relatively high abundance ( namely Propionibacteriaceae , Streptococcaceae and Staphylococcaceae ) , dog paws and forehead harbor a more even mixture of taxa commonly found in a variety of host-associated environments including mammalian gut ( Enterobacteriaceae , Fusobacteriaceae ) , mouth ( Porphyromonadaceae , Veillonellaceae ) , and skin ( Propionibacteriaceae , Staphylococcaceae ) , as well as free-living environments such as soil and water ( e . g . , Hyphomicrobiaceae and Sphingomonadaceae ) ( Tables 6 , 8 , and 9 ) . This evenness and diversity of taxa found on dog skin may reflect frequent exposure of these sites to many different sources of microbes , or behavioral differences . The dog gut and tongue communities , on the other hand , harbor microbial communities that are somewhat similar in diversity and composition yet distinct from those in the human counterparts ( Figure 5 ) . Collectively , these data suggest that our pets not only harbor a diverse microbial community , but also shed a diverse set of microbiota that may in turn influence our own microbial composition . 10 . 7554/eLife . 00458 . 014Table 8 . Summary of taxon abundances ( % ) on the external body sites of dogsDOI: http://dx . doi . org/10 . 7554/eLife . 00458 . 014TaxonBack left pawBack right pawFront left pawPaws averagedForeheadActinobacteria Corynebacteriaceae1 . 81 . 30 . 71 . 21 . 0 Microbacteriaceae4 . 04 . 24 . 34 . 21 . 8 Micrococcaceae2 . 22 . 62 . 42 . 41 . 7 Nocardioidaceae3 . 33 . 43 . 23 . 31 . 5 Propionibacteriaceae3 . 03 . 93 . 13 . 34 . 5Bacteroidetes Bacteroidaceae2 . 11 . 92 . 62 . 23 . 2 Porphyromonadaceae2 . 81 . 61 . 72 . 05 . 7 Prevotellaceae1 . 11 . 83 . 02 . 02 . 4 Flavobacteriaceae2 . 62 . 52 . 52 . 53 . 3 Flexibacteraceae1 . 61 . 71 . 71 . 71 . 5Firmicutes Staphylococcaceae1 . 42 . 01 . 01 . 51 . 2 Streptococcaceae1 . 21 . 61 . 51 . 42 . 8 Lachnospiraceae1 . 10 . 91 . 71 . 31 . 2 Veillonellaceae0 . 30 . 51 . 80 . 90 . 7Fusobacteria Fusobacteriaceae2 . 31 . 71 . 82 . 03 . 8Alphaproteobacteria Bradyrhizobiaceae0 . 60 . 80 . 50 . 61 . 0 Hyphomicrobiaceae1 . 11 . 21 . 11 . 20 . 7 Methylobacteriaceae0 . 40 . 70 . 30 . 51 . 0 Sphingomonadaceae5 . 36 . 65 . 25 . 76 . 1Betaproteobacteria Comamonadaceae1 . 91 . 91 . 51 . 82 . 6 Oxalobacteraceae1 . 51 . 81 . 51 . 61 . 3 Neisseriaceae1 . 31 . 01 . 41 . 22 . 8Gammaproteobacteria Enterobacteriaceae2 . 25 . 15 . 94 . 42 . 5 Oceanospirillaceae2 . 31 . 03 . 12 . 10 . 3 Pasteurellaceae2 . 42 . 52 . 32 . 46 . 9 Moraxellaceae1 . 91 . 41 . 61 . 62 . 2 Pseudomonadaceae7 . 14 . 85 . 55 . 83 . 4Family level abundances >1% are shown . The front right paw was sampled but failed to amplify and is therefore not shown . 10 . 7554/eLife . 00458 . 015Table 9 . Summary of taxon abundances ( % ) present on the palms for each age groupDOI: http://dx . doi . org/10 . 7554/eLife . 00458 . 015InfantsChildren/ AdolescentsAdultsSeniorsActinobacteria Corynebacteriaceae0 . 7 ( 0 . 6 ) 3 ( 2 . 3 ) 4 . 2 ( 3 . 9 ) 4 . 3 ( 4 . 0 ) Micrococcaceae4 . 7 ( 3 . 9 ) 3 . 9 ( 3 . 9 ) 2 . 8 ( 2 . 6 ) 2 . 7 ( 2 . 8 ) Propionibacteriaceae ( 0 . 00016 ) 3 . 1 ( 2 . 2 ) 11 ( 11 ) 27 ( 27 ) 20 ( 15 . 9 ) Bacteroidetes Bacteroidaceae0 . 5 ( 0 . 6 ) 1 . 6 ( 1 . 4 ) 1 . 6 ( 3 . 3 ) 3 . 4 ( 8 . 7 ) Flavobacteriaceae1 ( 1 . 4 ) 1 . 9 ( 1 . 4 ) 2 . 7 ( 2 . 0 ) 3 . 8 ( 4 . 1 ) Porphyromonadaceae2 ( 2 . 1 ) 1 . 9 ( 1 . 5 ) 1 . 4 ( 0 . 7 ) 0 . 6 ( 1 . 0 ) Prevotellaceae5 . 6 ( 4 . 2 ) 4 . 2 ( 5 . 7 ) 3 . 1 ( 3 . 0 ) 2 . 1 ( 1 . 7 ) Firmicutes Carnobacteriaceae* ( 1 . 15 × 10-11 ) 6 . 4 ( 5 . 8 ) 5 ( 3 . 5 ) 1 . 7 ( 1 . 9 ) 1 ( 0 . 7 ) ClostridialesFamilyXI . IncertaeSedis0 . 2 ( NA ) 1 . 5 ( NA ) 1 ( NA ) 1 . 4 ( NA ) Lachnospiraceae* ( 8 . 57 × 10-5 ) 0 . 3 ( 0 . 5 ) 0 . 9 ( 1 . 3 ) 1 ( 1 . 5 ) 3 . 1 ( 7 . 2 ) Lactobacillaceae0 . 1 ( NA ) 0 . 2 ( NA ) 1 . 5 ( NA ) 4 . 2 ( NA ) ( Ruminococcaceae* ) ( 1 . 17 × 10-5 ) NA ( 0 . 2 ) NA ( 0 . 7 ) NA ( 0 . 8 ) NA ( 3 . 5 ) Staphylococcaceae3 . 2 ( 2 . 1 ) 5 . 1 ( 7 . 2 ) 6 . 7 ( 7 . 3 ) 2 . 8 ( 2 . 2 ) Streptococcaceae* ( 3 . 91 × 10-10 ) 49 ( 54 ) 27 ( 26 ) 15 ( 16 ) 13 ( 9 . 1 ) Veillonellaceae* ( 1 . 76 × 10-7 ) 5 . 5 ( 4 . 5 ) 2 . 2 ( 2 . 1 ) 1 . 7 ( 2 . 0 ) 1 . 9 ( 2 . 1 ) Fusobacteria Fusobacteriaceae1 . 6 ( 1 . 6 ) 1 . 8 ( 1 . 2 ) 1 . 4 ( 1 . 2 ) 1 ( 1 . 0 ) Betaproteobacteria Comamonadaceae* ( 9 . 90 × 10-5 ) 0 . 7 ( 0 . 3 ) 0 . 7 ( 1 . 0 ) 1 . 5 ( 1 . 7 ) 3 . 6 ( 3 . 7 ) Neisseriaceae2 . 6 ( 2 . 4 ) 3 . 5 ( 2 . 3 ) 1 . 6 ( 1 . 1 ) 2 ( 1 . 6 ) Gammaproteobacteria Moraxellaceae1 . 5 ( 0 . 7 ) 1 . 2 ( 1 . 7 ) 3 . 2 ( 2 . 7 ) 3 . 3 ( 2 . 7 ) Pasteurellaceae* ( 0 . 017 ) 2 . 6 ( 3 . 3 ) 4 . 4 ( 2 . 9 ) 1 . 8 ( 1 . 6 ) 1 . 2 ( 1 . 1 ) Pseudomonadaceae* ( 0 . 049 ) 0 . 2 ( 0 . 2 ) 0 . 6 ( 0 . 7 ) 1 . 1 ( 1 . 0 ) 2 . 4 ( 3 . 0 ) ( Enterobacteriaceae* ) ( 0 . 027 ) NA ( 0 . 6 ) NA ( 0 . 8 ) NA ( 0 . 8 ) NA ( 2 . 7 ) *A significant effect of age ( p<0 . 05 after Bonferroni correction; exact p-values are shown in parentheses ) . Shown only for the right palm ( left palm showed similar trends ) . Infants were considered to be individuals aged 0–12 months , children/adolescents as 1–17 years , adults as 18–59 years and seniors as ≥60 years . Abundances for the left palm are shown in parentheses . Family level abundances of greater than 1% were subjected to ANOVA analysis in QIIME . Taxa present at >1% on the left palm but <1% on the right are shown in parentheses . 10 . 7554/eLife . 00458 . 016Figure 5 . Variation within and between the communities of skin , oral , and fecal samples from humans and dogs . Panel ( A ) shows a PCoA plot of all the body habitats , using unweighted UniFrac distances of human and dog samples , rarefied at 5000 sequences/sample . Panels ( B–D ) show select body habitats from the full plot . Panel ( E ) shows a summary of the taxa shaded by relative abundance at the phylum level broken down by specific body habitat; the seven most abundant taxa are shown in the legend . DOI: http://dx . doi . org/10 . 7554/eLife . 00458 . 016 Given that recent studies in gnotobiotic and other animal models show pervasive effects of the microbiota on metabolism , immunity , and other aspects of our biology , it is intriguing to consider that who we cohabit with , including companion animals , may alter our physiological properties by influencing the consortia of microbial symbionts that we harbor in and on our various body habitats , and in particular , our skin habitats . One example relates to the hygiene hypothesis , which posits that a broad range of microbial exposures helps educate our developing immune systems to tolerate a variety of environmental antigens , thereby reducing risk for atopic disorders such as asthma and food allergies . Recent studies link early exposure to pets to decreased prevalence of allergies , respiratory conditions , and other immune disorders in later stages of development ( Havstad et al . , 2011 ) and skin microbes in particular are now receiving more focus as important players in immune regulation ( Naik et al . , 2012 ) . Given the potential of skin as a collector and integrator of shared environmental bacteria as demonstrated in this study , identifying exactly how such communities can be functionally affected by environmental exposures may help us better understand how they may be deliberately manipulated in order to prevent or treat disease . Epidemiologic studies of the impact of environmental factors on physiological variations and disease predispositions would be enhanced by integrating microbiological surveys , including time series studies during the first years of postnatal life . These efforts would be timely as we seek to understand the impact of Westernization on human biology and to delineate , from an anthropologic perspective , how different cultural traditions and lifestyles relate to our microbial ecology ( Benezra et al . , 2012 ) . Each sample was processed using methods and procedures described in previous publications ( Hamady et al . , 2008; Caporaso et al . , 2011 ) . DNA was extracted from each swab using the MOBIO PowerSoil DNA isolation kit ( MO BIO Laboratories , Inc . , Carlsbad , CA ) according to manufacturer instructions with modifications . For each sample , the V2 region of bacterial 16S rRNA genes was amplified in triplicate reactions using the primers F27 ( 5′-AGAGTTTGATCCTGGCTCAG-3′ ) and R338 ( 5′-TGCTGCCTCCCGTAGGAG T-3′ ) barcoded with a unique12-base error-correcting Golay code for multiplexing . PCR reactions contained 13 μl MO BIO PCR water , 10 μl 5 Prime Hot Master Mix , 0 . 5 μl each of the forward and reverse primers ( 10 μM final concentration ) , and 1 . 0 μl genomic DNA . Reactions were held at 94°C for 3 min to denature the DNA , run for 35 cycles of amplification at 94°C for 45 s , 50°C for 60 s , and then 72°C for 90 s , and completed with a final extension step of 10 min at 72°C . Amplicons were processed using the MO BIO Ultra Clean-htp 96-well PCR clean up kit and quantified using Picogreen dsDNA reagent in 10-mM Tris buffer ( pH 8 . 0 ) . Equal amounts of amplicons from each reaction for a given sample were pooled , followed by gel purification and ethanol precipitation . Multiplex DNA sequencing was performed with a Illumina GAIIx instrument located in the Center for Genome Sciences and Systems Biology at Washington University School of Medicine . The resulting DNA sequences , OTU table , and associated sample metadata have been deposited in the QIIME database ( http://www . microbio . me/qiime/ ) under the study ID 979 ( Song_2012_family_study ) . Sequence data were processed with QIIME v1 . 4 . 0-dev ( Caporaso et al . , 2010 ) as previously described ( Yatsunenko et al . , 2012 ) . Sequences were demultiplexed and quality filtered using default QIIME parameters , and 16S rRNA operational taxonomic units ( OTUs ) were picked using a closed reference OTU picking procedure ( QIIME script pick_reference_otus_through_otu_table . py ) . Briefly , sequences were clustered against the Greengenes database ( reference collection , 2011 release ) ( http://greengenes . lbl . gov/ ) at 97% identity and those failing to match within this threshold were discarded . Taxonomy was assigned to the retained clusters ( OTUs ) based on the Greengenes reference sequence and the Greengenes tree was used for all downstream phylogenetic community comparisons . For the 1076 samples , the number of sequences per sample ranged from 1 to 300 , 473 , with a mean of 54 , 475 sequences per sample ( total: 58 , 615 , 414 ) . Such large variability in the number of sequences per sample is typical for studies employing high-throughput sequencing methods , the causes of which have not yet been systematically tested to our knowledge . To standardize sequence counts across samples , samples with <5000 sequences per sample were removed . Remaining samples were rarefied to 5000 sequences and further filtered by eliminating samples that had a high probability of being mislabeled ( e . g . , labeled skin , but likely a tongue or fecal sample; detected using the script supervised_learning . py ) . The remaining 969 samples were used for all downstream analyses . Phylogenetic diversity ( PD ) was computed and rarefaction analyses were conducted using the QIIME scripts multiple_rarefaction . py , alpha_diversity . py and collate_alpha . py . Because there are no generally accepted methods for ‘denoising’ Illumina sequence data , alpha diversity estimates such as PD and OTU counts may be overestimated due to sequencing error . However , overestimation should not affect relative differences in diversity . Analyses of community similarity ( β-diversity ) were performed by calculating pairwise distances using the phylogenetic metric UniFrac ( Lozupone and Knight , 2005 ) . The resulting distance matrices were used for principle coordinates analyses ( PCoA ) . We examined the effect of pet ownership , gender , and the cohabitation of children on the bacterial diversity ( measured as phylogenetic diversity [PD] ) of each body site using a linear mixed effects model including age group , family membership , family size , sequencing lane , and primer plate as random factors . Variability between left and right palms was also controlled for in the palm model . For each body site , we began with the full model including all random and fixed factors , and fixed factors were subsequently removed in a step-wise manner using the function ‘drop1’ in R . Akaike's Information Criterion ( AIC ) values and a chi-square test were used to select the best model for each body site . An increase in the AIC value indicates that the removed factor significantly worsened the fit of the model . All modeling was performed using the function ‘lmer’ in the R-package ‘lme4’ ( Bates et al . , 2008 ) . Differences in alpha diversity between groups were subsequently tested using a t-test with Monte Carlo simulations on the dataset rarefied to 5000 sequences ( compare_alpha_diversity . py in QIIME v1 . 5 . 0-dev ) . To describe changes in the microbial community with age , distances were calculated between each participant and all participants within specified age groups ( the core groups of adults were considered 30–45 years old and elderly participants ≥65 years old ) , averaged for each participant , and then plotted against their age using the QIIME script categorized_dist_scatterplot . py . A linear regression model was fitted to the distance plots using R . Analyses of differences in the number of shared phylotypes between groups were performed using the Wilcoxon test in R . For all body sites , we tested for differences in taxon abundances across the age groups using an analysis of variance ( ANOVA ) ( otu_category_significance . py in QIIME ) . Infants were considered to be individuals aged 0–12 months , children/adolescents as 1–17 years , adults as 18–59 years and seniors as ≥60 years . Most of the subjects in the child/adolescent category were between the ages of 1 and 6 years , none were between 7 and 11 years , and five participants were between the ages of 12 and 17 years . Due to the low sample size in the latter age range , post-pubescent subjects/teens were not split into a separate category . Exclusion of these five subjects from the analysis did not significantly affect the results . In all appropriate analyses , p values were adjusted for the number of comparisons made using the Bonferroni method .	The human body is home to many different microorganisms , with a range of bacteria , fungi and archaea living on the skin , in the intestine and at various other sites in the body . While many of these microorganisms are beneficial to their human hosts , we know very little about most of them . Early research focused primarily on comparing the microorganisms found in healthy individuals with those found in individuals suffering from a particular illness . More recently researchers have become interested in more general issues , such as understanding how these collections of microorganisms , which are also known as the human microbiota or the human microbiome , become established , and exploring the causes of similarities and differences between the microbiota of individuals . We now know that the communities of microorganisms found in the intestines of genetically related people tend to be more similar than those of people who are not related . Moreover , the communities of microorganisms found in the intestines of non-related adults living in the same household are more similar than those of unrelated adults living in different households . We also know that the range of microorganisms found in the intestine changes dramatically between birth and the age of 3 years . However , these studies have focused on the intestine , and little is known about the effect of relatedness , cohabitation and age on the microbiota at other body sites . Song et al . compared the microorganisms found on the skin , on the tongue and in the intestines of 159 people—and 36 dogs—in 60 families . They found that co-habitation resulted in the communities of microorganisms being more similar to each other , with those on the skin being the most similar . This was true for all comparisons , including human pairs , dog pairs and human–dog pairs . This suggests that humans probably acquire many of the microorganisms on their skin through direct contact with their surroundings , and that humans tend to share more microbes with individuals , including their pets , with which they are in frequent contact . Song et al . also discovered that , unlike what happens in the intestine , the microbial communities on the skin and tongue of infants and children were relatively similar to those of adults . Overall , these findings suggest that the communities of microorganisms found in the intestine changes with age in a way that differs significantly from those found on the skin and tongue .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2013	Cohabiting family members share microbiota with one another and with their dogs
Endophytic insects provide the textbook examples of herbivores that manipulate their host plant’s physiology , putatively altering source/sink relationships by transferring cytokinins ( CK ) to create ‘green islands’ that increase the nutritional value of infested tissues . However , unambiguous demonstrations of CK transfer are lacking . Here we show that feeding by the free-living herbivore Tupiocoris notatus on Nicotiana attenuata is characterized by stable nutrient levels , increased CK levels and alterations in CK-related transcript levels in attacked leaves , in striking similarity to endophytic insects . Using 15N-isotope labeling , we demonstrate that the CK N6-isopentenyladenine ( IP ) is transferred from insects to plants via their oral secretions . In the field , T . notatus preferentially attacks leaves with transgenically increased CK levels; plants with abrogated CK-perception are less tolerant of T . notatus feeding damage . We infer that this free-living insect uses CKs to manipulate source/sink relationships to increase food quality and minimize the fitness consequences of its feeding . Insect herbivores are under constant pressure from their host plants: they must adapt to toxic or anti-digestive defense compounds whose levels often dramatically increase in response to insect feeding; and their food source has low nitrogen to carbon ratios and a dietary value which decreases as leaves mature and senesce . Some herbivorous insects have developed strategies to overcome the low nutritional contents of their host plants and have evolved specialized mechanisms to tolerate , or even co-opt toxic plant defense metabolites for their own uses , in an apparent evolutionary arms race ( Strong et al . , 1984; Després et al . , 2007; Heckel , 2014 ) . Phytophagous insects can be categorized as either endophytic or free-living depending on the relationships that they establish with their host plant . This distinction is not binary and many transitional forms exist even within the same taxa . Consequently , the large differences in herbivorous lifestyles has selected for plant defense responses that counter different herbivory strategies ( Kessler and Baldwin , 2002; Schuman and Baldwin , 2016 ) . Free-living insects are mobile on their host plants , moving among plants , and frequently among different plant species . As a consequence of this mobility , they can freely choose tissues that are most nutritious or least defended , but the most nutritious tissues are often highly defended , resulting in a potential trade-off for herbivores ( Ohnmeiss and Baldwin , 2000; Brütting et al . , 2017 ) . To avoid herbivore-induced defenses , free-living insects often move to other plant parts or even other host plants in response to defense activation , and the advantages of such movement are readily seen when induced defenses are abrogated ( Paschold et al . , 2007 ) or experimentally manipulated ( van Dam et al . , 2000 ) . In contrast , endophytic insects develop more intimate relationships with their host plants as they are sedentary and spend a large portion of their life cycle within plant tissues . They have evolved strategies to overcome many of the plant defenses by hijacking plant metabolism and reprogramming plant physiology in their favor ( Giron et al . , 2016 ) . Often the only viable plant defense is the ‘scorched earth’ response , whereby infested tissues are abscised from the plant ( Fernandes et al . , 2008 ) . To date , the best-studied examples of endophytic plant-manipulating species , featured in most textbooks of plant physiology , are the gall-forming insects and leaf-miners . Gall-forming organisms , which include not only several orders of insects but also mites , nematodes and microbes , promote abnormal plant growth by reprogramming the expression of plant genes , to create novel organs that provide favorable environments for the exploiter ( Stone and Schönrogge , 2003; Shorthouse et al . , 2005 ) . Advantages for the gall-formers range from an improved nutritional value , with reduced defense levels , to protection from diseases , competitors , predators , parasitoids and unfavorable abiotic conditions ( Hartley , 1998; Stone and Schönrogge , 2003; Allison and Schultz , 2005; Harris et al . , 2006; Saltzmann et al . , 2008; Nabity et al . , 2013 ) . Manipulations of leaf-mining larvae do not result in the formation of new macroscopic structures like galls but they are often revealed during senescence of host tissues , where ‘green islands’ appear around the active feeding sites ( Engelbrecht , 1968; Engelbrecht et al . , 1969; Giron et al . , 2007; Kaiser et al . , 2010 ) . Such green islands maintain a high level of photosynthetic activity typical of non-senescent leaves , thus providing nutrition for the larvae which feed on them ( Behr et al . , 2010; Body et al . , 2013; Zhang et al . , 2016 ) . In this way , green islands reflect a battle between plant and infesting insect during the nutrient recovery phase that precedes abscission . The host plant tries to recover nutrients from the senescent leaf , whereas the insect tries to maintain a nutritious environment so as to complete its development . The most likely effectors used by insects to manipulate a plant’s normal physiological response to wounding are phytohormones , since significant levels of some well-known wound-responsive phytohormones , including cytokinins ( CKs ) , abscisic acid ( ABA ) and auxins , have been found in the body and salivary secretions of a number of gall-forming insects ( Mapes and Davies , 2001; Straka et al . , 2010; Tooker and De Moraes , 2011; Yamaguchi et al . , 2012; Tanaka et al . , 2013; Takei et al . , 2015 ) , as well as in the bodies and labial glands of leaf-mining larvae ( Engelbrecht et al . , 1969; Body et al . , 2013 ) . Amongst these phytohormones , CKs deserve additional discussion due to their role in the formation of green islands ( Engelbrecht , 1968; Engelbrecht , 1971; Engelbrecht et al . , 1969; Giron et al . , 2007; Kaiser et al . , 2010; Body et al . , 2013; Zhang et al . , 2017 ) . CKs are adenine derivatives which play a key role in the regulation of plant growth and development ( Sakakibara , 2006 ) . They are known for their capacity to increase photosynthetic activity ( Jordi et al . , 2000 ) , determine sink strength ( Mok and Mok , 2001 ) and inhibit senescence ( Richmond and Lang , 1957; Gan and Amasino , 1995; Ori et al . , 1999 ) . More recently , CKs have been shown to regulate herbivory-induced defense signaling ( Schäfer et al . , 2015b; Schäfer et al . , 2015c; Brütting et al . , 2017 ) . The long history of investigating CKs in the formation of green islands dates back to the late 1960’s , to reports of increased levels of CKs in affected tissues ( Engelbrecht , 1968; Engelbrecht et al . , 1969 ) . In the last decade , studies on the leaf-mining larvae of Phyllonorycter blancardella identified CKs as the causative factors for the ‘green island’ phenomenon ( Giron et al . , 2007; Kaiser et al . , 2010; Body et al . , 2013; Zhang et al . , 2017 ) . These studies suggested that insects could be the source of phytohormones used to manipulate plant physiological responses . However , a clear demonstration of the ability of insects to transfer CKs to a host plant remains elusive . To assess whether an insect actively transfers CKs to manipulate plant physiology , we studied the interactions between the well-established ecological model-plant Nicotiana attenuata and one of its most abundant specialist herbivores , Tupiocoris notatus . N . attenuata is a wild diploid tobacco species native to southwestern North America . T . notatus is a free-living , 3–4 mm mirid bug ( Miridae , Heteroptera ) specialized to tobacco species and a few other solanaceous plants including Datura wrightii . It is a piercing-sucking cell-content feeder that damages the surface of the leaves without removing foliar material . Its feeding behavior is in sharp contrast with the feeding behavior of a well-studied specialist herbivore of N . attenuata , the lepidopteran Manduca sexta , whose chewing larvae cause extensive tissue damage and a well characterized defense response ( Baldwin , 1998; Kessler and Baldwin , 2001; Kessler et al . , 2004; Steppuhn et al . , 2004; Zavala et al . , 2004; Schuman et al . , 2012 ) . When plants are attacked by M . sexta , specific insect-derived fatty acid-amino acid conjugates elicit a defense response regulated by a burst of jasmonic-acid ( JA ) ( Baldwin , 1998; Halitschke et al . , 2001; Kessler et al . , 2004 ) . This jasmonate burst triggers the accumulation of defense metabolites like nicotine , caffeoylputrescine , diterpene-glycosides and tripsyin-proteinase inhibitors . It has also strong effects on the regulation of primary metabolism ( Voelckel and Baldwin , 2004 ) : sugars , starch and total soluble proteins readily decrease in the attacked leaves ( Ullmann-Zeunert et al . , 2013; Machado et al . , 2015 ) , as does photosynthesis ( Meza-Canales et al . , 2017 ) . In contrast , infestation with T . notatus in nature , surprisingly , does not decrease plant fitness ( Kessler and Baldwin , 2004 ) , despite resulting in damage to large portions of photosynthetically active leaf area . Tissues around T . notatus feeding sites have increased rates of photosynthesis per chlorophyll content that may compensate for the damage caused by herbivore feeding , resulting from an active ingredient of the oral secretion of T . notatus which remains to be identified ( Halitschke et al . , 2011 ) . We previously observed increased damage by T . notatus in tissues that were enriched in CKs through the transgenic manipulation of N . attenuata CK metabolism , using plants expressing a dexamethasone ( DEX ) -inducible construct driving transcription of the CK biosynthesis gene , isopentenyltransferase ( IPT , i-ovipt ) . Individual DEX-treated leaves of field-grown plants suffered more damage from T . notatus than did mock-treated leaves . This led to the hypothesis that increased CK levels promote better nutritional quality , which in turn increases T . notatus feeding damage ( Schäfer et al . , 2013 ) . Here , we report that T . notatus adults and nymphs contain high concentrations of two CKs . When confined to feeding on single N . attenuata leaves , concentrations of CKs increase in attacked leaves throughout the feeding period , with consequences for nutrient concentrations . Using 15N-labeled tracers , we demonstrate that T . notatus transfer CKs to the leaves on which they feed . Finally , we analyzed how changes to CK metabolism in plants affected T . notatus feeding preferences . We conclude that CK-dependent manipulation of plant metabolism is not only a strategy used by gall-forming insects or leaf-miners , but also employed by this free-living insect , which directly transfers CKs at feeding sites to manipulate its host plant . To characterize the defensive response of N . attenuata to mirid attack , we analyzed jasmonate hormones and defense metabolites that are known to be induced by M . sexta , as well as T . notatus feeding ( Kessler and Baldwin , 2004 ) . Continuous feeding by T . notatus ( Figure 1a ) causes visible damage to N . attenuata leaves ( Figure 1b ) and triggers defense responses in attacked leaves ( Figure 1c–j ) . Three days of T . notatus feeding induced levels of the defense metabolites nicotine and caffeoylputrescine ( CP ) , as well as trypsin proteinase inhibitor activity ( TPI ) ( Figure 1c–e ) . T . notatus feeding also elevated the levels of jasmonic acid ( JA ) , its precursor cis- ( + ) −12-oxophytodienoic acid ( OPDA ) and its bioactive isoleucine conjugate ( JA-Ile ) ( Figure 1f–h ) . Interestingly , there was also a significant increase in salicylic acid ( SA ) , but no influence on abscisic acid ( ABA ) ( Figure 1i , j ) . JA and JA-Ile levels triggered by T . notatus feeding remained elevated for up to six days when mirids were confined to feed on a single leaf ( Figure 1—figure supplement 1a–c ) . Their concentrations remained higher than controls even when T . notatus were free to move to other parts of the plant , although they steadily decreased over the six days ( Figure 1—figure supplement 1d–f ) These results demonstrate that N . attenuata’s response to T . notatus involves activation of JA signaling and downstream defense responses . Feeding by M . sexta is detrimental to N . attenuata fitness . It causes reduction of photosynthesis in attacked leaves ( Halitschke et al . , 2011; Meza-Canales et al . , 2017 ) and a decrease in sugar and total soluble protein ( TSP ) contents ( Ullmann-Zeunert et al . , 2013; Machado et al . , 2015 ) . In contrast , T . notatus feeding seems to increase photosynthetic activity in attacked leaves , when accounting for tissue damaged by the feeding ( Halitschke et al . , 2011 ) . We measured the impact of continuous T . notatus feeding over several days on the nutritional quality of the attacked leaves . We analyzed TSPs , sugar and starch levels , as well as measuring photosynthetic rates and chlorophyll contents of leaves over a period of 144 hr . Visibly heavily damaged leaves did not show significant decreases in nutrient levels when mirids were confined to feed on a single leaf with a small plastic cage ( Figure 2a ) . TSP levels decreased with time in a clipcage but mirid feeding did not have a significant influence ( Figure 2b ) . Furthermore , we did not observe any significant changes in starch , sucrose , glucose or fructose ( Figure 2c–f ) . Although we did not observe changes in carbohydrate levels , photosynthesis was significantly reduced in attacked leaves ( Figure 2—figure supplement 1b ) . In contrast , mirid feeding had no effect on chlorophyll contents ( Figure 2—figure supplement 1c ) . When entire plants were heavily infested ( Figure 2—figure supplement 2a ) , changes in nutrient levels in the plant became apparent only for TSP levels , which decreased after mirid feeding ( Figure 2—figure supplement 2b ) . Conversely , levels of starch , sucrose , glucose and fructose were not affected by mirid feeding ( Figure 2—figure supplement 2c–f ) . Both chlorophyll contents and photosynthetic rates significantly decreased after T . notatus whole-plant attack ( Figure 2—figure supplement 3a–c ) . In summary , when only twenty mirids were allowed to feed on a single leaf the overall nutritional quality was not altered , although the feeding damage was visibly severe . In contrast , during a more extreme mirid infestation in which entire plants were severely attacked , TSP levels of attacked leaves decreased , but sugar and starch contents remained unchanged . However , no overall apparent increased photosynthetic activity was observed . An allocation of nutrients from unattacked to attacked tissue may explain the observation that even heavy T . notatus feeding only marginally influenced nutrient levels in attacked leaves . If this inference is correct , then mirid feeding likely influences the source/sink relationships of the host plant . Cytokinins ( CKs ) are known to regulate source/sink relationships and stabilize nutrient levels in tissues fed on by endophytic insects . Recently , we showed that these phytohormones also play a role in plant defense , since M . sexta herbivory , wounding , and JAs can increase the levels of cZ-type CKs in N . attenuata ( Schäfer et al . , 2015c; Brütting et al . , 2017 ) . As we did not see a strong decrease in nutrients after mirid feeding , it was especially interesting to investigate CK metabolism during T . notatus attack . When entire plants were attacked , mirid feeding significantly increased the accumulation of NaCKX5 transcripts , which code for a CK oxidase/dehydrogenase responsible for CK degradation ( Figure 3a ) . Transcript levels of NaZOG2 , which codes for a CK glucosyltransferase responsible for CK inactivation , as well as transcripts of NaLOG4 ( Figure 3—figure supplement 1b ) , which is involved in CK biosynthesis , also increased after mirid feeding . In contrast , transcript levels of the isopentenyltransferase NaIPT5 , which catalyzes the rate-limiting step of CK biosynthesis , were reduced after mirid feeding ( Figure 3—figure supplement 1c ) . T . notatus feeding did not change levels of the CK response regulator NaRRA5 ( Figure 3—figure supplement 1d ) . The levels of the different types of CKs varied depending on time and whether mirids attacked single leaves or entire plants . When entire plants were infested with T . notatus , overall leaf CK contents gradually increased over time ( Figure 3b ) . Levels of cis-zeatin ( cZ ) ( Figure 3—figure supplement 2a ) , cis-zeatin riboside ( cZR ) ( Figure 3—figure supplement 2d ) , trans-zeatin ( tZ ) ( Figure 3—figure supplement 2b ) and trans-zeatin riboside ( tZR ) ( Figure 3—figure supplement 2e ) were significantly higher after T . notatus attack . In contrast , levels of N6-isopentenyladenine ( IP ) remained unaffected by mirid feeding ( Figure 3—figure supplement 2c ) and levels of N6-isopentenyladenosine ( IPR ) decreased in attacked leaves ( Figure 3—figure supplement 2f ) . This decrease was significant in the first 24 hr after the initiation of mirid attack and disappeared at later harvest times ( p<0 . 05 in TukeyHSD post hoc test ) . When mirids were only allowed to feed on a single leaf , we did not observe changes in levels of summed CKs over the whole time series ( Figure 3—figure supplement 3b ) . However , Bonferroni-corrected t-tests of single time point revealed increased levels at the last time point harvested , after 144 hr of feeding ( tt: p=0 . 026 ) . The changes in individual CKs only partially overlapped with those observed during whole-plant feeding . There was a significant increase in cZ ( Figure 3—figure supplement 3c ) , but levels of cZR decreased ( Figure 3—figure supplement 3f ) . IP levels , which did not change during whole-plant feeding , were significantly higher overall after single-leaf feeding , although pairwise comparisons for each time point did not reveal significant changes at any given time point . tZ , tZR and IPR remained unaffected by mirid feeding when only single leaves were attacked ( Figure 3—figure supplement 3d , g , h ) . Interestingly , in both experiments overall CK levels remained unchanged or were increased , despite the concomitant increases in transcripts of genes related to CK degradation . From these results , we infer that CKs are involved in the observed nutritional stability during mirid feeding; this hypothesis prompted a more detailed analysis of the origin of these CKs . Mirid attack enhanced the levels of cZ and cZR as was previously found for M . sexta herbivory , wounding , and JA application ( Schäfer et al . , 2015c; Brütting et al . , 2017 ) ; but in contrast to these other types of elicitations , long-term mirid feeding and the associated JA accumulation did not decrease IP levels . This was particularly surprising given that CK degradation and inactivation processes appeared to have been activated by mirid feeding . We analyzed CK levels in T . notatus to determine whether these insects could themselves provide a source of CKs . We found very high levels of IP and IPR in extracts from the insect bodies ( Figure 3c ) . While concentrations of IPR were comparable to those in leaves ( around 1 pmol per g fresh mass ( FM ) ) , concentrations of IP exceeded those of leaves by up to three orders of magnitude: while levels in leaves ranged from 0 . 01 to 0 . 1 pmol g FM−1 , levels in insects were usually between 1 and 5 pmol g FM−1 and attained values as high as 16 pmol g FM−1 . Insects collected from N . attenuata plants in their natural habitat at a field site in Utah , USA , also contained high amounts of IP: in a pooled sample of ten insects , we measured 18 . 26 pmol IP per g FM-1 . Mirids contained high IP and IPR levels in their bodies independently of their sex , developmental stage , or food source ( Figure 3—figure supplement 4 ) . The sole significant difference was that IP concentration in nymphs was about half as high as in adult males and females , but nymphs still had concentration several times those found in leaves ( Figure 3—figure supplement 4a ) . To evaluate if CKs levels remained stable when T . notatus was no longer feeding on its host plant , we reared insects for five days either on artificial diet ( containing no CKs ) or on plants . Insects raised on artificial diet had IP levels in their body that were not different from levels in insects raised on plants; IPR levels were also unchanged ( Figure 3—figure supplement 4b ) . Although the source of CKs in T . notatus remains unknown , we hypothesize that IP and IPR found in T . notatus body could be used by the mirid to counter the decrease in IP levels in attacked leaves that is commonly observed in response to long-term JA elicitation or M . sexta feeding . To evaluate whether T . notatus could transfer CKs to the plant , we conducted 15N- labeling experiments . We grew plants in hydroponic culture with 15N-labeled KNO3 as the only source of nitrogen . We furthermore created a stock of T . notatus insects that were 15N-labeled by raising them for an entire generation on 15N-grown plants . We then performed two different types of experiments to trace the origin of CKs in T . notatus attacked leaves: we either used 15N-grown plants that we exposed to 14N-labeled insects ( Figure 4 and Figure 4—figure supplement 1 ) or we used 14N-grown plants and exposed them to 15N-labeled insects ( Figure 4—figure supplements 2 and 3 ) . CKs are adenine derivatives that contain five nitrogen atoms . Therefore , CKs produced by 15N-labeled plants or insects harbored five 15N and are readily distinguished from 14N-labeled CKs by mass spectrometry ( Figure 4—figure supplements 4 and 5 ) . In the first approach , we used a low-infestation setup by placing 20 15N-labeled T . notatus adults in a small cage on the leaf of a 14N-grown plant for five days . After four days of continuous feeding , we found detectable amounts of 15N-labeled IP ( and IPR ) in the leaves ( Figure 4—figure supplements 2 and 3 ) : around 2 . 35 fmol [15N5]-IP per g FM , which represent the 3 . 3% of the [15N5]/[14N5]-IP ratio ( Figure 4—figure supplement 2d ) . [15N5]-labelled IP and IPR could only have originated from the insects , as the natural abundance of 15N is below 0 . 4% , and IP ( or IPR ) with five 15N would occur about once in a trillion molecules . From these values , one mirid feeding on a leaf for five days could account for a transfer of at least 0 . 12 fmol IP per g FM−1 ( Supplementary file 1 ) , assuming that CK transport , degradation or conversion to other CK forms can be excluded . In the reverse experiment , we used 15N-grown plants and insects raised on 14N-grown plants ( Figure 4 and Figure 4—figure supplement 1 ) . We placed 15N-grown plants in cages where T . notatus were reared on 14N-grown plants . These 15N-grown plants were switched to a new cage with infested 14N-grown plants once per day to ensure that they were always attacked by 14N-labeled insects and to prevent the accumulation of 15N in the 14N-labeled insects . After 5 days , an average of 48% of the [15N5]/[14N5]-IP ratio was 14N labeled and therefore originating from the insects ( Figure 4 ) . In this stronger induction setup , IPR transfer from the insect to the plant was also already detected after 24 hr and accounted to 19% of the [15N5]/[14N5]-IP ratio after 5 d ( Figure 4—figure supplement 1 ) . To evaluate how IP and IPR were transferred to the leaf during feeding , we analyzed the CK contents of the oral secretions and frass of T . notatus , which we considered the most likely means of transfer . Mirids were fed on sugar solutions covered with parafilm , which allowed the insects to penetrate the film with their stylets while preventing evaporation and preventing either insects or their frass from being immersed in the liquid . We then measured CKs in the sugar solution , which contained substances transferred by the oral secretions , as well as in the surface wash , which contained insect excretions ( frass ) . We found large amounts ( high signal intensity ) of IP mainly from the oral secretions ( from the sugar solution that mirids had fed on ) and much lower amounts in the frass of the mirids ( from the surface wash ) ( Figure 5 ) . IPR was found in oral secretions and in frass in similar amounts ( Figure 5—figure supplement 1 ) . These results clearly demonstrate that T . notatus is able to transfer CKs ( mainly IP ) to its host plant . The most likely means of transfer would be via the salivary secretion produced during feeding , although we cannot rule out a smaller contribution of feces , which are sticky and tend to cover infested leaves . In nature , T . notatus feeds on young N . attenuata tissues , such as younger stem leaves and young growing leaves . This feeding pattern was inferred from the damage distributions observed on plants in both nature and the glasshouse ( Figure 6—figure supplement 1a ) , as well as in two-choice assays ( Figure 6—figure supplement 1b ) . The young leaves preferred by T . notatus are typically rich in CKs ( Brütting et al . , 2017 ) . To evaluate how CK metabolism affects the interaction of N . attenuata with T . notatus , we used transgenic N . attenuata plants that were either enhanced in CK production ( i-ovipt ) or silenced in CK perception ( irchk2/3 ) . Transgenic i-ovipt plants that contain a dexamethasone ( DEX ) -inducible promotor system coupled to an IPT gene were produced as previously described ( Schäfer et al . , 2013; Schäfer et al . , 2015b ) , and allowed a DEX-mediated induction of CK overproduction . irchk2/3 plants , fully characterized in Schäfer et al . , ( 2015b ) are silenced for two of three CK receptors . T . notatus prefers leaves of i-ovipt plants which have been treated with DEX and therefore have higher levels of CKs ( Figure 6a ) . If T . notatus is given the choice between empty vector ( EV ) and irchk2/3 plants , mirids show a strong preference for EV plants , as shown in the lower damage levels on irchk2/3 plants ( Figure 6b ) . Furthermore , we found pronounced differences in the reaction of the plants to the damage caused by T . notatus feeding . Mirid attack caused necrotic lesions in irchk2/3 plants , comparable to a pathogen-induced hypersensitive response , whereas this did not occur in WT , EV or i-ovipt plants ( Figure 6c ) . To better understand the feeding preferences of T . notatus , we measured nutrient levels in irchk2/3 , DEX-induced i-ovipt plants and EV plants ( Figure 7 ) . Starch and sucrose did not differ among the lines ( Figure 7c , f ) . However , i-ovipt plants had higher concentrations of protein , free amino acids , glucose and fructose than did irchk2/3 plants ( Figure 7a , b , d , e ) . The i-ovipt plants tended to have higher nutrient levels than did EV plants but the results were only statistically significant for glucose concentrations ( Figure 7d ) . In contrast , irchk2/3 plants tended to have lower nutrient levels compared to EV , but these were only significantly lower for fructose concentrations ( Figure 7e ) . From these results , we conclude that CKs play a dual role in the T . notatus-N . attenuata interaction: as important determinants of tissue palatability for T . notatus by enhancing nutrient contents , but also as important tolerance factors that allow plants to suffer negligible or lower fitness consequences of mirid attack than they would otherwise . We unambiguously demonstrated that T . notatus transfers two types of CKs , IP and its riboside IPR , to N . attenuata providing the first clear demonstration of CK transfer from an insect to a plant . IP has been generally considered one of the most active natural CKs based on classical activity assays ( Gyulai and Heszky , 1994; Sakakibara , 2006 ) and high concentrations of IP have been previously reported in plant-manipulating endophytic insects , like leaf miners and gall-formers ( Engelbrecht , 1968; Engelbrecht et al . , 1969; Mapes and Davies , 2001; Straka et al . , 2010; Yamaguchi et al . , 2012; Body et al . , 2013; Tanaka et al . , 2013 ) . We also showed that oral secretions , and in much lower amounts , frass of T . notatus , contained IP and IPR , thus providing a possible means of transfer . Concentrations of total CK content of N . attenuata leaves steadily increased during long-term T . notatus feeding , consistent with the observation that mirids transferred CKs to plants . The overall reconfiguration of the transcriptional activity of genes involved in CK degradation , inactivation , and biosynthesis upon T . notatus feeding did not correlate with the apparent changes in CK concentrations . This suggests that N . attenuata might activate a type of CK detoxification in response to mirid feeding and CK introduction . Additional support for the existence of a mechanism that counter-balances mirid-injected CKs comes from the observation that the leaf concentration of IP and IPR , the two CKs transferred by T . notatus , were unaffected or only slightly changed in plants by mirid feeding . This was surprising , considering that we estimated that after five days of whole-plant infestation , roughly half of the total IP in attacked leaves originated from mirids . Understanding to what extent the observed changes in cytokinin signaling result from mirid-mediated CK transfer is further complicated by the fact that CK levels respond as part of the herbivory-inducible defense signaling ( Schäfer et al . , 2015a; Schäfer et al . , 2015c; Brütting et al . , 2017 ) . The dual role of CKs in plant growth and defense highlights the complexity of the fine regulation of CKs needed to regulate plant physiological responses . Not only CK quantities , but also CK structures , and the hormonal balance with other phytohormones may influence changes in metabolism upon insect feeding ( Giron et al . , 2013 ) . Demonstrable effects on host plants induced by endophytic insects include alterations of plant morphology , changes in the nutritional quality of the affected tissues and the inactivation of plant defenses surrounding the attack sites ( Giron et al . , 2016 ) . Whereas alterations of plant morphology are associated with only some endophytic insects , for example gall-formers , control of the nutritional quality of the infested tissues seems to be a common feature of all endophytic insect-plant interactions . N . attenuata leaves maintain their nutritional quality despite being heavily damaged by T . notatus feeding: only total soluble proteins ( TSPs ) decreased with heavy infestation , as in the whole-plant experiments , whereas concentrations of glucose , fructose , sucrose and starch remained unchanged . Previous studies in N . attenuata showed that wounding and application of oral secretions ( OS ) of M . sexta as well as M . sexta feeding reduced glucose and fructose concentrations by inhibiting soluble invertases . Such reductions are JA-dependent , and abrogated in transgenic lines impaired in JA production ( Machado et al . , 2015 ) . A negative influence of jasmonates on plant primary metabolism has also been suggested by studies in a number of other plants ( Babst et al . , 2005; Skrzypek et al . , 2005; van Dam and Oomen , 2008; Hanik et al . , 2010; Tytgat et al . , 2013 ) . Hence , the fact that T . notatus feeding activated JA-signaling , but did not negatively influence soluble monosaccharide concentrations , suggested that an additional counterbalancing alteration in the primary metabolism of N . attenuata occurs during T . notatus feeding . Similar to what has been observed for carbohydrates , wounding and M . sexta OS application results in a 91% reduction of total soluble proteins ( TSPs ) in young rosette leaves ( Ullmann-Zeunert et al . , 2013 ) , the same leaf stage used in this study . After 144 hr of continuous T . notatus infestation , during which proteins should be heavily depleted by mechanical cell-content damage – in contrast to the minor damage associated with OS elicitation ( Halitschke et al . , 2001 ) – TSP reductions were only ca . 75% . More surprisingly , a smaller T . notatus infestation ( twenty mirids confined on a single leaf ) did not change TSP contents at all during 144 hr of continuous feeding . These results are consistent with microarray analysis that compared expression patterns induced by T . notatus and M . sexta , which revealed that mirid-specific transcriptional responses occurred largely in primary metabolism ( Voelckel and Baldwin , 2004 ) . Thus , during T . notatus feeding , plant’s primary metabolism seems to be influenced by mechanisms different from the classical JA-mediated herbivory and wound responses . In contrast to the apparent sugar and starch homeostasis observed during T . notatus feeding and to the observations of Halitschke et al . ( 2011 ) , we observed a decrease in photosynthetic rates during continuos mirid feeding . Reduced photosynthetic rate is a general response observed in a number of plant-insect interactions ( Zhou et al . , 2015 ) as well as in N . attenuata . Wounding and elicitation with M . sexta OS rapidly decrease photosynthetic CO2 assimilation and this reduction is mediated by the JA-precursor OPDA ( Meza-Canales et al . , 2017 ) . We think that the discrepancy between our work and results from Halitschke et al . ( 2011 ) likely results from differences in the experimental protocols used: ( 1 ) we used a very heavy infestation , ( 2 ) the leaf area used to measure the photosynthetic rates of mirid-attacked leaves included both damaged and undamaged areas , ( 3 ) we did not normalize photosynthetic rates to intact undamaged leaf tissue . In any case , the reduction in the overall photosynthetic rates observed during T . notatus feeding was not consistent with unchanged starch and sugar levels and , together with the observation that T . notatus transfers CK during feeding , suggests the inhibition of senescence and/or transport of nutrients to the attacked leaves . Manipulation of plant defenses is another phenomenon often observed in endophytic insect-host plant interactions ( Giron et al . , 2016 ) . We showed that T . notatus feeding activated N . attenuata JA-dependent defense pathways in a way consistent with previous studies; the increases in defense metabolites induced by T . notatus feeding were comparable to those elicited by M . sexta attack ( Kessler and Baldwin , 2004 ) . T . notatus is well-adapted to the specialized metabolism of N . attenuata; it prefers wild-type plants which are less susceptible to invasion by other herbivores , rather than those impaired in JA biosynthesis with reduced defense metabolites ( Fragoso et al . , 2014 ) . Insects counter the presence of toxic metabolites in their host plants by detoxification or sequestration of toxic substances ( Heckel , 2014 ) , and the detoxification ability of T . notatus is suggested by the observation that it accumulates transcripts encoding detoxification enzymes in response to JA-dependent defenses ( Crava et al . , 2016 ) . This fact , togheter with the finding that defense pathways were not down-regulated during the T . notatus feeding , point out that down-regulation of plant defense may not benefit T . notatus as much as its manipulation of its host’s nutritional status . Choice assays demonstrated that T . notatus is attracted to plant tissues with enhanced CK levels , both when CK levels were naturally higher , as in young plant tissues , and when CKs are experimentally increased using DEX-inducible transgenic plants ( Schäfer et al . , 2013 ) . When CK perception was impaired as in the irchk2/3 line , mirids preferred WT or EV plants over the transgenic plants as shown by their different damage levels . This preference for higher CK levels and against irchk2/3 plants could either be a direct effect of CKs or – more likely – an indirect effect of CK-related processes . A direct attraction to CKs in insects has been discussed ( Robischon , 2015 ) but to our knowledge there is no direct evidence that insects perceive CKs . Consistent with the second hypothesis , CK levels were not reduced in irchk2/3 plants compared to those of the EV line ( Schäfer et al . , 2015c ) . Thus , we infer that T . notatus prefers metabolites positively associated with the CK pathway . These might be molecules produced either by N . attenuata primary or specialized metabolism . For example , T . notatus is attracted to quercetin ( Roda et al . , 2003 ) , and some related phenolic compounds are influenced by CK levels ( Schäfer et al . , 2015b; Brütting et al . , 2017 ) . Consistent with the primary metabolism hypothesis , we showed that nutrient levels of transgenic lines correlated with T . notatus preference . This inferred preference for nutrients is also consistent with T . notatus damage distribution on whole plants , which is concentrated on young , CK-rich and nutrient-rich tissues . These tissues are also better defended ( Brütting et al . , 2017 ) suggesting a possible trade-off between palatability and anti-digestive effects of the diet . However , specialized detoxification mechanisms likely allow T . notatus to feed with impunity on otherwise well-defended tissues ( Crava et al . , 2016 ) , thus allowing T . notatus’s feeding choice to reflect the nutritional quality , rather than the defensive status , of its host . Our results provide evidence that a free-living insect transfers CKs which manipulates its host plant’s metabolism , likely for its own benefit . CK-mediated plant manipulation strategies have only been known so far from endophytic insects ( Giron et al . , 2016 ) . The low mobility and intimate associations of endophytic insects with their host plants provides the selective environment for the evolution of mechanisms that allow them to manipulate host plant physiology and/or morphology . Species known for CK-dependent manipulation of host plants are not closely related to each other , and span several orders: Lepidoptera , Hymenoptera , Hemiptera and Diptera . The most studied examples are Lepidopteran leaf-miners which cause the green island phenomenon , like Phyllonorycter blancardella ( Giron et al . , 2007; Kaiser et al . , 2010; Body et al . , 2013; Zhang et al . , 2017 ) or Stigmella argentipedella ( Engelbrecht , 1968; Engelbrecht , 1971; Engelbrecht et al . , 1969 ) . Among gall-forming organisms , two species of the genus Bruggmannia are also capable of producing green islands ( Fernandes et al . , 2008 ) . Other gallers that manipulate plant morphology can be found among hymenopterans , such as gall-wasps , dipterans such as gall-midges and gall-flies , and hemipterans such as psyllids and gall-aphids . Among these , a role for CKs in gall formation has been shown for the dipterans Eurosta solidaginis ( Mapes and Davies , 2001 ) and Rhopalomyia yomogicola ( Tanaka et al . , 2013 ) , the hymenopteran Dryocosmus kuriphilus ( Matsui et al . , 1975 ) and sawflies of the genus Pontania ( Yamaguchi et al . , 2012 ) and the hemipterans Pachypsylla celtidis ( Straka et al . , 2010 ) and the galling-aphid Tetraneura nigriabdominalis ( Takei et al . , 2015 ) . The fact that species from different orders have developed similar mechanism of plant manipulation suggests either an ancient evolutionary origin or a convergent evolutionary trait . We propose that such CK-dependent manipulation is more widespread than previously thought , and is also shared with free-living insects like T . notatus . CKs transferred by T . notatus could originate from its host plant and be sequestered by the insect but also they could be synthesized by the insect itself or its associate endosymbionts . In fact , CKs are also produced by organisms other than plants , like fungi ( Chanclud et al . , 2016 ) , bacteria ( Costacurta and Vanderleyden , 1995 ) and nematodes ( Siddique et al . , 2015 ) . It is thought that IP and IPR can be derived from tRNA , and this suggests that the substrate for CK biosynthesis is shared by all organisms ( Persson et al . , 1994 ) , virtually enabling insects to produce CKs . The most recent studies on CK-mediated manipulation of plant physiology by insects suggested a role of endosymbiotic bacteria in CK production ( Kaiser et al . , 2010; Giron et al . , 2013; Zhang et al . , 2017 ) . Antibiotic feeding experiments have revealed that endosymbionts like Wolbachia are the most likely producers of CKs in the leaf-miner P . blancardella ( Kaiser et al . , 2010; Body et al . , 2013 ) . Yet , this bacterium is unlikely responsible for CK production in T . notatus , as Wolbachia was not be detected in mirids from the same glasshouse colony used in our experiments , and was identified only very rarely from insects collected from the field ( Adam et al . , 2017 ) . Free-living phytophagous insects are thought not to manipulate their host plant’s physiology to enhance the nutritional quality of their diet , as they are free to move to the best feeding locations on a plant . This work provides evidence of the ability of a free-living insect to introduce CKs into their host during feeding to maintain a better nutritional environment . We suggest that this mechanism may be commonly found in other free-living species and that it combines the benefits of the two different lifestyles: the ability to move , hide and choose the best feeding locations , and to manipulate the host plant via CK-transfer . Clarifying the details of the origins of the T . notatus-transferred CKs and studying their role in nature will provide new insights into the complex interactions that occur during plant-herbivore interactions . All chemicals used were obtained from Sigma‐Aldrich ( St . Louis , MO , US ) ( http://www . sigmaaldrich . com/ ) , Merck ( Darmstadt , Germany ) ( http://www . merck . com/ ) , Roth ( Karsruhe , Germany ) ( http://www . carlroth . com/ ) or VWR ( Radnor , PE , US ) ( http://www . vwr . com ) , if not mentioned otherwise in the text . CK standards were obtained from Olchemim ( Olomuc , Czech Republic ) ( http://www . olchemim . cz ) , dexamethasone ( DEX ) from Enzo Life Sciences ( Farmingdale , NY , US ) ( http://www . enzolifesciences . com/ ) , HCOOH for ultra-performance LC from Fisher Scientific ( Hampton , NH , US ) ( http://www . fisher . co . uk/ ) or from Honeywell Riedel-de HaënTM ( Morris Plains , NJ , US ) ( http://www . riedeldehaen . com/ ) , and GB5 from Duchefa ( Haarlem , The Netherlands ) ( http://www . duchefa-biochemie . nl/ ) . We used the 31st inbred generation of Nicotiana attenuata ( Torr . ex S . Wats . ) originating from the ‘Desert Inn’ population from the Great Basin Desert ( Washington County , UT , US ) as wild-type ( WT ) plants ( Baldwin et al . , 1994 ) . Transgenic plants were generated from WT N . attenuata , as described by Krügel et al . , 2002 ) by Agrobacterium-mediated transformation . Empty vector transformed plants ( EV ) ( line A-04-266-3 ) were used as controls in experiments that included other transgenic lines . The transgenic line irchk2/3 was transformed with a construct harboring inverted-repeat gene fragments to silence the expression of two of the three known CK receptor homologs NaCHK2 and NaCHK3 ( Chase Domain Containing Histidine Kinase 2 and 3 ) and was previously characterized ( Schäfer et al . , 2015b ) . We used line A-12–356 , which had a silencing efficiency of about 50% . The i-ovipt line ( A-11–92 x A-11–61 ) contains a gene encoding the rate-limiting step of CK biosynthesis , the isopentenyltransferase ( IPT ) from Agrobacterium tumefaciens ( Tumor morphology root; Tmr ) . The IPT gene is controlled by the pOp6/LhGR expression system , which allows transcriptional up-regulation in a specific tissue by the application of DEX ( Schäfer et al . , 2013 ) . Application of DEX to the leaves of the plant induces the transcription of IPT which locally increases CK levels . DEX was dissolved in lanolin paste with 1% DMSO at a final concentration of 5 µM . For control treatments , we used 1% DMSO in lanolin . The lanolin paste was applied to the petioles of the leaves 24 hr prior to other treatments as previously described ( Schäfer et al . , 2013 ) . Seeds were sterilized and germinated on Gamborg B5 media as described by Krügel et al . ( 2002 ) with modifications as previously described ( Brütting et al . , 2017 ) . For soil growth conditions , ten days after germination , plants were first transplanted to TEKU pots and 10 days later into 1 L pots . For hydroponic growth conditions , the plants were transferred after 12 days into 50 mL hydroponic culture single pots and 10 days later into 1 L hydroponic containers . Conditions for hydroponic culture were previously described ( Ullmann-Zeunert et al . , 2012 ) as were conditions for soil growth ( Brütting et al . , 2017 ) . For the damage determination experiment in the glasshouse , single plants were grown in 4 L pots . Plants were maintained under glasshouse conditions ( 22–27°C; ca . 60% RH , 16:8 light:dark regime ) as previously described ( Adam et al . , 2017 ) . To prevent T . notatus infestation of the main glasshouse facility of the MPI-COE , we maintain the colony in a separate glasshouse located in Isserstedt , Germany , approximately 7 km from the main glasshouse facility . Plants were germinated in the main glasshouse facility and transferred to the Issersted glasshouse just before plants began to flower , when plants had main stems about 25 cm tall . In both glasshouses , plants were maintained at comparable growth conditions . After transferring plants to the Isserstedt glasshouse , we allowed for at least two days of acclimation before initiating experiments . The colony of T . notatus ( Distant , 1893 ) ( Figure 1 , 1A ) originated from insects caught in the vicinity of the Brigham Young University/Max Planck field station at Lytle Ranch Preserve in the Great Basin Desert ( Washington County , UT , US ) and was annually refreshed with insects caught from the same field site . The colony was maintained in cages made of acrylic glass ( 2 × 1 × 1 m ) with a fine mesh for air circulation . Cages were maintained under the same glasshouse growth conditions that were used for the cultivation of N . attenuata ( 27°C; ca . 60% RH , 18:8 light:dark regime ) . We fed insects with hydroponically grown WT N . attenuata plants . Fresh plants were provided weekly and remained in the cages for several weeks to allow nymphs to hatch from eggs laid in the older plants . Insects were collected from the cage for experiments using an insect exhauster . Prior to being clip-caged onto N . attenuata leaves , insects were anaesthetized with CO2 . For the artificial diet we dissolved amino acids ( L-alanine , 50 mg; L-arginine , 30 mg; L-cysteine , 20 mg; glycine , 20 mg; L-histidine , 30 mg; L-leucine , 30 mg; L-lysine , 20 mg; L-phenylalanine , 30 mg; L-proline , 80 mg; L-serine , 100 mg; L-tryptophan , 500 mg; L-tyrosine , 10 mg; L-valine , 40 mg; L-asparagine , 200 mg; L-aspartic acid , 200 mg; L-glutamine , 500 mg; L-glutamic acid , 300 mg; L-isoleucine , 20 mg; L-methionine , 10 mg; L-threonine , 100 mg ) , sugars ( glucose , 400 mg; fructose , 150 mg; sucrose , 800 mg ) and vitamins ( Vanderzant Vitamin Mix , 650 mg ) in 40 mL water and sterile filtered the solution . Additionally , we prepared an agar solution ( 1 g Agar-Agar in 60 mL water ) which was sterilized by autoclaving . After cooling the liquid agar solution to approximately 60°C in a water bath , we added the nutrient solution and aliquoted the diet under sterile conditions in single 0 . 5 mL micro-centrifuge tubes , where it solidified . These tubes were stored at 4°C until use . For experiments , T . notatus were placed in plastic boxes ( 10 × 6 × 6 cm ) covered with paper tissue and sealed with a perforated lid . In each box , 15 to 20 mirids were placed with a tube containing the artificial diet as the sole food source in addition to a source of water . The diet was exchanged with fresh diet every day . The shaded boxes were kept in the glasshouse . After 5 days the surviving mirids were collected , flash-frozen in liquid nitrogen and stored at −80°C until CK extraction . T . notatus oral secretions were collected as previously described ( Halitschke et al . , 2011 ) with minor modifications . In brief , we placed 15 to 20 mirids in a single plastic box ( 10 × 6 × 6 cm ) covered with paper tissue and sealed with a perforated lid . In each box , we placed an inverted scintillation vial lid filled to the brim with sugar solution ( ~3 mL , 40 mM glucose ) as the sole food and water source . Lids were covered by a stretched thin layer of Parafilm ( Neenah , WI , US ) which allowed for stylet penetration . After 24 hr , we collected the lids , removed the sugar solution with a syringe and carefully dissolved ( with MeOH ) the frass spots deposited on the parafilm . As a control , similarly packaged sugar solutions in boxes lacking mirids maintained under the same conditions were used . Frass and sugar solution samples originating from the exposure to approximately 100 mirids were pooled . Pooled sugar solutions were freeze-dried overnight . Prior to CK extraction , extraction buffer was used to dissolve the evaporated sugar solution . To measure the defense responses of N . attenuata to T . notatus feeding , whole plants were exposed to mirid attack in the T . notatus rearing cages and damaged lamina from the first ( lowest ) stem leaf were collected after three days of feeding . Control plants were placed in a similar but mirid-free cage under same conditions . Collected lamina pieces were flash-frozen in liquid nitrogen and stored at −80°C until analysis . For kinetic analysis of CKs , JA , JA-Ile , primary metabolites and photosynthetic rates during T . notatus attack , we used two different experimental setups which differed in the area damaged by T . notatus , and the number of T . notatus used to inflict the damage to the N . attenuata plants . In the first setup , only one leaf per plant was exposed to T . notatus . We enclosed twenty adults on the first ( lowest ) stem leaf in a round plastic clip-cage ( 7 cm dimeter , 5 cm height ) . Clip-cages had holes covered with a fine mesh for air ventilation . Control plants received empty clip-cages to control for the effects of caging leaves . We also sampled leaves from plants without clip-cages . Before sampling , mirid mortality was scored , and samples with more than 50% mortality were discarded . Control and damaged leaf lamina were harvested at nine time-points from separate plants ( 0 , 1 , 3 , 24 , 48 , 72 , 96 , 120 and 144 hr ) , flash-frozen in liquid nitrogen and kept at −80°C until analysis . In the second setup , the entire aboveground plant was exposed to mirids , by placing the plants directly into the T . notatus rearing cage; control plants were placed in a similar , empty cage . Damaged lamina from the first ( lowest ) stem leaf were sampled at seven time-points from separate plants ( 0 , 24 , 48 , 72 , 96 , 120 and 144 hr ) . Both experiments were started in the morning ( 09:00 – 12:00 ) and each harvest time represented at least three replicate plants . For each experiment , phytohormone concentrations ( CKs , JA and JA-Ile ) , sugars ( sucrose , fructose and glucose ) , starch , total soluble proteins , photosynthetic rates and chlorophyll contents were measured . Caffeoylputrescine and nicotine were extracted and determined by UHPLC-ToF-MS by analyzing extracted ion chromatograms as previously described ( Schäfer et al . , 2015c; Brütting et al . , 2017 ) . For extraction , 80% MeOH ( v/v ) was used for approximately 100 mg of frozen and ground leaf material from each sample ( exact tissue masses were recorded ) . Values are presented as peak area * g FM−1 . TPI activity was determined using a radial diffusion assay ( Jongsma et al . , 1994; van Dam et al . , 2001 ) with approximately 50 mg of frozen and ground leaf lamina ( exact tissue masses were recorded ) . TPI activity was normalized to leaf protein content . The protein content of the extracts used for the TPI assay was determined using a Bradford-assay ( Bradford , 1976 ) on a 96-well microtiter plate . Total soluble proteins were extracted from 50 mg of frozen ground leaf lamina ( exact tissue masses were recorded ) in 0 . 5 ml 0 . 1 M M Tris-HCl ( pH 7 . 6 ) following the protocol described by Ullmann-Zeunert et al . ( 2012 ) . Protein concentrations were measured using a Bradford assay ( Bradford , 1976 ) on a 96-well microtiter plate using bovine serum albumin ( BSA ) as standard . Free amino acids were extracted from leaf lamina by acidified MeOH extraction [MeOH:H2O:HCOOH 15:4:1 ( v/v/v ) ] and analyzed by liquid chromatography coupled to a triple quadrupole MS ( Bruker EVOQ Elite , Bruker Daltonics , Bremen , Germany; www . bruker . com ) , as previously described ( Schäfer et al . , 2016 ) . Glucose , fructose , sucrose and starch were determined following the protocol described by Machado and colleagues ( Machado et al . , 2015 ) . Briefly , 100 mg plant tissues ( exact tissue masses were recorded ) were extracted first with 80% ( v/v ) ethanol and later twice with 50% ( v/v ) ethanol , each by incubation for 20 min at 80°C . Supernatants from all extractions were pooled , and sucrose , glucose and fructose were quantified enzymatically as previously described ( Velterop and Vos , 2001 ) . The remaining pellets were used for an enzymatic determination of starch content ( Smith and Zeeman , 2006 ) . Net CO2 assimilation rate was measured with a LI-COR LI-6400/XT portable photosynthesis system ( LI-COR Inc . , Lincoln , NE , US ) . All measurements were conducted using a 2 cm2 chamber , at constant CO2 ( 400 µmol CO2 mol air−1 ) , light ( 300 μmol m−2 s−1 PAR ) , temperature ( 25–26°C ) and relative humidity ( 20–40% ) . Measurements of photosynthetic rates of leaves with clip-cage were specifically done on the area included in the clip-cage . We measured photosynthetic rates of control leaves and leaves damaged by T . notatus . Leaves with clip-cages were analyzed in the covered area shortly after removal of the clip cage . Chlorophyll was quantified using a Minolta SPAD Chlorophyll meter 502 . Chlorophyll content is displayed in arbitrary SPAD units . Each sample value is the mean of chlorophyll content measured at three different random spots from each analyzed leaf . Leaves with clip cages were analyzed in the covered area shortly after the removal of the clip cage . RNA was extracted with TRIzol ( Thermo Fisher Scientific , Waltham , MA , US ) , according to the manufacturer’s instructions . cDNA was synthesized by reverse transcription using oligo ( dT ) primer and RevertAid reverse transcriptase ( Thermo Fisher Scientific ) . Real-time qPCR was performed using actin as a standard on a Mx3005P qPCR machine ( Stratagene , San Diego , CA , US ) using a qPCR Core kit for SYBR Green I No ROX mix ( Eurogentec , Seraing , Belgium ) . The primer sequences are provided in Supplementary file 2 . Levels of defense signaling compounds after three days of T . notatus herbivory were quantified as described by Kallenbach et al . , 2010 ) . JA , JA-Ile , OPDA and SA were analyzed as described by Kallenbach and colleagues ( Kallenbach et al . , 2010 ) and ABA as described by Dinh and colleagues ( Ðinh et al . , 2013 ) . Kinetics of CKs , JA and JA-Ile during 144 hr of T . notatus attack was measured as previously described ( Schäfer et al . , 2016 ) . In brief , phytohormones were extracted from ca . 100 mg of fresh ground leaf material ( exact sample masses were recorded ) using acidified methanol and purified on reversed phase and cation exchange solid-phase extraction columns . The measurements were done via liquid chromatography coupled to a triple quadrupole MS ( Bruker EVO-Q Elite ) equipped with a heated electrospray ionization source . The method was extended for the detection of 15N-labeled CKs . The parent → product ion transitions for 15N labeled CKs are listed in Supplementary file 3 . Chromatograms of IP , [D6]-IP , [15N5]-IP as well as IPR , [D6]-IPR and [15N5]-IPR are shown in Figure 4—figure supplement 4 and 5 . The same extraction method was used for CK extraction from T . notatus , using approximately 10 mg of ground material ( five pooled samples of ca . 20 adults; exact sample masses were recorded ) . N . attenuata plants with more than 98% of their total nitrogen content as 15N were obtained following the protocol described by Ullmann-Zeunert and colleagues ( Ullmann-Zeunert et al . , 2012 ) . Briefly , twelve days after germination , plants were transferred into individual 50 mL hydroponic culture containers with Ca ( 15NO3 ) 2 as the sole nitrogen source . Ten days later , the plants were transferred to 1 L hydroponic culture chambers with the same 15NO3- concentration as of K15NO3 . Once per week , the plants were fertilized with 1 mM K15NO3 and the containers were maintained at 1 L with distilled water . To generate 15N labeled T . notatus , we reared insects for an entire generation on fully 15N-labelled N . attenuata plants . Two-hundred adult females were transferred to a 47 . 5 × 47 . 5 × 93 cm insect cage with four 15N-labelled N . attenuata plants in the early elongation stage of growth . Females were allowed to lay eggs for four days and subsequently removed . 15N-labelled plants were fertilized once a week ( as described above ) , and after three weeks two fresh 15N-labelled plants were added to the insect cage . One week after the first adults emerged , the 15N labeled mirids were collected and used for the cytokinin transfer experiment . Two types of experiments were conducted to quantify the transfer of cytokinins from mirids to N . attenuata plants . In the first , twenty 15N-labelled T . notatus adults were quickly anesthetized with CO2 prior to clip-caging on a lower stem leaf of a 14N-grown plant , with one clip-cage per plant . We collected and froze in liquid nitrogen the leaf lamina corresponding to the area included in the clip-cage and thus damaged by mirids at different time-points: 0 , 3 , 6 , 24 , 48 , 72 , 96 and 120 hr . Each harvest time included at least three replicate plants . Samples were stored at −80°C until analysis . In the second type of experiment , five 15N-labelled N . attenuata plants were placed in the mirid rearing cage . One lower stem or rosette leaf per plant was harvested at 0 , 3 , 6 , 24 , 48 , 72 , 96 and 120 hr , thus each harvest time represented five replicate plants . Plants were transferred once per day to a different cage to ensure that mirids did not accumulate 15N-labeled metabolites . We separated the leaf lamina from the mid-rib and froze the lamina in liquid nitrogen , and stored the samples at −80°C until analysis . In the field , the damaged area on different leaf types was estimated as % of the total leaf area . The proportion of damaged leaf area for each leaf was visually estimated and the leaves were grouped into three different types ( Figure 6—figure supplement 1a ) . Finally , we calculated average leaf damage for all leaf types: rosette leaves , the first ( oldest ) three stem leaves and all younger stem leaves and side branches . Similarly , T . notatus damage distribution within the plant was evaluated under controlled conditions in the glasshouse . In this experiment , a total of seven WT plants were used to form four replicates . The replicates consisted of three pairs of WT plants and one single plant , where each replicate was placed in one cage , meaning that one cage represented one replicate with no matter if there were one or two plants inside the cage . The plants in each cage were exposed to adults of T . notatus ( n = 10 insects/plant ) for one week . T . notatus infestations continued for an additional two weeks , where in the last week , five insects/plant were added to each cage . T . notatus damage was estimated from high resolution pictures of 15 leaves per plant at standardized rosette , mid stem , and young stem positions . Using Photoshop ( Adobe ) , the damage was evaluated and expressed as a percentage of total damage per plant . For choice assays conducted in the field between young and fully mature leaves , we collected insects from native populations at our field station in Utah , USA . Ten to fifteen T . notatus adults were placed in a plastic cup . The cup was connected to two other plastic cups ( Figure 6—figure supplement 1b ) , one enclosing a fully mature stem leaf and the other enclosing young , growing leaves ( apical meristem and young leaves which had not yet completed the sink-source transition ) . To prevent desiccation , leaf petioles were submerged in water in a 2 mL plastic microcentrifuge tube . As the insects are night-active , after one night ( 12 hr ) during which the insects could choose between the two containers , we counted the number of mirids in each . For choice assays between WT and irchk2/3 plants , we placed plants in a large mesh-enclosed cage in the glasshouse ( 3 × 4 × 1 . 6 m ) into which 500 T . notatus were released . The damage on each plant ( as described above ) was estimated 10 days later . Data from choice assays on i-ovipt plants were taken from a previously published dataset ( Schäfer et al . , 2013 ) . Plants were either treated with pure lanolin ( LAN ) as a control or with DEX-containing lanolin as described above . We treated the first ( oldest ) ten stem leaves of a flowering plant and placed one DEX- and one LAN-treated plant in one 47 . 5 × 47 . 5×93 cm insect cage . About 100 T . notatus adults were added to the cage , and the damaged leaf area was estimated after 10 days . The average damage level from all 10 treated leaves from each plant was counted as one replicate . Data were analyzed using R 3 . 3 . 1 ( 2016-06-21; http://www . r-project . org ) . Statistical tests and number of replicates , as well as transformations to data in order to meet assumptions of a test ( homoscedasticity , normality ) , are provided in the figure legends . Normality of data sets was assessed by Shapiro–Wilk tests and homoscedasticity by Levene's test . If not mentioned otherwise , time course data were analyzed with ANCOVA with mirid feeding as factor and time as continuous explanatory variable . If the response variable was not linearly dependent on time we used two-way ANOVAs ( TWA ) with mirid feeding and time as factors . In all the analyses of experiments with the data from clip-cages , we only used data from control clip-cages and clip-cages with mirids . Differences were considered significant when p<0 . 05 .	Many insects use plants for food and for shelter . To protect themselves , plants often develop defense mechanisms that deter or debilitate their attackers , such as producing toxins or storing nutrients away from the attacked tissues . But some insects manage to counter the plants’ defense responses . Such species are often less mobile and spend a large part of their life in a restricted area of the plant , for example , inside plant tissues . Also known as ‘endophytic’ animals , these insects can even manipulate the signaling system in a plant , such as a class of plant hormones called cytokinins , which help plants to grow and to develop seeds and nutrient-storing fruits or young leaves . Researchers have previously assumed that endophytic animals target cytokinins because they are restricted to living in certain areas within the plant , and – unlike ‘free-living’ insects – lack access to other , potentially more nutritious feeding sites . By modifying cytokinins , the location-bound insects could create their own ‘nutrient pool’ . Until now , it was unclear how insects transfer cytokinins to a plant and if this ability was restricted to endophytic insects . To investigate this further , Brütting , Crava et al . studied the response of coyote tobacco plants infested with a free-living insect , the sap-sucking bug Tupiocoris notatus . The experiments revealed that the insects’ bodies contained large quantities of a type of cytokinin , even when insects were raised on artificial diets . Brütting , Crava et al . then developed a method to clearly distinguish cytokinins present in the insects from those produced by the plants to test whether T . notatus can transfer these plant hormones during feeding . The results showed that similar to endophytic insects , T . notatus injects cytokinins into the attacked leaves , presumably to create a stable nutritious environment . Insects represent the largest and most diverse group of organisms on Earth , including many crop pests . Despite their detrimental impact on the environment and the health of the farmers , pesticides are used predominantly for pest control . A better understanding of how insects use cytokinins to increase the nutritional value of the leaves may help us to find ways to increase the crop’s tolerance to insect attacks .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology", "plant", "biology" ]	2018	Cytokinin transfer by a free-living mirid to Nicotiana attenuata recapitulates a strategy of endophytic insects
Over the past century , soybean oil ( SBO ) consumption in the United States increased dramatically . The main SBO fatty acid , linoleic acid ( 18:2 ) , inhibits in vitro the growth of lactobacilli , beneficial members of the small intestinal microbiota . Human-associated lactobacilli have declined in prevalence in Western microbiomes , but how dietary changes may have impacted their ecology is unclear . Here , we compared the in vitro and in vivo effects of 18:2 on Lactobacillus reuteri and L . johnsonii . Directed evolution in vitro in both species led to strong 18:2 resistance with mutations in genes for lipid biosynthesis , acid stress , and the cell membrane or wall . Small-intestinal Lactobacillus populations in mice were unaffected by chronic and acute 18:2 exposure , yet harbored both 18:2- sensitive and resistant strains . This work shows that extant small intestinal lactobacilli are protected from toxic dietary components via the gut environment as well as their own capacity to evolve resistance . While antibiotics can cause lasting alterations to the microbiome ( David et al . , 2014a; Dethlefsen et al . , 2008; Dethlefsen and Relman , 2011; Jakobsson et al . , 2010 ) , dietary perturbations rarely do so ( Sonnenburg et al . , 2016 ) . In humans and in mice , the gut microbiome can be quickly altered by diet but community composition generally recovers within days ( Carmody et al . , 2015; David et al . , 2014a; David et al . , 2014b; Zhang et al . , 2012 ) . Resilience to dietary perturbation may be direct , as gut microbes functionally adapt to diet , or indirect through buffering by the gut habitat . During the 20th century , the greatest dietary change in the United States was in the consumption of soybean oil ( SBO ) , which increased from less than 0 . 001 kg/person/year to 12 kg/person/year ( Blasbalg et al . , 2011 ) . Conventional ( ‘commodity’ ) soybean oil , frequently labeled as ‘vegetable oil’ , is a mixture of triglycerides composed of five long chain fatty acids ( FAs ) , with linoleic acid ( 18:2 ) comprising over 50% of the FAs . After triglycerides are hydrolyzed by lipases active in the saliva , stomach , and upper duodenum , free FAs and monoglycerides are absorbed in the small intestine ( Mansbach et al . , 2000 ) . The microbiota of the human small intestine is exposed to FAs during this process ( El Aidy et al . , 2015; Kishino et al . , 2013 ) : therefore , an increase in the concentration of specific FAs has the potential to reshape microbial communities and select for microbes that thrive in the novel environment . Linoleic acid and the other two major unsaturated FAs in SBO , oleic acid ( 18:1 ) , and alpha-linolenic acid ( 18:3 ) , are known to be bacteriostatic and/or bactericidal to small intestinal bacteria as non-esterified ( free ) fatty acids in vitro at concentrations found in the small intestine ( Kabara et al . , 1972; Kankaanpää et al . , 2001; Kodicek , 1945; Nieman , 1954 ) . The primary modes of killing include permeabilization of cell membranes ( Greenway and Dyke , 1979 ) and interference with FA metabolism ( Zheng et al . , 2005 ) . Affected microbes are predominantly Gram-positive bacteria including the genus Lactobacillus ( Nieman , 1954 ) . Lactobacilli are particularly important as they are considered beneficial members of the human small intestine ( Walsh et al . , 2008; Walter et al . , 2007; Walter et al . , 2011 ) . They have been shown to be growth inhibited by the specific FAs present in SBO ( Boyaval et al . , 1995; De Weirdt et al . , 2013; Jenkins and Courtney , 2003; Jiang et al . , 1998; Kabara et al . , 1972; Kankaanpää et al . , 2001; Kodicek , 1945; Raychowdhury et al . , 1985 ) . It is interesting to note that the human-associated L . reuteri underwent a population bottleneck that coincides with the increase in SBO consumption in the U . S . and is far less prevalent than it was in the past ( Walter et al . , 2011 ) . Despite its decline , L . reuteri and other lactobacilli persist in the small intestine of Western individuals , suggesting mechanisms to counter the inhibitory effects of FAs in vivo . Here , we explored mechanisms of microbiome resistance to toxic dietary components with a focus on linoleic acid ( 18:2 ) toxicity to L . reuteri and L . johnsonii . Using an in vitro evolution assay , we assessed the capacity for these species to develop 18:2 resistance . To assess resistance in the host , we fed mice from two vendors diets high or low in 18:2 for 10 weeks and exposed their intestinal microbes to acute dosing of 18:2 via gavage . Lactobacilli populations were quantified in live-only and whole cell fractions obtained from the small intestine , and isolates from mice were assessed for resistance in vitro . We confirmed the previously reported in vitro toxicity of long chain FAs towards L . reuteri by performing disc diffusion assays with the individual free FAs of soybean oil ( SBO ) using L . reuteri ATCC 53608 . We observed growth inhibition of this strain by free 18:1 , 18:2 , and 18:3 ( Figure 1A ) , and this inhibition occurred in the presence of completely hydrolyzed SBO ( Figure 1—figure supplement 1 ) . The two saturated free FAs 16:0 and 18:0 and glycerol did not interfere with growth . To determine if the inhibitory concentration of 18:2 was comparable to concentrations in the mammalian digestive tract , we performed a cell permeability assay using propidium iodide with L . reuteri ATCC 53608 over a 10-fold dilution range from 0 . 01 to 1000 μg/ml of 18:2 . We observed that 18:2 permeabilized the cells with an estimated inhibitory concentration 50 ( IC50 ) of 20 μg/ml ( p < 0 . 001 ) ( Figure 1B ) . This IC50 concurs with our estimates of the concentration of 18:2 present in a mouse consuming a SBO diet ( 11 to 28 μg/ml for a mouse on a 7% by weight SBO diet , see Materials and methods ) and with previous estimates of mammalian physiological relevant concentrations of unsaturated FAs ( Kankaanpää et al . , 2001; Kodicek , 1945 ) . Thus , physiological levels of 18:2 were toxic to L . reuteri in vitro . We next assessed 40 strains of L . reuteri for 18:2 resistance in liquid culture . These 40 strains were previously isolated from humans , pigs , rodents ( mice , rats ) , birds ( chicken , turkey ) , and sourdough and stemmed from six different continents ( Supplementary file 1 ) ( Böcker et al . , 1995; Oh et al . , 2010 ) . We quantified how the strains grew in 18:2 by taking the mean of the ratios for cells growing in 18:2 to cells growing in medium alone for each of the last three OD600 measurements at hours ~ 4 , 6 , and 8 during the growth assay ( see Materials and methods for a discussion on why this approach was used ) . L . reuteri strains have been shown to be host-specific and form host-specific clades ( Walter et al . , 2011 ) . We observed that the basal , rodent-associated strains on average were inhibited by 18:2 more strongly than the other strains ( Kruskal-Wallis test , p < 10−4 ) ( Figure 2 ) . However , we observed considerable variation within host sources , and the human-associated strains were no more resistant to 18:2 than the strains derived from pig , poultry , or sourdough . Moreover , within human strains , 18:2 resistance did not relate to L . reuteri isolation site ( Figure 2—figure supplement 1A ) . There also did not appear to be a clear relationship of 18:2 resistance with L . reuteri clades as defined by Oh et al . , 2010 ( Figure 2—figure supplement 1B ) . Overall , we observed variation in L . reuteri 18:2 resistance regardless of source . To directly test if 18:2 resistance could evolve in L . reuteri through exposure to 18:2 , we isolated an 18:2-sensitive L . reuteri strain ( LR0 ) from the jejunum of a conventionally-raised mouse ( see Materials and methods ) . We seeded five cultures with LR0 and passaged them twice daily from a growth-dampening concentration of 18:2 up to a growth-inhibitory concentration over a period of six weeks ( Figure 3A and Figure 3—figure supplement 1 ) . We also evolved five cultures of L . johnsonii strain ( LJ0 ) obtained from the same mouse . We selected L . johnsonii based on its high abundance in mouse small intestinal microbiota ( see below ) . At the end of the passaging regime , all of the evolved lactobacilli populations showed smaller zones of inhibition around 18:2 and 18:3 in a disc diffusion assay compared to their respective starting strains ( Figure 3B and Figure 3—figure supplement 2A ) . We tested isolates LR2-1 from population LR2 and LJ41072 from population LJ4 in liquid culture supplemented with 18:2 to confirm their 18:2 resistance ( Figure 3C and D ) . To characterize the mutations these populations acquired , we sequenced all five of the L . reuteri populations , four of the five L . johnsonii populations ( the fifth was lost ) , the evolved isolates LR2-1 and LJ4107 , and the starting strains LR0 and LJ0 , using 300 bp paired end sequencing on an Illumina MiSeq . For the populations , we achieved approximately 500X coverage , and for the isolates , 50X coverage ( Supplementary file 2 ) . Mutations were called in the populations and isolates by aligning sequencing reads to the assembled genome for the respective starting strain ( LR0 or LJ0 ) . After requiring mutations have a minimum frequency of 10% in a population and confirming all mutations were not due to potential mismapping , we observed 30 mutational events in 15 genes across the five L . reuteri populations and 35 mutational events in 21 genes in the four L . johnsonii populations . ( Supplementary file 3 and 4 ) . In each population , a few mutations had swept the entire population ( Tables 1 and 2 , Supplementary file 3 and 4 ) . Both the L . reuteri and L . johnsonii populations bore high frequency variants ( >60% ) in genes relating to FA metabolism , ion transport , and the cell membrane/wall . In the L . reuteri populations , we found high frequency variants in ( i ) FA biosynthesis transcriptional regulator FabT ( Eckhardt et al . , 2013 ) , ( ii ) two related tyrosine-protein kinases involved in exopolysaccharide synthesis , EpsD , and EpsC ( Minic et al . , 2007 ) , ( iii ) an HD family hydrolase , ( iv ) a hypothetical protein , and ( v ) in the region upstream of an ammonium transporter that may respond to acid stress ( Wall et al . , 2007 ) . In the L . johnsonii populations , high frequency mutations were present in ( i ) two distinct intracellular lipases , ( ii ) a putative membrane protein gene , ( iii ) the potassium efflux system KefA/small-conductance mechanosensitive channel , which protects against growth defects in acidic conditions ( Cui and Adler , 1996; McLaggan et al . , 2002 ) , ( iv ) the glycosyltransferase LafA , which affects the lipid content of the cell wall and membrane ( Webb et al . , 2009 ) , ( v ) a TetR family transcriptional regulator , and ( vi ) the ribonucleotide reduction protein NrdI . All but two of the above mutations are non-synonymous or cause protein truncations . The other two mutations are intergenic and thus may alter the expression of the downstream gene . The isolate LR2-1 contained both of the mutations present at high frequencies in the total LR2 population as well as an additional mutation in a hypothetical protein , which was present in the LR2 population at 45% ( Supplementary file 3 ) . Similarly , LJ41072 had all of the high frequency mutations present in its source population ( LJ4 ) and one additional mutation in LafA , which was mutated in 39% of the LJ4 population ( Supplementary file 4 ) . We observed no overlap in the specific genes mutated in L . reuteri and L . johnsonii . Only a subset of the genes mutated in one species are present in the other species ( EpsD , EpsC , FIG00745602 , LafA ) and in no case was the same mutation already present in the opposite species . Although the specific genes mutated differed between the two species , they are associated with similar functions , suggesting that Lactobacillus species can evolve 18:2 resistance through changes relating to lipid metabolism , acid stress , and the cell wall/membrane . To confirm the role of these genes in fatty acid resistance , we generated these mutations individually in a fatty acid sensitive background . The human derived L . reuteri ATCC PTA 6475 ( also called MM4-1A ) is amenable to recombineering ( van Pijkeren and Britton , 2012 ) . Of the genes mutated in L . reuteri , only FabT and the hydrolase gene are present in this strain . The amino acid sequences , but not the nucleotide sequences of these genes are identical between our mouse strain and L . reuteri 6475 . We created the LR2 18 bp deletion in FabT , the LR5 SNP in FabT , and the LR5 SNP in the hydrolase gene . The latter two were accompanied by several surrounding synonymous mutations as recombineering is orders of magnitude more efficient when multiple consecutive mutations are made due to the avoidance of the mismatch repair system ( van Pijkeren and Britton , 2012 ) . The specific mutations made are indicated in the recombineering oligos in Figure 3—source data 1 . Note that these oligos match the reverse strand of the chromosome . The LR2 18 bp deletion and the LR5 SNP in FabT present alone were able to enhance 18:2 resistance in L . reuteri 6475 , similar to that observed for the total LR2 and LR5 populations ( Figure 3—figure supplement 2B ) . The LR5 SNP in the hydrolase gene , however , was not sufficient to render the strain observably 18:2 resistant by a disc diffusion assay . We cannot rule out the possibility that the additional synonymous mutations we created in this strain impacted the phenotype or that mutation of the hydrolase gene enhances resistance in the background of a strain mutated for FabT . These results verify the role of the fatty acid transcriptional regulator FabT in L . reuteri 18:2 resistance . Given that 18:2 resistance can evolve in vitro , we asked if L . reuteri and L . johnsonii could survive either a chronic or acute exposure to 18:2 in vivo . For the chronic exposure , 3 week-old male C57BL/6J mice from Jackson Laboratories were fed ad libitum for 10 weeks a low fat ( LF , 16% kcal from SBO ) or high fat ( HF , 44% kcal from SBO ) diet , wherein all of the fat was derived from SBO ( Figure 4—source data 1 ) . For the acute exposure , at the end of the 10 weeks , we gavaged ( delivered to the stomach ) mice with 6 mg 18:2 per gram mouse weight ( e . g . , double the 18:2 consumed by mice daily on the LF diet ) or saline . At 1 . 5 hr post-gavage , when gavaged 18:2 is observed in the bloodstream ( Figure 4—figure supplement 1 ) , mice were sacrificed , and the small intestine contents were collected ( Figure 4A ) . To assess how the gavage impacted the microbiome of the jejunum , where the bulk of fat absorption occurs ( Alfin-Slater and Aftergood , 2012; Borgstrom et al . , 1962 ) , we sequenced the V4 region of 16S rRNA genes derived from DNA obtained from propidium monoazide ( PMA ) treated and untreated aliquots of each jejunal luminal sample . The PMA or ‘live-only’ aliquot , is depleted of DNA from cells with compromised membranes . In addition to live cells , the untreated or ‘total’ aliquot includes DNA from live as well as cells permeabilized by 18:2 and dead cells . This approach allowed us to gauge which taxa were still alive after the 18:2 treatment . The effect of 18:2 on the microbial community was evident from analysis of the live cells but not for the total cell population: microbiomes within a diet-group could be distinguished by gavage treatment only when the live-only aliquot was analyzed ( live-only; weighted UniFrac , n = 23 for LF diet: adonis , pseudo-F = 4 . 78 , 15% of variance explained , p = 0 . 022; n = 21 for HF diet: adonis , pseudo-F = 7 . 84; 28% of variance explained , p = 0 . 003; also see Figure 4—figure supplement 2A and B ) . This observation suggests that 18:2 compromised select microbes , thereby decreasing their abundances and altering the abundances of other live microbes . Such differences in microbial abundances due to the 18:2 gavage should be evident by directly comparing the total and live-only aliquots for each sample . Indeed , we observed that for both diets , although the jejunal contents from saline-gavaged mice showed differences in the live-only and total diversity , this difference was greater in mice gavaged with 18:2 ( p < 0 . 01 for LF diet , p < 10−7 for HF diet , Kruskal-Wallis tests ) ( Figure 4B ) . Hence , while compromised cells exist in the saline control animals , 18:2 caused additional cells to be compromised . We note that this difference ( beta-diversity distance ) was greater for the HF than the LF diet samples ( Figure 4B ) , suggesting that the HF-diet conditioned microbiome was disrupted to a greater extent by 18:2 than the LF-diet microbiome . The LF versus HF SBO diets themselves , on the other hand , had little effect on the microbiome . While the mice on the HF diet gained significantly more fat mass ( p = 1 . 53*10−4 , mean in HF diet group = 0 . 043 , mean in LF diet group = 0 . 028 , 95% CI = ( 0 . 0076 , 0 . 0218 ) , two-sample , two-tailed t-test on epididymal fat pad mass ) , we observed no differences between the total microbiome composition of the jejuna of mice on the two diets ( PERMANOVA on the total cell population , p > 0 . 5 ) . Using a Kruskal-Wallis test , with FDR < 0 . 1 , we observed that OTU 363731 mapping to Akkermansia muciniphila was 60-fold enriched in the HF diet . These results imply that the level of SBO and compensatory reduction in carbohydrates in the HF diet was not sufficient to greatly alter the microbiome . OTUs 692154 and 592160 , taxonomically assigned by Greengenes to L . reuteri and L . johnsonii , respectively , were the two most abundant lactobacilli OTUs in all samples . These OTUs displayed comparable relative abundances in the two diets ( total aliquot; Kruskal-Wallis test and ANOVA on a linear mixed model to include cage effects , p values > 0 . 05 , Figure 4C and D ) . These L . reuteri and L . johnsonii OTUs were present in the 18:2 , live-only microbiota in both sets of mice ( Figure 4C and D ) , suggesting these taxa survived the 18:2 acute treatment regardless of the dietary fat content . Note we detected L . reuteri OTU 692154 at very low levels in the microbiota of mice housed in three out of six LF diet cages and in two out of six HF diet cages ( Figure 4—figure supplement 2C ) . Comparison of the relative abundance of these two OTUs in the total and live-only microbiota revealed these lactobacilli ( with the exception of L . reuteri in the LF diet ) enriched 2- to 5-fold ( ANOVA on a linear mixed model to include cage effects and Kruskal-Wallis tests , p values < 0 . 01 ) after 18:2 gavage . Furthermore , the live-only microbiota of HF diet mice had an enrichment of 11 lactobacilli OTUs after 18:2 gavage ( 5- to 9-fold enrichment compared to control gavage , Kruskal-Wallis , FDRs < 0 . 1 , Figure 4—figure supplement 2D ) at the expense of Allobaculum spp . Similar enrichment of live lactobacilli after the 18:2 gavage was observed for the LF diet , although no OTU passed our significance threshold . These observations suggest that lactobacilli resist acute 18:2 exposure particularly in the context of a high-18:2 diet . To confirm that the Lactobacillus population was not reduced by the 18:2 gavage and that any changes in their relative abundances were due to die-offs of other bacteria , we quantified their levels in total and live cell fractions by qPCR . We determined the difference in the copy number of Lactobacillus 16S rRNA sequences in the total and live-only samples normalized to the equivalent difference for total Eubacteria . We observed no difference between the saline and 18:2 gavage samples for either diet ( Figure 4—figure supplement 2E , two-sample , two-tailed t-test , p values > 0 . 1 ) . All live-only to total relative copy numbers were close to 1 , as expected if the Lactobacillus population was not reduced by 18:2 exposure . To determine if our findings were limited to our specific mouse experiment , we repeated the chronic 18:2 exposure with two additional sets of mice originating from Taconic Biosciences and an F2 generation of mice from Jackson Laboratories . In these two additional sets of mice , 16S rRNA gene sequence diversity analysis of jejunal contents showed that the same two OTUs annotated as L . reuteri and L . johnsonii were again the predominant lactobacilli , although these are extremely unlikely to be the same lactobacilli strains present in our first study . In Taconic mice , L . reuteri and L . johnsonii were detected in the jejunum after 10 weeks on both diets ( Figure 4—figure supplement 3A and B ) . In F2 Jackson mice , L . johnsonii was detected after 10 weeks on both diets ( Figure 4—figure supplement 3C ) , whereas L . reuteri was only present in LF diet mice ( Figure 4—figure supplement 3D ) . L . reuteri , however , was not observed in fecal samples from week 0 ( Figure 4—figure supplement 3E ) . As all mice were similarly handled , the diets sterilized , and the mice bred in the same facility , L . reuteri may have invaded the LF mice , though we cannot rule out the possibility of L . reuteri existing below detection . Nevertheless , these additional studies support the notion that lactobacilli populations are minimally impacted by chronic dosing of 18:2 . Our results in mice suggest that L . reuteri and L . johnsonii survived chronic and acute exposure to 18:2 either directly , by 18:2 resistance , or indirectly , through an unknown aspect of life within the mouse gut . To assess the direct resistance of these lactobacilli to 18:2 , we established a collection of L . reuteri and L . johnsonii isolates derived from the upper ileum ( as a proxy for the jejunum ) of mice on both HF and LF diets . We determined the ability of these isolates to grow in liquid culture amended with 18:2 . While most isolates were sensitive to 18:2 , we observed that L . reuteri isolates recovered from the HF-diet fed mice were on average more resistant to 18:2 than L . reuteri isolated from the LF-diet fed mice ( 113 isolates from 15 mice in eight cages Kruskal-Wallis , p < 0 . 05 , Figure 4E and Figure 4—figure supplement 4A ) . This observation is consistent with the hypothesis that chronic exposure to a diet high in 18:2 promotes resistance in the resident L . reuteri population . Next , we sought to relate the in vitro resistance of the L . reuteri isolates to the in vivo changes in L . reuteri populations before and after acute 18:2 exposure . To do so , we assessed the enrichment of L . reuteri OTU 692154 in the live jejunal aliquot post 18:2 gavage: we considered the rarified sequence counts for this OTU in the live-only aliquot ( i . e . , in PMA-treated samples ) in mice gavaged with 18:2 normalized by the equivalent sequence counts for the OTU in saline gavaged co-caged mice . A resulting log10 ratio greater than 0 indicates that live L . reuteri OTU 692154 had greater relative abundance counts in mice gavaged with 18:2 compared to same-cage controls gavaged with saline , signifying that other OTUs had been depleted . We observed no correlation between the ability of these strains to grow in vitro in 18:2 and their abundance in mice gavaged with 18:2 for mice on either diet ( Figure 4—figure supplement 4B and C ) . Note that we cannot exclude the possibility that the isolation procedure favored susceptible strains , and thus is not representative of the in vivo population . With this caveat in mind , these results indicate that while chronic exposure to 18:2 can result in L . reuteri strains with higher 18:2 resistance , the mouse gut environment protects susceptible strains . We partially replicated these findings with L . johnsonii: all isolates were sensitive to 18:2 , but L . johnsonii from the HF diet-fed mice were more strongly inhibited by 18:2 than those isolated from the LF-fed mice ( 159 isolates from 22 mice in 12 cages; Kruskal-Wallis , p < 0 . 001 , Figure 4E ) . Therefore , the results for L . johnsonii are similar to those of L . reuteri , with a lack of congruence between the response of the population in vivo and the resistance of isolates in vitro . We sequenced to 50X coverage an isolate of L . reuteri resistant to 18:2 , derived from a HF diet mouse ( strain LRHF , Supplementary file 2 , Figure 4—figure supplement 5 ) . Although we cannot be certain that HF diet-isolated L . reuteri share a common ancestor with those present in LF diet mice , we compared LRHF to LR0 , the 18:2-susceptible isolate from a LF diet mouse and used in the in vitro evolution assay . The comparison revealed 71 mutations in 60 genes with functions predominantly in DNA metabolism , energy metabolism , and environmental response ( Table 3 , Supplementary file 3 ) . None of the genes mutated in the in vitro evolution assay differed between LR0 and LFHF . LRHF exhibited mutations in a sodium-hydrogen antiporter gene and a peroxide stress ( PerF ) gene , both of which may represent adaptation to an acidic environment caused by exposure to FAs . Of potential relevance to FA exposure , we observed mutations in a membrane-bound lytic murein transglycosylase D precursor involved in the production of the peptidoglycan layer ( Vollmer et al . , 2008 ) and the fructosyltransferase Ftf involved in the production of exopolysaccharide ( Sims et al . , 2011 ) . These results suggest that exposure to 18:2 in vivo does not invoke selection on the same genes that are implicated in 18:2 resistance in vitro . A drastic change in dietary macronutrient composition has the capacity to restructure the microbiome within a day ( David et al . , 2014b; Faith et al . , 2011; Turnbaugh et al . , 2009 ) and is one of the most influential contributors to microbiome composition ( Carmody et al . , 2015 ) . Here , we consider how the gut microbiome is influenced by diet from the perspective of a single FA known to be toxic to gut microbes: specifically , the interaction between lactobacilli and linoleic acid ( 18:2 ) . In accord with previous reports , we observed 18:2 to inhibit the growth of most naturally-derived lactobacilli in vitro . However , in the mouse gut , L . reuteri and L . johnsonii persisted through both chronic and acute exposures to 18:2 . L . reuteri isolates derived from mice on a diet high in 18:2 included some that were more resistant to 18:2 . This observation suggests that 18:2 resistance has the potential to be selected in a host . In vitro , L . reuteri and L . johnsonii both evolved 18:2 resistance through mutations in the cell wall/membrane and fat metabolism genes . Collectively , these data indicate that the host gut environment protects gut microbes from the inhibitory effects of FAs , but that these microbes can also evolve resistance , providing additional resilience . The mutations our 18:2 in vitro adapted lactobacilli strains acquired are consistent with the known bacteriostatic and bactericidal mechanisms of 18:2: by increasing membrane fluidity and permeability ( Greenway and Dyke , 1979 ) potentially leading to cell lysis or leakage ( Galbraith and Miller , 1973b; Parsons et al . , 2012 ) , by blocking absorption of essential nutrients ( Nieman , 1954 ) , and by inhibiting FA synthesis ( Zheng et al . , 2005 ) and oxidative phosphorylation ( Galbraith and Miller , 1973a ) . Lactobacilli are also capable of combating 18:2 toxicity by converting 18:2 to conjugated 18:2 and subsequently a monounsaturated or saturated fatty acid ( Jenkins and Courtney , 2003; Kishino et al . , 2013 ) . We did not recover any mutations in genes known to be involved in the production of conjugated 18:2 . Despite the toxicity of 18:2 towards lactobacilli , mouse-associated L . reuteri and L . johnsonii were present at equivalent relative abundances in mice fed diets high or low in 18:2 . Moreover , these microbes survived a gavage of 18:2 equal to double what mice normally encounter in their daily diet . Our results are consistent with the findings of Holmes and colleagues , who analyzed the fecal microbiomes of mice on 25 different SBO diets varying in their macronutrient ( fat , protein , carbohydrate ) composition . Their results demonstrate that fat has only a minor effect on microbiome structure ( Holmes et al . , 2017 ) . In contrast , in microbial systems engineered for waste processing , concentrations of linoleic acid within the range predicted to be consumed by animals can cause failure of the desired microbial biodegradation processes ( Lalman and Bagley , 2000 ) . The resistance of lactobacilli to linoleic acid in the mouse host is therefore inferred to be dependent on the complexity of the gut habitat . In mice , lactobacilli colonize both the small intestine and forestomach ( Walter et al . , 2007 ) . While lingual lipases exist in mice ( DeNigris et al . , 1988 ) , fat digestion occurs primarily in the small intestine . As a result , forestomach microbes should not be exposed to a high concentration of free FAs , and SBO itself is not toxic . A gavage of 18:2 , on the other hand , exposes forestomach microbes to free 18:2 . Lactobacilli may be protected from this direct exposure by their capacity to form a dense biofilm on non-mucus secreting stratified epithelial cells ( Frese et al . , 2013 ) . In the human host , other aspects of the small intestinal habitat likely buffer the microbiota . The decline of L . reuteri in Western populations may never be fully explained . In the 1960’s and 1970’s prior to the emergence of SBO as a major dietary fat source , L . reuteri was recovered from the intestinal tract of 50% of subjects surveyed and was considered a dominant Lactobacillus species of the human gut ( Reuter , 2001 ) . Today , however , it is found in less than 10% of humans in the USA and Europe ( Molin et al . , 1993; Qin et al . , 2010; Walter et al . , 2011 ) , yet it is present at a reported 100% prevalence in rural Papua New Guineans ( Martínez et al . , 2015 ) . Moreover , human L . reuteri strains show very little genetic variation ( Duar et al . , 2017; Oh et al . , 2010 ) , and one human associated lineage of L . reuteri appears to have arisen approximately when SBO consumption increased ( Walter et al . , 2011 ) . These observations raise the question of whether a change in dietary habits drove the decline in the prevalence of L . reuteri in Western populations . In humans , L . reuteri forms neither high gastric populations nor biofilms ( Frese et al . , 2011; Walter , 2008 ) , thus human-derived L . reuteri strains may have survived increased exposure to 18:2 by developing resistance . Indeed , we did observe that some human L . reuteri strains are resistant to 18:2 , but not all . While the increase in SBO consumption may have conspired with other facets of modernization to reduce the prevalence of L . reuteri in Western populations , it did not appear to have resulted in a selective sweep of 18:2 resistant L . reuteri . The mechanistic underpinnings of how dietary components shape the composition of the gut microbiome need to be further elucidated if manipulation of the microbiome for therapeutic applications is to succeed . Dietary components have the potential to inhibit microbes directly through their toxicity , or indirectly by promoting the growth of other , more fit , microbes . While FAs are generally toxic to many lactobacilli , this work suggests that toxicity is greatly reduced when lactobacilli are host-associated . Future work in this area will elucidate how the host environment protects gut microbes from otherwise toxic dietary components such as FAs , and the ways specific strains within the microbiome can be resilient to such stresses . Supplementary file 1 details the naturally derived L . reuteri strains from various hosts and countries . The L . reuteri strain ( LR0 ) and L . johnsonii strain ( LJ0 ) used in the in vitro 18:2 evolution assay were isolated from the jejunum contents of a mouse originally purchased from Taconic Biosciences ( Hudson , NY , USA ) and maintained on the low fat soybean oil diet for 6 weeks since weaning , and strain LRHF was isolated from a parallel mouse on the high fat soybean oil diet for 6 weeks since weaning ( see Mouse care section for further details ) . Lactobacilli were cultured in MRS liquid medium ( Criterion , Hardy Diagnostics , Santa Maria , CA , USA ) or on MRS agar plates ( Difco , BD , Sparks , MD , USA ) , pH-adjusted to 5 . 55 using glacial acetic acid . All liquid cultures and plates were incubated at 37°C in an anoxic chamber ( Coy Lab Products , Grass Lake , MI , USA ) supplied a gas mix of 5% H2 , 20% CO2 , and 75% N2 . We plated 100 μl of a dense , overnight culture of L . reuteri strain ATCC 53608 on an agar plate and applied sterile Whatman paper ( Buckinghamshire , UK ) discs to the surface of the culture plate . To each disc , we added 10 μl of each test compound or control . Compounds tested were alpha-linolenic acid ( 18:3 ) ( ≥99% , L2376 , Sigma Aldrich , St . Louis , MO , USA ) , linoleic acid ( 18:2 ) ( ≥99% , L1376 , Sigma Aldrich ) , oleic acid ( 18:1 ) ( ≥99% , O1008 , Sigma Aldrich ) , stearic acid ( 18:0 ) ( ≥98 . 5% , S4751 , Sigma Aldrich ) , palmitic acid ( 16:0 ) ( ≥99% , P0500 , Sigma Aldrich ) , 0 . 85% NaCl ( saline ) , DMSO , glycerol , all afore-mentioned FAs mixed ( FA mix ) , the FA mix with glycerol , and soybean oil ( Wegmans , NY , USA ) . FAs were dissolved in DMSO to a concentration of 50 mg/ml , except for stearic acid , which was dissolved to a concentration of 5 mg/mL due to its lower solubility . For the FA mix , the five FAs were mixed in the ratio that these FAs are present in soybean oil: 14% 16:0 , 4% 18:0 , 23% 18:1 , 52% 18:2 , 6% 18:3 . For the FA mix with glycerol , glycerol was mixed with the FA mix to a molar mass ratio of 0 . 1 ( e . g . , the molar mass ratio of glycerol in the total molar mass of soybean oil ) . For testing glycerol alone , the same amount of glycerol used in the FA mix with glycerol was used , and the total volume was brought up to 10 μl with DMSO . Plates were dried for 20 min at 37°C before being turned agar side up and incubated overnight . First , we centrifuged 5 mL of an overnight culture of L . reuteri ATCC 53608 at 10 , 000 rcf for 10 min and resuspended the pellets in 30 mL of 0 . 85% NaCl solution . Then we centrifuged 1 mL aliquots of the resuspended culture at 15 , 000 rcf for 5 min . The resulting pellets were resuspended in 0 . 85% NaCl solution to a total volume of 1 mL in the presence of 18:2 , 18:3 , 0 . 85% NaCl , or ethanol . We diluted FAs in 100% ethanol in a ten-fold dilution series ranging from 0 . 01 to 1000 μg/ml . We incubated samples at room temperature for 90 min on a rocking platform ( setting 6; VWR , Radnor , PA , USA ) and inverted the samples by hand every 20 min to ensure adequate mixing . After exposure to the FA , we washed the cells by centrifuging at 15 , 000 rcf for 5 min , and resuspending the pellets in 1 mL 0 . 85% NaCl; we repeated this wash a second time . To measure the permeability of the cells , we stained samples using the Live/Dead BacLight Bacterial Viability Kit ( L7007 , Invitrogen , Life Technologies , Grand Island , NY , USA ) according to the manufacturer’s instructions . We measured fluorescence from propidium iodide and SYTO9 on a BioTek Synergy H1 Hybrid Reader ( BioTek Instruments , Inc . , VT , USA ) . At each FA concentration , fluorescence was read in triplicate ( technical replicates ) . We used the drc package ( Ritz et al . , 2015 ) in R ( Team , 2016 ) for dose-response modeling and statistical analyses . The mice in this study consumed on average 2 . 7 grams of mouse food per day . Therefore , mice on a 23% by weight soybean oil mouse diet ( the 44% by calorie HF diet ) , consume 0 . 62 grams SBO . In SBO , fatty acids comprise 90% of the molar mass . As 52% of the fatty acids in SBO are 18:2 , therefore in a day , a mouse consumes ~0 . 3 grams 18:2 . We estimated that the transit time of fat from feeding and into the bloodstream is approximately 1 . 5 hr ( Figure 4—figure supplement 1A ) . Using the approximation that food is consumed continuously over the course of the day , we expect 18 μg of 18:2 to pass through the small intestine in a 1 . 5 hr period . The volume of the small intestine is between 200 and 500 μl ( McConnell et al . , 2008 ) and therefore approximately 36 to 91 μg/ml 18:2 will pass through the small intestine in a transit time . For the 7% by weight SBO diet ( 16% by calorie LF diet ) , 11 to 28 μg/ml 18:2 will pass in a transit time . We inoculated a Lactobacillus reuteri or L . johnsonii colony grown 1 to 2 days on an MRS agar plate into a well containing 300 μl MRS liquid medium on a sterile 2 ml 96 well polypropylene plate ( PlateOne , USA Scientific , FL , USA ) . We covered the plate with Breathe-Easy polyurethane film ( USA Scientific , FL , USA ) and incubated the plate overnight at 37°C in an anoxic chamber ( Coy Lab Products , Grass Lake , MI , USA ) supplied a gas mix of 5% H2 , 20% CO2 , and 75% N2 . Following overnight growth , we split the cultures 100-fold into a new 96 well plate , whereby each overnight culture was diluted into a well containing MRS medium and to a well containing MRS medium plus 1 mg/ml linoleic acid . To emulsify the FA in solution , prior to and following inoculation , we vortexed the 2 ml plate on a Multi-Tube Vortexer ( VWR , PA , USA ) for 30 s at setting 3 . 5 . We then transferred the entire plate to a 300 μl Microtest Flat Bottom non-tissue treated culture plate ( Falcon , Corning , NY , USA ) . We measured the OD600 of the plate on a BioTek Synergy H1 Hybrid Reader ( BioTek Instruments , Inc . , VT , USA ) at approximately 0 , 2 , 4 , 6 , and 8 hr . For growth curves of strains LR0 , LR2-1 , LRHF , LJ0 , and LJ41072 , cultures were read in triplicate ( technical replicates ) . We quantified how well the strain grew in 18:2 compared to the without 18:2 control by analyzing the last three time points of the growth assay . We used this approach over fitting a doubling time because , in the first few points of the growth curve , the OD values in wells with cells and 18:2 were lower than those in inoculation-control wells ( e . g . , with 18:2 , but lacking cells ) . Hence , for the first few time points when subtracting the OD600 of medium with 18:2 , without cells from the OD600 of medium with 18:2 , with cells , we obtained negative OD600 values . As well , the time spent in log phase varied among the strains and proper modeling of log to late-log phase could not be achieved without significant trimming and manipulation of the data . At these final three time points , we determined the ratio of the ‘blanked OD600s’ for the strain growing in MRS medium with linoleic acid to the strain growing in MRS medium alone:OD600MRS with 18:2OD600MRS We excluded time points in which the OD600 in MRS medium alone was less than 0 . 1 ( i . e . strain did not grow ) . We determined the mean of the above ratios for the last three time points . All negative normalized cell densities were confirmed to result from negative values in the OD600 of cells growing in 18:2 . For the naturally derived L . reuteri strains , we tested strains in triplicate to sextuplet ( biological replicates ) and averaged replicate normalized cell densities . For each L . reuteri and L . johnsonii strain isolated from mice on the SBO diet , we tested eight isolates from two mice per cage . The sample sizes for the SBO diet mice isolated strains were upper bounded by the observation that the microbiomes of these mice were dominated by one or few L . reuteri/johnsonii OTUs . Sixty-two isolates were tested between 2 and 5 times ( biological replicates ) and normalized cell densities were averaged across replicates . Statistical analyses were completed using kruskal . test in the R stats package ( Team , 2016 ) . For L . reuteri strain LR0 and L . johnsonii strain LJ0 , both originating from a mouse on the LF SBO diet for 6 weeks , we inoculated a single colony into 5 ml MRS and grew the cultures overnight . The following day , we diluted the overnight cultures for LR0 and LJ0 100-fold , separately , into five 5 ml MRS medium supplemented with 5 mg/ml 18:2 . These five cultures became the five populations evolved for L . reuteri or L . johnsonii and we refer to them as LR1-5 and LJ1-5 , respectively . We passaged these cultures twice daily using a 100-fold dilution . We omitted an emulsifier ( DMSO or ethanol ) from this assay to avoid the possibility of the lactobacilli adapting to the emulsifier rather than to 18:2 . As a result , we needed to use a relatively high concentration of 18:2 . To promote and maintain emulsification of the FA , we rigorously vortexed the tubes every few hours throughout the day . After seven days , we increased the concentration of 18:2 to 6 mg/ml . Each subsequent week , we increased the concentration by 1 mg/ml until reaching a final concentration of 10 mg/ml . Each week , we froze a 20% glycerol stock of each population at −80°C . We excluded L . johnsonii population #1 , LJ1 , from further study due to contamination . We isolated genomic DNA from approximately 30 μl cell pellets frozen at −20°C using the Gentra Puregene Yeast/Bact . Kit ( Qiagen , MD , USA ) . For isolates , we grew a single 50 ml log to late-log phase culture from a single colony . For populations , we inoculated five 10 ml cultures directly from glycerol stock , grew the cultures to log to late-log phase , and thoroughly mixed the replicate cultures together before pelleting to aid in representing the diversity of the original population structure . We grew 18:2-adapted isolates and populations in MRS medium with 10 mg/ml 18:2 , and non-adapted isolates in MRS medium . We used the Gentra Puregene Yeast/Bact . kit following the optional protocol adjustments: a 5 min incubation at 80°C following addition of the Cell Lysis Solution , a 45 min to 60 min incubation at 37°C following RNase A Solution addition , and a 60 min incubation on ice following addition of Protein Precipitation Solution . DNA was resuspended in Tris-EDTA and further purified using the Genomic DNA Clean and Concentrator−25 ( Zymo Research , CA , USA ) . We quantified isolated DNA using the Quant-iT PicoGreen dsDNA Assay Kit ( Thermo Fisher Scientific MA , USA ) . Lastly , to ensure we had obtained large molecular weight DNA , we ran the DNA on a 1% sodium borate agarose gel ( Agarose I , Amresco , OH , USA ) . We prepared barcoded , 350 bp insert libraries using the TruSeq DNA PCR-Free Library Preparation Kit ( Illumina , CA , USA ) . We fragmented starting genomic DNA ( 1 . 4 μg ) using the recommended settings on a Covaris model S2 ( Covaris , MA , USA ) . The barcodes used for each library are indicated in Supplementary file 2 . We submitted these barcoded libraries to the Cornell University Institute of Biotechnology Resource Center Genomics Facility where they were quantified by digital PCR using a QX100 Droplet Reader ( Bio-Rad Laboratories , CA , USA ) , pooled ( Supplementary file 2 ) , and pair-end sequenced on an Illumina MiSeq 2 × 300 bp platform using reagent kit V3 ( Illumina , CA , USA ) . Resulting reads from libraries sequenced on multiple MiSeq runs were merged for further analyses . To generate reference genomes for the ancestor strains used in the in vitro evolution assay , we assembled paired-end sequences for L . reuteri LR0 and L . johnsonii LJ0 using SPAdes v3 . 7 . 1 ( Nurk et al . , 2013; Prjibelski et al . , 2014 ) with k-mers 21 , 33 , 55 , 77 , 99 , and 127 using the ‘careful’ option to reduce mismatches and indels . To select and order contigs , we aligned the assembled genomes against the closest complete genome available: NCC 533 for L . johnsonii and TD1 for L . reuteri as determined by a whole genome alignment using nucmer in MUMmer ( Kurtz et al . , 2004 ) . The assembled genomes we aligned against the NCC 533 or TD1 genome using ABACAS . 1 . 3 . 1 ( Assefa et al . , 2009 ) with the ‘nucmer’ program . Next , we aligned previously unaligned contigs using promer . We merged these sets of aligned contigs into one file and contigs with low coverage , less than 20 , were removed . Finally , we ordered these filtered contigs using promer without the maxmatch option ( -d ) to prevent multiple reference-subject hits . For the LR0 genome , we identified a contig representing a plasmid from the assembly and included it in the set of assembled contigs . We uploaded these assembled genomes to RAST ( Aziz et al . , 2008; Brettin et al . , 2015; Overbeek et al . , 2014 ) for annotation ( see Supplementary file 2 for details on the assembled genomes ) . First , we manually identified variant alleles in an isolate from L . reuteri population LR2 , LR2-1 , and an isolate from L . johnsonii population LJ4 , LJ41072 , using the Integrative Genomics Viewer ( Robinson et al . , 2011; Thorvaldsdóttir et al . , 2013 ) . We used the variants in these isolates to calibrate the allele detection methods applied to the whole populations . Next , we identified variant alleles in the populations by aligning the paired-end sequence reads to the ancestor genome ( LR0 or LJ0 ) using BWA-MEM ( Li and Durbin , 2009 ) . We marked duplicate sequences using Picard 2 . 1 . 1 ( http://broadinstitute . github . io/picard ) and utilized Genome Analysis Toolkit ( GATK ) ( McKenna et al . , 2010 ) , and the GATK Best Practices recommendations ( DePristo et al . , 2011; Van der Auwera et al . , 2013 ) to accurately select true variants . This pipeline realigns indels and recalibrates and filters base calls using the known alleles identified in the isolates using a BQSR BAQ gap open penalty of 30 . We used the GATK HaplotypeCaller to call alleles with the maxReadsInRegionPerSample option set utilizing the observed coverage binned across the genome by the GATK DepthOfCoverage script . We applied the following options for populations and isolates: pcr_indel_model was set to ‘NONE’ , stand_call_conf was set at ‘10’ , stand_emit_conf at ‘4’ . For populations only , we set sample_ploidy at ‘10’ and for isolates , ‘1’ . After we had separately processed all populations and isolates , we jointly called alleles across the entire set of populations and isolates using GenotypeGVCFs with sample_ploidy at ‘10’ , stand_call_conf at ‘10’ , and stand_emit_conf at ‘4’ . We filtered these results to remove alleles with frequencies less than 10% and to remove alleles in genes annotated with ‘mobile element protein’ , ‘transposase’ , ‘phage’ , or ‘RNA’ . In addition , the ancestor genomic reads were mapped onto the ancestor genome to aid in the removal of poorly mapping reads . We removed alleles discovered in the evolved isolates and populations that were also present at frequencies greater than 0 . 5 in the aligned ancestor reads against the reference . The remaining alleles we manually checked using IGV to remove any alleles in regions of the genome with abnormally high coverage , compared to the directly adjacent regions , likely representing genomic repeat regions . Filtered and unfiltered reads are presented in Supplementary file 3 and 4 . We used PredictProtein ( Yachdav et al . , 2014 ) to predict the cellular location and structure of hypothetical and putative proteins and SignalP 4 . 0 ( Petersen et al . , 2011 ) to predict signal peptide sequences . To test the role of the mutations discovered in the in vitro evolution experiment on fatty acid resistance , we recreated the L . reuteri mutations in the recombineering strain PTA 6475 using the procedure described by van Pijkeren and Britton ( 2012 ) . Briefly , L . reuteri ATCC PTA 6475 ( BioGaia AB , Sweden ) bearing the plasmid pJP042 , which has inducible RecT and is selectable with 5 µg/ml erythromycin , was induced with 10 ng/ml peptide pheromone ( SppIP ) ( Peptide 2 . 0 , VA , USA ) at OD600 0 . 55–0 . 65 . After washing the cells in 0 . 5 M sucrose , 10% glycerol , we electroporated the cells with 100 µg of the recombineering oligo targeting the FabT or hydrolase gene and 40 µg of oligo oJP577 ( van Pijkeren and Britton , 2012 ) , which targets rpoB , rendering the cells rifampicin-resistant . We electroporated in 0 . 2 cm Gene Pulser cuvettes ( Bio-Rad , CA , USA ) using a Bio-Rad Gene Pulser Xcell with conditions 2 . 5 kV , 25 µF , and 400 Omega . We recovered cells for 2 hr at 37°C and then plated the cells on MRS supplemented with 25 µg/ml rifampicin and 5 µg/ml erythromycin . We screened resulting colonies using either a restriction digest or primers specific to the mutation through mismatch amplification mutation analysis-PCR ( MAMA-PCR ) ( Figure 3—source data 1 ) . For screening by restriction digest , we first amplified the FabT or hydrolase gene by colony PCR in 8 µl reactions: a small amount of a colony , 100 nM f . c . of each primer ( see Figure 3—source data 1 ) , and 1x Choice Taq Mastermix ( Denville Scientific , MA , USA ) . PCR conditions were 94°C for 10 min , 35 cycles of 94°C for 45 s , 56 or 58 . 5°C ( see Figure 3—source data 1 ) for 1 min , and 72°C for 30 s , followed by a final extension at 72°C for 10 min . Reactions were held at 10°C and stored at 4°C . Following , the PCR products were digested in 16 µl reactions at 37°C for 1 hr: 8 µl PCR product , 0 . 2 µl ( four units ) MfeI ( NEB , MA , USA ) , and 1x CutSmart Buffer ( NEB ) . For screening by MAMA-PCR , PCRs were carried out as before except an additional primer specific to the mutation was included . We confirmed that the mutations were correct by Sanger sequencing ( GENEWIZ , NJ , USA ) the entire FabT or hydrolase gene using PCR conditions and primers previously described . The pJP042 plasmid was lost from cells by passaging in MRS . All animal experimental procedures were reviewed and approved by the Institutional Animal Care and Usage Committee of Cornell University protocol 2010–0065 . The 16% and 44% SBO diets were custom designed by and purchased pelleted , irradiated , and vacuum packed from Envigo ( formerly Harlan Laboratories , Inc . , Madison , WI , USA , www . envigo . com ) . We stored open , in-use diet bags at 4°C and unopened , bags at −20°C . See Figure 4—source data 1 for the diet compositions . The increase of SBO in the HF diet was compensated by a decrease in cornstarch ( carbohydrate ) . Also , the amounts of protein ( casein ) , vitamins , and minerals were increased in the HF diet to prevent nutritional deficiencies from arising: HF diet fed mice consume a more calorically dense diet and thus intake a smaller volume of food per body mass . We gavaged nine mice with 6 mg per gram mouse weight 18:2 . Every half hour following gavage , we euthanized a mouse by CO2 asphyxiation and collected blood by cardiac puncture . Blood was collected into EDTA coated tubes and stored on ice . Tubes were spun at 900 rcf at 4°C for 10 min and plasma was collected and stored at −80°C . We extracted lipids using the Bligh and Dyer method ( Bligh and Dyer , 1959 ) and quantified FA methyl esters on a Hewlett-Packard 5890 series II gas chromatograph with a flame ionization detector ( GC-FID ) using H2 as the carrier . See Su et al . , 1999 for further details . We used a linear mixed model to determine if the gavage treatments significantly altered the plasma levels of 18:2 and 18:3 in 1 . 5 hr . The model was fatty acid mass ~diet + gavage + total fatty acid mass + ( 1\|cage ) + ( 1\|GC run date ) + ( 1\|fatty acid extraction date ) + plasma vol + ( 1\|study ) , where the terms cage , GC run date , fatty acid extraction date , and study were handled as random effects and all others as fixed effects . GC run date refers to when the extracted fatty acids were run on the gas chromatograph , and plasma volume refers to the amount of mouse plasma used in the extraction . Models were run in R ( Team , 2016 ) using the lme4 package ( Bates et al . , 2015 ) with REML = FALSE and the control optimizer set to ‘bobyqa’ . Significance values were determined using a two-sample , two-tailed t-test ( t . test in the R stats package ( Team , 2016 ) on the least squares means estimates data from the predict R stats function run on the model . In this study , we used three sets of male C57BL/6 mice bred in three different facilities: Jackson Laboratories ( Bar Harbor , ME , USA ) , Taconic ( Hudson , NY , USA ) , and an F2 generation of mice originally purchased from Jackson Laboratories . At weaning ( 3 weeks of age ) , we split littermates into cages housing up to four mice and provided the mice either the LF ( 16% kcal SBO ) or HF ( 44% kcal SBO ) diet ( Figure 4—source data 1 ) . Littermates were split so to balance mouse weights within a cage and between the two diets . All mice were housed in the Accepted Pathogen Facility for Mice at Cornell University . In total , 24 mice were purchased directly from Jackson Laboratories and maintained in six cages on the LF diet and 24 mice in six cages on the HF diet; from Taconic , 12 mice in three cages on the LF diet and 12 mice in three cages on the HF diet; and the F2 mice from Jackson Laboratories were comprised of 11 mice in five cages on the LF diet and 15 mice in five cages on the HF diet . Sample sizes of five mice per group have been successful in delimiting diet-driven microbiome composition differences ( Turnbaugh et al . , 2008 ) . The three different sets of mice were maintained at distinct time periods with the goal of ensuring our findings were not specific to a given base-microbiota . Up to four mice were co-caged . We stocked cages with Pure-o-cel ( The Andersons , Maumee , Ohio , USA ) , cotton nestlets , and plastic igloos so to avoid the introduction of exogenous fat . Food was placed in the cages and not on the wire racks to minimize loss and crumb buildup of the diets as the HF SBO diet does not maintain pelleted form . Twice weekly , we completely replaced cages and food . We weighed the amount of new food provided . To obtain mouse weights , we weighed mice in plastic beakers at the same approximate time of day twice weekly . We collected fresh fecal samples once weekly from the beakers into tubes on dry ice , which were later stored at −80°C . Mice were handled exclusively inside of a biosafety cabinet . We changed personal protective equipment and wiped all surfaces with a sterilant between cages to prevent cross-contamination . To measure food consumption , we filtered food crumbs out of the used bedding using a large hole colander followed by a fine mesh sieve , weighed the recovered food , and subtracted this amount from the known amount of food provided . After 10 weeks on the SBO diets , we gavaged the Jackson Laboratory mice with saline ( 0 . 85% NaCl ) or 18:2 . The Taconic mice were gavaged with phosphate buffered saline ( PBS ) or 18:2 , and the F2 mice from Jackson Laboratories with PBS , 18:2 , or 18:3 . The volume gavaged was 6 mg per gram mouse weight . The amount of FA gavaged is roughly double the amount of 18:2 consumed by mice on the LF diet each day , and more than half of the 18:2 consumed per day by mice on the HF diet . Within a cage , we gavaged half of the mice with a FA and the other half with saline/PBS , selecting which mouse received which gavage so to balance mouse weights between gavage groups . Following gavage , we moved mice to a fresh cage supplied with water , but lacking food . After 1 . 5 hr , we euthanized mice by decapitation and harvested small intestine contents ( see below ) . To harvest the jejunal contents , we divided mouse small intestines into three equivalent pieces . For Jackson Laboratory mice , we flushed the middle segment , the jejunum , with 10 ml anoxic 0 . 85% NaCl using a blunt , 18G , 1 . 5 inch needle into a 15 ml conical tube that we immediately placed on ice . After flushing , we quickly shook the tube and split its contents roughly equally into a second 15 ml conical tube . One of the tubes we covered with foil to which we added 12 . 5 μl of propidium monoazide ( PMA ) ( Biotium , Fremont , CA , USA; f . c . 50 μM from a 2 mM stock dissolved in DMSO ) . Which tube received PMA , the original or the second , we alternated between mice . To the other tube , we added 12 . 5 μl DMSO . To allow the PMA time to enter permeabilized cells , we placed all tubes on ice on a rocking platform for 5 min . To activate the azido group in PMA and cause DNA damage , we removed the foil from the tubes , placed the tubes horizontally on ice , and exposed the tubes for 5 min to a 650W halogen bulb ( Osram 64553 C318 , Danvers , MA , USA ) positioned 20 cm from the samples . We frequently rotated the tubes during these 5 min to ensure equal light exposure across the whole sample . We immediately spun these tubes at 4500 rcf for 5 min at 4°C . After we discarded the supernatant , we flash froze the tubes on liquid N2 , placed them on dry ice , and later stored the tubes at −80°C . We also flushed the upper half of the last segment of the small intestine , the ileum , with MRS medium and 20% glycerol , immediately placed the glycerol stock on dry ice , which we later stored at −80°C . For the other mice , we flushed the jejunum with 10 ml anoxic PBS ( pH 7 . 4 ) and did not use a PMA treatment . The small intestine contents for these mice were pelleted as described above . We used the PowerSoil DNA isolation kit ( Mo Bio Laboratories , Carlsbad , CA , USA ) to extract DNA from these jejunal pellets frozen in 2 ml tubes containing 0 . 1 mm glass beads ( Mo Bio Laboratories , Carlsbad , CA , USA ) . We eluted the DNA on the spin filter using 50 μl Solution C6 and stored the DNA at −20°C . We conducted blank extractions in parallel . We processed mouse fecal pellets in a similar manner . We quantified DNA samples and blank extractions using the Quant-iT PicoGreen dsDNA Assay Kit . For each sample , we performed two 50 μl PCRs to amplify the V4 region of the 16S rRNA gene using primers 515F ( f . c . 100 nM ) , Golay barcoded 806R ( f . c . 100 nM ) ( Caporaso et al . , 2012 ) , 5 Prime Mix ( Quanta Biosciences , CA , USA ) or Classic++ Taq DNA Polymerase Master Mix ( TONBO biosciences , CA , USA ) , and 25 ng of DNA . PCR conditions were 94°C for 3 min , 30 cycles of 94°C for 45 s , 50°C for 1 min , and 72°C for 1 . 5 min , followed by a final extension at 72°C for 10 min . Reactions were held at 4°C and stored at −20°C . We combined the two 50 μl PCRs and purified DNA using Mag-Bind E-Z Pure ( OMEGA Bio-tek , GA , USA ) following the manufacturer’s instructions and eluting with 35 μl TE . We measured DNA concentrations using PicoGreen . We pooled 100 ng of amplicon DNA from each sample together and sequenced the pool using the Illumina MiSeq 2 × 250 bp platform at the Cornell Biotechnology Resource Center Genomics Facility . We processed , filtered , and analyzed the 16S rRNA gene amplicon data from all studies using QIIME 1 . 9 . 0 ( Caporaso et al . , 2010 ) . Paired-end reads were joined using join_paired_ends . py running the fastq-join method and requiring at least 200 bp of sequence overlap . Joined reads were demultiplexed using split_libraries_fastq . py requiring a Phred quality cutoff of 25 to remove ambiguous barcodes and low quality reads . Reads were clustered into operational taxonomic units ( OTUs ) using open-reference OTU picking at 97% sequence identity to the Greengenes database version 13 . 8 ( DeSantis et al . , 2006 ) . We focused our analyses on the two most abundant lactobacilli OTUs: OTU 692154 identified as L . reuteri and OTU 592160 as L . johnsonii as denoted by the Greengenes assignment . We confirmed these assignments by sequencing the full 16S rRNA gene of lactobacilli isolates ( see below ) . Except where noted , for all subsequent analyses , we rarified data to 40 , 000 sequences per sample . We calculated beta-diversity using the weighted UniFrac metric implemented in QIIME 1 . 9 . 0 . We performed adonis ( PERMANOVA ) with 10 , 000 iterations and beta-diversity plots with the ordplot function using a t-distribution in the phyloseq package ( McMurdie and Holmes , 2013 ) . We identified OTUs differentiating samples by first filtering OTU tables to only include those OTUs present in at least 25% of samples and with at least one sample having at least 100 counts of that OTU . To the filtered OTU tables , we applied a Kruskal-Wallis test with an FDR cutoff of 10% using the group_significance . py script in QIIME . We created heatmaps of OTUs passing with FDR < 0 . 1 using the make_otu_heatmap . py script in QIIME . To detect L . reuteri in the fecal pellets of F2 mice from Jackson Laboratories , samples with at least 10 , 000 sequences were used ( sequencing depth was lower for the fecal pellets ) , and data were not rarefied so to maximize detection of L . reuteri . We determined the copy numbers of the lactobacilli 16S rRNA gene and total Eubacterial 16S rRNA gene in the PMA and non-PMA treated jejunal aliquots by quantitative real-time PCR ( qPCR ) using the LightCycler 480 platform and the SYBR Green I Master kit ( Roche Diagnostics Corporation , Indianapolis , IN , USA ) . We utilized the lactobacilli and Eubacterial primers described by Oh et al . , 2012 . PMA treatment reduces the total amount of DNA extracted by removing DNA from any dead cells . Thus , using the same mass of DNA for the PMA and non-PMA aliquots would result in quantifying copy numbers relative to the total amount of DNA assayed , similar to the relative abundances determined from the 16S rRNA gene sequencing . Therefore , we fixed the amount of DNA used for all non-PMA samples to 10 ng . Thus , 10 μl qPCRs consisted of 10 ng of DNA for the non-PMA aliquots and equal volume for the PMA aliquot , each qPCR primer at 500 nM , and 5 μl of SYBR Green I Master mix . Cycling conditions were 5 min at 95°C followed by 45 cycles consisting of 10 s at 95°C , 20 s at 56°C for the Eubacterial primers and 61°C for the lactobacilli primers , and 30 s at 72°C after which fluorescence from SYBR Green was read . Melting curve analysis was used to determine whether each qPCR reaction generated a unique product . Cycle threshold ( Ct ) values were calculated using the absolute quantification/2nd derivative max function available on the LightCycler 480 software . All reactions were run in triplicate , and the mean Ct values were used in subsequent calculations . To determine if the Lactobacillus population decreased due to the 18:2 gavage , we calculated the difference in lactobacilli copy number between the PMA ( live-only cells ) and non-PMA ( total cells ) aliquots relative to that for Eubacteria . That is , 2ΔCtLacto ( PMA−non . PMA ) 2ΔCtEubac ( PMA−non . PMA ) If the Lactobacillus population is not affected by the 18:2 gavage , no difference should be observed between the saline and 18:2 gavage samples . Significance values between gavage groups were calculated using two-sample , two-tailed t-tests . Moreover , this ratio is expected to be close to one if lactobacilli were not specifically killed by the 18:2 gavage . For each cage , we split the mice according to which gavage they received ( 18:2 or saline ) and we took the mean of the rarefied sequence counts for OTU 692154 ( L . reuteri ) . Then we calculated the log10 of the ratio of the 18:2 mean rarefied sequence counts to the mean saline relative abundance sequence counts:log10mean 18:2 counts per cage for OTUmean saline counts per cage for OTU We streaked the glycerol stocks of mouse ileum contents onto MRS agar plates . One or two colony morphologies were present on nearly all plates: lowly abundant bright cream , round colonies present on most plates , and abundant flatter , dull white colonies present on all plates . We determined the species identity of these colony morphologies by full length 16S rRNA gene sequencing using primers 27F ( f . c . 1 nM ) and 1391R ( f . c . 1 nM ) ( Turner et al . , 1999 ) , 10 μl of Classic++ Hot Start Taq DNA Polymerase Master Mix ( Tonbo Biosciences , CA , USA ) , and a small amount of a single bacterial colony in a 25 μl reaction . PCR conditions were 94°C for 3 min , 38 cycles of 94°C for 45 s , 50°C for 1 min , and 72°C for 1 . 5 min , followed by a final extension at 72°C for 10 min . We purified PCRs using Zymo DNA Clean and Concentrator−5 ( Zymo Research , CA , USA ) and submitted samples to Cornell University Institute of Biotechnology Sanger sequencing facility . Returned sequences were assembled using Sequencher version 5 . 4 . 6 ( DNA sequence analysis software , Gene Codes Corporation , Ann Arbor , MI , USA , http://www . genecodes . com ) and aligned against National Center for Biotechnology Institute’s nr database . The lactobacilli raw sequencing reads and the assembled genomes for strains LR0 and LJ0 are available under BioProject accession PRJNA376205 at National Center for Biotechnology Institute . The RAST genome annotations for these genomes are available in Supplementary file 5 and 6 . The 16S rRNA gene amplicon data are available under the study accession PRJEB19690 at European Nucleotide Archive . Code to generate figures , mutational analysis pipelines , and relevant raw data are available at https://github . com/sdirienzi/Lactobacillus_soybeanoil ( Di Rienzi , 2017; copy archived at https://github . com/elifesciences-publications/Lactobacillus_soybeanoil ) .	Though “you are what you eat” may just be a figure of speech , it is clear that what we eat does affect our own cells and the microbes that live in our gut . During the 20th century , the American diet changed dramatically and now includes a lot more vegetable oil from soybeans , the main component of which – a fat called linoleic acid – is toxic to many microbes . Among the microbes inhibited by linoleic acid is a beneficial bacterium called Lactobacillus reuteri . This microbe has become less common in Western populations , and the timing of its decline approximately follows when the consumption of soybean oil began increasing . However , L . reuteri and other related microbes still exist in people who eat a Western diet . This suggests that these bacteria must be protected from linoleic acid in the gut , or that they can become resistant to this toxic molecule . Now , Di Rienzi et al . report that both protection in the gut and resistance could explain how L . reuteri can persist in the presence of linoleic acid . First , experiments in the laboratory showed that these microbes could indeed become resistant to linoleic acid , either by gaining mutations in genes involved in creating fats , by growing in an acid , and by forming a cell wall . Further experiments involving mice then showed that the gut protects also L . reuteri from this molecule: linoleic acid did not inhibit L . reuteri within the mouse , but those same L . reuteri were inhibited when grown outside of a mouse . Di Rienzi et al . went on to recover some resistant L . reuteri from mice , implying that there is a mix of resistant and non-resistant strains in the mouse gut . However , notably , the resistant bacteria recovered from the mice did not have mutations in the genes that had been identified from the earlier experiments . Together these findings show that gut bacteria have several means of surviving the high levels of potentially toxic fat molecules . Also , the specific finding that linoleic acid does not inhibit L . reuteri within the gut may help scientists to understand how a high fat diet affects microbes; for example , it is possible that the decrease in carbohydrates or protein that occurs in high fat diets may explain why such diets cause microbes to be lost . Lastly , and on a practical level , linoleic acid-resistant L . reuteri may in the future be used as a probiotic in foods rich in vegetable oil .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2018	Resilience of small intestinal beneficial bacteria to the toxicity of soybean oil fatty acids
Drosophila central neurons arise from neuroblasts that generate neurons in a pair-wise fashion , with the two daughters providing the basis for distinct A and B hemilineage groups . 33 postembryonically-born hemilineages contribute over 90% of the neurons in each thoracic hemisegment . We devised genetic approaches to define the anatomy of most of these hemilineages and to assessed their functional roles using the heat-sensitive channel dTRPA1 . The simplest hemilineages contained local interneurons and their activation caused tonic or phasic leg movements lacking interlimb coordination . The next level was hemilineages of similar projection cells that drove intersegmentally coordinated behaviors such as walking . The highest level involved hemilineages whose activation elicited complex behaviors such as takeoff . These activation phenotypes indicate that the hemilineages vary in their behavioral roles with some contributing to local networks for sensorimotor processing and others having higher order functions of coordinating these local networks into complex behavior . A major obstacle to understanding how the central nervous system ( CNS ) translates sensory inputs into appropriate motor outputs is the CNS's staggering cellular complexity . Even the small CNS of Drosophila has on the order of 100 , 000 neurons ( Power , 1943; Chiang et al . , 2011 ) , each of which is capable of making connections with dozens of synaptic partners . To understand such a complex system , it is helpful to parse its elements into relevant units organized in a functional hierarchy . This approach can be seen in the analysis of the vertebrate brainstem and spinal cord , which are the major sites of sensorimotor processing for locomotor behaviors . In the spinal cord , for example , identified progenitor cells generate discrete pools of interneurons , which in turn have defined roles in producing functional circuits ( Grillner and Jessell , 2009 ) . The functional organization of the spinal cord in vertebrates appears to be conserved , in that the progenitor cells and transcription factors that characterize different interneuron classes are conserved from fish through mammals ( Lupo et al . , 2006 ) despite the marked differences in modes of locomotion . The hindbrain of the zebra fish is similarly structured from clusters of cell types that are recruited in predictable ways to assemble functional circuits ( Koyama et al . , 2011 ) . Like the vertebrate spinal cord , the ventral nervous system ( VNS ) of insects is the site of sensorimotor patterning for complex behaviors such as walking , jumping , and flight . Years of work on grasshoppers and other orthopteroid insects ( summarized in Burrows , 1996 ) have defined clusters of spiking and nonspiking interneurons that transform sensory input into changes in leg position during the maintenance of stance ( Burrows , 1996 ) and organize more complex behaviors such as walking ( Buschges et al . , 2008 ) and flight ( Robertson et al . , 1982 ) . As in vertebrates , these clusters follow a developmental logic in which specific clusters arise from the same neuronal stem cell ( neuroblast [NB] ) . In the grasshopper , for example , NB 4-1 has been shown to produce the cluster of midline spiking interneurons that receive exteroceptor input and shape the receptive fields of leg motoneurons ( Shepherd and Laurent , 1992 ) . Similarly , the unpaired medial NB produces the pool of local , midline GABAergic neurons that respond to sound ( Thompson and Siegler , 1991 ) . The finding that specific stem cells generate pools of functionally similar interneurons has important evolutionary implications because the insect VNS develops according to a conservative plan based on a segmental set of 30 paired and one unpaired NBs that has changed little through the 350 million years of insect evolution ( Thomas et al . , 1984; Truman and Ball , 1998 ) . Each NB is uniquely identifiable and characterized by its position in the array , its pattern of molecular expression ( Broadus and Doe , 1995 ) , and the suite of early neurons that it produces ( Bossing et al . , 1996; Schmid et al . , 1999 ) . Comparative studies between a basal , primitively flightless insect and flying insects showed that the number of thoracic neurons roughly doubled with the evolution of flight . This increase did not come from adding new NBs , but rather was correlated with a subset of the NBs making more neurons by extending their proliferative phase ( Truman and Ball , 1998 ) . This mixed proliferative response associated with the evolution of flight is consistent with the idea that particular neuronal lineages have become adapted to specific behavioral functions , but this has yet to be comprehensively evaluated . The systematic decoding of the behavioral roles of the segmental set of NBs can best be carried out in Drosophila because of its life history and the genetic tools available . Its metamorphic life history is advantageous because the great majority of central neurons arise during a second neurogenic period as the larva grows . During this postembryonic phase , each NB divides repeatedly , with the smaller product of each division being a ganglion mother cell ( GMC ) , which then terminally divides to produce two neurons . By the start of metamorphosis each NB and its immature progeny form a discrete cell cluster from which one or two axon bundles project to specific regions in the larval neuropil ( Truman et al . , 2004 ) . Each axon bundle identifies the neurons of a hemilineage , a set of neuronal ‘cousins’ that includes either the ‘A’ ( Notch-on ) or ‘B’ ( Notch-off ) daughter from the GMC division ( Truman et al . , 2010 ) . In Drosophila , 25 of the 30 embryonic NBs in a hemisegment generate postembryonic lineage . Of the 50 possible hemilineages , 17 are removed by programmed cell death so that a segmental unit of the adult thoracic CNS contains neurons from only 33 hemilineage-based pools . These account for 90–95% of the neurons in the adult thoracic CNS . Hemilineages are also units for the early molecular diversity within the VNS . Expression of many early transcription factors is restricted along hemilineage lines ( Lacin et al . , 2014 ) and many of these transcription factors have homologs involved in fate determination in the vertebrate spinal cord ( Thor and Thomas , 1997 ) . For example , the homeodomain protein Dbx1 controls the difference between the V0 and V1 fates in the mouse spinal cord ( Pierani et al . , 2001 ) , while the Drosophila homolog , dbx , controls differentiation of particular VNS cell types through interactions with even-skipped and hb9 ( Lacin et al . , 2009 ) . We developed a set of genetic tools to allow us to examine the form and function of most of the thoracic hemilineages in Drosophila . We find that most hemilineages are composed of pools of similar interneurons , confirming that neuronal classes are indeed based on a hemilineage plan . Stimulation of interneurons in a hemilineage pool using the temperature-sensitive cation channel dTRPA1 ( Hamada et al . , 2008 ) elicits a characteristic , often unique , behavioral response from each . The complexity of the elicited behavior reflects the complexity of the hemilineage's projection pattern and the neuropil region to which it projects . These findings suggest that the hemilineages provide a functional as well as an anatomical ground plan for the thoracic nervous system , and that thorax-mediated behaviors occur by combinations of simple movements elicited by relatively simple ventral hemilineages , which are then orchestrated by a hierarchy of increasingly complex dorsal hemilineages . The neurons in a hemilineage are designated by their NB of origin and whether they are the Notch-on ( A ) or Notch-off ( B ) daughters of the GMC division ( Truman et al . , 2010 ) . For example , hemilineages 1A and 1B are the respective A and B daughters of NB 1 . We are using the postembryonic designations of the NBs ( from Truman et al . , 2004 ) rather than their embryonic names ( i . e . , Schmid et al . , 1999 ) because there is controversy over the correspondence of the two maps ( Birkholz et al . , 2015; Lacin and Truman , in preparation ) . Working with isolated hemilineages has been difficult because the members are cousins rather than sisters . Consequently , a clonal approach , such as MARCM ( Lee and Luo , 1999; Yu et al . , 2010a ) , serves to mark the 2 daughters of the GMC division , but does not label all the neurons in one hemilineage at the exclusion of those in the other . Therefore , we devised approaches to mark the secondary neurons in a hemilineage and then track them through metamorphosis to assess their form and function in the adult . Of the 33 major thoracic hemilineages , hemilineages 0A , 4A , 16B and 17A are not addressed in this study . Since each hemilineage can be identified in the larva based on cluster position and trajectory of its axon bundle ( Truman et al . , 2010 ) , we screened the larval expression patterns of the Rubin GAL4 collection ( Jenett et al . , 2012 ) based on over 7000 cis-regulatory regions ( CRMs; see Pfeiffer et al . , 2008 ) , for lines that drove expression in single hemilineages ( Li et al . , 2014 ) . The larval pattern was then maintained by using a transiently expressed recombinase to remove a transcriptional stop cassette from an Actin5C -FRT>-stop-FRT>-LexA::p65 transgene and thereby extending expression into the adult stage . The successful application of this strategy , though , required that the recombinase activity was limited to the larval growth period when the desired hemilineage pattern was expressed . Because of the diversity of expression patterns in the driver lines , we explored three different methods for implementing this strategy ( summarized in Figure 1 ) . The results of the first method ( see Figure 1A ) is illustrated in Figure 2 . In the last instar larva , line R24B02-GAL4 drives expression in the cells of hemilineage 12A from the subesophageal through the A1 segments ( Figure 2A ) . However , this expression pattern wanes during metamorphosis and is replaced by a different set of cells in the adult CNS ( Figure 2B ) . To maintain lineage expression into the adult , we used R24B02-GAL4 to drive pJFRC180-20XUAS-IVS-Flp2::PEST , hereafter referred to as UAS-Flippase , to remove the transcriptional stop from an Actin5C-FRT>-dSTOP-FRT>-GAL4 construct . As shown in Figure 2C , when combined with the reporter construct , pJFRC2-10XUAS-IVS-mCD8::GFP ( Pfeiffer et al . , 2010 ) , expression in the hemilineage 12A cells was maintained into the adult . A complication , though , was that expression was observed in other cells , presumably ones in which the stop cassette had been removed in earlier larval or embryonic stages , when this CRM drives a different expression pattern . To circumvent this early expression we used a GeneSwitch strategy ( Osterwalder et al . , 2001; Roman et al . , 2001 ) in which we fused a Drosophila codon-optimized ligand-binding domain of the human progesterone receptor to the DNA-binding domain of GAL4 . R24B02-GeneSwitch should then drive expression only after larvae are treated with a progesterone mimic , such as RU486 . As seen in Figure 2D , E feeding larvae RU486 during the third instar resulted in strong expression of immature 12A interneurons . Removal of the drug at the start of metamorphosis was followed by a waning of GFP expression so that it was lost by adult emergence ( Harris , 2012 ) . Feeding adults with RU486 reinduced GFP expression but only in the expected adult pattern and not in the 12A interneurons . Consequently , the RU486 given to feeding larvae is effectively cleared from the animal during metamorphosis ( this study ) . We then used the R24B02-GeneSwitch line in conjunction with UAS-flippase , Actin5C-FRT>-dSTOP-FRT>-GAL4 , and pJFRC2-10XUAS-IVS-mCD8::GFP to reveal the adult morphology of the hemilineage 12A neurons . Without treatment with RU486 in the larva , we saw no GFP expression at any stage ( Figure 2D , F ) , but feeding them with the drug during the third larval instar resulted in expression in the 12A interneurons that persisted through metamorphosis and continued to be robust in the adult ( Figure 2G , H ) . Importantly , this adult expression pattern did not also contain the adult-specific cells characteristic of R24B02 because the RU486 was cleared from the system before these cells started to express late in metamorphosis . 10 . 7554/eLife . 04493 . 003Figure 1 . The different strategies that were used to establish lines that showed selective expression in defined hemilineages . The strategies are based on a screen through the large collection of enhancer lines built from cis regulatory modules ( CRMs ) of CNS expressed genes . ( A ) For CRMs whose thoracic expression is confined to a hemilineage , gene-switch constructs are combined with feeding larvae the progesterone mimic ( RU486 ) in the last larval stage . The larval expression of flippase then promotes the excision of a STOP cassette from another trans-gene allowing a constitutive promotor ( Actin5C ) to drive continual expression following excision . ( B ) When the larval expression pattern includes functional larval neurons as well as a hemilineage , expression in the larval neurons is blocked by including a nSynaptobrevin-GAL80 gene . Gene switch cannot be used in this context because it is not suppressed by GAL80 . ( C ) Spatial and temporal specificity is accomplished using a conditional flippase that is the human progesterone receptor ligand-binding domain ( hPR ) fused to Flippase . Exposure of third instar larvae to RU486 then confines the flip event to the last larval stage . See text for more details . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 00310 . 7554/eLife . 04493 . 004Figure 2 . Strategy for clean expression of hemilineages into the adult stage . Z-projections of confocal stacks showing the expression pattern driven by R24B02 used in various genetic combinations . Arrowheads show the t1 and t2 clusters of hemilineage 12A . ( A , B ) Pattern shown by R24B02-GAL4 driving pJFRC2-10XUAS-IVS-mCD8::GFP ( pJFRC2 ) in larval ( A ) and adult ( B ) stages . The hemilineage 12A clusters are prominent at the end of larval life , but do not express in the adult . ( C ) Adult VNS of a cross of Actin5C>dSTOP>GAL4 , UAS-Flippase; pJFRC2 to R24B02-GAL4 . The persisting expression in the hemilineage 12A clusters is badly obscured by many ‘off-target’ cells ( e . g . , red arrowheads ) presumably arising from embryonic and adult expression patterns in this line . ( D , E ) R24B02-GeneSwitch flies crossed to pJFRC2 and either maintained without hormone ( D ) or fed on 1 mM RU486 food for 24 hr as third instar larvae . ( F–H ) Adult nervous systems of flies of the genotype R24B02-GeneSwitch , Actin5C>dSTOP>GAL4 , UAS-Flippase , pJFRC2 and either raised without hormone mimic ( F ) or fed RU486 food during the third larval stage ( G , H ) . H shows a higher power view of the hemilineage 12A cells found in T1 , T2 and A1; this hemilineage dies in T3 . Green: GFP; magenta: N-cadherin . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 004 A second strategy ( see Figure 1B ) to remove extraneous expression from hemilineage lines was based on the fact that the arrested secondary neurons do not express terminal differentiation products such as synaptic vesicle proteins . We first confirmed that a driver for such a gene—R57C10-GAL4 that carries an 872-bp promoter fragment from the nSynaptobrevin gene ( nSyb; Pfeiffer et al . , 2008; Pfeiffer et al . , 2012 ) —drives expression in the functioning primary neurons , but not in the clusters of arrested immature secondary neurons ( Figure 3A , inset ) . This promoter fragment was then used to drive expression of GAL80 , an inhibitor of GAL4 activity ( Yun et al . , 1991; Traven et al . , 2006 ) , in the construct R57C10-GAL80-6 , hereafter called nSyb-GAL80 . When crossed into lines that contained a mixture of functioning larval neurons and clusters of arrested , postembryonic cells ( Figure 3B ) , nSyb-GAL80 suppressed expression in the mature neurons , leaving only the expression in the arrested immature neurons ( Figure 3C ) . When nSyb-GAL80 was used in conjunction with a GAL4 diver , UAS-flippase and Actin5C-FRT>-dSTOP-FRT>-LexAp::65 , we were able to fix—or ‘immortalize’—the expression pattern of that GAL4 pattern specifically in the immature neurons , now as a pattern of LexA activity , which was observed using pJFRC19-13XLexAop2-IVS-myr::GFP ( Pfeiffer et al . , 2010 ) . However , significant off target expression was observed ( Figure 3D , E ) because we did not have temporal control over the timing of the flippase activity . 10 . 7554/eLife . 04493 . 005Figure 3 . nSyb-GAL80 suppresses GAL4 expression specifically in mature neurons . Yellow arrowheads show location of hemilineage 12B cell clusters in the thoracic segments . ( A ) In the third-instar larva , R57C10-GAL4 , a nSynaptobrevin promoter fusion-GAL4 drives expression in primary neurons , but not secondary neurons . Inset: single optical slice showing colocalization of anti-GFP ( green ) and the pan-neural marker anti-elav ( magenta ) in primary neurons ( e . g . , white arrowhead ) , but not in immature secondary neurons ( e . g . , blue arrowhead ) . ( B ) R15D11-GAL4 drives expression in secondary hemilineage 12B ( yellow arrowheads ) and various primary neurons . ( C ) R15D11-GAL4 with nSyb-GAL80: expression is suppressed in primary neurons , and only 12B expression remains . ( D ) When used to drive UAS-Flippase in a flip-on immortalization strategy , R15D11-GAL4 yields expression in hemilineage 12B , but also in numerous off-target cells ( red arrowheads ) . Genotype: w; UAS-Flippase ( attP40 ) /pJFRC19-13XLexAop2-IVS-myr::GFP ( attP40 ) , Actin5Cp4 . 6>dsFRT>LexAp65 ( su ( Hw ) attP5 ) ; R15D11-GAL4 ( attP2 ) . ( E ) Same genotype as ( D ) , but with nSyb-GAL80 on the X chromosome ( su ( Hw ) attP8 ) . Expression in off-target cells is much reduced . A small amount of off-target expression is observed in secondary neurons and descending neurons ( e . g . , red arrowheads ) . ( F ) Adult VNC expression pattern using both nSyb-GAL80 and UAS-Flp-Switch to restrict expression to the targeted cells in ( C ) . No off-target expression is observed . In panels D–F , myr::GFP concentrates in processes rather than in the cell bodies and therefore the cell clusters are difficult to see in Z-projections . E , F: insets are partial projections showing t1 cell body clusters . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 00510 . 7554/eLife . 04493 . 006Figure 3—figure supplement 1 . UAS-hPR-flp is fully active in the presence of RU486 , but inactive without drug . ( A ) R20B05-GAL4 drives UAS-GFP expression in all secondary neurons . ( B ) A larva bearing the genotype pJFRC2-UAS>STOP>GFP ( attP18 ) /w; +; UAS-hPR-flp ( VK00005 ) /R20B05-GAL4 and fed on RU486 food for 24 hr prior to dissection . GFP expression is present in nearly all of the expected cells , indicating that the STOP cassette has been excised by UAS-hPR-flp with high efficiency . ( C ) A larva of the same genotype as ( B ) , but never treated with RU486 . No GFP expression is observed , meaning that the hPR-flp was inactive in the absence of drug . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 006 To make a more precise tool ( depicted in Figure 1C ) , we fused the codon-optimized ligand-binding domain of the human progesterone receptor to the Flp recombinase , making recombinase activity dependent on the presence of RU486 ( Figure 3F ) . Figure 3 ( Figure 3—figure supplement 1 ) shows a test of this construct , pJFRC108-20XUAS-IVS-hPR::Flp-p10 , which we call UAS-Flp-Switch . Expression of GFP in larval nervous systems carrying R20B05-GAL4 , pJFRC177-10XUAS-FRT>-dSTOP-FRT>-myr::GFP ( Nern et al . , 2011 ) , and the UAS-Flp-Switch showed very strong expression that was conditional on feeding larvae on RU486-containing food during the third larval instar ( Harris , 2012 ) . Using the Flp-Switch system in conjunction with GAL80 suppression ( Figure 1C ) , we could then obtain clean expression in hemilineage 12B cells in the adult ( Figure 3F ) . By tracking hemilineages through metamorphosis , we were able to define which neuroglian-positive tracts were used by each hemilineage in the adult ( Harris , 2012 ) . With this knowledge , we rescreened the GAL4 collection and in a few instances found a line that cleanly showed expression in an adult hemilineage . We then used a combination of these few lines as well as ones that were generated using a combination of the nSyb-GAL80 and Flp-Switch approaches to obtain coverage for most of the thoracic hemilineages . The genotypes selected for accessing each of the hemilineages in the adult are listed in Supplementary file 1 , which also assesses off-target expression in the lines . Each thoracic hemilineage has a unique projection pattern that allows it to be distinguished from the others . The only exceptions are hemilineages 20A and 22A , which are from neighboring NBs . These two hemilineages cannot be readily distinguished in the larva ( Truman et al . , 2004 ) and we were unable to differentiate them in our analysis . Consequently , they are considered as a single unit . An important diagnostic feature of each hemilineage is the neuroglian-positive tracts in which it runs . A detailed analysis of the larval-to-adult transformation of this tract system is described in Shepherd et al . ( 2015 , in revision ) , so only a superficial treatment will be given here . The immortalized driver lines show the neuropil regions that are targets of a given hemilineage , but for the more complex hemilineages , that overlap either bilaterally or intersegmentally , a detailed analysis of the anatomy needs to be supplemented by a clonal approach that has been used sparingly here and will be dealt with in detail elsewhere ( Shepherd , Sahota , Court , Harris , Truman and Williams , in preparation ) . The thoracic neuropil can be roughly divided into the paired , ventrolateral leg neuropils in each segment and a dorsal region , the tectulum ( Power , 1943 ) , which has lost obvious segmental organization and is most expanded in the T2 region . To understand the behavioral function of each hemilineage , we activated the neurons in decapitated flies and observed their response . Decapitated flies maintain a good stance and show little spontaneous leg or wing movements except for occasional grooming bouts of the front legs or of the hind legs over the wings ( Video 1 ) . Cells in the respective hemilineages were excited using the temperature-sensitive TRPA1 channel expressed in the corresponding neurons and the decapitated flies subjected to a linear heat ramp and videos were acquired through the warming process for 45 to 55 s ( see ‘Materials and methods’ ) . Decapitated flies lacking the TRPA1 effector show no response to the heat ramp , maintaining their posture with occasional bouts of grooming movements . The behavioral responses of the decapitated flies were divided into six behavioral categories . ( 1 ) Changes in posture: this category includes tonic changes in posture or in leg or joint position . ( 2 ) Uncoordinated leg movements: the flies showed phasic leg movements that had no obvious inter- or intrasegmental coordination . The decapitated flies typically stayed in place or , if they moved , their course was erratic . ( 3 ) Walking: this involved translocation of the animal over more than a body length during the test period . The movements could be forwards , sideways or backwards , but the trajectories were typically smooth . ( 4 ) Wing waving: this category includes bouts of low frequency wing movements such as wing extension , wing flicking , or wing scissoring . ( 5 ) Wing buzzing: these movements include high frequency wing movements that appeared as a blur on the 60 fps video frames . The wings could be extended laterally in the flight or singing positions or could remain over the back . During buzz episodes the decapitated fly might or might not become airborne . ( 6 ) Takeoff: these typically involved the decapitated fly launching into the air within one to two frames of the normal video ( span of ∼30–60 ms ) . In the lines in which this response was common , we used high-speed video to confirmed the role of the t2 legs in producing the takeoff jump . 10 . 7554/eLife . 04493 . 007Video 1 . Control , decapitated flies of the genotype R24B02-GAL4 subjected to a 24° to 37°C heat ramp and recorded at 60 frames per second ( fps ) . This video related to Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 007 The lineage 1 cluster in the larva is located in the anterolateral region of the segment and has a 1A bundle that projects in the ventral anterior ( vA ) commissure to the contralateral leg neuropil and a 1B bundle that projects to the leg neuropil in the next anterior segment . During metamorphosis the somata of the 1A and 1B clusters are pulled apart so that the 1B somata become situated at the posterior edge of the next anterior segment ( Figure 4E ) . The 1B interneurons are local interneurons that have arbors in the ventromedial and dorsolateral regions of the leg neuropil ( Figure 4F ) . The adult 1A interneurons , by contrast , are projection neurons that travel through the vA commissure and form one of the most ventral tracts in the VNS ( Figure 4A , B ) . After crossing the midline , the 1A bundle makes a distinctive posterior ‘hook’ ( Figure 4A ) and then bifurcates into dorsal and ventrolateral projections ( Figure 4B , Figure 4—figure supplement 1 ) . The ipsilateral and contralateral arbors project into similar neuropil areas so that the dorsal and ventral arbors from the right and left clusters appear to overlap . A synaptotagmin::GFP fusion protein ( nSyt::GFP ) ( Zhang et al . , 2002; Pfeiffer et al . , 2010; Seelig and Jayaraman , 2013 ) localizes primarily in the ventral arbors ( Figure 4C ) in the intermediate layers of the leg neuropil , showing the primary output regions of these cells . 10 . 7554/eLife . 04493 . 008Figure 4 . The anatomy and behavioral consequences of stimulating hemilineages 1A through 5B . Each hemilineage is depicted as a projected confocal Z-stack of the VNS ( T1 , T2 , T3 and the fused abdominal neuromeres ) ( A , E , G , K , M , P ) and a transverse projection through segment T2 ( B , F , H , L , N , Q; bracketed region in dorsal view ) . yellow arrowheads: hemilineage cell body clusters; white arrowheads: main neurite bundle entering neuropil or crossing midline; magenta asterisk: major off-target expression; green: GFP; magenta: neuroglian . Pictures are video frames of groups of decapitated flies that express TRPA1 in the particular hemilineage and are exposed to a heat ramp to stimulate the neurons . ( A–D ) Hemilineage 1A . ( A , B ) The 1A neurons are located dorsolaterally in each thoracic segment , project across the midline and have ventral ( v ) arbor in the leg neuropil and dorsal ( d ) arbor in the tectulum neuropil . Arrow: characteristic posterior hook of the ventral arbor . ( C ) T2 transverse projection showing that nSynaptotagmin::GFP ( nSyt , magenta ) localizes to the ventral arbor . ( D ) The response of activating the 1A neurons in decapitated flies with a 24–37°C heat ramp over a 50 s period . Images show the position of marked flies at the beginning and end of the ramp and the path each moved during the period . ( E , F ) Hemilineage 1B . The 1B neurons are located in the posterior vertrolateral region of each segment and send arbors in to ventral and dorsal regions of the ipsilateral leg neuropil . ( G–J ) Hemilineage 2A . ( G , H ) The 2A neurons are situated ventromedally in the anterior third of the ganglion; they project dorsally and arborize throughout the ipsilateral tectulum . ( I ) T2 transverse projection showing that nSyt-GFP ( magenta ) localizes in the more lateral parts of the arbor . ( J ) Activation of the 2A neurons by the head ramp results in buzzing of the outstretched wings ( 30°C; arrow ) . ( K , L ) Hemilineage 3A . The 3A interneurons are in a posterior ventrolateral cluster; they enter the leg neuropil near the leg nerve and ramify through most of the ventral half of the leg neuropil . The line had substantial sensory expression ( * ) . ( M–O ) Hemilineage 3B . ( M , N ) The 3B interneuron clusters are in posterior T1 and T2 and project dorsally to ramify through the dorsal part of the tectulum . ( O ) Thermal activation of the 3A primarily evokes flicking and scissoring movements of the wings ( arrows ) . ( P–S ) Hemilineage 5B . ( P , Q ) The 5B clusters are positioned ventrolaterally in the posterior part of each thoracic neuromere . They project dorsally and across the intermediate posterior commissure to arborize in dorsal and medial regions of the neuropil . ( R , S ) Thermal activation of the 5B interneurons cause decapitated flies to splay out their legs ( R , 37°C ) and tethered intact flies to crouch onto a Styrofoam ball they are holding ( S , 32°C ) . Genotypes: for most genotypes see Supplementary file 1; for the remainder: ( C ) LexAop-RFP ( su ( Hw ) attP8 ) /w; R22G11-LexA ( attP40 ) /+; LexAop-nSyt-GFP ( su ( Hw ) attP1 ) /+ . ( I ) UAS-nSyt-GFP ( attP18 ) /w; +; R50G08-GAL4 ( attP2 ) /UAS-HA ( VK00005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 00810 . 7554/eLife . 04493 . 009Figure 4—figure supplement 1 . Dorsal ( A ) and transverse ( B ) view of the adult VNS showing a MARCM clone for the T2 lineage 1 . The 1B siblings have been pulled anteriorly into T1 where they branch through the ipsilateral leg neuropil . The 1A siblings are projection neurons with ventral ( v ) arbor in both the ipsi- and contralateral leg neuropils and bilateral dorsal ( d ) arbors that extend into the dorsal , tectulum neuropil . : arbor from another lineage . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 009 Since our best 1B line showed expression in a dorsal hemilineage in addition to the 1B neurons , we did not use it for behavioral observations . Activation of 1A interneurons via TRPA1 evoked forward locomotion . As the temperature ramped up , the decapitated flies started to walk , with the total distance covered ranging from just over a body length to greater than ten lengths during the trial period ( Figure 4D; Video 2 ) . Locomotion was also occasionally interrupted by bouts of grooming . The behavior displayed by these flies was similar to that described as locomotion for decapitated flies exposed to biogenic amines in Yellman et al . ( 1997 ) . During locomotion the leg movements were erratic and not organized in the tripod gait typical of walking in intact flies ( Strauss and Heisenberg , 1990 ) , but there was clear intersegmental coordination of the limbs . The net movement is always forwards , although the trajectory turns due to uneven step sizes , as observed by Yellman et al . ( 1997 ) . Rarely , a fly would takeoff during the heat ramp . 10 . 7554/eLife . 04493 . 010Video 2 . Behavioral effects of exciting the neurons in hemilineage 1A . This video related to Figure 4 . Decapitated flies expressing TRPA1 under control of R22G11 are subjected to a 24° to 37°C heat ramp and recorded at 60 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 010 The 2A interneurons in the larva represent the surviving hemilineage from lineage 2 . They are situated as paired clusters on either side of the midline at the anterior margin of each thoracic neuromere . Their axons project dorsally and then spread laterally over the ipsilateral dorsal neuropil . This anatomy is conserved in the adult , with the 2A interneurons extending broadly within the ipsilateral part of the tectulum ( Figure 4G–I ) . Arbor from all three segments show a degree of convergence into the T2 region . Activation of the 2A interneurons drove high frequency wing movements ( 54% , n = 25 ) . The decapitated flies typically stood in place but as the temperature increased , they abruptly initiated high frequency flapping with the wings extended laterally in the flight position ( Figure 4J; Video 3 ) . This wing buzzing was usually maintained for the remainder of the heat ramp , but it was occasionally interrupted by a bout of wing grooming . A few of the buzzing flies eventually went airborne . We do not know if this was accomplished by a jump via their T2 legs or if their tarsi simply lost contact with the substrait . 10 . 7554/eLife . 04493 . 011Video 3 . Behavioral effects of exciting the neurons in hemilineage 2A . This video related to Figure 4 . Decapitated flies expressing TRPA1 under control of R50G08 are subjected to a 24° to 37°C heat ramp and recorded at 60 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 011 In the larva , the lineage 3 interneurons are in a ventromedial cluster at the posterior border of the segment . The axons of the A and B daughters project dorsally to the mid-neuropil , where the 3A neurons enter the ipsilateral leg neuropil , while the 3B daughters continue into the ipsilateral dorsal neuropil . In the adult , the cell bodies of hemilineage 3A remain ventral , just posterior to the leg nerve insertion and their arbors extend through intermediate levels of the leg neuropil ( Figure 4K , L ) . The cluster of 3B interneurons is severely reduced or absent in T3 . In T1 and T2 the arbors of the cells converge in the tectulum to form a complex arbor ( Figure 4M , N ) in the very dorsal layer of this neuropil . The 3A line had major contamination from sensory neurons and was not used for behavioral observations . The effects of activating the 3B interneurons were seen using the R23B05-GAL4 line driving TRPA1 ( Figure 4O , Video 4 ) . As the temperature ramped up , the decapitated flies showed a subtle repositioning of the legs , with some repetitive , poorly coordinated leg movements and occasional grooming bouts . They also performed occasional wing flicking and wing scissoring movements but did not show high frequency wing buzzing . 10 . 7554/eLife . 04493 . 012Video 4 . Behavioral effects of exciting the neurons in hemilineage 3B . This video related to Figure 4 . Decapitated flies expressing TRPA1 under control of R23B05 are subjected to a 24° to 37°C heat ramp and recorded at 60 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 012 In the larva , hemilineage 5A undergoes programmed cell death ( Truman et al . , 2010 ) leaving only the 5B interneurons . The hemilineage 5B somata are ventrolateral , and their neurites project through the intermediate posterior ( iP ) commissure and arrest in an ascending tract . In the adult , the cell bodies are pulled towards the midline . Their bundled axons cross the midline in the iP commissure , and form a major ipsilateral arbor just prior to crossing ( Figure 4P , Q ) . The contralateral arbors of the cells make a compact projection that extends up and down the medial region of the neuropil and into the neck connective . When the 5B interneurons were activated , the decapitated flies showed a progressive repositioning of their limbs so that they are closer to the ground by the end of the thermal ramp ( Figure 4R , Video 5 ) . Spontaneous grooming appeared to be suppressed and we observed no walking or wing associated behaviors . Tethered intact flies that were gripping a Styrofoam ball responded to the temperature ramp by lowering their body position , thereby drawing the ball closer to them ( Figure 4S ) . 10 . 7554/eLife . 04493 . 013Video 5 . Behavioral effects of exciting the neurons in hemilineage 5B . This video related to Figure 4 . Decapitated flies expressing TRPA1 under control of R86D02 are subjected to a 24° to 37°C heat ramp and recorded at 60 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 013 In the larva , both lineage 6 hemilineages project across the midline with the 6A bundle forming the dorsal posterior ( dP ) commissure and 6B bundle crossing in the iP commissure . During metamorphosis the cell bodies of the two hemilineages separate into distinct clusters , but stay ventral ( Figure 5B , F ) , as described in Brown and Truman ( 2009 ) . Hemilineage 6A clusters are found in neuromeres S3 through A1 ( Figure 5A ) and form one of the dorsalmost arbors of the tectulum . The 6A clusters show an ipsilateral arbor that is largely confined to the segment of origin and a distal , contralateral arbor that converges from S3 through A1 into the T2 region of the tectulum ( Figure 5A , B ) . nSyt::GFP localizes to the lateral portions of the contralateral arbor ( Figure 5C ) . The neurons of 6B cluster are found in T1 to T3 but typically are missing from segment A1 . The 6B bundle crosses in the iP commissure without an ipsilateral arbor and then branches profusely in the tectulum neuropil and the dorsal-most region of the leg neuropils ( Figure 5E , F ) . As with the 6A neurons , the T1 and T3 clusters tend to converge onto T2 . 10 . 7554/eLife . 04493 . 014Figure 5 . The anatomy and behavioral consequences of stimulating hemilineages 6A through 9A . Organization of panels and general symbols are as in Figure 4 . ( A–D ) Hemilineage 6A . ( A , B ) The 6A neurons are in prominent ventromedial clusters in segments T1 through A1 . They ramify through the lateral regions of the tectulum neuropil with a concentration in the dorsal T2 area . ( C ) T2 transverse projection showing that nSyt-GFP ( magenta ) localizes in the lateral parts of the arbor . ( D ) Video frames of decapitated flies at early ( 1 s ) and intermediate ( 31 s ) times in the heat ramp . Middle frames are multiple exposures taken over 1 s periods showing phasic , repetitive movements of single limbs ( 15–16 s , arrow ) , and jerky movements of the entire fly ( 30–31 s , arrow ) . ( E–G ) Hemilineage 6B . ( E , F ) The 6B neurons are located in the posterior medial region of each segment and project through a posterior commissure to arborize in the tectulum and dorsal regions of the leg neuropils . ( G ) Video frames showing the position of decapitated flies at an intermediate ( 20 s ) and late ( 36 s ) portion of the heat ramp as they are starting to move . The respective dots mark the anterior margin of the thorax at the 2 times . The middle frame shows the position of the fly at the start ( white dot ) and end ( red dot ) of the sequence and at 2 s intervals in between; arrow shows direction of movement . ( H–J ) Hemilineage 7B . ( H , I ) The 7B neurons are in prominent ventrolateral clusters in segments T1 through A1 . They ramify through the lateral regions of the tectulum and send a prominent projection into the T2 leg neuropil ( arrow ) . ( J ) Frames of a high-speed video of a decapitated fly taking off during heating . It shows the expected sequence of wing elevation ( 30 ms ) , jump ( 38 ms ) , and flapping ( 45 ms ) . ( K–M ) Hemilineage 8A . ( K , L ) The 8A cluster is situated in the anterolateral region of each segment and projects into the lateral leg neuropil . ( M ) . Early and late frames during the heat ramp showing only minor positional changes through the period . ( N , O ) Hemilineage 8B . ( N , O ) The 8B clusters are in the anterolateral region of each segment and project through an anterior commissure to spread widely through the thoracic neuromeres . A prominent subset of the t3 cells form a bowtie-shaped structure ( arrow ) that receives input from haltere afferents . Transverse section ( O ) is of this input area . ( P–R ) Hemilineage 9A . ( P , Q ) The 9A clusters assume an anterolateral position and project into the region of the ventral leg neuropil that receives input from proprioceptors . ( R ) . Video frames from early and late stages of the heat ramp . The decapitated flies responded by splaying out their legs . Genotypes: for most genotypes see Supplementary file 1; for the remainder: ( C ) nSyb-GAL80 ( su ( Hw ) attP8 ) /LexAop-RFP ( su ( Hw ) attP8 ) ; UAS-flp ( attP40 ) /act>STOP>LexA ( attP40 ) ; R35A03-GAL4 ( attP2 ) /LexAop-nSyt-GFP ( su ( Hw ) attP1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 01410 . 7554/eLife . 04493 . 015Figure 5—figure supplement 1 . Dorsal ( A ) and transverse ( B ) view of the adult VNS showing a MARCM clone for the T3 lineage 7 . From the cell cluster ( arrowhead ) the cells project dorsally to elaborate a dense arbor ( i ) in the ipsilateral tectulum neuropil and then cross via an anterior commissure and extend an anterior output arbor ( c ) . They also send a send a second projection that crosses the midline and extends ventrally into leg neuropil ( cv ) . Bracket shows level of transverse projection . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 015 Activation of the 6A interneurons via the heat ramp started with poorly coordinated movements of the legs that often pitched the fly forward onto its anterior thorax ( Figure 6D , Video 6 ) . The movements were not organized into any particular direction and became very erratic towards the end ( Figure 6D , 30-31s ) . The decapitated flies occasionally showed wing flicking and high frequency buzzing of the wings but the wings were usually only partially spread rather than in the extended flight position . 10 . 7554/eLife . 04493 . 016Figure 6 . The anatomy and behavioral consequences of stimulating hemilineages 10B through 14A . Organization of panels and general symbols are as in Figure 4 . ( A–C ) Hemilineage 10B . ( A , B ) The 10B clusters are ventromedially located in the anterior part of the segment . They project across an anterior commissure and produce a dorsal ( d ) arbor that runs longitudinally , and a ventral ( v ) arbor that extends into medial leg neuropil . ( C ) Video frames showing the position of decapitated flies at intermediate ( 20 s ) and late ( 36 s ) portion of the heat ramp . The respective dots mark the anterior margin of the thorax at the 2 times . The middle frame shows the position of the fly at the start ( white dot ) and end ( red dot ) of the sequence and at 2 s intervals in between; arrow shows direction of movement which was generally backward . ( D–F ) Hemilineages 11A and B . ( D , E ) These hemilineages are laterally located only in T1 ( 11A ) and T2 ( 11A and 11B ) . They ramify primarily in the T2 tectulum neuropil and have projections into the leg neuropils of T1 and T3 ( arrows ) . ( F ) Frames of a high-speed video of a decapitated fly taking-off during heating . Flapping begins prior to the jump . ( G–J ) Hemilineage 12A . ( G , H ) The ventrolateral 12A clusters are on the posterior border of segments T1 and T2 . They project dorsally and arborize through most of the dorsal tectulum . ( I ) T2 transverse projection showing that nSyt-GFP ( magenta ) localizes to medial regions of the 12A projection . ( J ) Video frames showing progression of behaviors of decapitated flies during the heat ramp . 27°C: flies quiet; 30°C , some walking and showing lateral wing waving ( arrow ) ; 35°C: flies showing wing buzzing ( blur ) although wings ( arrows ) usually not extended in flight position . ( K–N ) Hemilineage 12B . ( K , L ) The 12B clusters are ventrally located at the posterior border of the neuromere . They project across a posterior commissure and arborize widely through the contralateral leg neuropil . ( M , N ) Video frames of dorsal and lateral views of decapitated flies subjected to a heat ramp . At elevated temperatures the flies often showed tonic extensions of the T2 and T3 legs . ( O , P ) Hemilineage 13A . ( O , P ) The somata of the 13A neurons are spread over the anterior ventrolateral region of the neuromere . Their arbors extend through most of the ventral half of the leg neuropil . ( Q–T ) Hemilineage 13B . ( Q , R ) The 13B clusters are pulled to the anterior midline and their axons cross the midline in a very ventral anterior commissure and the cells branch through the ventrolateral leg neuropil . ( S , T ) Video frames of dorsal and lateral views of decapitated flies early and late in the heat ramp . At high temperatures the legs are extended laterally and often are elevated from the substrate ( arrow ) . ( U , V ) Hemilineage 14A . ( U , V ) The 14A clusters are also pulled to the midline and their axons cross in a ventral commissure . They project through most of the ventral and lateral leg neuropil . This line also expressed in occasional other lineages ( ) . Genotypes: for most genotypes see Supplementary file 1; for the remainder: ( I ) nSyb-GAL80 ( su ( Hw ) attP8 ) /act>STOP>LexA ( attP18 ) ; LexAop-RFP ( attP40 ) /UAS-flp ( attP40 ) ; LexAop-nSyt-GFP ( su ( Hw ) attP1 ) /R24B02-GAL4 ( attP2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 01610 . 7554/eLife . 04493 . 030Figure 6—figure supplement 1 . tsh-GAL80 eliminates expression specifically in the VNC . ( A ) The lineage 11 genotype drives GFP expression in the VNC in lineage 11 ( yellow arrowheads , arrows ) and occasionally in descending axons ( * ) . ( B ) The same genotype as in ( A ) , but with the addition of tsh-GAL80 . Expression is eliminated in lineage 11 , but remains in the descending neurons . Genotypes A: nSyb-GAL80 ( su ( Hw ) attP8 ) /w; UAS-flp ( attP40 ) /+; R26B05-GAL4 ( attP2 ) /nSyb-LexA ( attP2 ) , LexAop>STOP>GFP ( VK00005 ) . ( B ) nSyb-GAL80 ( su ( Hw ) attP8 ) /w; UAS-flp ( attP40 ) /tsh-GAL80; R26B05-GAL4 ( attP2 ) /nSyb-LexA ( attP2 ) , LexAop>STOP>GFP ( VK00005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 03010 . 7554/eLife . 04493 . 017Video 6 . Behavioral effects of exciting the neurons in hemilineage 6A . This video related to Figure 5 . Decapitated flies expressing TRPA1 under control of R35A03 are subjected to a 24° to 37°C heat ramp and recorded at 60 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 017 Activation of the 6B interneurons in the decapitated flies elicited a mixture of leg-related movements ( Figure 5G; Video 7 ) . Some showed a curving , forward movement while others pivoted in place . Most flies did not move far enough ( >one body length ) to be classified as walking . There were no wing movements except for those associated with occasional grooming bouts . 10 . 7554/eLife . 04493 . 018Video 7 . Behavioral effects of exciting the neurons in hemilineage 6B . This video related to Figure 5 . Decapitated flies expressing TRPA1 under control of R46C11 are subjected to a 24° to 37°C heat ramp and recorded at 60 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 018 Lineage 7 has an anterior cell cluster located in ventrolateral region of neuromeres T1 through A1 and consists of the surviving 7B hemilineage ( Truman et al . , 2010 ) . The 7B neurite bundle crosses the midline in the iA commissure and then turns anteriorly . The basic morphology is much the same in the adult ( Figure 5H ) , as described in Brown and Truman ( 2009 ) . The neurite projects dorsally , forming a bushy ipsilateral ( proximal ) arbor in the tectulum , then crosses the midline to form an ascending tract that extends through the neck connective ( Figure 5H ) . The arbor in the T2 hemineuromere includes prominent projections into the dorsolateral region of the T2 leg neuropil ( Figure 5I ) . The T1 and T3 versions also send a branch into their respective leg neuropils ( Figure 5 , Figure 5—figure supplement 1 ) , but these branches are not as robust as the T2 version . The gradual activation of the 7B interneurons by the heat ramp resulted in little walking behavior . The decapitated flies continued spontaneous grooming with occasional wing flicking behavior until they abruptly launch themselves into the air ( Video 8 ) . If they landed back on the hotplate , they then repeated the behavior . High speed video analysis ( Figure 5J , Video 9 ) shows that their behavior corresponds to the normal takeoff sequence described by Card and Dickinson ( 2008 ) : first the wings are raised , the mesothoracic legs then extend in a jump , and finally the wings depress and the fly begins to flap . 10 . 7554/eLife . 04493 . 019Video 8 . Behavioral effects of exciting the neurons in hemilineage 7B . This video related to Figure 5 . Decapitated flies expressing TRPA1 under control of R65A12 are subjected to a 24° to 32°C heat ramp and recorded at 60 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 01910 . 7554/eLife . 04493 . 020Video 9 . High-speed video showing the behavioral effect of exciting the neurons in hemilineage 7B . This video related to Figure 5 . Decapitated flies expressing TRPA1 under control of R65A12 are subjected to a 24° to 32°C heat ramp and recorded at 3000 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 020 In the larva , the lineage 8 neurons are in a ventrolateral cell cluster in the anterior part of each thoracic segment . The 8A neurons project to the dorsal region of the ipsilateral leg neuropil while the 8B neurons extend across the iA commissure , just anterior to the 7B bundle . In the adult , the 8A somata are found in approximately the same location; their bundled neurites make a pronounced lateral bend after they enter the neuropil and form prominent arbors that ramify through the lateral portion of the ipsilateral leg neuropil ( Figure 5K , L ) . The 8B cluster in the adult contains multi-part , midline-crossing , intersegmental arbors ( Figure 5N ) . Proximal arbors include an intrasegmental lateral arbor and an ascending medial arbor in the dorsal third of the ipsilateral neuropil . The hemilineage also makes symmetric ascending projections in the upper part of the dorsal neuropil . The T2 and T3 hemilineages send projections to T1 , while the T1 hemilineage sends projections up the neck connective . A prominent component of the T3 lineage are the contralateral haltere interneurons ( cHINs; Strausfeld and Seyan , 1985 ) , with their distinctive ‘bowtie’ arbor ( Figure 5N , O ) . These neurons receive sensory inputs from the haltere nerve , on their lower , intrasegmental ipsilateral arbor , and then make an ascending contralateral projection . The homologous lower arbor is found in T2 where the wing nerve inserts , and may receive wing sensory inputs . A similar , but reduced , arbor is also found in T1 . Our line for the 8B interneurons had some expression from other thoracic hemilineages and was not used for behavioral studies . Activation of the 8A interneurons using line R69H11-GAL4 had minimal effects on the behavior of the decapitated flies ( Figure 5M , Video 10 ) . As the temperature ramped up , we saw little effect on spontaneous behavior . The flies continued to show bouts of grooming; they also showed fidgety repositioning of their legs , but no walking or wing movements . 10 . 7554/eLife . 04493 . 021Video 10 . Behavioral effects of exciting the neurons in hemilineage 8A . This video related to Figure 5 . Decapitated flies expressing TRPA1 under control of R69H11 are subjected to a 24° to 32°C heat ramp and recorded at 60 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 021 In the larva , the lineage 9 cluster is the dorsal-most cluster in the anterior half of the neuromere , and the surviving 9A neurons project to the medial region of the ipsilateral leg neuropil ( Truman et al . , 2010 ) . In the adult , the 9A somata are found in the middle third of the VNS . They are local leg interneurons that arborize in the ventral leg neuropil ( Figure 5P , Q ) , as described in Brown and Truman ( 2009 ) . These neurons overlap with afferents from leg chordotonal organs ( Harris , 2012 ) . Activation of the 9A interneurons resulted in a subtle change in posture in the decapitated flies as their legs became gradually more splayed out during the course of the heat ramp ( Figure 5R , Video 11 ) . There was very little locomotion and bouts of spontaneous grooming were occasionally seen . 10 . 7554/eLife . 04493 . 022Video 11 . Behavioral effects of exciting the neurons in hemilineage 9A . This video related to Figure 5 . Decapitated flies expressing TRPA1 under control of R52E12 are subjected to a 24° to 37°C heat ramp and recorded at 60 frames per second ( fps ) . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 022 In the larva , only hemilineage 10B survives ( Truman et al . , 2010 ) . The 10B cell body clusters are just lateral to those of lineage 2 and their neurite bundle projects across the iA commissure . The hemilineage 10B cell bodies remain anteromedial in the adult . Their primary neurites cross the midline in an iA bundle and form a finger-like ventral arbor , which starts medial and projects ventrally almost to the leg nerve insertion , and an intersegmental dorsal arbor running just lateral to the midline ( Figure 6B ) . The dorsal arbor continues up the neck connective ( Figure 6A ) . The behavioral responses of decapitated flies to the activation of the 10B interneurons were dominated by leg movements ( Figure 6C; Video 12 ) . These movements were somewhat erratic , often causing the flies to make pivoting movements around a leg . When there was net movement , it was usually backwards . Wing flicking and wing buzzing occasionally occurred during walking . Wings were typically held back , rather than in the flight position , during these movements . Rarely a fly would go airborne during one of these episodes . 10 . 7554/eLife . 04493 . 023Video 12 . Behavioral effects of exciting the neurons in hemilineage 10B . This video related to Figure 6 . Decapitated flies expressing TRPA1 under control of R13B08 are subjected to a 24° to 32°C heat ramp and recorded at 30 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 023 We were unable to construct lines to target hemilineages 11A and 11B independently , but R26B05-GAL4 targets the whole lineage . Lineage 11A is found in T1 and T2 , whereas lineage 11B is found only in T2 . Based on this , we attributed the parts of the T2 arbor to 11A vs 11B , based on 11A projections in T1 . In the adult , the lineage 11 cell bodies are in the dorsoposterior margin of T1 and T2 . The lineage 11 neurons show a complex projection that is predominantly in T2 and largely confined to the tectulum neuropil ( Figure 6D , E ) . However , a branch of the 11A arbors from the T1 cells project into the T1 leg neuropil while the corresponding branch from the T2 cells extends into the T3 leg neuropil . There is no equivalent branch for the T2 neuropil . Similar to the 7B interneurons , the activation of the lineage 11 interneurons using R26B05-GAL4 evoked takeoff behavior . High-speed video analysis showed that the behavioral sequencing differed from normal takeoff in that the decapitated flies commenced wing flapping prior to the jump ( Figure 6F , Video 13 ) . The R26B05 line used to activate the lineage 11 cells occasionally showed expression in a bundle of descending interneurons ( Figure 6—figure supplement 1A ) and we were concerned that their severed axons might be responsible for driving the takeoff behavior in the decapitated flies . Consequently , we used teashirt-GAL80 ( tsh-GAL80; Clyne and Miesenböck , 2008 ) to suppress thoracic expression , but leaving expression in the descending interneurons ( Figure 6—figure supplement 1B ) . These decapitated flies showed a severe reduction in the number of takeoffs when subjected to the heat ramp ( 9% of tsh-GAL80 flies [N = 22] vs 100% [N = 13] for R26B05-GAL4 flies lacking the tsh-GAL80 ) . 10 . 7554/eLife . 04493 . 024Video 13 . High-speed video showing the behavioral effect of exciting the neurons in hemilineages 11A and B . This video related to Figure 6 . Decapitated flies expressing TRPA1 under control of R26B05 are subjected to a 24° to 37°C heat ramp and recorded at 6400 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 024 In the larva , the paired lineage 12 cell body clusters are in the posterior ventromedial region of the thoracic neuromeres . Their neurite bundles project dorsally and then split , with the 12B neurons projecting contralaterally through the iP commissure while the 12A neurons continue into dorsal neuropil . The 12A neurons die in T3 . In the adult , the cell body clusters for the 12A and 12B hemilineages remain ventral but move slightly more lateral . Hemilineage 12A arborizes throughout the dorsal tectulum neuropil . The arbors from the T1 and T2 clusters converge on T2 and extensively overlap ( Figure 6G , H ) . nSyT::GFP localizes to the medial portions of the dorsal arbors , plus a small region in the lateral part of the lower arbors ( Figure 6I ) . The mature 12B neurons have no ipsilateral arbor and cross the midline in the pI commissure to arborize in the medial region of the contralateral leg neuropil ( Figure 6K , L ) . Activation of the 12A interneurons produced a sequence of behaviors as the temperature ramped up ( Figure 6J; Video 14 ) . The decapitated flies started with walking behavior that was later joined by wing flicking and by the lateral extension and vibration of a single wing ( Figure 6J , 30°C ) , resembling the courtship singing of the male . Eventually the frequency of wing vibrations became extreme but the wings were usually not extended in the flight position; often one vibrating wing was partially or fully extended to the side while the other was vibrating while over the back ( Figure 6J , 35°C ) . These flies would occasionally go careening into the air , but there did not appear to be an organized jump preceding going airborne . 10 . 7554/eLife . 04493 . 025Video 14 . Behavioral effects of exciting the neurons in hemilineage 12A . This video related to Figure 6 . Decapitated flies expressing TRPA1 under control of R24B02 are subjected to a 24° to 37°C heat ramp and recorded at 60 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 025 In contrast to the 12A cells , the activation of the 12B interneurons had little effect on some decapitated flies but evoked a tonic postural change in others . The T2 and T3 legs underwent an extreme extension and the legs froze in this position ( Figure 6M , N , Video 15 ) . These flies either stayed rigidly upright or toppled over onto their side . The movements of the T1 legs were variable . 10 . 7554/eLife . 04493 . 026Video 15 . Behavioral effects of exciting the neurons in hemilineage 12B . This video related to Figure 6 . Decapitated flies expressing TRPA1 under control of R15D11 are subjected to a 24° to 37°C heat ramp and recorded at 60 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 026 In the larva , both hemilineages of lineage 13 contribute to the immature leg neuropils; the 13A neurons project to the ipsilateral leg and 13B neurons project across the vA commissure to the contralateral leg neuropil . In the adult , the 13A neurons insert into the neuropil near the entry of the leg nerve . Their arbors extend along the edges of the leg neuropil and ramify through its ventral half ( Figure 6O , P ) . The cell bodies of 13B cluster are pulled medially during metamorphosis and sometimes are pulled across the midline ( Figure 6Q ) . Their axons cross the midline in the vA commissure and constitute the ventral-most bundle ( Figure 6R ) . The 13B neurons branch extensively through the ventral region of the leg neuropil . We did not have a clean enough line to assess the effects of activating the 13A interneurons . Activation of the 13B cells in decapitated flies evoked a tonic postural change . As the temperature increased they extended their legs progressively more to the side so that by the end of the heat ramp they were often resting on their coxae , with their tarsi elevated into the air ( Figure 6S , T; Video 16 ) . 10 . 7554/eLife . 04493 . 027Video 16 . Behavioral effects of exciting the neurons in hemilineage 13B . This video related to Figure 6 . Decapitated flies expressing TRPA1 under control of R41G09 are subjected to a 24° to 32°C heat ramp and recorded at 60 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 027 In lineage 14 , most of the 14B neurons die , so the cluster is comprised almost exclusively of hemilineage 14A neurons ( Truman et al . , 2010 ) . During metamorphosis the cell clusters of the 14A neurons are pulled to the anterior midline . Their axons form an extreme ventral bundle that is just dorsal and anterior to the 13B neurons . The 14A neurons have no ipsilateral arbor and arborize throughout the ventral half of the leg neuropil and also laterally in the region where the motoneurons arborize ( Figure 6U , V ) . Our best line for 14A interneurons had major contamination with the 2A hemilineage , so we do not have behavioral data for this hemilineage . This is one of the two major lineages that make leg motoneurons ( Figure 7A , B ) . The adult composition of this lineage has been extensively characterized ( Baek and Mann , 2009; Brierley et al . , 2012 ) and they primarily supply muscles of the distal leg segments . 10 . 7554/eLife . 04493 . 028Figure 7 . The anatomy and behavioral consequences of stimulating hemilineages 15B through 24B . Organization of panels and general symbols are as in Figure 4 . ( A–C ) Hemilineage 15B . ( A , B ) The larger of the two motor lineages . Cell bodies are located in the anterolateral region of the segment and the neurons innervate primarily muscles to distal leg segments . ( C–E ) Hemilineage 18B . ( C , D ) The 18B clusters are located in the dorsoanterior regions of T2 and T3 . Their axons cross in an anterior commissure and have discrete contralateral ( c ) , intermediate ( i ) and ventral ( v ) projections . The ventral projections extend into the leg neuropil . ( E ) Frames of a high-speed video of a decapitated fly taking-off during heating . Flapping often began prior to wing raising causing the wings to bend ( 8ms , 25ms ) . ( F–I ) Hemilineage 19A . ( F , G ) The 19A clusters are posterior dorsolateral in the segment and project arbor into the ventral leg neuropil and also to a convergence point at the midline just below the tectulum . This convergence point shows nSyt-GFP ( magenta ) localization ( H , arrow ) . ( I ) Video frames from early and midway through the heat ramp . With increased temperature the flies extend their T2 legs and begin to wave them . The frames for a one second period ( 18–19 s ) are superimposed to illustrate movements of the T2 legs ( arrow ) and stability of the others . ( J , K ) Hemilineage 19B . ( K , L ) The major representation of the 19A cells are in the T2 cluster located posterior dorsolateral region of the segment , with a smaller cluster in T3 . The axons cross in a posterior commissure and show medial ( m ) and lateral ( l ) anterior projections that are confined to the tectulum . The line had massive contamination from haltere afferents ( * ) . ( L–N ) Hemilineages 20A , 22A . ( L , M ) The cell bodies for the 20A , 22A clusters are in the posterior ventrolateral region of the segment and the neurons innervate the middle third of the leg neuropil . ( N ) Video frames of decapitated flies showing that they splay out their legs as the temperature rises . ( O–Q ) Hemilineage 21A . ( O , P ) The 21A clusters are in the posterior ventrolateral region of the segment . They project dorsomedially and then arborize over most of the dorsal two-thirds of the leg neuropil . ( Q ) Video frames of decapitated flies during the heat ramp . Eventually the flies become immobile with their legs frozen at unusual angles ( 21 s ) . The frames for an intermediate , one second period ( 10–11 s ) are superimposed to show the transient hyperkinetic leg movements . ( R–T ) Hemilineage 23B . ( R , S ) The 23B clusters are in the posterior dorsolateral region of the segment . The neurons produce an extensive ipsilateral arbor and then cross the posterior commissure for their output arbor . ( T ) Stimulation of the 23B neurons produce uncoordinated leg movements and the flies basically stay in place ( 36°C ) . ( U–W ) Hemilineage 24 . ( U , V ) The smaller of the two motor lineages . Cell bodies are located in the anterodorsolateral region of the segment and the neurons innervate primarily proximal leg muscles . ( W ) Video frames of decapitated flies subjected to a heat ramp . At elevated temperatures the flies typically showed repetitive leg movements . Genotypes . For most genotypes see Supplementary file 1; for the remainder: ( I ) UAS-nSyt-GFP ( attp18 ) /w; UAS-RFP ( attP40 ) /+: R32E04_GAL4 ( attP2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 02810 . 7554/eLife . 04493 . 029Figure 7—figure supplement 1 . Dorsal ( A ) and transverse ( B ) view of the T2 region of the adult VNS showing a lineage 19 MARCM clone . The 19A siblings project ventrally into the ipsilateral leg neuropil while the 19B siblings project across the posterior commissure ( white arrowhead ) . ( C , D ) Z-projections of ventral ( C ) and dorsal ( D ) regions of the clone that captures most of the arbors of the 19A and 19B hemilineages , respectively . The 19A cells have primarily arbor in their leg neuropil but they also have a prominent projection to the midline . The 19B cells have strong lateral anterior projections ( lat ) that are both ipsilateral and contralateral as well as weaker , symmetrical medial projections ( med ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 029 Our best 15B lines had contamination from some of the interneuron hemilineages in the thorax . Consequently , we do not have behavioral observations for this hemilineage . Lineage 18 is missing from T1 and only the 18B hemilineage persists in T2 and T3 ( Truman et al . , 2010 ) . In the larva , the 18B clusters are situated at the anteriodorsal margin of the segment and they send their axons across the iA commissure and into a longitudinal tract . The adult projection pattern is complex as the cells project through much of the tectulum neuropil ( Figure 7C , D ) . The T2 neurons show a concentration of arbor in the dorsal-most regions of the tectulum and also project a ventral arbor into the dorsolateral regions of the leg neuropils , a region occupied by the dendrites of the leg motoneurons . The T3 arbor is similar to that of T2 , but reduced , especially with regard to its projection to the leg neuropil . In response to the activation of the 18B neurons by the heat ramp , the decapitated flies typically initiated walking and that was sometimes accompanied by jerky wing movements with the wings remaining partially folded . After a latent period ( varying from milliseconds to seconds ) , the fly would jump and the wing movements would transition into flapping , but not consistently in that order ( Figure 7E; Video 17 ) . 10 . 7554/eLife . 04493 . 031Video 17 . Behavioral effects of exciting the neurons in hemilineage 18B . This video related to Figure 7 . Decapitated flies expressing TRPA1 under control of R27A09 are subjected to a 24° to 37°C heat ramp and recorded at 60 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 031 In the larva , the lineage 19 cluster is situated dorsolaterally at the posterior border of each thoracic neuromere . The 19A neurons descend into the ipsilateral leg neuropil , while the 19B neurons project across the iP commissure and turn anteriorly . The adult morphology is similar . The 19A and 19B cell clusters remain in the posteriodorsal region of each segment . The 19A neurons project into the ipsilateral leg neuropil , where they form a major projection extending through lateral and ventral leg neuropil and a medial projection that extends to the midline . The medial projections in T1 and T2 converge on the midline in T2 ( Figure 7F , G ) . nSyt::GFP localizes to the most medial ( distal ) part of the medial projection ( Figure 7H ) . The hemilineage 19B cluster in T1 is greatly reduced in the larva , and after metamorphosis the 19B cluster in T3 is also very small , presumably through cell death during metamorphosis . The adult T2 neurons make a robust projection across the iP commissure and arborize dorsally in the tectulum neuropil . The 19B neurons have both medial and lateral arbors that project anteriorly ( Figure 7J , K; Figure 7—figure supplement 1 ) . Our best 19B line had massive contamination from the haltere afferents and so we have no behavioral data for this lineage . During activation of the 19A interneurons the decapitated flies stayed in place and showed no movements of their T1 or T3 legs , but their T2 legs show incessant waving movements that continue for the duration of the thermal activation ( Figure 7I; Video 18 ) . 10 . 7554/eLife . 04493 . 032Video 18 . Behavioral effects of exciting the neurons in hemilineage 19A . This video related to Figure 7 . Decapitated flies expressing TRPA1 under control of R32E04 are subjected to a 24° to 32°C heat ramp and recorded at 60 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 032 In the larva , 20A and 22A are adjoining lineages in the posterior half of the neuromere . Their first few B progeny become motoneurons but the remainder die ( Truman et al . 2010 ) . The A progeny of both NBs send very short projections to the ventral leg neuropil , and it is difficult to tell them apart . We were unable to recover lines that distinguished them , and have treated them as a single entity in the adult . The cell body clusters for 20A/22A are found at the posteriolateral border of each thoracic segment ( Figure 7L ) . The clusters project anteriorly into the mid region of the leg neuropil and then make a robust lateral branch and a thinner medial branch that extends into dorsal leg neuropil ( Figure 7M ) . Activation of the 20A/22A interneurons evoked a simple change in posture in the decapitated flies . As the temperature increased , the flies extended their legs until they had assumed a splayed out posture ( Figure 7N; Video 19 ) . Occasional grooming bouts occurred during and after the change in posture . 10 . 7554/eLife . 04493 . 033Video 19 . Behavioral effects of exciting the neurons in hemilineages 20A and 22A . This video related to Figure 7 . Decapitated flies expressing TRPA1 under control of R24G06 are subjected to a 24° to 32°C heat ramp and recorded at 60 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 033 In the larva , lineage 21 is situated next to 20 and 22 but the 21A neurons project into the medial region of the immature leg neuropil , rather than staying lateral . In the adult , the cell body cluster of the 21A neurons remains close to the 20A/22A cells . They project dorsally into the middle of the ipsilateral leg neuropil and send branches through most of the leg neuropil ( Figure 7O , P ) . Activation of the 21A interneurons using the R51H05-GAL4 line induced uncoordinated leg movements that lacked both intralimb and interlimb coordination . The leg movements were incessant during the stimulation but , typically , the flies did not move from their place ( Figure 7Q; Video 20 ) . 10 . 7554/eLife . 04493 . 034Video 20 . Behavioral effects of exciting the neurons in hemilineage 21A . This video related to Figure 7 . Decapitated flies expressing TRPA1 under control of R51H05 are subjected to a 24° to 32°C heat ramp and recorded at 30 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 034 In the larva , only the hemilineage 23B siblings survive ( Truman et al . , 2010 ) . The hemilineage 23B cell bodies are located ventrolateral in the larva but are displaced dorsally during metamorphosis . The adult 23B neurons project ventrally in parallel with the axons of the 19B neurons and cross the iP commissure . The 23B cluster sends off a robust arbor just prior to crossing the midline ( Figure 7R ) . This ipsilateral arbor extends through the ventralmost portions of the leg neuropil , where it appears to overlap with multiple classes of leg sensory neurons ( Figure 7S ) . The contralateral arbor extends to dorsal leg neuropil in adjacent segments . Activation of the 23B interneurons using line R77C10-GAL4 caused a progressive higher frequency of intersegmental limb movements but the intralimb coordination of limb joints appeared to be quite poor ( Figure 7T; Video 21 ) . 10 . 7554/eLife . 04493 . 035Video 21 . Behavioral effects of exciting the neurons in hemilineage 23B . This video related to Figure 7 . Decapitated flies expressing TRPA1 under control of R77C10 are subjected to a 24° to 32°C heat ramp and recorded at 60 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 035 Hemilineage 24A ( Brown and Truman , 2009; lineage B of Baek and Mann , 2009 ) is also a motor lineage in which all of the B daughters are leg motoneurons . It is a smaller motor lineage than hemilineage 15B and its cells innervate more proximal leg segments ( Baek and Mann , 2009 ) . The adult morphology of the 24B motoneurons have been described in detail ( Baek and Mann , 2009; Brierley et al . , 2012 ) , and their dendrites occupy the intermediate region of the leg neuropil ( Figure 7U , V ) . The 24B motoneurons were activated using the R15A03-GAL4 line . Stimulation of the 24B motoneurons produced repetitive movements of the legs although these typically resulted in no net movement of the fly ( Figure 7W; Video 22 ) . By the end of the ramp the legs were often held at awkward angles . Rare wing movements occurred but these were usually associated with grooming attempts . 10 . 7554/eLife . 04493 . 036Video 22 . Behavioral effects of exciting the motorneurons in hemilineage 24B . This video related to Figure 7 . Decapitated flies expressing TRPA1 under control of R22G11 are subjected to a 24° to 32°C heat ramp and recorded at 30 fps . Full genotype in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 036 The toolset described here can be used to dissect broad gene expression patterns in the VNS into subsets of cells that come from a common NB , and/or to assign the NB of origin to a neural class of interest . As proof of principle , we demonstrated that a previously identified subset of the fruitless ( fru ) -expressing neurons involved in courtship song are all produced in hemilineage 12A and cleanly isolated this fru+ cluster . The fru-expressing neurons are necessary and sufficient to drive most aspects of male courtship behavior , and their anatomy , genetics , and function have been dissected in detail ( e . g . , Manoli and Baker , 2004; Yu et al . , 2010b; Meissner et al . , 2011 ) . The fru expression pattern can be divided into ∼100 groups of cells , including ∼40 in the VNS ( Yu et al . , 2010 ) . von Philipsborn et al . ( 2011 ) subsequently identified lines targeting those groups using a screen of ∼1000 GAL4 lines intersected with fru-flippase . The largest cell cluster in the VNS , called vPR6 , projects to a structure called the thoracic triangle and appears to be involved in shaping the courtship song ( von Philipsborn et al . , 2011 ) . The latter GAL4 screen recovered 5 lines that each captured 2–5 of these cells in the male and none in the female . No single line hit all of the cells in the vPR6 cluster since Yu et al . ( 2010b ) estimated that it should contain approximately 6–10 cells in the male and 4–6 cells in the female . The location and projection path of the vPR6 cells suggested that they were part of hemilineage 12A . We used the hemilineage 12A combination ( including the 12A driver R24B02-GAL4 , nSyb-GAL80 , UAS-flippase , and pJFRC40-13XLexAop2-FRT>-STOP-FRT>-myr::GFP [Pfeiffer et al . , 2010; Nern et al . , 2011] ) to restrict 13XLexAop2>myr::GFP expression in the VNS to the neurons of hemilineage 12A , then used fruP1 . LexA ( Mellert et al . , 2010 ) to drive GFP expression in the subset of 12A neurons that were part of the fru pattern . This technique reliably captured a group of approximately 12 cells in males and 3–5 cells in females . The collected arbor shape of the fru+ hemilineage 12A neurons in both sexes matched the digitally masked arbor assigned to the vPR6 cluster ( Figure 8 ) ( Yu et al . , 2010b ) . Thus , it is possible to identify the developmental origin of secondary neurons of interest and genetically isolate those neurons using the toolkit presented here . 10 . 7554/eLife . 04493 . 037Figure 8 . Generation of a vPR6-specific line by intersecting hemilineage 12A and fru-LexA . ( A , B ) Parental expression patterns . ( A ) The fru-LexA expression pattern . The female thoracic pattern is a subset of the male pattern , and many of the neurons have dimorphic arbors . ( B ) Hemilineage 12A . Genotype: nSyb-GAL80 ( su ( Hw ) attP8 ) /+; UAS-flp ( attP40 ) /+; R24B02-GAL4/nSyb-LexA , LexAop>STOP>GFP . ( C ) The intersection of A and B , isolating the vPR6 neurons . Genotype: nSyb-GAL80 ( su ( Hw ) attP8 ) ; UASflp ( attP40 ) /LexAop>STOP>GFP ( attP40 ) ; R24B02-GAL4 ( attP2 ) /fru-LexA . Insets: the arbors and numbers of cells match digital representations of the complete vPR6 pattern in males ( left ) and females ( right ) , adapted from Yu et al . 2010a . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 037 Our study focused on the anatomy and function of the hemilineage clusters of the thoracic CNS . We developed new genetic tools that allowed us to target most of the 31 interneuron clusters so that we could examine their form and function after metamorphosis . We find that the initial projections that these hemilineage groups make in the larva ( Truman et al . , 2004 ) prefigure their target regions in the adult CNS . Two motor hemilineages ( 15B and 24B ) and fifteen interneuron hemilineages form the immature ‘leg neuropil’ of the larva ( Truman et al . , 2004 ) and all of these make neurons exclusively for the adult leg neuropil . The only mature leg hemilineage that resides outside of the larval leg neuropil is hemilineage 23B whose mature neurons have dendrites in the most ventral regions of the leg neuropil where they overlap the leg afferents . They may be excluded from the immature leg neuropil because their inputs do not grow into the CNS until metamorphosis is well underway . The neurons in many of the leg hemilineages constitute a relatively homogeneous group with arbors in restricted regions of the leg neuropil . Presumably each of these is dedicated to a particular step in sensory-motor processing for the leg . On the sensory side , the dendrites of the above-mentioned 23B neurons and the neurons of the 9A group overlap the terminals of exteroceptors and proprioceptors , respectively ( Harris , 2012 ) and are likely involved with dealing with primary sensory input . More complex computations are likely carried out by the 14A interneurons . Their NB is the persisting NB4-1 NB from the embryo ( Birkholz et al . , 2015 ) , and the homologous NB4-1 in grasshoppers produces the cluster of midline spiking interneurons ( Shepherd and Laurent , 1992 ) that conveys information from the somatosensory map to the myotopic motor map ( Burrows and Newland , 1993; Burrows , 1996 ) . The neurons generated by NB 4-1 in flies are anatomically similar to the grasshopper neurons and we expect that the fly neurons are likewise involved in maintaining spatial information during the transformation from the sensory ( Murphey et al . , 1989 ) to motor maps ( Brierley et al . , 2012 ) in the leg . Outside of these spatially restricted leg hemilineages , other leg hemilineage clusters have a rather diffuse projection through the leg neuropil , as in the case of 3A and 4B , or they have arbors that also extend beyond their respective leg neuropil , such seen for the 1A and 19A neurons . The tectulum neuropil is built from the dorsal-projecting lineages in the larva . They arise as segmentally discrete units but as metamorphosis approaches the segmental hemilineages begin to converge on the T2 region . The hemilineage groups that project to this region are either intrasegmental , joining left and right neuropils , or intersegmental . For some of the dorsal-projecting hemilineages the members in the cluster appear relatively homogeneous ( e . g . , 6A and 6B ) , as seen for most of the leg hemilineages . In other instances , though , such as the 12A and 8B groups , the hemilineage cluster is more complex and contains obvious subclasses of interneurons . The thoracic hemilineages , then , vary in the heterogeneity of neurons that they contain . We speculate that the more homogenous hemilineages are collections of parallel components , which provide parallel but overlapping channels in the circuits that control sensory to motor transformations . The more heterogeneous hemilineages likely have more diversified integrative functions . The recent characterization of lineages in the fly brain ( Ito et al . , 2013; Yu et al . , 2013 ) shows many brain lineages that contain a greater diversity of cell types than we see for the VNS hemilineages . Nevertheless , some of the brain lineages also appear to be simple , rather homogeneous collections of neurons and we would expect these to also represent parallel processing units . This idea is supported by a projection neuron hemilineage from the lateral antennal lobe ( lAL ) NB . Overall , this hemilineage contains five classes of projection neurons but they are all involved in receiving primary sensory input ( olfactory , gustatory , or primary antennal mechanosensory ) and relaying it to higher centers within or around the mushroom bodies ( Lin et al . , 2012 ) . We stimulated the neurons in the hemilineages via TRPA1 activation to assess the functional roles of these interneuron groups . Would the stimulation of neurons in different hemilineage groups result in distinctive behavioral responses , and was there any evidence for a functional hierarchy amongst the various groups ? As summarized in Supplementary file 1 , most of our lines had some degree of contamination with expression in neurons that were outside of our target hemilineage . We did not use lines that showed expression in sensory neurons ( such as hemilineages 3A and 19B ) , or that had expression in other hemilineage clusters ( e . g . , the hemilineage 14A line ) . We used lines that had expression in a few abdominal neurons per neuromere or weak expression in a few scattered thoracic neurons per neuromere . In all cases , the predominant expression in the line was from the target hemilineage . The problem of extraneous brain contamination was removed by decapitation , although we had concerns about severed descending axons that might still be activated by the heat ramp . Where this was a potential issue ( i . e . , presence of descending axons in the line ) we circumvented the problem by ageing the animals for 24 hr after the decapitation to allow the severed axons to degenerate prior to behavior testing , or by blocking thoracic expression using tsh-GAL80 ( Figure 7; Figure 6—figure supplement 1 ) and showing that the behavioral response disappeared . We collected behavioral data for 22 of the 33 hemilineage groups . Another issue was how much of a given hemilineage is captured in the various driver lines ? We do not have quantitative data on this question , but comparison of the driver lines expression with MARCM clones for the various lineages ( Brown and Truman , 2009; Harris , 2012; Figure 4—figure supplement 1 , Figure 5—figure supplement 1 , Figure 7—figure supplement 1 ) showed that all of the neuropil features evident in the clones were also found in the lines . Therefore , the cellular diversity in the various hemilineage groups is captured in the driver lines that we used . Our behavior testing system was quite artificial since it utilized decapitated flies . Despite lacking their heads , these flies showed a complex array of behaviors during the tests . They provide insights in how the circuitry of the thorax can operate when freed from the descending influences of the brain and subesophageal ganglia . Using TRPA1 expression and a heat ramp to activate the hemilineage groups resulted in both tonic and sequential types of behavioral responses . Tonic responses were the more common and are defined as a persistent behavioral response that began at some point during the heat ramp and were then maintained for the remainder of the stimulus . This type of response was seen for changes in posture ( such as seen for hemilineages 9A or 13B ) for sustained walking ( as for the 1A or the 10B neurons ) , and for sustained wing buzzing ( the 2A interneurons ) . The less common pattern was for the decapitated flies to display an ordered sequence of behavioral responses that changed as the temperature increased . This was very evident for the stimulation of the 12A neurons that began with locomotion that was later joined by lateral wing waving and then finally by the onset of sustained wing buzzing . This behavioral sequencing might result from a progressive increase in neuronal firing rate of the cells in the group during the heat ramp , or by the successive recruitment of group members with higher firing thresholds . We have no data to favor either option at present . A number of observations come from the analysis of the responses to the activation of the different hemilineage groups ( Figure 9 ) . The response to the stimulation of the interneuron group was quite stereotyped and highly reproducible for only a few of the hemilineages . This response consistency was confined to some of the hemilineage groups that evoked postural changes . We do not have detailed transmitter information for all of the hemilineage groups but we do know that the 9A and 5B clusters are GABA-immunopositive neurons ( Harris , 2012 ) . Therefore , the stereotypy resulting from their activation may be due to imposing inhibition at a particular level of sensory-motor integration . The hemilineages that evoked more complex behaviors typically showed a range of behaviors , although there was usually a dominant category of response that characterized that particular hemilineage group . Therefore , for these more complex behaviors , the response to hemilineage group stimulation was probabilistic rather than deterministic , in that the activity of these neurons enhanced the probability that a particular behavior would be performed . A similar phenomenon was seen in the behavioral responses of larval Drosophila to TRPA1 activation of different sparse sets of central neurons ( Vogelstein et al . , 2014 ) . The larval behavioral responses also tended to be probabilistic in that a given set of neurons enhanced the probability of a behavioral response but their activity did not invariably lead to this response . Our discussion below focuses on the most frequently shown behaviors seen during the stimulation of each neuronal group . 10 . 7554/eLife . 04493 . 038Figure 9 . Relationship of hemilineages to neuron type and to classes of evoked behavior . ( A ) Summary of the range of behaviors in decapitated flies elicited by stimulation of the neurons in each of the hemilineage groups using TRPA1 expression and a heat ramp to activate the temperature-sensitive channel . Behaviors are divided into six categories explained in the text . Hemilineages are arranged according to the complexity of their behavioral responses . Most behavioral responses were sustained but a few hemilineages ( * ) showed a progression of behaviors during the heat ramp . Diagrams show the extent of the hemilineage's arbor in transverse views of the ventral nervous system at the level of the mesothorax ( T2 ) . The numbers of flies analyzed in each group ranged from 15 to 25 . ( B ) Registration of the hemilineage arbors to a common outline and then overlapping the hemilineage groups in which at least 50% of the individuals showed the indicated types of behavior . DOI: http://dx . doi . org/10 . 7554/eLife . 04493 . 038 Figure 9 orders the tested hemilineages in terms of the increasing behavioral complexity seen as the neurons were stimulated . Responses ranged from simple postural changes , through rhythmic movements of walking or flight , to the complex behavioral sequence involved in takeoff . This diversity in behavior supports a hierarchical relationship amongst the various interneuron pools with some involved in the patterning of simple movements and others having a more ‘upstream’ function of assembling these simple movements into complex movements or sequences of behavior . Figure 9 also relates the projection patterns of the various hemilineages to the most common behavior seen when these neuronal groups are stimulated . Not surprisingly , the hemilineage groups that evoke changes in leg posture have their arbors concentrated in the leg neuropils . Most are confined to the leg neuropil except for the 3B interneurons , that have major projections into the tectulum and also evoke wing movements , and the 5B neurons , that are potentially GABAergic and may have a general suppressive function . Hemilineages that evoke coordinated or uncoordinated leg movements similarly have a strong leg neuropil component but their arbors also project into the medial neuropil and into the tectulum neuropil . Strong projections into the tectulum neuropil were also evident for all of the hemilineages that evoked wing movements of either low or high frequency . Two of the three hemilineages that evoked frequent takeoff had a strong projection into the T2 leg neuropil ( the site of the jump motoneurons ) in addition to their extensive arbor in the tectulum . Curiously , the third takeoff lineage , lineage 11A/B , lacked a T2 arbor but had substantial projections into the T1 and T3 leg neuropils . Overall the anatomy of the hemilineages conformed to the generally held idea that the leg neuropils deal with leg function while the tectulum neuropil is involved with wing-related behavior such as singing or flight . In addition , though , the tectulum and dorsomedial neuropil may be generally involved in complex , coordinated behavior in the thorax . Hemilineages , such as 1A and 10B , that evoke walking behavior but little in the way of wing movement nevertheless have arbor extension into the dorsomedial neuropil and tectulum , suggesting that these regions may be generally involved in the coordination of complex thoracic behavior regardless of whether it involves legs or wings . Overall the hemilineages appear to have a modular function with cells in a given group being associated with particular behavioral responses . This notion is consistent with physiological data such as that from grasshoppers in which computational circuits are built from clusters of distinct types of interneurons ( Burrows , 1996 ) , at least some types of which have been shown to arise in the same lineage ( Shepherd and Laurent , 1992 ) . A similar strategy is seen in the vertebrate spinal cord in which different cell types arise from defined sets of precursors and the neurons from those cell types play specific roles in defined circuit motifs ( Grillner and Jessell , 2009 ) . Our assessment of the functional roles of the hemilineages is obviously incomplete . We assayed hemilineage function only under one condition—decapitated flies standing at rest and showing occasional spontaneous grooming bouts . Stimulation of the hemilineages under other conditions , for example steady-state flight , will undoubtedly reveal more complexity in the behavioral functions of some of the hemilineages . A fuller understanding of lineage function will require a larger palate of behavioral tests . A second caution comes from our demonstration that the vPR6 neurons , which are required for song production in the male ( von Philipsborn et al . , 2011 ) , are part of hemilineage 12A cluster ( Figure 7 ) . During the course of stimulation of the 12A interneurons the decapitated flies showed a range of behaviors , some of which mimicked the movements displayed during courtship singing , but these soon degenerated into generalized wing vibration movements . We would expect that in the more complex dorsal hemilineages there might be specializations of function within the cluster and that the stimulation of all of the neurons at once might set up conflict states that might be hard to interpret . The specialization of the function of the vPR6 neurons suggest that some neurons within a hemilineage can be adapted for specialized behaviors that are , nevertheless , related to the basic function of the hemilineage as a whole . The segmental CNS of insects provides an unparalleled system in which to study the evolution and diversification of the nervous system . The advantage of the insect system lies in a stereotyped array of segmental NBs that has undergone little change through the course of insect evolution . Based on position , molecular markers and the types of early progeny that they produce , the homologous stem cells can be identified in insects as diverse as silverfish , grasshoppers and Drosophila ( Thomas et al . , 1984; Truman and Ball , 1998 ) . Not only have the NBs been highly conserved , but their early neuronal progeny are also highly conserved as first elegantly shown by Thomas et al . ( 1984 ) . These early-born cells are the ‘primary’ neurons and their identities are determined by birth-order as encoded in a transcription factor progression within the cycling NB ( Isshiki et al . , 2001 ) . Many of these early-born cells serve a pioneer function , which explains the stereotyped pattern of tracts and commissures seen in the ventral neuropil throughout the insects ( Thomas et al . , 1984 ) . The later-born ‘secondary’ neurons are born after the NB starts to express grainyhead ( Almeida and Bray , 2005 ) and include late-born embryonic neurons as well as all of the neurons that are born during postembryonic life ( see Zhou et al . , 2009; for a discussion of primary vs secondary neurons ) . As summarized in Figure 9 , the hemilineage units of these secondary neurons typically include neurons of related phenotypes that may participate in functionally related components of behavior . The interesting question is whether these cell classes are likewise conserved across the insects ? Conservation in class character in widely different groups is well illustrated by the medial lineage . In grasshoppers the A hemilineage cells are GABAergic , engrailed-positive local interneurons and the B hemilineage cells are engrailed-negative , projection cells ( Jia and Siegler , 2002 ) , and the same is true in Drosophila ( JW Truman , unpublished ) . Similarly , a comparative analysis of the clusters of GABA-immunoreactive neurons in the VNS of diverse insects ( Witten and Truman , 1998 ) supports the contention that the properties of neurons in homologous hemilineages are highly conserved across much of the insects . Although one can homologize cell classes of secondary neurons across most of the insects , we think that it is unlikely that one could homologize individual neurons within a particular class . This then brings us to the question of the conservation of function for the neurons of the various hemilineage classes . In grasshoppers ventral interneurons that run in the transverse tract receive input from wing afferents ( Watson and Burrows , 1983 ) while in flies the 2A interneurons are the main neurons of the transverse tract and they can drive flight behavior . For the leg hemilineages , NB4-1 and NB3-1 in grasshopper make the medial and anteromedial spiking interneurons , respectively , of the leg circuit ( Burrows , 1996 ) , while in flies the homologous NBs ( Truman et al . , 2004 ) generate the 14A and 4B interneurons which are similar appearing local interneurons in the leg neuropil . Whether the functions of the neurons are exactly the same in the two species are as of yet unknown , but their involvement in leg function is clearly conserved . We expect that the functional roles of most of all of the hemilineages to be highly conserved through much of insect evolution . For the simpler hemilineages that contain units that work in parallel , addition or reductions in the number of neurons could enhance or reduce functionality by altering the spatial or temporal resolution of their level of the computational network . Another benefit of an increase in numbers of an interneuron type could be to provide a finer control over the speed of locomotion , as seen for the control of swimming speed in larval zebrafish ( McLean et al . , 2008 ) in which the increase in speed involves the recruitment of new interneurons and the repression of interneurons active at the slower speed . For higher level , more complex hemilineages , changes in neuronal properties/connectivity rather than numbers might be more important for generating behavioral changes . Our data from Drosophila provide the first systematic analysis of the hemilineages that make up a major section of the insect CNS . For the thoracic segments it provides a reference to which information from other insects can be compared . It is important to remember , though , that the flight system of Drosophila , with its mesothoracic wings and metathoracic halteres , is a highly derived system compared with more basal insect groups . Also , in the fly about a third of the thoracic hemilineages die leaving few if any members ( Truman et al . , 2010 ) . Consequently , some cell types that may have been important in the ancestral nervous system have been lost with the evolution of Drosophila's derived pattern of flight . The other important caveat is that our analysis includes only the postembryonic secondary neurons , and so we are lacking access to the embryonic-born neurons , a small but extremely important component of the adult CNS . Finally is the issue of whether the data from Drosophila have bearing on the functional organization of nervous systems outside of the arthropods , such as the vertebrate spinal cord . At this point in time we do not know . However , a thoracic unit of the fly thoracic nervous system has about 2500 pairs of neurons , most of which can be assigned to 33 hemilineage units . For many ( most ? ) of the hemilineages one may be able to distill the functional essence of each unit down to one or two sets of cellular characteristics that may reflect the ancestral roles of these cells before the neuronal numbers were expanded . This would then provide us with a basic ‘tool-kit’ of neuron types that insects have used to construct their VNS . These features and the knowledge of the other units on which they preferably synapse may allow us to hypothesize an ancestral wiring diagram for a primitive insect/arthropod VNS . With such a wiring diagram in hand we would then be better able to look for functional homologies in a primitive vertebrate spinal cord such as the lamprey . Flies were reared on standard cornmeal and molasses food at 25°C . All GAL4 lines are from the Rubin GAL4 collection ( Pfeiffer et al . , 2008; Jenett et al . , 2012 ) . Additional stocks used in this study: pJFRC180-20XUAS-IVS-Flp2::PEST in su ( Hw ) attP8 , pJFRC180-20XUAS-IVS-Flp2::PEST in attP40; Actin5C-FRT>-dSTOP-FRT>-GAL4 in attP18; R24B02-GeneSwitch in attP2; pJFRC2-10XUAS-IVS-mCD8::GFP in attP2 ( Pfeiffer et al . , 2010 ) ; R57C10-GAL80-6 in su ( Hw ) attP8; Actin5C-FRT>-dSTOP-FRT>-LexAp::65 in su ( Hw ) attP5; pJFRC19-13XLexAop2-IVS-myr::GFP in attP40 ( Pfeiffer et al . , 2010 ) ; pJFRC177-10XUAS-FRT>-dSTOP-FRT>-myr::GFP in attP18 ( Nern et al . , 2011 ) ; R57C10-LexA::p65 in attP2 ( Pfeiffer et al . , 2012 ) ; pJFRC109-13XLexAop2-FRT>-dSTOP-FRT>-dTRPA1 in VK00005; pJFRC108-20XUAS-IVS-hPR::Flp-p10 in VK00005; pJFRC26-13XLexAop2-IVS-dTRPA1-WPRE in attP40 and attP2 ( Liu et al . , 2012 ) ; pJFRC40-13XLexAop2-FRT>-STOP-FRT>-myr::GFP in attP40 and su ( Hw ) attP5 ( Pfeiffer et al . , 2010; Nern et al . , 2011 ) ; pJFRC67-3XUAS-IVS-Syt::GFP in attP18 ( Zhang et al . , 2002; Pfeiffer et al . , 2010; Seelig and Jayaraman , 2013 ) ; tsh-GAL80 ( Clyne and Miesenböck , 2008 ) . Standard molecular biology techniques , as previously described ( Pfeiffer et al . , 2010 ) , were used in generating the vectors detailed in this study . Drosophila codon-optimized GeneSwitch ( Wang et al . , 1994 , 1997; Burcin et al . , 1998; Burcin et al . , 1999 ) , a fusion of truncated human progesterone receptor ligand binding domain and the p65 activation domain , was synthesized by DNA2 . 0 ( Menlo Park , CA ) . To generate pBPGeneSwitchUw , a 5′–KpnI to 3-BamHI fragment of codon-optimized GAL4 DNA-binding domain ( Pfeiffer et al . , 2010 ) and a 5′-BamHI to 3′-HindIII codon-optimized GeneSwitch fragment were cloned , as a triple ligation , into 5′-KpnI to 3′-HindIII digested pBPGAL4 . 2::p65Uw ( Pfeiffer et al . , 2010 ) . GAL4 , GAL80 , LexA , and GeneSwitch drivers containing putative cis-regulatory modules from the Janelia Research Campus used in this study were generated as previously described ( Pfeiffer et al . , 2008 ) . All Constructs were sequence verified . pJFRC180-20XUAS-IVS-Flp2::PEST was generated by cloning a fusion of codon-optimized Flp2 and a C-terminal PEST sequence as a 5-XhoI to 3′-XbaI fragment into a similarly digested pJFRC7-20XUAS-IVS-mCD8::GFP vector ( Pfeiffer et al . , 2010; Nern et al . , 2011 ) . pJFRC108-20XUAS-IVS-hPR::Flp-p10 ( UAS-Flp-Switch ) was generated as follows: Codon-optimized GeneSwitch was used as template to PCR the truncated hPR ligand binding domain as a 5′-XhoI to 3′-KpnI fragment . Codon-optimized Flp1 variant was amplified from pJFRC150-20XUAS-IVS-Flp1::PEST ( Nern et al . , 2011 ) as a 5′-KpnI to 3′-XbaI fragment . The two fragments were digested and cloned , as a triple ligation , into 5′-XhoI to 3′-XbaI cut pJFRC82-20XUAS-IVS-Syn21-GFP-p10 ( Pfeiffer et al . , 2012 ) . Actin5C-FRT>-dSTOP-FRT>-GAL4 and Actin5C-FRT>-dSTOP-FRT>-LexAp::65 were generated in three steps: First , a 4290 bp fragment 5′ of the translation start site and encompassing the two promoters for Drosophila act5C ( Bond and Davidson , 1986; Chung and Keller , 1990 ) were cloned in a sequential manner: The distal promoter of act5C was PCR amplified from y; cn bw sp genomic DNA ( Adams et al . , 2000 ) and cloned as a 5′-FseI to 3′-KpnI/AgeI into pBDPGAL4U ( Pfeiffer et al . , 2008 ) . A second PCR was performed using y; cn bw sp genomic DNA to amplify a 5′-AgeI to 3′-AgeI fragment for the proximal promoter and cloned into the former construct yielding Actin5C-GAL4-hsp70T . Second , Actin5C-GAL4-hsp70T was then digested 5′-KpnI to 3′-NotI to replace the GAL4 with codon-optimized LexA::p65 ( Pfeiffer et al . , 2010 ) to generate Actin5C-LexA::p65-hsp70T . Third , both the Actin5C-GAL4 and–LexA::p65 constructs were digested with KpnI to insert a FRT flanked cassette consisting of hsp70 and SV40 transcriptional terminator sequences ( Nern et al . , 2011 ) to generate Actin5C-FRT>-dSTOP-FRT>-GAL4 and Actin5C-FRT>-dSTOP-FRT>-LexAp::65 , respectively . GeneSwitch animals were treated with the progesterone mimic mifepristone ( RU486 , Sigma-Aldrich , St . Louis , MO ) . Either 1 . 5 or 7 mg RU486 was dissolved in 1 . 5 ml 70% ethanol and then mixed with 15 ml of melted fly food , which was then allowed to solidify and fed to larvae . For experiments in which adults were fed RU486 , 10–20 flies were placed in a food vial with RU486 , allowed to feed for 3 days , then collected and dissected . For surface application of RU486 , parents were allowed to lay eggs in a food vial for a few days , then transferred to a fresh vial . Approximately 4 days after egg-laying , 60 μl of a ∼10 mM RU486 stock solution ( 10 mg RU486 dissolved in 2 ml 95% ethanol ) was applied to the surface of the food . At 24 hr after treatment , any larvae that had wandered and/or pupariated were discarded to ensure that test animals had fed on RU486 for at least 24 hr . At 48 hr after treatment , the subsequent wandering larvae and pupae ( which had all fed on RU486 for 24–48 hr ) were collected and transferred to an untreated food vial . These animals were then dissected at various times thereafter . Tissues were dissected in PBS ( phosphate-buffered saline , pH 7 . 8 , Cellgro by MediaTech , Inc . , Manassas , VA ) and fixed in 4% buffered formaldehyde overnight at 4°C . Fixed tissues were rinsed in PBS-TX ( PBS with 1% Triton X-100 , Sigma-Aldrich ) , then incubated overnight at 4°C in a cocktail of 10% normal donkey serum ( Jackson ImmunoResearch , West Grove , PA ) , 1:1000 rabbit anti-GFP ( Jackson ImmunoResearch ) , 1:40 rat anti-N-Cadherin ( Developmental Studies Hybridoma Bank , Iowa City , IA ) , and 1:40 mouse anti-Neuroglian ( Developmental Studies Hybridoma Bank ) . Tissues were then rinsed in PBS-TX and incubated overnight at 4°C with 1:500 AlexaFluor 488-conjugated donkey anti-rabbit , AlexaFluor 594-conjugated donkey anti-mouse , and AlexaFluor 649-conjugated donkey anti-rat ( all from Invitrogen , Grand Island , NY ) . Tissues were then washed in PBS-TX , mounted onto poly-lysine-coated coverslips , dehydrated through an ethanol series , cleared in xylenes , and mounted in DPX mountant ( Sigma–Aldrich ) . Nervous systems were imaged on a Zeiss LSM 510 confocal microscope at 40× with optical sections taken at 2 μm intervals . LSM files were contrast-enhanced as necessary and z-projected using ImageJ ( http://rsbweb . nih . gov/ij/ ) . 3 to 4 days after eclosion , adult females were lightly CO2-anesthetized , sorted , and returned to food vials , where they were allowed to recover for 2 days . Females were then chilled in an iced vial , transferred to a cold plate ( Teca , Chicago , IL ) at 2°C , and quickly decapitated using microscissors in batches of 5–20 . Flies were on the cold plate for less than 3 min . Decapitated flies were brushed back into a food vial and allowed to recover for at least 1 hr . The heat-activated cation channel TRPA1 shows some activity at 25°C , and is fully active in the range of 27°C–32°C ( Hamada et al . , 2008 ) . For TRPA1 activation , batches of 5–10 decapitated flies were tapped onto a sheet of paper , and any flies that were unable to right themselves were discarded . The paper was then placed on a hot plate ( Teca ) and ramped from 24°C to maximum activation temperature . For flies reared at 21°C , the maximum activation temperature was 32°C and the ramp took 45 s . For flies reared at 25°C , the maximum activation temperature was 37°C and the ramp took 55 s . Reference temperatures taken from the hot plate's internal sensor were manually marked in the videos . For all lineages , low-speed ( 60 fps ) videos were taken using a Dragonfly Express digital camera ( Point Grey Research , Canada ) controlled by the MATLAB Image Acquisition ( ImAq ) tool in the Image Acquisition Toolbox . Videos were taken from above , with a field of view diameter of approximately 50 mm . Recordings lasted 60 s , encompassing the entire temperature ramp and a period of time at maximum activation temperature . For lineages that induced flight-related phenotypes , high-speed ( 1000–6000 fps ) videos were taken with a Phantom v9 high-speed digital camera ( Vision Research , Wayne , NJ ) using the Phantom software . Flies were recorded individually or in groups of less than five , with a field of view approximately 5 mm in diameter . Recordings were saved from the camera's memory buffer from the period preceding a predetermined behavioral trigger-point , such as takeoff . Low-speed footage ( 60 fps ) was played back using VirtualDub ( http://virtualdub . org/ ) and/or ImageJ ( http://rsbweb . nih . gov/ij/ ) . The behavior of each fly during the heat ramp was annotated manually . High-speed footage was annotated manually using Phantom software . Behaviors were scored throughout the temperature ramp .	The legs and wings of insects are borne on the middle body segments , which make up the thorax . The nervous system inside of the thorax is part of the insect equivalent of the spinal cord and contains clusters of interneurons that relay signals between the sensory nerves , the brain and the muscles . This enables the insect to perform complex actions such as walking and flying . The thoracic interneurons are produced by a fixed set of stem cells . Each stem cell makes neurons in a pair-wise fashion by producing a sequence of neural progenitor cells , each of which then divides to produce two different types of daughter neurons . All of the daughter neurons of the same type are said to belong to the same hemilineage , and in the fruit fly Drosophila , the majority of the interneurons in the thorax are from one of 33 hemilineages . Each interneuron cluster in the insect thorax is made up of cells from a single hemilineage . Harris et al . developed genetic tools that allow the different hemilineages in the Drosophila thorax to be labeled , and used this to create a set of flies that allows the role of the different clusters to be investigated . Each fly type was modified so that increasing the temperature activated a heat-sensitive channel in the neurons of a single hemilineage , and Harris et al . recorded the behavioral response this produced . Each hemilineage caused the fly to move in a distinctive way when stimulated , and many of these movements were unique to a single cluster . Furthermore , the hemilineages can be divided into different groups based on their complexity . Activating the simplest group of hemilineage clusters produces simple movements such as leg twitches and stretches . Another group of hemilineages are then able to organize these movements into more complicated behaviors , such as walking . The third , most complex , hemilineages can coordinate several complex actions to enable the flies to perform very complicated tasks , like take off for flight . These findings suggest that hemilineages act as the basic modules of the nervous system in the fly thorax . Furthermore , the flies and techniques developed by Harris et al . will provide valuable resources for future studies into the organization and function of the nervous system .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2015	Neuron hemilineages provide the functional ground plan for the Drosophila ventral nervous system
Male germ cells of all placental mammals express an ancient nuclear RNA binding protein of unknown function called RBMXL2 . Here we find that deletion of the retrogene encoding RBMXL2 blocks spermatogenesis . Transcriptome analyses of age-matched deletion mice show that RBMXL2 controls splicing patterns during meiosis . In particular , RBMXL2 represses the selection of aberrant splice sites and the insertion of cryptic and premature terminal exons . Our data suggest a Rbmxl2 retrogene has been conserved across mammals as part of a splicing control mechanism that is fundamentally important to germ cell biology . We propose that this mechanism is essential to meiosis because it buffers the high ambient concentrations of splicing activators , thereby preventing poisoning of key transcripts and disruption to gene expression by aberrant splice site selection . It has been a long recognised hallmark of mammalian gene expression patterns that there are extremely high levels of transcription and transcriptome complexity in testicular cells ( de la Grange et al . , 2010; Licatalosi , 2016; Soumillon et al . , 2013; Pan et al . , 2008; Wang et al . , 2008; Clark et al . , 2007; Grosso et al . , 2008; Yeo et al . , 2004 ) . These high gene expression levels are thought to result from epigenetic changes that favour relaxed patterns of gene expression during meiosis – the unique form of division used to generate sperm and eggs ( Soumillon et al . , 2013 ) . Many nuclear RNA-binding proteins are differentially expressed during and immediately after meiosis ( Grellscheid et al . , 2011; Schmid et al . , 2013 ) . These include the nuclear RNA binding protein RBMXL2 ( also known as hnRNP GT ) that is expressed only during and immediately after male meiosis , but not in the preceding spermatogonial cells ( Ehrmann et al . , 2008 ) ( Figure 1A ) . Consistent with an important function in germ cell development , genetic studies have identified point mutations within infertile men in the human chromosome 11 RBMXL2 gene ( Westerveld et al . , 2004 ) . A further connection to male infertility is that RBMXL2 belongs to the same gene family as RBMY , which was historically the first human Y chromosome infertility gene identified in the search for the AZF ( AZOOSPERMIA FACTOR ) gene ( Ma et al . , 1993 ) . RBMXL2 evolved ~65 million years ago via retrotransposition of an mRNA from the X chromosome located RBMX gene ( Figure 1A ) . An RBMXL2 gene is found in all placental mammals , which is consistent with a fundamental role in germ cell biology . This role remains to be identified , but RBMXL2 protein has an N-terminal RNA Recognition Motif ( abbreviated RRM , Figure 1B ) . RBMX and RBMY proteins are also nuclear RNA binding proteins that are very similar to RBMXL2 ( 73 . 2% and 36 . 8% overall identity to RBMX and RBMY , respectively; 93 . 7% and 77 . 2% identity within the RRMs , Figure 1—figure supplement 1 ) . The RNA binding specificity of RBMXL2 protein is unknown , but both RBMX and RBMY proteins bind to AA dinucleotide-containing RNA sequences ( Cléry et al . , 2011; Moursy et al . , 2014; Nasim et al . , 2003 ) . The RBMXL2 , RBMX and RBMY proteins interact with and modulate the splicing activity of Tra2β and SR proteins in vitro ( Figure 1B ) ( Cléry et al . , 2011; Moursy et al . , 2014; Liu et al . , 2009; Nasim et al . , 2003; Elliott et al . , 2000a ) , suggesting a role in splicing control . Maintaining proper ratios of mRNA splice isoforms can be critical in normal development ( Kalsotra and Cooper , 2011 ) , where changes in isoforms can have effects on encoded proteins ranging from major to subtle . Alternative splicing is known to be critical for germ cell development . For example , deletion of the splicing regulator protein PTBP2 within germ cells affects mRNA isoforms important for cell-cell communication with Sertoli cells ( Hannigan et al . , 2017 ) . Some alternative splicing events in the testis are conserved between humans and mice so may control fundamental aspects of germ cell biology ( Schmid et al . , 2013 ) . However many alternative splicing patterns are not conserved between humans and mice ( Kan et al . , 2005 ) . RBMX protein is also reported to control transcription ( Takemoto et al . , 2007 ) , affect DNA double strand repair and mitotic sister chromatid cohesion ( Adamson et al . , 2012; Matsunaga et al . , 2012 ) , and to bind to m6A methylated RNA ( Liu et al . , 2017 ) . The function of RBMXL2 and why this RNA binding protein has been conserved across placental mammals are not known . A major factor limiting understanding of endogenous RBMXL2 functions has been the absence of a reliable mouse model . Development of a mouse model is also critical to test the importance of this wider family of RNA binding proteins in germ cell development . Men carrying the AZFb deletion on their Y chromosomes are missing RBMY genes and undergo meiotic arrest . However , it is unclear if RBMY loss is causing this phenotype , because the deletion interval encompasses several other genes which could contribute to male infertility ( Elliott , 2000; Vogt et al . , 1996 ) . Within the meiotic and immediately post-meiotic germ cells which express RBMXL2 , the RBMX and RBMY genes are transcriptionally inactivated within a heterochromatic structure called the XY body ( Wang , 2004 ) . Meiosis thus provides a genetically tractable window to probe RBMXL2 function where there should be no redundancy effects possible with either RBMX or RBMY . Hence to discover what RBMXL2 does in the germline we have made a conditional Rbmxl2 gene knockout mouse . Analysis of this knockout mouse reveals that RBMXL2 protein is essential for meiosis and has a major role in protecting the meiotic transcriptome from aberrant selection of cryptic splice sites that are normally ignored by the spliceosome . Our data suggest this fundamentally important process operates so efficiently in meiosis that it has been previously undetected , yet is critical to avoid male infertility caused by aberrant splicing of key meiotic transcripts . To test how important RBMXL2 protein is for male germline development we made a conditional mouse model in which we flanked the Rbmxl2 open reading frame with LoxP sites . Since Rbmxl2 is only expressed in the testis ( Elliott et al . , 2000b ) , we chose to delete the entire Rbmxl2 open reading frame to create a null allele . We achieved this by crossing our conditional model with a mouse strain expressing Cre recombinase under control of the ubiquitous Pgk promoter ( experimental details are provided in the Materials and methods ) . We confirmed deletion of this genomic region in homozygous Rbmxl2 gene knockout ( Rbmxl2-/- ) mice by Southern blotting ( Figure 1—figure supplement 2 ) and the specific absence of the 50 KDa RBMXL2 protein from knockout testes by Western blotting ( Figure 1C ) . Rbmxl2-/- mice developed comparably to their wild type littermates but their testes were much smaller ( Figure 2A and B ) . This small testis phenotype correlated with a severe disruption of testicular histology . Adult Rbmxl2-/- mice contained cells undergoing meiosis but almost no post-meiotic cells ( Figure 2C , meiotic spermatocytes are abbreviated Spc , and post-meiotic round spermatids are abbreviated Rtd ) . The epididymis dissected from Rbmxl2-/- mice were completely devoid of sperm . Four wild type mice tested had an average of 7 . 07 ± 1 . 39×106 epididymal sperm ml−1 , compared with zero in four Rbmxl2-/- mice ( Figure 2D and E ) . No effect on female fertility was observed in Rbmxl2-/- mice ( not shown ) . Very rarely a few round spermatids were observed in adult Rbmxl2-/- testis sections ( Figure 2—figure supplement 1A ) , although no elongated spermatids were detected . The presence of these round spermatids indicates that meiosis can occasionally complete in the absence of RBMXL2 protein . This is consistent with loss of RBMXL2 causing either ( 1 ) a developmental block in meiosis from which a few cells can escape; or alternatively ( 2 ) a slow attrition effect , in which the Rbmxl2-/- testis phenotype is caused by post-meiotic stages dying in the adult testis . To differentiate between these two possibilities we histologically analysed testes at 21 days postpartum ( 21dpp ) , at which point germ cells in the wild type ( Rbmxl2 +/+ ) mice have just started to enter the post-meiotic round spermatid stage . At 21dpp there were still significantly fewer round spermatids in the Rbmxl2-/- sections compared to wild type ( Figure 2—figure supplement 1B ) , even though there would have been less time for round spermatid cell death compared with the adult . This result thus supports a strong meiotic block in Rbmxl2-/- mice , with occasional completion of meiosis rather than a gradual attrition of round spermatids . The above data demonstrate that an Rbmxl2 gene , which is conserved in all placental mammals and specifically expressed in male meiosis , is essential for mouse spermatogenesis somewhere within meiotic prophase . Staining of nuclear spreads with antibodies specific to the meiotic chromosome proteins SYCP1 and SYCP3 more precisely showed that Rbmxl2-/- mice arrest germ cell development during the diplotene substage of meiotic prophase ( Figure 2F and Figure 2—figure supplement 2 ) . Metaphase I nuclei were only detected in wild type mice and not in Rbmxl2-/- mice ( N = 3 wild type and N = 3 Rbmxl2-/-testes , 180 and 138 spermatocyte nuclei scored for wild type and Rbmxl2-/- respectively ) . No significant differences in the frequency of earlier stages of meiotic prophase were detected between Rbmxl2-/- and wild type testes . The presence of germ cell populations up to diplotene in Rbmxl2-/- testes was further confirmed by analysis of histological sections stained with Periodic Acid Schiff ( PAS ) ( Figure 2—figure supplement 3A–D ) . Staining of adult testis sections with PAS or antibodies specific to the apoptotic marker activated Caspase three further showed that adult Rbmxl2-/- germ cells die via apoptosis ( Figure 2—figure supplement 3E–F ) . Analysis of diplotene nuclear spreads revealed further abnormalities in the Rbmxl2-/- testis . There was a decreased number of H3K9me3-marked centromere clusters ( Takada et al . , 2011 ) , with each individual cluster also containing more centromeres than in wild type testis ( Figure 2—figure supplement 4A–C ) . H3K9me3 staining was also present at the sex body in a proportion ( 82% ) of wild type diplotene spermatocytes , but essentially absent in mutant diplotene spermatocytes ( 1 . 5% of diplotene spermatocytes stained , 66 Rbmxl2-/- and 66 wild type nuclei scored , n = 3 for both ) ( Figure 2—figure supplement 4D ) . These defects in centromere clustering and H3K9me3 modification of the sex body were still detectable but less severe at pachytene ( Figure 2—figure supplement 4D ) . Asynapsis was rarely observed in either control or mutant pachytene stage spermatocytes , indicating that the defect causing diplotene arrest does not significantly impact either synaptonemal complex formation or meiotic homology searching ( Figure 2—figure supplement 4E ) ( N = 3 , 109 and 94 nuclei scored for wild type and Rbmxl2-/- respectively ) . In summary , the above mouse phenotype showed that deletion of the ancient RNA binding protein RBMXL2 induces progressive defects in male mouse meiotic prophase . These defects culminate with a major block during diplotene that prevents entry into metaphase I , but with defects becoming already apparent during pachytene . Since RBMXL2 is a nuclear RNA binding protein we predicted that the above phenotype could be associated with a primary molecular defect in generating or processing RNAs in the Rbmxl2-/- testis . To test this we analysed wild type and Rbmxl2-/- testes using RNAseq . Based on the knockout phenotype above , and because the adult wild type testis contains additional more advanced germ cells that are missing from the adult Rbmxl2-/- testis , we analysed testes at 18 days post partum ( 18dpp ) during the first synchronised wave of mouse spermatogenesis . Such 18dpp wild type mouse testes contain germ cells between spermatogonia all the way through to diplotene , with 60% of cells engaged in meiotic prophase , but no post-meiotic cells ( Bellvé et al . , 1977 ) . Gene expression analysis of this RNAseq data ( Anders and Huber , 2010 ) showed overall patterns of transcription were similar between wild type and Rbmxl2-/- 18dpp testes ( Figure 3A , and Figure 3—source data 1 ) . Only 45 genes showed a fold change greater or equal to two between the wild type and Rbmxl2-/- testes , with an adjusted p value of less than 0 . 05 , and only 23 of these changes were for known protein coding genes ( Figure 3—source data 1 ) . The strongest difference in overall gene expression between the wild type and Rbmxl2-/- backgrounds was for the Rbmxl2-/- gene itself , as expected since the Rbmxl2-/- gene is deleted from the knockout mouse . We also detected strong expression changes within the Fsip2 gene , particularly for Fsip2 exon 16 and downstream exons that were expressed only in the wild type background ( Figure 3—figure supplement 1A and B ) . Mutations in Fsip2 correlate with defects in human sperm flagella – cellular structures which develop after meiosis ( Martinez et al . , 2018 ) . RNAseq analysis detected more subtle expression changes in Cul4a , Slc9c1 and Tex15 , each of which are required for mouse male fertility ( Smith et al . , 2018 ) ( Figure 3—source data 1 ) ( Kopanja et al . , 2011 ) . Genes encoding transcription factors ( Myf6 and Nxn ) and a signalling protein ( Cyr61 ) that controls apoptosis ( Jun and Lau , 2011 ) also changed expression . Mouse phenotype information at the Mouse Genome Database ( MGD ) ( Smith et al . , 2018 ) indicate that each of these latter three genes have important roles in normal development but not specifically of the testis ( Figure 3—source data 2 ) . Sixteen ( 36% of the total detected ) gene expression changes in the Rbmxl2-/- testis were for predicted non coding RNAs of unknown function ( Figure 3—source data 1 ) . We carried out further bioinformatic analysis to specifically search for mis-regulated splicing events in the Rbmxl2-/- testes ( Vaquero-Garcia et al . , 2016 ) . A total of 237 high-confidence , mis-regulated local splicing variations were identified in 186 genes ( Figure 3B , and Figure 3—source data 3 ) . Using RT-PCR to distinguish splice isoforms 27 of these splicing changes were experimentally tested , validating 23/27 splice isoform switches ( Figure 3—source data 4 ) . Some genes had more than one splicing event controlled by RBMXL2 ( e . g . the Catsperb gene had three events ) . Gene Ontology ( GO ) analysis showed that a number of the genes regulated at the splicing level by RBMXL2 have established roles in spermatogenesis , meiosis and germ cell development ( Figure 3—source data 5 ) . However , amongst the complete set of regulated genes there was no significant enrichment of particular GO terms . This is consistent with RBMXL2 regulating splicing of a functionally diverse group of genes . Analysis of knockout phenotypes provided by the MGD ( Smith et al . , 2018 ) indicated that whole gene deletion of 25/186 RBMXL2-regulated genes cause male infertility . These latter target genes must thus have a key role in germ cell development ( Figure 3C , Figure 3—source data 6 , Figure 3—source data 7 ) . Genetic deletion of some other RBMXL2 target genes cause either embryonic or neonatal lethality ( 33/186 genes ) , developmental ( 19/36 genes ) or pleiotropic defects ( 43/186 genes ) . Cell type mis-splicing of these latter genes during meiosis could thus cause severe phenotypic effects on germ cell biology . Fourteen ( 33% ) of the genes originally identified to have changed overall expression levels between wild type and Rbmxl2-/- testes also changed splicing patterns ( Figure 3C , and Figure 3—source datas 1 and 3 ) . These included splice variants in Esco1 ( encoding a protein involved in chromatid cohesion ) , Slc39a8 ( that encodes the transporter protein responsible for cadmium toxicity in the testis ) ( Dalton et al . , 2005 ) ; and the Slc9c1 gene ( annotated on the MGD as essential for male fertility [Smith et al . , 2018] ) . The RNA binding specificity of RBMXL2 protein was unknown . Thus we used high throughput sequencing cross linking immunoprecipitation ( HITS-CLIP ) to enable us to correlate splicing changes detected within the Rbmxl2-/- 18dpp mouse testis with global RBMXL2 protein-RNA interactions ( Grellscheid et al . , 2011 ) . Antibodies specific to mouse RBMXL2 immuno-precipitated a radiolabelled RNA protein adduct of the known size of RBMXL2 protein ( 50 KDa ) after treatment with high concentrations of RNase ( Figure 3—figure supplement 2A ) ( Elliott et al . , 2000b ) . Lower concentrations of RNase were used to retrieve an average tag length of 40 nucleotides . Enriched motif analysis of the sequenced RBMXL2 CLIP tags showed that each of the top 10 5-mers contained the dinucleotide AA ( Figure 3D ) . Interestingly , AA is also the dinucleotide bound by RBMX ( Moursy et al . , 2014 ) , which is consistent with over 90% shared sequence identity within the RRMs of these proteins ( Figure 1—figure supplement 1 ) . These intragenic cross-linked sites mapped to the mouse genome most frequently within introns , consistent with RBMXL2 being involved in nuclear RNA-processing events ( Figure 3—figure supplement 2B ) . Next we searched for CLIP tags mapping to regions near the splicing events controlled by RBMXL2 , which would suggest direct regulation . We observed enrichment of RBMXL2 binding both within alternative exon sequences and in regions proximal to regulated splice sites compared to non-regulated events ( Figure 3—figure supplement 3A ) . Consistent with these enriched binding occurrences , motif maps of the top pentamers identified by CLIP ( Figure 3D ) showed regions of enrichment proximal to regulated versus non-regulated splice junctions ( Figure 3—figure supplement 3B ) . Overall , the above bioinformatics analysis indicated an accumulation of defective splice isoform patterns in the Rbmxl2-/- testis . MAJIQ captures local splicing variations ( LSVs ) that involve both known and un-annotated ( de-novo ) splice sites , junctions and exons . These LSVs can correspond to both classical binary splicing events and more complex events involving three or more junctions . Since classical binary events ( e . g . skipped exons ) are easier to inspect and visualise with the RNAseq reads on the UCSC genome browser , we focused on the binary events for further investigation . This showed that 60% of the 87 most easily visualised classical splicing events controlled by RBMXL2 involve the altered selection of de-novo splice sites ( defined as not currently annotated on the most recent build of the Mus musculus genome GRCm38/mm10 , Figure 3—source data 3 ) . These de-novo splicing variations controlled by RBMXL2 are putatively cryptic events as they insert novel internal exons that were flanked by consensus GT-AG 5' and 3' splice sites . Moreover , these de-novo splice sites showed low expected percent spliced in ( E ( PSI ) ) values in wild type 18 dpp testes and across a panel of twelve mouse tissues , further suggesting they are cryptic events only included in the absence of RBMXL2 and not in wild type mice ( Figure 4—figure supplement 1 ) . Detailed investigation indicated 84% of such cryptic exons were either not multiples of three or introduced stop codons into their mRNAs , meaning their insertion into mRNAs would disrupt protein reading frames and interfere with meiotic protein expression . Splicing inclusion of cryptic terminal exons could also severely impact patterns of meiotic gene expression . We detected high inclusion of a cryptic terminal exon within the 5′ UTR of the Kdm4d gene ( Figure 4A and B , note also increased splicing of an already annotated upstream Kdm4d alternative exon within the Rbmxl2-/- testis ) . We confirmed splicing inclusion of this Kdm4d cryptic exon in the Rbmxl2-/- testes using RT-PCR ( primer positions are shown in Figure 4A ) . Kdm4d encodes a histone demethylase protein that is important for normal patterns of germ cell apoptosis in the testis ( Iwamori et al . , 2011 ) . Exon two contains the entire CDS ( Coding DNA Sequence , Figure 4A ) of the Kdm4d gene . Analysis of exon junction read numbers using Sashimi plots confirmed different nuclear processing pathways are used for Kdm4d in wild type and knockout testes ( Figure 4—figure supplement 2A ) . There were 14-fold more exon junction reads connecting Kdm4d exon one to the cryptic exon ( 140 reads ) , when compared to exon junction reads joining the cryptic exon to Kdm4d coding exon 2 . This pattern is consistent with splicing inclusion of the cryptic exon being connected with use of an associated polyA site . Consistent with this , some individual RNAseq reads extended past the 5′ splice site of the Kdm4d cryptic exon and then terminated downstream following a consensus polyadenylation site ( AATAAA ) sequence ( Figure 4C ) . Both bioinformatics analysis ( Figure 3—source data 1 ) and qPCR ( Figure 4—figure supplement 2B ) detected reduced Kdm4d gene expression levels in each of 3 replicate Rbmxl2-/- testes compared to their wild type equivalents , but these trends were not statistically significant due to biological variability between individual mice . We also detected increased selection in the Rbmxl2-/- testes of another 17 splicing events that mapped to terminal exons ( Figure 3—source data 3 ) . These included an already annotated upstream terminal exon in the Lrrcc1 gene ( annotated on the MGD as essential for mouse fertility ) ( Smith et al . , 2018 ) ; and a terminal exon in the Slc39a8 gene ( Figure 4—figure supplement 3A and B ) . In addition to splicing of entire cryptic exons , loss of RBMXL2 protein also activated splicing selection of individual cryptic 5′ and 3′ splice sites within introns that increased the length of already annotated exons ( Figure 3—source data 3 ) . This kind of changed splicing pattern was observed for the Gm5134 mRNA that encodes a solute transporter important for normal metabolism ( Figure 4D and E ) . The aberrant splice isoform of the Gm5134 transcript made in the Rbmxl2-/- testis uses a cryptic 3′ splice site within intron 10 , increasing the size of exon 11 by 1353nt . As well as cryptic splice sites within introns , some exon-located cryptic splice sites were activated within the Rbmxl2-/- testes , including within genes important for testis development ( Figure 3—source data 6 ) . Amongst the genes known to be important for meiosis , we observed a dip in RNAseq density for the interior of Meioc ( meiosis specific gene with coiled coil domain ) exon five in the Rbmxl2-/- testis transcriptome ( Figure 5A ) ( Abby et al . , 2016; Soh et al . , 2017 ) . This dip in RNAseq read density was caused by high levels of cryptic splice site activation within Meioc exon 5 . Sashimi plots showed that in the Rbmxl2-/-testes ~ 28% of the splice junctions joined Meioc exon four to an exon-internal cryptic 3′ splice site within Meioc exon 5 ( Figure 5—figure supplement 1A ) . A further cryptic 5′ splice site within Meioc exon five was also used in 22% of the Rbmxl2-/- testis transcripts . Use of both cryptic splice sites converted Meioc exon five into an exitron ( an exon with internal splice sites ) ( Marquez et al . , 2015; Sibley et al . , 2016 ) . RT-PCR analysis confirmed both these defective Meioc exon five splicing patterns within the Rbmxl2-/- testes ( Figure 5B–C ) . The cryptic splice sites within Meioc exon five are hardly ever used in the wild type 18dpp testis transcriptome . In silico analysis showed that the internal cryptic 5′ splice site within Meioc exon five is weak compared to 5′ splice sites for other known alternative exons ( 7 . 5th percentile by SROOGLE , compared to 57th percentile for the downstream annotated 3′ splice site ( Figure 5—figure supplement 1B ) ( Schwartz et al . , 2009 ) . The first two nucleotide positions downstream of the Meioc exon 5- cryptic internal 5′ splice site are GC , so diverge from the GT consensus which is found flanking almost all eukaryotic introns . Similarly , the cryptic internal 3′ splice site within Meioc exon five is relatively weak ( 8th percentile of the average 3′ splice site strength for an alternative exon , compared to 41st percentile for the upstream annotated 3′ splice site ) . Meioc exon five is also directly bound by RBMXL2 protein , evidenced by three internal HITS-CLIP tags ( Figure 5A ) . These include a CLIP tag directly overlapping the Meioc exon 5 internal cryptic 5' splice site , as well as CLIP tags upstream of the internal 3′ splice site . Meioc exon five is highly conserved across species suggesting it encodes a functionally important part of Meioc protein ( Figure 5A ) . Exitrons frequently modify protein coding capacity of mRNAs by removing peptide coding segments from mRNAs ( Marquez et al . , 2015; Sibley et al . , 2016 ) . Both the cryptic splicing events in Meioc exon five remove RNA segments within the Meioc CDS that are multiples of 3 . These events shorten the CDS just upstream of but largely not including the coding information for the MEIOC domain itself ( http://pfam . xfam . org/family/PF15189 ) . We thus predicted that a shorter Meioc protein isoform may be produced in the Rbmxl2-/- testis . Consistent with this , Western blotting with a Meioc specific antiserum ( Abby et al . , 2016 ) detected a 40–50 KDa Meioc protein within the Rbmxl2-/- testis , but not in wild type or heterozygote Rbmxl2 ±testes ( Figure 5D ) . This 40–50 KDa molecular weight corresponds to an expected Meioc protein size following complete exitron removal . Although shorter Meioc protein isoforms were only detected in the Rbmxl2-/- testes , in each genotype we detected a protein doublet just above 100 KDa corresponding to the reported size of full length Meioc protein ( Abby et al . , 2016 ) ( Figure 5D ) . We further analysed splicing patterns during the first wave of mouse spermatogenesis to test whether appearance of the Meioc splicing defects corresponded with the known expression window of RBMXL2 protein in meiotic prophase ( Ehrmann et al . , 2008 ) . Aberrant splice isoforms of Meioc exon five only appeared within Rbmxl2-/- testes from 14dpp , and were barely visible at any time point within similar age wild type testes . We performed similar experiments to analyse the dynamics of Brca2 splicing in the first wave of mouse spermatogenesis . The RNAseq data predicted utilisation of an upstream cryptic 5′splice site within the 4809 nucleotide Brca2 exon 11 ( Figure 3—source data 3 ) that removes 1263 nucleotides from Brca2 exon 11 in the Rbmxl2-/- testes ( Figure 5—figure supplement 2A ) . RT-PCR analysis during the first wave of spermatogenesis showed Brca2 exon 11 splicing was normal at 13dpp in the Rbmxl2-/- testes , with defects appearing only during and after 14dpp ( Figure 5—figure supplement 2B ) . These time-course data indicate that splicing defects in both Meioc and Brca2 appear and become progressively worse during meiotic prophase in the Rbmxl2-/- testes within genes that are expressed and processed normally in earlier germ cell types . An implication of this is that the aberrant splice isoforms detected in 18dpp whole testis RNA must be significantly diluted by signals from germ cells from earlier developmental stages . Other unusually long exons in genes important for germ cell development were also mis-spliced in the Rbmxl2-/- testes including exons in the Alms1 and Esco1 genes ( Figure 3—source data 3 ) . The Alms1 gene encodes a protein involved in microtubule organisation during cell division , and its knockout disrupts spermatogenesis resulting in a small testis phenotype ( Smith et al . , 2018 ) . In the Rbmxl2-/- testis an upstream 5′splice site is selected within Alms1 exon 5 , which is is unusually long at 1546nt ( Figure 5F ) . This splice switch removes 1407 nucleotides from the Alms1 mRNA , and coding information for 469 amino acids from the predicted Alms1 reading frame . The above data indicate that loss of RBMXL2 protein within spermatocytes activates utilisation of cryptic splice sites . We next tested if cryptic splice site selection would be reciprocally repressed by ectopic expression of RBMXL2 protein . The genomic region spanning the Kdm4d cryptic exon and its flanking intron sequences were cloned into a minigene ( Figure 6A ) . We detected ~40% splicing inclusion of the cryptic Kdm4d exon after co-transfection of this minigene into HEK293 cells ( that do not normally express RBMXL2 ) with an expression vector encoding GFP ( Figure 6B lane 1 ) . Levels of Kdm4d cryptic exon splicing inclusion were repressed 6-fold upon co-transfection of a construct encoding an RBMXL2-GFP fusion protein . We also tested the activity of an RBMXL2 fusion protein without the RNP1 motif of the RRM ( RBMXL2 ΔRNP1 ) . Consistent with RNA-protein contacts not being critical for splicing repression of this cryptic exon , RBMXL2 ΔRNP1 could still repress Kdm4d cryptic exon splicing , although slightly less efficiently than the full length RBMXL2 fusion protein . Interestingly , splicing of the Kdm4d cryptic exon was also efficiently silenced by ectopic co-expression of an RBMX-GFP fusion protein ( Figure 6B lane 7 ) . The above results show that RBMXL2 represses Kdm4d cryptic exon splicing , and is consistent with RBMXL2 providing a direct functional replacement for RBMX during meiosis . iCLIP experiments in adult mouse testis further identified the Kdm4d cryptic exon as directly binding the splicing activator protein Tra2β ( Dalgliesh and Elliott , unpublished data ) . Consistent with this , Tra2β over-expression in HEK293 cells strongly increased Kdm4d cryptic exon splicing inclusion ( Figure 6B lane 4 ) . Ectopic activation by Tra2β indicated that although this Kdm4d exon cryptic exon is effectively ignored by the spliceosome in wild type testis , cryptic exon splicing is activated in response to increased expression of a splicing activator protein . Importantly , splicing activation by Tra2β was efficiently supressed when the minigene was co-transfected at the same time with both RBMXL2 and Tra2β ( Figure 6B , lanes 5 and 6 ) . We addressed if other cryptic exons repressed by RBMXL2 might also possess the capacity to be activated by splicing regulators that bind to their associated splicing enhancer sequences . In silico analysis ( Schwartz et al . , 2009 ) of a panel of cryptic exons that were activated in the Rbmxl2-/- testes showed that these exons had significantly weaker 3′ splice sites , but similar Exonic Splicing Enhancer ( ESE ) sequence content compared to their immediately adjacent downstream exons ( Figure 6C ) . These other cryptic exons that are repressed by RBMXL2 may be likewise poised for splicing during meiosis , because they bind to splicing activator proteins such as Tra2β . This study finds that loss of the ancient RBMXL2 protein completely prevents sperm production in the mouse , largely because of a developmental block during meiosis . Rbmxl2-/- mouse germ cells develop as far as meiosis , thus escaping the major developmental checkpoints that operate during meiotic pachytene ( de Rooij et al . , 2003 ) , but then undergo cell death by apoptosis after reaching diplotene . Our data further show that RBMXL2 protein is important to protect the meiotic transcriptome from a spectrum of splicing defects ( Figure 7A ) . These include classic manifestations of cryptic splicing that would be strongly deleterious to gene expression ( Marquez et al . , 2015; Sibley et al . , 2016 ) , such as insertion of novel cryptic exons and premature polyadenylation sites that would disrupt protein coding sequences , and the selection of cryptic splice sites that would shorten or extend exon lengths . Previous work reporting poor conservation of splicing patterns between species suggested that alternative splicing might not control fundamental aspects of germ cell biology ( Kan et al . , 2005 ) . In contrast , the work presented here suggest a generic requirement for cryptic splice site suppression during meiosis that could explain the conservation of an Rbmxl2 gene in all placental mammals , even if individual cryptic splice sites are not conserved between species’ genomes ( Elliott et al . , 2000b ) . Cells in meiotic prophase might be particularly susceptible to cryptic splice site poisoning because of their high levels of transcription , altered splicing regulator expression and relaxed chromatin environments ( de la Grange et al . , 2010; Licatalosi , 2016; Soumillon et al . , 2013; Pan et al . , 2008; Wang et al . , 2008; Clark et al . , 2007; Grosso et al . , 2008; Yeo et al . , 2004 ) . In silico data suggest that the cryptic exons repressed by RBMXL2 would be poised for splicing inclusion in conditions of high concentrations of splicing activators because of their strong ESE contents . We suggest a model where RBMXL2 protein may buffer the activity of splicing activator proteins to prevent such cryptic splice site selection from happening in meiosis ( Figure 7B ) . This model is also consistent with previously reported data showing that RBMXL2 , RBMX and RBMY proteins physically interact with and antagonise the splicing activity of Tra2β and SR proteins ( Liu et al . , 2009; Nasim et al . , 2003; Venables et al . , 2000; Elliott et al . , 2000a ) ( Figure 7B ) . Data presented in this study also show that Kdm4d cryptic exon splicing activation by Tra2β is blocked in vitro by co-expression of RBMXL2 . As an alternative and not mutually exclusive model , RBMXL2 protein-RNA binding may sterically block recognition of some cryptic splice sites during meiosis , including the cryptic 5′ splice site within Meioc exon five which was utilised in the Rbmxl2-/- testis ( Figures 5A and 7B ) . However , no HITS-CLIP tags for RBMXL2 were detected near the Kdm4d cryptic exon . Furthermore , a version of RBMXL2 deleted for the key RNP1 motif involved in RNA-protein interactions could still repress Kdm4d cryptic exon splicing ( albeit slightly less efficiently than the wild type RBMXL2 protein ) . Thus buffering protein interactions between RBMXL2 and proteins like Tra2β may be potentially more important in repressing cryptic splicing in meiosis than direct RBMXL2 RNA-protein interactions . Loss of RBMXL2 disrupts expression of many downstream target genes in parallel , both at splicing and transcriptional levels . This makes it difficult to resolve whether the phenotype detected in the Rbmxl2-/- mouse testes is caused by aberrant expression of a single target RNA , or is a compound phenotype involving multiple genes . Some genes affected by RBMXL2 protein deletion are annotated as important for spermatogenesis on the MGD ( Smith et al . , 2018 ) . These include Alms1 ( Arsov et al . , 2006; Collin et al . , 2005 ) ; Brca2 which is required for meiotic prophase ( Sharan et al . , 2004 ) ; Kdm4d , which helps control apoptosis during male germ cell development ( Iwamori et al . , 2011 ) ; and Cul4a that encodes an important ubiquitin ligase needed for DNA damage response . Two factors complicate direct comparison of MGD phenotypes with those after RBMXL2 knockout . Firstly , conventional knockouts will report a phenotype the first time a gene is needed in a developmental pathway . In contrast RBMXL2 is only expressed from the onset of meiosis , so any defects in target gene expression that occur without RBMXL2 could result in a later phenotype . For example , Meioc knockout phenotype analysis has concentrated on entry into meiosis ( Abby et al . , 2016; Soh et al . , 2017 ) . Our data show that Meioc exon five is still spliced normally on meiotic entry in the Rbmxl2-/- mouse ( at 12dpp ) , but becomes compromised later on during meiotic prophase ( when RBMXL2 is expressed ) . It is possible that Rbmx and Rbmy will provide an equivalent function to Rbmxl2 in spermatogonia and early spermatocytes , and that it is only when meiotic sex chromosome inactivation initiates during pachytene that germ cells will depend on Rbmxl2 . Consistent with this , in transfected cells RBMX was able to suppress a cryptic splice exon in the Kdm4d with similar efficiency to the RBMXL2 protein . Secondly , the MGD phenotype data ( Smith et al . , 2018 ) is after gene knockout , whereas Rbmxl2 deletion changes patterns of splice isoforms . High levels of gene expression and alternative splicing have also been observed in the nervous system , and this is another anatomic site where pathological cryptic exon inclusion has been reported . Neuronal cryptic exon inclusion occurs following depletion of the splicing repressor protein TDP43 , and might contribute to neuron death in neurological diseases like ALS and Alzheimer’s disease ( Sun et al . , 2017 ) . Cryptic splicing has also been observed in cultured human cells after depletion of the splicing repressor proteins hnRNP C , hnRNPL and PTB ( Ling et al . , 2016; McClory et al . , 2018; Zarnack et al . , 2013 ) , and in mouse oocytes depleted for the splicing factor SRSF3 ( Do et al . , 2018 ) . Taking all these results into consideration we suggest that RBMXL2 has a key role controlling the meiotic transcriptome , and suggest that human male infertility caused by loss of RBMXL2 or its paralog RBMY may be associated with germ cell type-specific cryptic splicing . A targeting construct in which the Rbmxl2 open reading frame was flanked by LoxP sites was made using standard molecular biology techniques , and electroporated into ES129 cells . Positive clones were injected into blastocysts to create chimaeras , and bred to yield agouti pups heterozygous for the targeted locus ( Ozgene , Perth , Australia ) . The original mice containing the Neomycin gene ( Figure 1—figure supplement 1A ) were crossed to FlpE mice to remove the Neo gene and to generate the Rbmxl2 LoxP conditional allele ( Figure 1—figure supplement 1B ) . Mice containing the Rbmxl2 LoxP conditional allele were crossed with mice expressing PGKCre , resulting in deletion of the Rbmxl2 reading frame ( Figure 1—figure supplement 1C ) . Genetic structures of the wild type and targeted alleles were confirmed by Southern blotting . Genomic DNA was prepared from mouse tails , cut with BamHI , run on an agarose gel and blotted onto a Hybond N nylon membrane ( GE Healthcare , Little Chalfont , UK ) . Blots were probed using an internal probe Enp ( generated using primers enpF 5′-ACTGTTGATTCCCCTTCCAAC-3′ and enp R 5′- ACTCCTGCCTGTGATTGGTC-3′ , and α- 32 P labelled by random priming ) . The correct insertion of the targeting vector in the genome was confirmed by cutting genomic DNA with PshA1 and EcoRV and hybridizing with 5′ ( generated from genomic DNA using 5′-agcattcagcaaaggctcac-3′ and 5′- ttaaaactgagggagactgc-3′ ) and 3′ ( generated from genomic DNA using 5′-actgcatagttgtagccatc-3′ and 5′- tgcattctctttaggctcatttc-3′ ) probes respectively . Animal research was carried with the approval of the Newcastle University animal research ethics committee and the UK Government Home Office ( Home Office project Licence Number PIL 60/4455 ) . Testis/body weight ratios , and sperm counts were measured on a C57Bl6 background . In order to determine sperm counts , the cauda epididymis was dissected in PBS and the sperm were counted in a haemocytometer . Mice were back crossed onto the C57Bl6 background for eight generations , and we used back crossed male mice for subsequent analysis . Testes were homogenized in RLTplus buffer from Qiagen before purification with the Rneasy plus kit ( Qiagen , 74134 ) . The RNA was then re-purified with the kit before Dnase digestion ( Ambion ) . Paired-end sequencing was done for six samples in total ( three biological replicates of 18dpp wild-type and Rbmxl2-/- testis ) using 75 bp reads . Libraries were prepared using TruSeq Stranded mRNA Library Prep Kit ( Illumina ) and sequenced; 75 bp single reads were sequenced on a NextSeq 500 ( Illumina , using the mid output v2 150 cycles kit ) . The base quality of the raw sequencing reads were checked using FastQC . Trimmomatic ( v0 . 32 ) was used to remove adapters and to trim the first twelve bases and the last base at position 76 ( Bolger et al . , 2014 ) . Reads were aligned to the UCSC D3c . 2011 ( GRCm38/mm10 ) assembly of the mouse genome using STAR ( v2 . 5 . 2b ) ( Dobin et al . , 2013 ) . Alternative splicing events were assessed using MAJIQ and VOILA software packages ( Vaquero-Garcia et al . , 2016 ) . Briefly , uniquely mapped , junction-spanning reads were used by MAJIQ to construct splice graphs for transcripts from a custom Ensembl transcriptome annotation and to quantify PSI ( within conditions ) and ΔPSI ( between conditions ) for all local splicing variations ( LSVs ) . The captured LSVs include classical alternative splicing events ( e . g . cassette exons , alternative 5’ splice sites , etc . ) as well as more complex variations ( Vaquero-Garcia et al . , 2016 ) . LSVs with an expected change of greater than 10% were then visualized using VOILA to produce splice graphs , violin plots representing PSI and ΔPSI quantifications , and interactive HTML outputs for changes between wild type and Rbmxl2-/- . Splicing patterns of identified target genes were analysed using the UCSC and IGV mouse genome browsers ( Karolchik et al . , 2014; Thorvaldsdóttir et al . , 2013 ) . A panel of 10 cryptic exons from the Catsperb ( two exons ) , Ccdc60 , Kdm4d , Lrrcc63 , 4933409G03Rik , Tcte2 , Ube2e2 , Fam178b genes ) and their immediately downstream exons were monitored for strength of splicing sequences using Sroogle ( Schwartz et al . , 2009 ) . A panel of 19 exons were analysed for length and translation termination codons content ( 1700074P13Rik , 4833439L19Rik , ccdc60 , Dnah12 , Hmga2 , Map2k5 , Rbks , Tcte2 , Ube2e2 , 4933409G03Rik , 4930500J02Rik , 4932414N04Rik , Fam178b , Ccdc144b , Ston1 , Usp54 , Lca5l , and Catsperb and Lca5l ) . The levels of different mRNA splice forms were detected in DNAse treated total RNA isolated from 18dpp mouse testes . cDNA was synthesised using superscript Vilo ( Invitrogen ) and PCR amplified , followed by either capillary gel electrophoresis ( Qiaxcel ) or agarose gel electrophoresis ( Grellscheid et al . , 2011 ) . Gene-specific primers specific for different splice isoforms are given in Figure 3—source data 4 . HITS-CLIP was performed as previously described ( Licatalosi et al . , 2008 ) using an antibody specific to RBMXL2 ( Ehrmann et al . , 2008 ) . In short , for the CLIP analysis a single mouse testis was sheared in PBS and UV crosslinked . After lysis , the whole lysate was treated with DNase and RNase , followed by radiolabelling and linker ligation . After immunoprecipitation with purified antisera specific to RBMXL2 ( Ehrmann et al . , 2008 ) , RNA bound RBMXL2 was separated on SDS-PAGE . A thin band at the size of 65 kDa ( RBMXL2 migrates at around 50 kDa and MW of 50 nt RNA is about 15 kDa ) was cut out and subject to protein digestion . RNA was recovered and subject to Illumina sequencing , and the reads were aligned to the mouse genome . HITS-CLIP tags mapped to 22 , 438 protein coding genes . HITS-CLIP data were further analysed to extract most frequently occurring pentamer nucleotide sequences within genes ( ncRNA , ORF , 3’UTR , 5’UTR and intron ) . A window between +50 and −50 of the cross link site was analysed ( to avoid crosslinking bias ) , and after correction for average pentamer occurrence in control sequences ( as described in Wang et al . , 2010 ) . Each pentamer was only counted once in the analysed window . The Kdm4d cryptic exon and flanking intron sequences were PCR-amplified from mouse genomic DNA using the cloning primers Kdm4dF ( 5′-AAAAAAAAGAATTCCCACACAGCAAAACCCTCTC-3′ ) and Kdm4dR ( 5′-AAAAAAAAGAATTCGCCACCTTTTGCTATTCCTTT-3′ ) . The PCR products were digested with EcoR1 restriction enzyme and cloned into the pXJ41 vector using the Mfe1 site midway through the 757 nucleotide β-globin intron ( Bourgeois et al . , 1999 ) . Analysis of splicing patterns transcribed from minigenes was carried out in HEK293 cells as previously described ( Grellscheid et al . , 2011; Venables et al . , 2005 ) using primers within the β-globin exons of pXJ41 , pXJ41F 5′-GCTCCGGATCGATCCTGAGAACT-3′ ) and pXJ41R ( 5′-GCTGCAATAAACAAGTTCTGCT-3′ ) . For protein detection by immunohistochemistry , testes were fixed in Bouins , embedded in paraffin wax , sectioned and attached to glass slides . Sections were prepared and immunohistochemistry carried out as previously described ( Grellscheid et al . , 2011 ) . Primary antibodies were specific to caspase 3 , p17-specific ( Proteintech Europe Ltd ) . Protein detection by Western was as previously described ( Grellscheid et al . , 2011 ) using primary antibodies specific to RBMXL2 ( Ehrmann et al . , 2008 ) , Meioc ( Abby et al . , 2016 ) , and control antibodies specific to tubulin ( Sigma T6793 ) and GAPDH ( Abgent , San Diego CA , AP7873b ) . The tunica was removed from testes , and seminiferous tubules macerated with razor blades in a small volume of DMEM media . DMEM was added to 4 ml , and debris allowed to settle for 10 min . Cells were then pelleted from the suspension at 233 g for 5 min . The pellet was resuspended in 1–2 ml DMEM per animal . 5 drops of 4 . 5% sucrose were added to the centre of each clean glass slide using a Pasteur pipette . A single drop of the cell suspension was added to each slide from a height of 15–20 cm , followed by one drop of 0 . 05% Triton-X-100 . The slides were incubated in a humid chamber for 10 min , then eight drops of fixative ( 2% PFA , 0 . 02% SDS , pH 8 ) were added to each slide , and incubated in humid chamber for 20 min . Slides were dipped in water to wash and allowed to air dry before storing at −80°C ( Crichton et al . , 2017 ) . Slides were blocked for one hour in 0 . 15% BSA , 0 . 1% Tween-20 , 5% goat serum PBS , then incubated with primary antibodies for three hours at room temperature or 4°C overnight . Primary antibodies were used at the following concentrations , diluted in block solution: rabbit anti-Sycp3 ( 1:500 , Lifespan Cat# LS-B175 RRID:AB_2197351 ) , mouse anti-Sycp3 ( 1:500 , Abcam Cat# ab97672 RRID:AB_10678841 ) , guinea pig anti-Sycp1 ( 1:200 , from Howard Cooke ) , mouse anti-γH2AX ( 1:3000; Millipore Cat# 05–636 , RRID:AB_309864 ) , rabbit anti-H3K9me3 ( 1:500; Abcam Cat# ab8898 , RRID:AB_306848 ) , mouse anti-phospho H3 ( 1:1000; Abcam Cat# ab14955 , RRID:AB_443110 ) , human anti-centromere ( 1:50; Antibodies Incorporated , Cat#15-235-0001 ) , mouse anti-AcSmc3 ( 1:1000 , a gift from Katsuhiko Shirage [Nishiyama et al . , 2010] ) . Slides were washed three times for five minutes in PBS before incubating with 1:500 secondary antibodies and ng/ul DAPI ( 4' , 6-diamidino-2-phenylindole ) for one hour at room temperature in darkness . Slides washed again three times and mounted in 90% glycerol , 10% PBS , 0 . 1% p-phenylenediamine . Images were captured with Micromanager imaging software using an Axioplan II fluorescence microscope ( Carl Zeiss ) equipped with motorised colour filters . Centromere and H3K9me3 staining was imaged by capturing z-stacks using a piezoelectrically-driven objective mount ( Physik Instrumente ) controlled with Volocity software ( PerkinElmer ) . These images were deconvolved using Volocity , and a 2D image generated in Fiji . Nuclei were staged by immunostaining for the axial/lateral element marker SYCP3 . Pachytene nuclei were identified in this analysis by complete co-localisation of SYCP3 and either the transverse filament marker SYCP1 on all nineteen autosomes , or by the presence of nineteen bold SYCP3 axes in addition to two paired or unpaired sex chromosome axes of unequal length . Asynapsed pachytene nuclei were judged to be nuclei containing at least one fully synapsed autosome , and at least one autosome unsynapsed along >50% of its length ( Crichton et al . , 2017 ) . Quantification and statistical analyses were done using Graphpad ( http://www . graphpad . com/ ) . All scoring of immunostained chromosome spreads was performed blind on images after randomisation of filenames by computer script . Three Rbmxl2-/- and three wild type control animals were analysed for each experiment . Meiotic substaging analysis involved 179 nuclei from control animals , and 138 from Rbmxl2-/- mice . Diplotene phospho-H3 analysis involved 128 control and 157 Rbmxl2-/- nuclei , and the proportion of nuclei with bold centromeric signal was analysed using a Fisher’s test . Pachytene asynapsis was measured in 108 control and 92 Rbmxl2-/- nuclei , and the data analysed using a Fisher’s test . Centromere positioning and H3K9me3 staining was assessed across 73 control and 65 Rbmxl2-/-pachytene nuclei , and 66 diplotene control and 66 Rbmxl2-/- diplotene nuclei . These data were analysed using Wilcoxon Rank Sum test .	In humans and other mammals , a sperm from a male fuses with an egg cell from a female to produce an embryo that may ultimately grow into a new individual . Sperm and egg cells are made when certain cells in the body divide in a process called meiosis . Many proteins are required for meiosis to happen and these proteins are made using instructions provided by genes , which are made of a molecule called DNA . The DNA within a gene is transcribed to make molecules of ribonucleic acid ( or RNA for short ) . The cell then modifies many of these RNAs in a process called splicing before using them as templates to make proteins . During splicing , segments of RNA known as introns are discarded and other segments termed exons are joined together . Some exons may also be removed from RNAs in different combinations to create different proteins from the same gene . A protein called RBMXL2 is able to bind to RNA molecules and is only made during and after meiosis in humans and most other mammals . RBMXL2 can also bind to other proteins that are known to be involved in controlling splicing of RNAs , but its role in splicing remains unclear . To address this question , Ehrmann et al . studied the gene that encodes the RBMXL2 protein in mice . Removing this gene prevented male mice from being able to make sperm . Further experiments using a technique called RNA sequencing showed that the RBMXL2 protein helps to ensure that splicing happens correctly by preventing bits of exons and introns in mouse genes from being rearranged . These findings suggest that the gene encoding RBMXL2 is part of a splicing control mechanism that is important for making sperm and egg cells . The work of Ehrmann et al . could eventually help some couples understand why they have problems conceiving children . Male infertility is poorly understood , and not knowing its causes can harm the mental health of affected men . Furthermore , these findings may help researchers to understand the role of a closely related protein called RBMY that has also been linked to infertility in men , but is much more difficult to study .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2019	An ancient germ cell-specific RNA-binding protein protects the germline from cryptic splice site poisoning
The medial temporal lobes play an important role in episodic memory , but over time , hippocampal contributions to retrieval may be diminished . However , it is unclear whether such changes are related to the ability to retrieve contextual information , and whether they are common across all medial temporal regions . Here , we used functional neuroimaging to compare neural responses during immediate and delayed recognition . Results showed that recollection-related activity in the posterior hippocampus declined after a 1-day delay . In contrast , activity was relatively stable in the anterior hippocampus and in neocortical areas . Multi-voxel pattern similarity analyses also revealed that anterior hippocampal patterns contained information about context during item recognition , and after a delay , context coding in this region was related to successful retention of context information . Together , these findings suggest that the anterior and posterior hippocampus have different contributions to memory over time and that neurobiological models of memory must account for these differences . The medial temporal lobes ( MTL ) are known to play a key role in the formation of lasting memories , but there has been considerable debate about whether their involvement in memory retrieval is stable over time . Some models have suggested that the hippocampal formation ( HF ) is critical for supporting new memories , but that , over time , memories can be supported by neocortical areas alone . In one account , the shift from hippocampal to cortical representation reflects a transfer of the memory trace through a time-dependent process known as systems consolidation ( Squire , 1992; Alvarez and Squire , 1994 ) . This type of transfer is thought to preserve the quality and contents of the memory , although older memories may generally be weaker . Other accounts , however , have argued that changes in hippocampal vs cortical involvement are accompanied by the transformation of episodic memories , which require the HF , into semantic memories , which lack episodic context information and can be supported by cortex alone ( Nadel and Moscovitch , 1997; Winocur et al . , 2010 ) . Changes in the neural bases of memory have typically been described as unfolding over very long timescales , but some studies have documented changes even across relatively short delays . For instance , some functional magnetic resonance imaging ( fMRI ) studies have shown that , when both encoding and retrieval are controlled in the laboratory , retrieval-related activity in the HF declines from immediate to delayed test with even a 1-day delay ( Bosshardt et al . , 2005a; Takashima et al . , 2006 , 2009; but see Stark and Squire , 2000 ) . These findings are echoed by work demonstrating that a night of sleep , or even a brief nap , can alter the neural bases of memory ( Takashima et al . , 2006; Gais et al . , 2007; Sterpenich et al . , 2007 ) and have long-term consequences for memory ( Diekelmann and Born , 2010 ) . Although these findings are often interpreted as reflecting the early stages of memory systems consolidation , it has remained a challenge to separate changes in neural representation from concomitant changes in episodic quality or content . For instance , one study found that differences in HF activity for recent vs remote autobiographical retrieval could be explained by differences in memory vividness ( Gilboa et al . , 2004 ) . One way to control for these differences is to limit analysis to memories endorsed with high confidence or recollection ( Takashima et al . , 2006; Sterpenich et al . , 2009; Takashima et al . , 2009; Milton et al . , 2011 ) . However , even this approach could be insensitive to differences in the kinds of details that accompany recollection , such as information related to episodic context . The HF is especially involved in tasks that require retrieval of contextual details ( Davachi , 2006; Eichenbaum et al . , 2007; Montaldi and Mayes , 2010; Ranganath , 2010 ) , so differences between HF and cortical contributions to memory over time could be due to changes in contextual retrieval . Another challenge to understanding changes in the neural bases of memory is that current models have not accounted for heterogeneity of function within the MTL . In particular , the perirhinal cortex ( PRC ) and parahippocampal cortex ( PHC ) are critically involved in episodic memory , yet they affiliate with different large-scale cortical networks ( Libby et al . , 2012; Ranganath and Ritchey , 2012; Ritchey et al . , 2014 ) and are widely believed to be functionally distinct from each other and from the HF ( Davachi , 2006; Eichenbaum et al . , 2007; Montaldi and Mayes , 2010; Norman , 2010; Ranganath , 2010 ) . Nonetheless , current models have been vague with respect to their predictions for the PRC and PHC , either grouping them alongside the HF ( Squire , 1992; McClelland et al . , 1995; Nadel and Moscovitch , 1997 ) or ignoring them altogether ( Winocur et al . , 2010 ) . The PRC and PHC also functionally interact with different pathways along the longitudinal axis of the HF ( Kahn et al . , 2008; Poppenk and Moscovitch , 2011; Libby et al . , 2012 ) , suggesting additional heterogeneity within the HF . For instance , the anterior and posterior HF are thought to play different roles in memory for spatial context ( Moser and Moser , 1998; Fanselow and Dong , 2010; Poppenk et al . , 2013 ) , with coarse context coding in the anterior HF and specific place coding in the posterior HF ( Poppenk et al . , 2013; Evensmoen et al . , 2014 ) . Because memories can lose contextual specificity over time ( Wiltgen and Silva , 2007; Winocur et al . , 2007 ) , differences in the scale of context processing in the MTL might be associated with differences in MTL contributions to memory over time . Despite this heterogeneity , prior imaging studies have not systematically investigated time-dependent differences in recruitment of the anterior and posterior HF , PRC , and PHC during retrieval . Thus , an important next step is to clarify the roles of the anterior and posterior HF and parahippocampal areas in supporting memory over time . The goal of the present study was to use fMRI to examine changes in MTL activity during immediate and delayed item recollection . Across 2 days , participants encoded sentences , each of which described an association between an item and a room in a house , such that each item was associated with one of eight contexts ( Figure 1A ) . Immediately after the second encoding session , participants were scanned while completing an item recognition test . To evaluate responses related to item recollection and context memory , we first compared the overall magnitude of recollection-related activity for each delay , which allowed us to determine whether changes in MTL involvement are observed even when controlling for subjective recollection . Next , we leveraged a novel multi-voxel pattern similarity analysis approach ( Kriegeskorte , 2008 ) that measured the sensitivity of voxel patterns in MTL subregions to information about shared study context ( c . f . , [Hannula et al . , 2013; Hsieh et al . , 2014] ) , thereby providing an objective measure of context reactivation . This analysis allowed us to determine whether a region's continued involvement in recollection is related to its representation of context information in memory . Moreover , we used a region-of-interest ( ROI ) approach to separately examine the properties of the anterior and posterior HF , PRC , and PHC ( Figure 1B ) , thus shedding new light on the regional specificity of memory changes within the MTL . 10 . 7554/eLife . 05025 . 003Figure 1 . Task design and regions of interest . ( A ) Overview of the experimental design . Here , all fMRI analyses are conducted on data from the item recognition phase . ( B ) ROIs included in the main analyses , including the anterior hippocampus ( ant . HF ) , posterior hippocampus ( post . HF ) , perirhinal cortex ( PRC ) , and parahippocampal cortex ( PHC ) . Coronal MRI slices show manually-traced ROIs from a representative subject , resliced to functional resolution and warped into MNI space for display on a template brain . DOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 003 Behavioral data are presented in Table 1 . Item memory was evaluated by comparing discriminability for old vs new items in the item recognition test . Item memory was above chance for both immediate , t ( 27 ) = 22 . 16 , p < 0 . 001 , and delayed test , t ( 27 ) = 18 . 85 , p < 0 . 001 , and , not surprisingly , accuracy was higher for items tested immediately than after a delay , t ( 27 ) = 8 . 54 , p < 0 . 001 . This difference was observed for both recollection , t ( 27 ) = 9 . 51 , p < 0 . 001 , and familiarity , t ( 27 ) = 5 . 06 , p < 0 . 001 , contributions to item recognition . Context memory was evaluated by comparing discriminability of intact and recombined sentences in the associative recognition test . Participants successfully discriminated between intact and recombined sentences for both immediate , t ( 26 ) = 8 . 22 , p < 0 . 001 , and delayed test , t ( 26 ) = 8 . 17 , p < 0 . 001 , and again , associative recognition accuracy was higher for sentences tested immediately than after a delay , t ( 26 ) = 5 . 13 , p < 0 . 001 . Consistent with the idea that item recollection might involve the reactivation of associative information , associative recognition accuracy was higher for recognized items that were associated with a ‘remember’ response than for recognized items that were not , F ( 1 , 26 ) = 14 . 20 , p < 0 . 001 . The difference in source accuracy did not interact with delay , F ( 1 , 26 ) = 0 . 27 , p = 0 . 61 . 10 . 7554/eLife . 05025 . 004Table 1 . Behavioral resultsDOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 00410 . 7554/eLife . 05025 . 005Table 1—source data 1 . Behavioral data . DOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 005Item recognition‘R’ rate‘4’ or ‘5’ rated′RecollectionFamiliarityImmediate0 . 46 ± 0 . 220 . 40 ± 0 . 221 . 79 ± 0 . 430 . 44 ± 0 . 220 . 46 ± 0 . 18Delayed0 . 25 ± 0 . 190 . 43 ± 171 . 14 ± 0 . 320 . 22 ± 180 . 32 ± 0 . 13Novel0 . 04 ± 0 . 050 . 23 ± 0 . 09–––Associative recognition‘intact’ rated′% correct for ‘R’ responses% correct for ‘4’ or ‘5’ responsesImmediate intact0 . 78 ± 0 . 131 . 34 ± 0 . 5973 . 5 ± 16 . 668 . 6 ± 14 . 8Delayed intact0 . 64 ± 0 . 170 . 59 ± 0 . 3865 . 1 ± 18 . 457 . 7 ± 11 . 4Immediate recombined0 . 34 ± 0 . 14–––Delayed recombined0 . 43 ± 0 . 13–––Note: Summary statistics for individual subjects are contained in Table 1—source data 1 . During the item recognition test , participants were faster to correctly recognize items from the immediate list than from the delayed list , F ( 1 , 26 ) = 13 . 87 , p = 001 . Importantly , this difference did not interact with memory status , F ( 1 , 26 ) = 0 . 06 , p = 0 . 80: that is , the delay effect was observed both for recognized items accompanied by a ‘remember’ response ( immediate: 1 . 35 ± 0 . 23 s; delayed: 1 . 40 ± 0 . 24 s ) and for those that were not ( immediate: 1 . 64 ± 0 . 35 s; delayed: 1 . 70 ± 0 . 29 s ) . The lack of interaction suggests that response time differences cannot account for delay-dependent activation changes that are specific to recollection . During the associative recognition test , participants were also faster to correctly recognize intact sentences from the immediate list ( 2 . 01 + 0 . 37 s ) than from the delayed list ( 2 . 24 + 0 . 51 s ) , t ( 26 ) = 3 . 59 , p = 0 . 001 . We first tested for delay-dependent differences in recollection-related activity during item recognition . Mean activity estimates were extracted from anatomical ROI masks of the anterior HF , posterior HF , PRC , and PHC ( Figure 1B ) , then compared with a memory ( recollection , familiarity ) x delay ( immediate , delayed ) repeated-measures ANOVA . Within each of these ROIs , activity was greater for recollection than familiarity trials ( all Fs > 4 . 6 , all ps < 0 . 046; Figure 2 ) . This recollection effect was stable over time in bilateral PRC , PHC , and anterior HF ( all interaction Fs < 1 . 67 , ps > 0 . 21 ) . However , in the posterior HF , the recollection effect interacted with delay ( left: F ( 1 , 18 ) = 5 . 68 , p = 0 . 028; right: F ( 1 , 18 ) = 3 . 93 , p = 0 . 063 ) , such that the posterior HF was more active for immediate than delayed recollection trials ( left: F ( 18 ) = 5 . 61 , p = 0 . 029; right: F ( 18 ) = 5 . 75 , p = 0 . 028 ) , with no concomitant change in familiarity trials , Fs < 1 . 26 , ps > 0 . 27 . No region showed delay-dependent changes in familiarity estimates ( Supplementary file 1 ) . The apparent difference in delay effects between the anterior and posterior HF was borne out as an ROI by delay interaction , F ( 1 , 18 ) = 5 . 80 , p = 0 . 027 , indicating that these areas are dissociable on the basis of their contributions in recollection over time . 10 . 7554/eLife . 05025 . 006Figure 2 . Recollection-related activity in the MTL . Univariate estimates of recollection-related activity ( i . e . , the difference in activation for recollection and familiarity trials ) for left hemisphere MTL ROIs ( similar results hold for right hemisphere ROIs; see Figure 2—figure supplement 1 ) . Asterisk ( * ) denotes a significant interaction between delay ( immediate , delayed ) and memory status ( recollection , familiarity ) , p < 0 . 05 . Error bars denote the standard error of the mean . See Figure 2—figure supplement 2 for results from a model in which the number of recollection and familiarity trials were matched between delays . Summary statistics for individual subjects are contained in Figure 2—source data 1 , and group-averaged activity estimates for all conditions can be found in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 00610 . 7554/eLife . 05025 . 007Figure 2—source data 1 . Activation estimates for MTL ROIs . DOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 00710 . 7554/eLife . 05025 . 008Figure 2—figure supplement 1 . Recollection-related activity in right-hemisphere MTL ROIs . DOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 00810 . 7554/eLife . 05025 . 009Figure 2—figure supplement 2 . Recollection-related activity in a model controlling for the number of trials between delays . DOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 009 Because recognition accuracy was higher for items tested immediately than after a delay , findings of delay-dependent differences might be confounded by differences in the numbers of recollection and familiarity trials contributing to the analysis . To control for this potential confound , all comparisons were re-analyzed using a model in which trials were randomly sampled to match numbers of recollection and familiarity across delays . These analyses replicated the MTL ROI findings described above ( Figure 2—figure supplement 2 ) . Many studies have shown that recollection is also associated with enhanced activation within an extended neocortical network outside of the MTL ( Spaniol et al . , 2009; Ranganath and Ritchey , 2012; Rugg and Vilberg , 2013 ) , sometimes referred to as the ‘core recollection network’ ( Johnson and Rugg , 2007 ) . To test whether recollection-related activity in this network was modulated by delay , we conducted ROI analyses for the retrosplenial cortex , posterior cingulate , precuneus , angular gyrus , and medial prefrontal cortex . In the left hemisphere , all of these regions showed a main effect of memory , all Fs > 9 . 5 , ps < 0 . 007 , with no significant interactions with delay , Fs < 2 . 21 , ps > 0 . 15 ( Figure 3A ) . In the right hemisphere , recollection-related activity in the right precuneus and posterior cingulate declined across the delay ( Figure 3—figure supplement 1 ) . However , recollection effects in these regions were weaker in general , consistent with previous findings that cortical activity associated with recollection of verbal materials tends to be strongest in the left hemisphere ( Yonelinas et al . , 2005 ) . Exploratory whole-brain , voxel-wise comparisons revealed that both immediate and delayed recollection were associated with activity in the recollection network ( corrected p < 0 . 05; Figure 3B ) , with no significant delay-dependent differences in recollection-related activity ( corrected p < 0 . 05 ) . To better define where recollection-related activity was insensitive to delay , we identified regions that were conjointly involved in immediate and delayed recollection ( corrected joint p < 0 . 05 ) while excluding voxels showing even small delay differences ( liberally defined at p < 0 . 05 uncorrected ) . All neocortical regions within the recollection network contained clusters that survived this approach ( Figure 3—figure supplement 2 ) . These results suggest that , for the most part , recollection-related responses in the neocortical recollection network were maintained across the delay . 10 . 7554/eLife . 05025 . 010Figure 3 . Recollection-related activity in the cortical recollection network . ( A ) Univariate estimates of recollection-related activity , that is , the difference in activation for recollection and familiarity trials , for cortical ROIs in the recollection network . Results for left-hemisphere ROIs are shown ( see Figure 3—figure supplement 1 for right-hemisphere ROIs ) . Error bars denote the standard error of the mean . Note that although the precuneus appears to show a reduction in recollection-related activity over time , the interaction was not significant . Summary statistics for individual subjects are contained in Figure 3—source data 1 . ( B ) Voxel-wise maps of recollection-related activity , that is , the difference between recollection and familiarity trial activity , thresholded to display significant clusters ( voxel-wise p < 0 . 001 , cluster-corrected p < 0 . 05 ) . Maps are displayed separately for immediate and delayed recollection . Surface images were rendered in Caret using the PALS atlas ( left hemisphere shown; see Figure 3—figure supplement 1 for right hemisphere ) . Peaks are reported in Supplementary file 2 . The conjunction of immediate and delayed recollection-related activity is shown in Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 01010 . 7554/eLife . 05025 . 011Figure 3—source data 1 . Activation estimates for cortical recollection network ROIs . DOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 01110 . 7554/eLife . 05025 . 012Figure 3—figure supplement 1 . Recollection-related activity in the right hemisphere of the cortical recollection network . DOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 01210 . 7554/eLife . 05025 . 013Figure 3—figure supplement 2 . Conjunction of immediate and delayed recollection-related activity . DOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 013 Results from these univariate analyses indicate that differences in brain activity associated with immediate and delayed recollection varied across MTL subregions . The anterior HF and cortical MTL areas maintained their contributions to recollection over time , whereas posterior HF effects were sensitive to delay . Thus , even when memories at both delays were endorsed with recollection , there were changes in posterior HF involvement in memory recognition . Our next analyses tested whether MTL activity patterns during item recognition carried information about the context ( i . e . , room ) that had been associated with the item at encoding . As depicted in Figure 4A , multi-voxel patterns within each ROI were estimated for every recollection trial , and pairs of trials were compared by calculating the similarity between their associated voxel patterns . Similarity values were then summarized according to whether the items had shared context information during encoding ( same room from the same study list: for example , ‘the apple is in the bedroom’ and ‘the pencil is in the bedroom’ ) or had not shared information ( different rooms from the same study list: for example , ‘the apple is in the bedroom’ and ‘the chair is in the kitchen’ ) . Because there was no context information present during the item recognition phase , any pattern similarity differences between these pair types must be ascribed to the reactivation of context information from memory . Thus , context similarity was defined as the difference in pattern similarity between same-room and different-room pairs , evaluated separately for each ROI . Note that because the locations were typical rooms in a house , this form of context retrieval may reflect a mixture of both spatial and semantic context information ( i . e . , remembering the general location or semantic features of the room associated with the item ) . 10 . 7554/eLife . 05025 . 014Figure 4 . Context similarity in anterior MTL during recollection . ( A ) Schematic of the pattern similarity analysis procedure . ( B ) Estimates of pattern similarity ( Pearson's r ) for same-room and different-room pairs are plotted for the left anterior HF and left PRC ( see Figure 4—figure supplement 1 for other regions ) . Asterisk ( * ) denotes a significant effect of context similarity , that is , the difference in similarity for same- and different-room pairs , p < 0 . 05 . Error bars denote the standard error of the mean . The cross ( ✝ ) denotes a marginally significant effect , p < 0 . 08 . Summary statistics for individual subjects are contained in Figure 4—source data 1 . A non-parametric randomization test confirmed that same-room similarity was greater than what was likely to be observed by chance ( Figure 4—figure supplement 2 ) . Furthermore , these effects were observed only for recollection trials , not familiarity trials ( Figure 4—figure supplement 3 ) . ( C ) The relationship between associative recognition accuracy ( d′ ) and context similarity in the left anterior HF . DOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 01410 . 7554/eLife . 05025 . 015Figure 4—source data 1 . Pattern similarity estimates . DOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 01510 . 7554/eLife . 05025 . 016Figure 4—figure supplement 1 . Context similarity effects in all MTL ROIs . DOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 01610 . 7554/eLife . 05025 . 017Figure 4—figure supplement 2 . Randomization test confirming context similarity effects in the anterior MTL . DOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 01710 . 7554/eLife . 05025 . 018Figure 4—figure supplement 3 . Context similarity in the anterior MTL during familiarity . DOI: http://dx . doi . org/10 . 7554/eLife . 05025 . 018 Because only a few studies ( Staresina et al . , 2012; Hannula et al . , 2013; Hsieh et al . , 2014 ) have shown that MTL multi-voxel patterns carry information about previously learned context information , even at immediate recall , our first analyses sought to establish the presence of pattern information related to context similarity during immediate recollection , when context memory was strongest . In the left anterior HF , pattern similarity was significantly greater among same-room pairs than different-room pairs , t ( 18 ) = 2 . 34 , p = 0 . 015 ( Figure 4B ) , and a similar trend was observed in the left PRC , t ( 18 ) = 1 . 73 , p = 0 . 051 . There were no significant effects for any of the other MTL regions , ts < 1 , ps > 0 . 18 ( Figure 4—figure supplement 1 ) . Context similarity effects were also absent from cortical regions outside of the MTL , either immediately or after a delay , ts < 1 . 8 , ps > 0 . 05 , suggesting that these effects were selective to the left anterior MTL . The MTL findings were verified with a randomization test showing that pattern similarity for same-room pairs exceeded what would be expected by chance if context information were randomly assigned ( Figure 4—figure supplement 2 ) . Additionally , when the same analysis was run with familiarity trials instead of recollection trials , no differences were observed between same and different context trial pairs , ts < 1 , ps > 0 . 2 ( Figure 4—figure supplement 3 ) . This distinction is consistent with the claim that pattern similarity differences were related to the reactivation of shared context information , which should be more evident during memory recollection . After determining that left anterior MTL regions showed evidence for context similarity at the immediate test , we next tested whether context similarity effects in these regions changed over time . The left PRC showed a significant main effect of context similarity across both delays , F ( 1 , 18 ) = 5 . 78 , p = 0 . 027 , and context similarity did not interact with delay , F ( 1 , 18 ) = 0 . 54 , p = 0 . 47 . In fact , the PRC showed a marginal context similarity effect for the delayed list as well , t ( 18 ) = 1 . 54 , p = 0 . 07 , suggesting that the sensitivity of this region to context information was stable over time . The anterior HF , on the other hand , showed neither a significant main effect , F ( 1 , 18 ) = 2 . 99 , p = 0 . 10 , nor interaction , F ( 1 , 18 ) = 1 . 13 , p = 0 . 30 . The absence of a clear main effect or interaction may be due , in part , to substantial variability in context similarity effects during delayed recollection . One possibility is that this variability is related to individual differences in retention of context information in memory . To test this hypothesis , we correlated individual subjects' context similarity estimates ( i . e . , the voxel pattern differences between same- and different-context trial pairs ) with their performance on the subsequent associative recognition test , based on the idea that the associative recognition test relied on the same kind of room information measured by the pattern similarity analysis . Indeed , participants who performed best on the associative recognition test also showed the largest context similarity effect in the left anterior HF during delayed recollection , r = 0 . 53 , t ( 16 ) = 2 . 53 , p = 0 . 022 ( Figure 4C ) . This relationship was significant even after controlling for individual differences in the number of recollection trials , t ( 15 ) = 2 . 81 , p = 0 . 013 . For the PRC , the correlation was in the same direction but not significant , r = 0 . 25 , t ( 16 ) = 1 . 02 , p = 0 . 32 . Finally , there was no significant relationship between anterior HF context similarity and associative recognition for the immediate list , r = −0 . 17 , t ( 16 ) = 0 . 71 , p = 0 . 49; however , there was little variability in the similarity estimates for these trials . These findings provide strong evidence that , after a delay , the context similarity analysis was picking up on meaningful associative information during the item recollection period . Moreover , they are consistent with the memory transformation account , in that anterior HF involvement in context coding over time was contingent on the retention of context information in memory . Both the standard consolidation model and memory transformation accounts predict that HF activity during retrieval should decline with increasing retention intervals , although they assign different explanations to this decline . To our knowledge , neither account makes explicit predictions about variability along the longitudinal axis of the HF . In the present study , the anterior and posterior HF had dissociable responses during immediate and delayed retrieval . Whereas recollection-related activity in the posterior HF decreased over time , activity in the anterior HF was relatively stable across the 1-day delay . In previous fMRI studies that used paradigms comparable to the one used here , HF results have been somewhat mixed: some studies have shown that HF activity was greater for early than delayed retrieval ( Takashima et al . , 2006; Sterpenich et al . , 2009; Takashima et al . , 2009; Yamashita et al . , 2009; Milton et al . , 2011; Watanabe et al . , 2012 ) , but others have reported no change ( Stark and Squire , 2000; Janzen et al . , 2008; Suchan et al . , 2008 ) or even the opposite effect ( Bosshardt et al . , 2005b; Gais et al . , 2007 ) . Some of these differences may be related to the sensitivity of the fMRI analysis to specific memory processes: some studies reporting no change used simple comparisons of targets and foils ( Stark and Squire , 2000 ) or recognized and forgotten trials ( Janzen et al . , 2008 ) , whereas studies that isolated activity for high-confidence , recollection-based or associative hits have often reported delay-dependence ( Takashima et al . , 2006; Sterpenich et al . , 2009; Takashima et al . , 2009; Yamashita et al . , 2009; Milton et al . , 2011; Watanabe et al . , 2012; but see Suchan et al . , 2008 ) . Although some delay effects have been localized to the anterior HF ( Takashima et al . , 2006; Milton et al . , 2011 ) , differences in the posterior HF have been commonly observed for studies using associative memory tasks ( Takashima et al . , 2009; Yamashita et al . , 2009; Watanabe et al . , 2012 ) . By investigating delay effects within anatomically restricted regions , the current approach might have improved our ability to detect differences in localization that might not have been readily apparent in a group analysis applying voxel-wise thresholds . Thus , the finding that changes in recollection-related activity were circumscribed to the posterior HF is consistent with the available literature , but would not have been predicted by extant models . Studies of autobiographical memory have examined differences between recent and remote memories across a more extended timescale ( Cabeza and St Jacques , 2007 ) , and some of these studies have compared the role of the anterior and posterior HF . For example , one study reported that the anterior HF showed a larger delay-dependent difference than the posterior HF during autobiographical memory retrieval ( Gilboa et al . , 2004 ) . Another study used multi-voxel pattern analysis to decode information about autobiographical events . This study reported that patterns in the posterior HF could be used to decode remote but not recent autobiographical memories during repeated retrieval events , whereas patterns in the anterior HF could be used to decode both recent and remote memories ( Bonnici et al . , 2012a , 2013 ) . At face value , these findings might seem to contradict the present results , but there are numerous differences between autobiographical memory tasks and paradigms that focus on laboratory-controlled events . In particular , the use of repeated retrieval events and long retention delays could complicate the interpretation of fMRI studies of autobiographical memory retrieval . During remote autobiographical retrieval , hippocampal activity could be related to the construction of a new memory for an old event , or to the retrieval of the original memory or its more recent reconstructions . Another issue is that differences between recent and remote autobiographical memories can be confounded by differences in vividness and context specificity . Indeed , in the study by Gilboa et al . ( 2004 ) , there were no significant delay-dependent differences in the HF after controlling for vividness . Although laboratory-controlled studies are not immune to this possible confound , the issue was mitigated in the present study by limiting analyses to memory retrieval accompanied by recollection and by identifying neural patterns associated with studied context information . An important question for future research is how changes in neural representation over a 1-day interval relate to the changes over longer intervals that are the focus of memory consolidation and transformation accounts . There is considerable evidence suggesting that early stages of systems consolidation can be initiated during the first night of sleep after learning ( Born and Wilhelm , 2012 ) , but little is known about how initial changes in hippocampal representation are related to longer-lasting changes . It is possible that the changes in activation magnitude and pattern similarity observed here reflect forgetting of certain aspects of context information , and that qualitatively different kinds of changes—particularly in cortical representation—might be apparent over a longer interval . The relative stability of recollection-related activity in the anterior HF may be related to its continued involvement in supporting context memory over time . Because recollection can be triggered by different kinds of associative details , including information about encoding context , we separately examined the sensitivity of the multi-voxel patterns to context information learned during encoding . Multi-voxel patterns in the left anterior HF were sensitive to context similarity during immediate recollection , and after a delay this effect was strongest for participants who could accurately retrieve context associations . According to one view of HF function , the anterior HF may be especially involved in representing episodic context at a global level ( Poppenk et al . , 2013 ) . In support of this view , the anterior HF contains cells with larger place fields than the posterior HF ( Kjelstrup et al . , 2008 ) , and it is involved in coding information about the coarse location of objects ( Evensmoen et al . , 2014 ) , the global position of landmarks ( Ekstrom et al . , 2011; Morgan et al . , 2011 ) , and the gist of memories ( Poppenk et al . , 2008; Gutchess and Schacter , 2012 ) . Taking this view into consideration , it may be that pattern similarity effects in the anterior HF reflect the activation of generalized context information ( e . g . , general features of the room associated with the item at study ) that was shared across many encoding events . As discussed below , the posterior HF might support more specific representations of context ( e . g . , remembering the exact location in which a specific item was imagined in the bedroom ) in a more time-limited way . Regardless of the scale of anterior HF representations , it is noteworthy that recollection-related activity in the anterior HF was relatively stable over the delay , and that the anterior HF continued to show evidence for context coding in participants who retained context information in memory , suggesting that the type of information carried by the anterior HF can be long lasting . These findings are compatible with the memory transformation account , which models posit that the HF should be involved in retrieval so long as its preferred form of mnemonic information is maintained . Irrespective of delay , an important finding of the present study was that multi-voxel patterns carried information about incidentally reactivated context associations during item recollection . A few previous studies have shown that neural patterns present during encoding are reactivated during item recognition ( Johnson et al . , 2009; Ritchey et al . , 2013 ) and cued recall ( Staresina et al . , 2012; Kuhl and Chun , 2014 ) , and that patterns within the MTL carry information related to study context ( Hannula et al . , 2013; Hsieh et al . , 2014 ) . For instance , during cued recall , patterns in the anterior HF , PRC , and PHC were shown to carry information about the study context ( Hannula et al . , 2013 ) . However , there has been little evidence that item recollection involves the spontaneous reactivation of context-related patterns even without an overt source decision , despite the assumption that recollection typically involves contextual retrieval . In one study , Johnson et al . ( 2009 ) showed that a classifier trained to discriminate among study contexts was sensitive to context information that was incidentally reactivated during recognition , and that for the posterior cingulate , context reactivation was especially apparent for items accompanied by recollection . However , this study did not report context reactivation effects within the HF . The present work expands on these findings to demonstrate that , during recollection , left anterior HF patterns carried information about the context associated with an item during encoding , even when context was not explicitly cued or re-presented . The context similarity effect in the HF was relatively small , perhaps because HF patterns are more sensitive to similarities in object–context associations than to context alone ( Hsieh et al . , 2014; Libby et al . , 2014 ) . The finding that context-related pattern similarity was predictive of associative recognition performance , however , provides converging evidence that hippocampal voxel patterns carry behaviorally relevant context information . Nevertheless , future work will be necessary to determine the sensitivity of hippocampal voxel pattern information to other kinds of context manipulations . The PRC had a similar response profile to the anterior HF , in that it was stable in its recollection-related activity over time . It also tended to show context similarity effects that persisted across the delay . The anterior HF is strongly connected with the PRC , which is part of an anterior temporal system thought to be important for processing , remembering , and assigning value to items ( Ranganath and Ritchey , 2012 ) . These two regions may work together to support memory for item or item–context associations , and context similarity effects in the anterior HF may reflect the retrieval of contextual information in response to an item cue . The PRC , on the other hand , has been linked to item recognition ( e . g . , [Davachi et al . , 2003; Ranganath et al . , 2003] ) , recollection of item associations ( Diana et al . , 2010 ) , and semantic processing ( e . g . , [Wang et al . , 2010; Clarke and Tyler , 2014] ) . Thus , the prior literature is most consistent with a role for the PRC in item processing , and the finding of a context similarity effect in the PRC was unexpected . However , there is some evidence that the PRC may additionally carry some information about context , such as the locations of items in space ( Burke and Barnes , 2014 ) . PRC lesions have been shown to disrupt some forms of context memory , including object–context associations ( Norman and Eacott , 2005 ) , positional changes in object arrays ( Norman and Eacott , 2005 ) , and contextual fear ( Bucci et al . , 2002 ) —although these impairments have typically been more circumscribed than those observed following PHC damage ( Norman and Eacott , 2005 ) . Based on this literature and the present data , we cannot rule out the possibility that like the anterior HF , the PRC is involved in the long-term storage of the association between an object and the general context in which it was encountered . Alternatively , it could be that the anterior HF and PRC are both sensitive to shared context information , but for different reasons: whereas the anterior HF might carry general representations of context , pattern similarity effects in the PRC might reflect the recollection of items associated with each room , through episodic associations learned during encoding or through existing semantic associations . In contrast to the anterior HF and PRC , the posterior HF was neither stable in its contributions to recollection over time nor did it show enhanced pattern similarity during retrieval of objects that shared the same contextual associations . As noted above , it is possible that the posterior HF encodes highly specific contextual details ( Poppenk et al . , 2013 ) , such as precise locations within a spatial context ( Kjelstrup et al . , 2008; Evensmoen et al . , 2014 ) or positions within a sequence ( Hsieh et al . , 2014 ) , which might be useful for disambiguating among related contexts . For instance , in one study , hemodynamic responses in the posterior HF were greater during the retrieval of precise rather than coarse location information associated with an object ( Evensmoen et al . , 2014 ) . It is possible that , here , the reduction in posterior HF activity across the delay reflects the forgetting of trial-specific context information . Prior work in rodents has shown that memories may lose their contextual specificity over time ( Wiltgen and Silva , 2007; Winocur et al . , 2007 ) , and that the loss of specificity is associated with diminished dependence on the dorsal HF ( which may be homologous to the posterior HF in humans ) during retrieval ( Wiltgen et al . , 2010 ) . In addition , if the posterior HF codes only specific context information , this could explain the absence of a context similarity effect in this region . By definition , the multi-voxel pattern similarity analysis used here depended on similarities among trials that had been associated with the same ‘room’ , and was therefore insensitive to trial-specific information . Some previous studies have used approaches that have enabled them to identify event- or scene-specific patterns , and these studies have shown that this kind of specific information can be decoded from multi-voxel patterns in the HF ( Chadwick et al . , 2011; Bonnici et al . , 2012b; Hsieh et al . , 2014 ) , and that the specificity of HF patterns is positively related to memory performance ( LaRocque et al . , 2013 ) . In this study , we were unable to directly measure trial-specific information in the posterior HF , but future studies could incorporate graded levels of context specificity in order to test its relation to response changes over time . The role of the MTL in memory has often been contrasted against the role of neocortical areas in memory , which are thought to increase or remain stable in their support of memory over time ( Squire , 1992; Nadel and Moscovitch , 1997; Winocur et al . , 2010 ) . However , most models have excluded the PRC and PHC from their discussion of cortical function ( [Squire , 1992; Nadel and Moscovitch , 1997; Winocur et al . , 2010; but see Norman and O'Reilly , 2003 ) and previous studies have not typically included these regions as ROIs ( but see Stark and Squire , 2000 ) . Here , we provide novel evidence that the roles of the HF and parahippocampal areas are dissociable in terms of their involvement in recognition memory over time . Whereas recollection-related activity in the posterior HF declined from immediate to delayed retrieval , activity in the PRC and PHC remained stable . One interesting point of divergence is between the posterior HF and PHC , which are strongly interconnected ( Ranganath and Ritchey , 2012 ) . Like the posterior HF , the PHC has been implicated in memory for specific context information ( Davachi , 2006; Eichenbaum et al . , 2007; Ranganath , 2010 ) , and the PHC has also been shown to carry information about scene-specific associations ( e . g . , Staresina et al . , 2012 ) . Here , patterns in both regions were insensitive to shared context information , yet the PHC was involved in memory recollection after a 1-day delay , whereas the posterior HF was not . Future work should address whether PHC representations simply remain more stable over time , as compared with those in the posterior HF , or whether there are other differences that might explain their different activation profiles . Beyond the MTL , across both delays , recollection-related activity was also observed in a network of regions including the retrosplenial cortex , posterior cingulate , precuneus , angular gyrus , and medial prefrontal cortex . This network is thought to be important for context memory and recollection ( Bar , 2004; Ranganath and Ritchey , 2012; Rugg and Vilberg , 2013 ) . Importantly , these findings suggest that most areas involved in recollection , whether they are within the MTL or not , maintain their involvement across a 1-day delay . Altogether , these findings clearly demonstrate that neurobiological models of memory must go beyond simple dichotomies between the MTL and neocortex to address the role of specific HF and parahippocampal areas . The present results raise several questions to be addressed in future studies . One question involves the role of the PRC and PHC in immediate and delayed recollection . Here , we found that both regions supported recollection immediately and after a 1-day delay , but that patterns in the PRC but not the PHC were sensitive to shared context information . Future work should address whether the PRC and PHC are similarly involved in memory tested after longer intervals and with measures that are more sensitive to the specificity of information retained in memory . Additionally , because memory was tested after a 1-day delay , we cannot disentangle changes attributable to active sleep-dependent processes ( Ellenbogen et al . , 2006 ) from more passive time-dependent changes . Finally , an important next step would be to compare the item–location associations used here with associations that may be less dependent on HF function , such as unitized associations ( Giovanello et al . , 2006; Quamme et al . , 2007 ) or emotional associations . In particular , emotional memories are forgotten more slowly than neutral memories ( e . g . , Sharot and Yonelinas , 2008 ) , and some evidence suggests that the persistence of emotional recollection is related to the function of anterior MTL structures ( Ritchey et al . , 2008 ) . In summary , this study provided novel evidence that regions within the MTL play different roles in supporting item recollection over time . The results highlight the need to revise existing models to incorporate differences between MTL areas . In particular , it will be important for models to distinguish between the anterior and posterior HF , which may show different changes in their contributions to memory over time . Data were acquired from 30 young adults ( 15 female; ages 18–31 years ) . Data from one participant was excluded due to button box issues during the scan , and data from another participant was excluded due to head motion and poor performance . Of the remaining 28 participants , 9 were excluded from fMRI data analyses due to insufficient variability in memory performance ( i . e . , fewer than 9 recollection or familiarity trials ) . Thus , the fMRI analyses included 19 participants ( 9 female ) . Due to technical problems , one of these participants completed the item recognition but not associative recognition task . Participants reported that they were native English speakers , free of neurological and psychiatric disorders , and eligible for MRI . Participants reported sleeping , on average , 7 . 42 hr ( range: 5–12 hr ) between the first and second session . Stimuli consisted of 252 nouns that referred to objects . For each participant , these words were randomly assigned to one of three lists ( N = 84 each ) : the Day 1 encoding list ( delayed list ) , the Day 2 encoding list ( immediate list ) , or the lure list for item recognition . During encoding , items were placed into sentences describing the location of the object , which could be in one of eight rooms in a house: bathroom , bedroom , den , dining room , kitchen , living room , office , and patio area . For example , on one trial , a possible sentence might read , ‘The apple is in the bedroom’ . Thus , the room associated with each item constituted its encoding context , which might include a mixture of spatial and semantic information about the room . The eight contexts were randomly assigned to either the immediate or delayed encoding list , such that only four contexts were presented on either day . There were two experimental sessions that occurred on consecutive days ( Figure 1A ) . On Day 1 , participants completed an encoding task . On Day 2 , participants completed another encoding task , an item recognition task , and an associative recognition task . Both encoding tasks took place in the same laboratory testing rooms . The recognition tasks took place in the scanner , with the item recognition task beginning as soon as the participant was positioned within the scanner , approximately 20 min after the end of the Day 2 encoding task . All fMRI analyses focus on the item recognition task . During the two encoding tasks , participants studied sentences in which trial-unique object nouns were paired with one of eight contexts ( Figure 1A ) . The set of items and contexts was different on each day . Different contexts were assigned to each day in order to avoid confounds related to contextual interference across days . During each encoding task , 84 sentences appeared on-screen for 5 s each , separated by jittered fixation intervals ( mean = 4 s , range = 2–10 s ) . Participants were instructed to rate on a continuum how well they were able to imagine the pairing on a 6-point scale , with 1 = not well and 6 = very well . Trial order was randomly determined for each participant . Immediately after each encoding task , participants were cued to group the four previously-studied rooms into two houses , based on random assignment . This grouping manipulation did not alter memory performance and will not be considered further . The item recognition task was designed to assess memory for the items studied during encoding . During this task , participants were presented with words from both encoding lists and the lure list ( Figure 1A ) . Words were presented for 2 s each , separated by jittered fixation intervals ( mean = 4 s , range = 2–10 s ) . Participants were asked to determine whether the word was old ( studied either day ) or new ( unstudied ) using a modified 6-point remember-know scale , including responses for definitely new , probably new , not sure , probably old , definitely old , and remember . For half of the participants , the scale was presented in reverse order . Participants were instructed that they should use the ‘remember’ response any time they could recall any kind of specific detail from when they initially studied that item , whereas the other memory responses reflected graded levels of memory confidence in the absence of a specific detail . We did not explicitly instruct the participants to remember the associated room during the item recognition phase; rather we emphasized that any type of detail would qualify for a ‘remember’ response . This was because we did not want participants to engage in a strategy in which they called to mind the rooms on every trial , which would have interfered with our ability to detect room information arising from memory . Trials from each list were evenly divided across three functional imaging runs , such that each run contained the same number of trials associated with each studied location . A unique sequence of trials and jittered fixation intervals was randomly determined for each participant . The associative recognition task was designed to assess memory for the item–context associations made during encoding . During this task , participants were presented with sentences that were either identical to sentences that they had studied during either encoding session ( ‘intact’ ) or sentences that were recombinations of items and contexts that were both previously studied but not as part of the same sentence ( ‘recombined’ ) . Of the 84 sentences studied on each day , 28 were presented as intact and 56 were presented as recombined . The items and contexts in the recombined sentences were always drawn from the same day list . Sentences were presented for 3 s each , separated by jittered fixation intervals ( mean = 4 s , range = 2–10 s ) . Participants were asked to rate whether the sentence was intact or recombined on a 6-point scale , including response for definitely recombined , probably recombined , guess recombined , guess intact , probably intact , and definitely intact . For half of the participants , the scale was presented in reverse order . Trials from each list were evenly divided across three functional imaging runs , and trial order was randomly determined for each participant . Behavioral analyses were based on the full available sample for each task . Item recognition performance was measured as the discriminability ( d′ ) between old items ( from the immediate or delayed list ) and new items . Item recognition was further broken down into estimates of recollection and familiarity according to the dual-process model of recognition memory . Recollection was defined as ( Rold − Rnew ) / ( 1 − Rnew ) , where Rold is the rate of ‘R’ responses to old items , and Rnew is the rate of ‘R’ responses to new items . Familiarity was defined as ( Fold/ ( 1 − Rold ) ) − ( Fnew/ ( 1 − Fnew ) ) , where Fold is the rate of ‘definitely old’ and ‘probably old’ responses to old items , and Fnew is the rate of ‘definitely old’ and ‘probably old’ responses to new items . Note that these process estimates were computed to allow comparison to previous studies . The primary findings of the study , however , do not depend on assumptions specific to the dual process model . Associative recognition performance was measured as the discriminability ( d′ ) between intact and recombined sentences , separately for the immediate or delayed encoding list . To determine the relation between item recognition and associative recognition , the proportion of correct associative recognition responses was calculated for items previously marked as recollected , familiar , or forgotten . All statistical comparisons on the behavioral data were conducted in R version 3 . 1 . 1 ( http://www . R-project . org ) . Scanning was performed on a Siemens Skyra 3T scanner system with a 32-channel head coil . High-resolution T1-weighted structural images were acquired using a magnetization prepared rapid acquisition gradient echo ( MPRAGE ) pulse sequence ( field of view = 25 . 6 cm , image matrix = 256 × 256 , 208 axial slices with 1 . 0 mm thickness ) . Functional images were acquired using a multi-band gradient echo planar imaging ( EPI ) sequence ( TR = 1220 ms; TE = 24 ms; FOV = 19 . 2 cm; image matrix = 64 × 64; flip angle = 67; multi-band factor = 2; 38 axial slices; voxel size = 3 . 0 × 3 . 0 × 3 . 0 mm ) . SPM8 ( http://www . fil . ion . ucl . ac . uk/spm/software/spm8/ ) was used to pre-process the images , including realignment , normalization , and smoothing . The high-resolution T1 image was skull-stripped via segmentation . Functional images were realigned , correcting for motion , and resliced . Resliced , native-space images served as the basis for the anatomical ROI analyses , in which manually segmented ROIs ( see ROI Definition ) were co-registered to the mean functional . For group voxel-wise analyses , the mean functional was co-registered to the skull-stripped anatomical image , moving all of the functional images in register with the anatomical image . At this point , the anatomical and functional images were warped to a group-derived template generated using diffeomorphic registration ( DARTEL ) and normalized to MNI space . Functional images were smoothed with a 6-mm Gaussian kernel . Skull-stripped anatomical images were also warped and smoothed for use as an explicit mask for subsequent functional analyses . Quality assurance included the identification of ‘suspect’ time-points via the Artifact Detection Tools ( ART; http://www . nitrc . org/projects/artifact_detect ) , defined as time-points marked by greater than 0 . 3 mm in movement or 1 . 3% global mean signal change . One participant was excluded from analysis due to excess motion ( >3 mm ) within the functional runs . The anterior HF ( HF head ) , posterior HF ( HF body and tail ) , PRC , and PHC were manually segmented on the MPRAGE coronal plane according to previously published guidelines ( Insausti et al . , 1998; Franko et al . , 2014 ) . In brief , the most posterior slice of the anterior HF was defined as the last slice containing the gyrus intralimbicus; the posterior HF immediately followed . The anterior extent of the PRC was defined as 2 mm anterior to the limen insula or the most anterior slice in which the collateral sulcus was visible , whichever was more anterior . The most posterior slice of the PRC was defined as 4 mm posterior to the anterior/posterior HF transition . The PHC immediately followed the PRC , and the posterior extent was defined as 2 mm posterior to the appearance of the posterior crus of the fornix . The PRC segmentation included the entire lateral bank and dorsal half of the medial bank of the collateral sulcus . The PHC segmentation included the medial bank of the collateral sulcus , extending to the most medial aspect of the parahippocampal gyrus . Some analyses also included a set of regions outside of the medial temporal lobes , including the retrosplenial cortex , posterior cingulate , precuneus , angular gyrus , and medial prefrontal cortex . These ROIs were labeled with FreeSurfer cortical parcellation tools ( http://surfer . nmr . mgh . harvard . edu/ ) using the Destrieux atlas ( Destrieux et al . , 2010 ) for the following labels: G_cingul-Post-ventral , G_cingul-Post-dorsal , G_precuneus , G_pariet_inf-Angular , and S_suborbital . Segmented brains were co-registered to the mean functional image and split into masks for each ROI ( see Figure 1B for an example set of MTL ROIs ) . Masks were filtered to exclude voxels with low signal , defined as having mean temporal SNR ( calculated across all functional runs ) more than 1 standard deviation below the ROI mean . For visualization in standard space , ROI masks were warped to MNI space and combined across subjects into probabilistic maps . For ROI analyses , models were run on unsmoothed functional images in native space . ROI summary statistics , including pattern similarity estimates , were extracted with in-house scripts ( Source code 1 ) in MATLAB 2009b ( The MathWorks , Inc . , Natick , MA ) , and statistical comparisons were conducted in R version 3 . 1 . 1 ( http://www . R-project . org ) . For voxel-wise analyses , models were run on smoothed functional images in standard MNI space , and statistical comparisons were conducted in SPM8 .	In 1953 , an American man called Henry Molaison underwent surgery to remove the medial temporal lobes of his brain in an effort to cure him of severe epilepsy . After the surgery , his epilepsy was indeed improved , but he was left without the ability to form new memories . His case is now seen as one of the first demonstrations of the medial temporal lobes being involved in memory . Beneath the surface of each medial temporal lobe is a structure called the hippocampus , which is essential for the formation of new memories . However , memories are not stored permanently within the hippocampus: instead they are transferred ultimately to the neocortex , which is the outer layer of the brain . Some neuroscientists believe that the content of memories remains unchanged during this transfer , whereas others argue that memories are stripped of their context—that is , details of when and where they were acquired—before they reach the neocortex . In a brain imaging experiment , Ritchey et al . have now attempted to distinguish between these two possibilities . Volunteers were asked to memorize sentences linking an object to a room , such as ‘the apple is in the bedroom’ , on two occasions 24 hr apart . Immediately after the second session , the volunteers were asked to complete memory tests while lying in the brain scanner . In one test the volunteer was shown a list of objects and asked to identify those objects they could recall seeing in either of the training sessions , and to identify objects they recognised as familiar , even if they could not specifically remember seeing these objects during training sessions . Analysis of the brain imaging data revealed that regions of the medial temporal lobes were more active when the volunteers recalled objects than when they recognised them as familiar . Moreover , for the ‘recall’ responses—in which the volunteers could still retrieve contextual information—the activity of the hippocampus depended on the age of the memories . The anterior ( front ) part of the hippocampus was active when subjects recalled either new memories or memories from 24 hr previously , whereas the posterior ( rear ) hippocampus was active only during the recall of new memories . In addition , patterns of activity observed in the anterior hippocampus could be used to determine which room was previously associated with the object . This suggests that contextual information is retained in the anterior hippocampus , but lost from the posterior hippocampus over time . Overall the results of Ritchey et al . highlight the fact that individual components of the medial temporal lobes , including hippocampal subregions , have distinct roles in the storage of memories , with these roles also changing over time . Moreover , the data lend support to the idea that contextual information may be lost from memories before they reach the neocortex .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Delay-dependent contributions of medial temporal lobe regions to episodic memory retrieval
Apolipoprotein A1 ( apoA1 ) is the major protein component of high-density lipoprotein ( HDL ) and has well documented anti-inflammatory properties . To better understand the cellular and molecular basis of the anti-inflammatory actions of apoA1 , we explored the effect of acute human apoA1 exposure on the migratory capacity of monocyte-derived cells in vitro and in vivo . Acute ( 20–60 min ) apoA1 treatment induced a substantial ( 50–90% ) reduction in macrophage chemotaxis to a range of chemoattractants . This acute treatment was anti-inflammatory in vivo as shown by pre-treatment of monocytes prior to adoptive transfer into an on-going murine peritonitis model . We find that apoA1 rapidly disrupts membrane lipid rafts , and as a consequence , dampens the PI3K/Akt signalling pathway that coordinates reorganization of the actin cytoskeleton and cell migration . Our data strengthen the evidence base for therapeutic apoA1 infusions in situations where reduced monocyte recruitment to sites of inflammation could have beneficial outcomes . CC chemokines are low-molecular-weight signalling proteins that mediate leukocyte trafficking in immune homeostasis and play a non-redundant role in monocyte/macrophage recruitment in pre-clinical models of inflammation ( Charo and Ransohoff , 2006; Mackay , 2008; White et al . , 2013 ) . Macrophage chemotaxis towards CC chemokines and other pathophysiological chemoattractants , such as N-formylmethionyl-leucyl-phenylalanine ( fMLF ) , complement component 5a ( C5a ) and chemerin , are mediated by Gi/o coupled G protein coupled receptors ( Allen et al . , 2007 ) . Chemokine-mediated recruitment of monocytes and macrophages , in addition to retention and activation , are attractive areas for the development of novel anti-inflammatory agents that could find application in a wide range of chronic inflammatory pathologies ( Asquith et al . , 2015; Schall and Proudfoot , 2011; Tabas and Glass , 2013; Zernecke and Weber , 2014 ) . Apolipoprotein A1 ( apoA1 ) is the major structural protein of high-density lipoprotein ( HDL ) and is widely recognized to have atheroprotective properties ( Fisher et al . , 2012 ) . The cellular mechanisms that account for the observed anti-inflammatory effects of apoA1 and HDL , including those on innate immune cell activation , are incompletely understood . Postulated mechanisms include but are not limited to: ( i ) dependence on ATP-binding-cassette ( ABC ) A1 , but independence of scavenger receptor class B type 1 ( SR-B1 ) ( Murphy et al . , 2008 ) ; ( ii ) inhibition of NF-kB and phosphatidylinositol-3-kinase ( PI3K ) signalling ( Bursill et al . , 2010 ) ; and ( iii ) increased Activator of Transcription Factor 3 ( ATF3 ) expression , which represses a subset of proinflammatory genes ( De Nardo et al . , 2014 ) . Notably , the bulk of studies that have shown anti-inflammatory effects of apoA1 or HDL involve relatively long-term treatment of macrophages in culture or chronic exposure in vivo ( Diederich et al . , 2001; Bursill et al . , 2010; Ansell et al . , 2003 ) . We thought it timely to focus on the acute effects of apoA1 treatment , in both in vitro and in vivo assays of monocyte/macrophage chemotaxis , a process of fundamental importance to the initiation and amplification of inflammation . We also explored the cellular mechanisms for anti-chemotactic effects and their dependence on cholesterol efflux . Chronic treatment of human monocytes with HDL and apoA1 has been reported to reduce chemotaxis to fMLF ( Diederich et al . , 2001 ) . In the current study , we wanted to test whether acute exposure to the human form of apoA1 ( referred to throughout as simply apoA1 ) could alter monocyte-derived cell chemotaxis to CC chemokines and in the process , to look for the earliest signalling events following addition of apoA1 . To address this question , we used a real-time chemotaxis assay ( Iqbal et al . , 2013 ) . Biogel elicited murine macrophages ( 4 × 105/well ) were pre-treated with apoA1 ( 40 µg/ml; a standard amount used in cholesterol efflux studies , e . g . Adorni et al . , 2007 ) for 60 min at 37°C , 5% CO2 and allowed to migrate towards 10 nM CCL2 . The representative trace in Figure 1A shows a clear reduction in chemotaxis towards CCL2 as quantified by slope and max-min analysis ( 67% reduction , p≤0 . 01 , Figure 1B–C ) . A similar reduction in chemotaxis ( 71% reduction , p≤0 . 01 ) was observed in response to 10 nM CCL5 ( Figure 1D–F ) . We also found that washing cells following acute exposure to apoA1 led to a similar impairment in migration as unwashed cells ( data not shown ) . We extended our studies to show that this reduction in chemotaxis was seen with other , non-chemokine , chemoattractants: a 52% reduction in macrophage chemotaxis was observed in response to 10 nM C5a peptide ( p≤0 . 05 , Figure 1G–I ) , and a 80% reduction in chemotaxis to 10 nM chemerin , a potent chemoattractant of antigen presenting cells ( p≤0 . 01 , Figure 1—figure supplement 1A–C ) . 10 . 7554/eLife . 15190 . 003Figure 1 . Acute pre-treatment with apoA1 inhibits macrophage chemotaxis . Biogel elicited macrophages were incubated with apoA1 ( 40 μg/ml ) for 60 min before being added to the upper chamber ( 4 × 105/well ) of a CIM-16 plate and allowed to migrate for 7–8 hr toward 10 nM CCL2 , CCL5 , or C5a . Representative traces are shown in panels ( A , D , G ) . Migration was measured with slope ( B , E , H ) and max-min analysis ( C , F , I ) . Data expressed as mean + SEM , n = 4–8 biological replicates with 3–4 technical replicates per condition . Statistical analysis was conducted by one-way ANOVA with Dunnett’s multiple comparison post-test . p , 0 . 05 , p , 0 . 01 , *p , 0 . 01 relative to CCL2 , CCL5 , or C5a alone . Biogel elicited macrophages from Cd68-GFP mice were incubated with either vehicle ( J and K ) or apoA1 ( 40 μg/ml ) ( L ) for 60 min before being seeded into ibidiµ-Slide Chemotaxis3D slides ( 1 . 8 × 104 cells/chamber ) . Migration of biogel elicited macrophages in the presence ( K and L ) or absence ( J ) of a CCL5 gradient ( indicated by the red triangle ) was followed by time-lapse microscopy and quantified by cell tracking . Data for FMI , Euclidean distance , total distance and velocity are summarised ( M ) and expressed as mean ± SEM of three independent experiments . Statistical analysis was conducted by an unpaired Students T test . p , 0 . 05 , CCL5 vs CCL5 + apoA1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15190 . 00310 . 7554/eLife . 15190 . 004Figure 1—figure supplement 1 . Short term incubations and different doses of apoA1 inhibit macrophage chemotaxis . Biogel elicited macrophages were incubated with apoA1 ( 40 μg/ml ) for 60 min before being added to the upper chamber ( 4 × 105/well ) of a CIM-16 plate and allowed to migrate for 7–8 hr toward 10 nM chemerin . Representative traces are shown in panel ( A ) . Migration was measured with slope ( B ) and max-min analysis ( C ) . Biogel elicited macrophages were pre-incubated with apoA1 ( 40 μg/ml ) for the indicated times before chemotaxis towards 10 nM CCL5 . Panel ( D ) shows a representative tracing , migration was measured with slope ( E ) and max-min analysis ( F ) . Biogel elicited macrophages were pre-incubated with various amounts of apoA1 ( 20–80 μg/ml ) as indicated before chemotaxis towards 10 nM CCL5 . Panel ( G ) shows a representative tracing , migration was measured with slope ( H ) and max-min analysis ( I ) . Data are expressed as mean ± SEM , n = 4–8 biological replicates with 3–4 technical replicates per condition . Statistical analysis was conducted by one-way ANOVA with Dunnett’s multiple comparison post-test . p<0 . 05 , p<0 . 01 , p<0 . 001 , *p<0 . 0001 relative to 10 nM chemerin or CCL5 alone . DOI: http://dx . doi . org/10 . 7554/eLife . 15190 . 004 To confirm and extend our real-time chemotaxis observations , we performed an alternate macrophage migration assay that utilizes time-lapse microscopy to visualize individual cell movement in a chemoattractant gradient . Biogel elicited macrophages were treated for 60 min with apoA1 ( 40 µg/ml ) , or vehicle prior to plating in chamber slides and stimulating cells with 10 nM CCL5 . Quantitative assessment of macrophage migratory behaviour was obtained by tracking individual cells using image analysis software ( representative traces shown in Figure 1J–L ) . Analysis revealed that macrophage migration towards CCL5 was significantly impaired ( p≤0 . 05 ) when cells were pre-treated with apoA1 as determined by the forward migration index ( FMI ) −0 . 090 v −0 . 012 , ( Figure 1M ) . Furthermore , treatment with apoA1 reduced the Euclidean distance and velocity of CCL5-stimulated macrophages to that of unstimulated controls , demonstrating a loss in cell directionality . Collectively , these findings support our observation that apoA1 directly impairs macrophage chemotaxis . The initial time and concentration used for apoA1 treatment were based on several previous reports ( Murphy et al . , 2011 , 2008; Diederich et al . , 2001 ) . We wanted to test whether similar inhibitory effects on macrophage chemotaxis could be observed with shorter incubations and different doses of apoA1 . Comparable to 60 min pre-treatment , both 20 min and 40 min were shown to significantly ( p≤0 . 001 ) reduce macrophage chemotaxis to CCL5 by 75–86% ( Figure 1—figure supplement 1D–F ) . Pre-treatment with 20 or 80 µg/ml of apoA1 for 60 min was also shown to reduce macrophage chemotaxis to CCL5 by 43% and 51% ( p≤0 . 05 ) , respectively ( Figure 1—figure supplement 1G–I ) . ApoA1 functions in the first step of reverse cholesterol transport , cholesterol efflux from macrophages , through both ABCA1-dependent and -independent mechanisms ( Phillips , 2014 ) . Given the strong inhibitory effect of apoA1 in our chemotaxis studies , we were interested to see if the effects were mediated through cholesterol efflux , and if so , whether they were dependent on the cholesterol transporter , ABCA1 . Our initial experiments used a recombinant apoA1 in which the Trp72 residue has been oxidized , thereby rendering apoA1 unable to efflux cholesterol via ABCA1-mediated processes ( Huang et al . , 2014 ) . Notably , the migration capacity of macrophages to CCL5 treated with either recombinant wild-type apoA1 ( rapoA1 ) or 5-OH-Trp72 rapoA1 ( ox-rapoA1 ) was significantly inhibited ( 90–94% reduction , p≤0 . 001 ) ( Figure 2A–C ) . We further demonstrated that the inhibitory effect of apoA1 to cell migration was independent of ABCA1-mediated cholesterol efflux by comparing the migratory ability of macrophages from C57BL/6 and Abca1-/- mice . We found that macrophages from both C57BL/6 and Abca1-/- mice have an equal capacity to migrate to CCL5 ( Figure 2D ) , and that migration of both is inhibited by 62–70% ( p≤0 . 05–0 . 001 ) following treatment with apoA1 for 60 min ( Figure 2D–F ) . Treatment of macrophages with the non-specific cholesterol depleting agent methyl β-cyclodextrin ( CyD ) also led to markedly reduced macrophage chemotaxis to CCL5 ( 91% ( p≤0 . 001 ) ; Figure 2G–I ) . Taken together , these results support a role for the contribution of a cholesterol-efflux pathway , independent of ABCA1 , mediating the inhibitory effect of apoA1 on macrophage chemotaxis . 10 . 7554/eLife . 15190 . 005Figure 2 . ApoA1 effects on macrophage chemotaxis are independent of ABCA1 , but dependent on cholesterol efflux . Biogel elicited macrophages from WT , Abca1-/- mice were pre-incubated with apoA1 , recombinant apoA1 ( rapoA1 ) or 5-OH-Trp72-rapoA1 ( ox-rapoA1 ) as described in Materials and methods before measuring chemotaxis towards 10 nM CCL5 . Panels ( A , D , G ) show representative traces , migration was measured with slope ( B , E , H ) and max-min analysis ( C , F , I ) . Biogel elicited macrophages from WT mice were pre-incubated with 3 mM methyl β-cyclodextrin ( CyD ) as described in Materials and methods before measuring chemotaxis towards 10 nM CCL5 . Data are expressed as mean + SEM , n = 4–8 biological replicates with 3–4 technical replicates per condition . Statistical analysis was conducted by one-way or two-way ANOVA with Dunnett’s multiple comparison post-test . p , 0 . 05 , p , 0 . 01 , p , 0 . 001 , **p , 0 . 0001 relative to 10 nM CCL5 alone . DOI: http://dx . doi . org/10 . 7554/eLife . 15190 . 005 Having shown significant effects of apoA1 on murine macrophage chemotaxis , we next tested the actions of apoA1 on human monocyte chemotaxis . Monocytes were isolated by CD14+ selection from human blood and pre-treated with apoA1 ( 40 µg/ml ) or vehicle prior to assessment of their migratory capacity . The representative trace in Figure 3A shows a reduction in human monocyte chemotaxis to CCL2 following apoA1 treatment . Slope ( Figure 3B ) and max-min analysis ( Figure 3C ) of monocytes from multiple donors confirmed a significant 43% reduction in chemotaxis ( p≤0 . 05 ) towards CCL2 . A similar significant inhibition in monocyte migration ( p≤0 . 05 ) was also observed in response to CCL5 ( 50% reduction ) and C5a ( 42% reduction ) ( Figure 3D–I ) . A similar degree of inhibition was observed with the ibidi chamber assay , with the human monocyte migration towards CCL5 significantly impaired ( p≤0 . 01 ) when cells were pre-treated with apoA1 ( representative traces shown in Figure 3J–L ) as determined by the forward migration index ( FMI ) −0 . 246 v 0 . 064 , ( Figure 3M ) . Furthermore , treatment with apoA1 significantly reduced the Euclidean distance , total distance , velocity and directionality of CCL5-stimulated human monocytes . 10 . 7554/eLife . 15190 . 006Figure 3 . Acute pre-treatment with apoA1 reduces human monocyte chemotaxis . CD14+ selected human monocytes ( 2 × 105 ) were pre-treated with apoA1 ( 40 μg/ml ) for 60 min before being added to the upper chamber of a CIM-16 plate and allowed to migrate for 2 hr toward 10 nM CCL2 , CCL5 or C5a . Panels ( A , D , G ) show representative traces , and migration was measured with slope ( B , E , H ) and max-min analysis ( C , F , I ) . Data are expressed as mean + SEM , n = 4 biological replicates with 3–4 technical replicates per condition . Statistical analysis was conducted with one-way ANOVA with Dunnett’s multiple comparison post-test . p , 0 . 05 , relative to chemoattractant alone . CD14+ selected human monocytes were pre-treated ( J and K ) with apoA1 ( 40 μg/ml ) ( L ) for 60 min before being seeded into ibidi µ-Slide Chemotaxis3D slides ( 1 . 8 × 104 cells/chamber ) . Migration of human monocytes in the presence ( K and L ) or absence ( J ) of a CCL5 ( 50 nM ) gradient ( indicated by the red triangle ) was followed by time-lapse microscopy and quantified by cell tracking . Data for FMI , Euclidean distance , total distance , velocity and directness are summarised ( M ) and expressed as mean ± SEM of two independent experiments . Statistical analysis was conducted by an unpaired Students T test . p , 0 . 05 , p , 0 . 05 , CCL5 vs CCL5 + apoa1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15190 . 00610 . 7554/eLife . 15190 . 007Figure 3—figure supplement 1 . Acute pre-treatment with apoA1 reduces PBMC:HUVEC interactions under flow . Isolated PBMCs ( 1 × 106 cells/ml ) were incubated with apoA1 ( 40 µg/ml ) or buffer ( control ) for 60 min prior to perfusion over TNF-α stimulated HUVECs . Panel ( A ) are representative images of PBMC:HUVEC interactions following treatment with apoA1 vs control ( Scale bar 20 µm ) . PBMC ( B ) rolling , ( C ) adhesion and ( D ) transmigration were quantified from six random fields/treatment . Results are expressed as percentage of control of two independent experiments . Statistical analysis was conducted with Mann-Whitney U test . 650 p , 0 . 05 , versus control; n = 4 donors . ( A ) Human CD14+ monocytes were isolated ( 1 × 106 ) and pre-treated with apoA1 ( 40 μg/ml ) for 60 min before being stimulated with 10 nM C5a for 5 min . Representative histograms from flow cytometry for ( E ) CD14 , ( F ) CD62L , ( G ) CD49b , ( H ) active CD11b , ( I ) total CD11b and ( J ) C5aR1 expression levels are shown from a total of four independent donors . DOI: http://dx . doi . org/10 . 7554/eLife . 15190 . 007 Previous studies have reported an inhibitory role for apoA1 on monocyte adhesion to endothelial cells ( Murphy et al . , 2008 ) . We therefore used an in vitro flow chamber assay to assess the effects of apoA1 on peripheral blood mononuclear cell ( PBMC ) :HUVEC interactions under flow . Pre-incubation of PBMCs with apoA1 considerably reduced cell capture as shown in the representative images ( Figure 3—figure supplement 1A ) . PBMC rolling , adhesion and transmigration were all significantly reduced by 20–45% following pre-treatment with apoA1 for 45 min ( Figure 3—figure supplement 1B–D ) . Integrins play a key role in the process of leukocyte recruitment ( Campbell and Humphries , 2011 ) , so we measured levels of active CD11b/ total CD11b , CD49b , CD62L and C5aR1 on human monocytes that had been treated for 45 min with apoA1 and then stimulated with C5a for 5 min . No differences were observed in the expression of any of these markers ( Figure 3—figure supplement 1E–J ) indicating that the effects of acute apoA1 treatment were independent of receptor and adhesion molecule modulation . To determine whether the inhibitory actions observed with acute apoA1 treatment on macrophage and human monocyte chemotaxis in vitro extends to monocyte/macrophage trafficking in vivo , we performed adoptive transfer studies with GFP+ monocytes . We previously reported that monocytes from Cd68-GFP mice express high levels of GFP transgene and that the level of expression increases as they differentiate into macrophages , greatly facilitating their detection and recovery ( Iqbal et al . , 2014 ) . Figure 4A summarizes the experimental design used to assess the effect of apoA1 treatment ex vivo on monocyte recruitment in vivo . Briefly , bone marrow monocytes were isolated by negative immunomagnetic selection ( typical purity of 65–70% GFP+ monocytes ( Figure 4B ) ) and then treated with either apoA1 ( 40 µg/ml ) or vehicle for 45 min . Monocytes were then washed and injected ( i . v ) into C57BL/6J recipient mice that had previously received 100 µg zymosan ( i . p ) to initiate a mild inflammatory response ( Iqbal et al . , 2014 ) . 10 . 7554/eLife . 15190 . 008Figure 4 . Acute exposure to apoA1 reduces the recruitment of adoptively transferred monocytes to sites of inflammation in vivo . Diagram of the experimental design presented in ( A ) . Monocytes were isolated from Cd68-GFP bone marrow using ‘no-touch’ negative immunomagnetic selection ( B ) . Isolated monocytes were characterized as 7/4high/Ly6Glow , with a typical yield of 65–70% monocytes , of which 70–80% were GFP positive . Isolated Cd68-GFP monocytes ( 1 × 106 ) treated for 45 min with either apoA1 ( 40 μg/ml ) or PBS and were adoptively transferred into C57BL/6J mice by i . v . injection 30 min after i . p . injection with 100 μg zymosan . Mice were euthanized at 16 hr , and peritoneal lavage and blood samples were collected . Representative flow cytometry plots of peritoneal lavage from C57BL/6J mice that received adoptively transferred GFP positive monocytes treated with or without apoA1 during ongoing zymosan-induced peritonitis ( C ) . Total monocytes and total adoptively transferred GFP+ monocytes were quantified in peritoneal lavage ( D–E ) and blood ( F–G ) . Data are expressed as mean + SEM of 8–10 mice from two independent experiments . Statistical analysis was conducted by paired Students T test . p , 0 . 01 , relative to control . DOI: http://dx . doi . org/10 . 7554/eLife . 15190 . 00810 . 7554/eLife . 15190 . 009Figure 4—figure supplement 1 . Acute exposure to apoA1 reduces the recruitment of adoptively transferred CD45 . 1 monocytes to sites of inflammation in CD45 . 2 recipient mice . ( A ) Diagram of the experimental design . ( B ) Monocytes were isolated from SJL/1 mice ( CD45 . 1 positive ) bone marrow using ‘no-touch’ negative immunomagnetic selection . Isolated monocytes were characterized as 7/4high/Ly6Glow , with a typical yield of 65–70% monocytes . A total of 1 × 106 isolated CD45 . 1+ monocytes treated with either apoA1 ( 40 μg/ml ) for 60 min or PBS were adoptively transferred into C57BL/6J mice ( CD45 . 2 positive ) by i . v . injection 30 min after i . p . injection with 100 μg zymosan . Mice were euthanized at 16 hr , and peritoneal lavage and blood samples were taken . ( B–E ) Total monocytes and total adoptively transferred CD45 . 1+ monocytes were quantified in peritoneal lavage and blood . Data are expressed as mean ± SEM of 8–10 mice from two independent experiments . Statistical analysis was conducted by paired Students T test . p<0 . 01 , relative to control . DOI: http://dx . doi . org/10 . 7554/eLife . 15190 . 009 Recruitment of GFP+ monocytes to the peritoneal cavity was assessed by flow cytometry , with representative plots shown in Figure 4C . Acute pre-treatment with apoA1 ex vivo reduced GFP+monocyte recruitment to the peritoneal cavity of recipient mice by 65% ( p≤0 . 01 ) when compared to vehicle treated cells ( Figure 4E ) . This occurred despite similar levels of GFP+ monocytes in the blood of mice that received adoptively transferred apoA1 or vehicle treated monocytes ( Figure 4G ) . Importantly , total monocytes recruited to the peritoneal cavity and circulating in the blood were equivalent at 16 hr in all groups of recipient mice following zymosan challenge ( Figure 4D and F ) , highlighting that the effects of treatment were restricted to only the monocytes pre-treated with apoA1 . In a parallel set of experiments , we used SJL/1 mice ( which express CD45 . 1 ) as bone marrow monocyte donors and C57BL/6J mice ( which express CD45 . 2 ) as recipients to confirm results obtained with Cd68-GFP monocytes , and observed a similar defect in recruitment following acute pre-treatment with apoA1 ( Figure 4—figure supplement 1 A–E ) . All experiments conducted thus far have focused on the effects of acute apoA1 treatment . We wanted to also investigate what effect chronic exposure to apoA1 may have on leukocyte recruitment in vivo . To address this we used mice carrying the human apoA1 transgene ( Apoa1Tg ) under the control of its natural promoter , which results in elevated circulating levels of total apoA1 , with >90% being human apoA1 ( Rubin et al . , 1991; Chajek-Shaul et al . , 1991 ) . Apoa1Tg and C57BL/6J control animals were injected ( i . p ) with 100 µg of zymosan to provide an inflammatory stimulus ( Iqbal et al . , 2014 ) . Peritoneal exudates 16 hr post-challenge contained predominately neutrophils and monocytes as identified as CD45+/CD115-/Ly6G+ and CD45+/CD115+/Ly6GChi/lo populations by flow cytometry , respectively ( representative plots shown in Figure 5A ) . Total cell recruitment in Apoa1Tg mice was significantly reduced by 27% ( p≤0 . 05 ) when compared to C57BL/6 controls , with a reduction in total neutrophils ( 26% , p≤0 . 05 ) and monocytes ( 30% , p≤0 . 01 ) ( Figure 5B–D ) . Elevated apoA1 levels significantly suppressed the recruitment of the Ly6Chi monocyte subset to peritoneal cavity ( 42% , p≤0 . 05 , Figure 5E ) . No changes in Ly6Clo subset or in total blood white cell counts were observed between any of the experimental groups ( Figure 5F–G ) . Chemokines CCL2 and CXCL1 were measured in the peritoneal exudates of mice from the 4 hr and 16 hr time-points , but no significant differences were observed ( CXCL1 was non detectable at 16 hr ) ( Figure 5H–J ) . Similar levels of chemokines , but reduced neutrophil and monocyte influx , suggests that changes in the intrinsic properties of the leukocytes were responsible for the reduction in innate immune cell recruitment . 10 . 7554/eLife . 15190 . 010Figure 5 . Apoa1Tg mice display reduced leukocyte recruitment in zymosan peritonitis . Apoa1Tg and WT control mice were injected i . p . with 100 μg zymosan . Mice were euthanized at 16 hr and peritoneal lavage samples were collected . Representative flow cytometry plots are shown in panel ( A ) of neutrophils ( Neut ) characterised as CD45+/CD115-/Ly6G+ and monocytes characterised as CD45+/CD115+/Ly6Chi/lo from peritoneal lavage fluid . Total cells ( B ) , neutrophils ( C ) total monocytes ( D ) , Ly6Chi monocytes ( E ) , Ly6Clo monocytes ( F ) within the peritoneal cavity following zymosan challenge are expressed . ( G ) Total circulating blood white cell counts from Apoa1Tg mice and WT controls are expressed as mean ± SEM of 3–9 mice per condition . ( H–J ) CXCL1 and CCL2 levels were measured in peritoneal lavage fluid by ELISA . Data are expressed as mean + SEM of 2–6 mice per genotype of two independent experiments . Statistical analysis was conducted by one-way ANOVA with Dunnett’s multiple comparison post-test . p , 0 . 05 , p , 0 . 01 compared to WT . DOI: http://dx . doi . org/10 . 7554/eLife . 15190 . 010 Given the consistent effects of apoA1 on monocyte and macrophage chemotaxis in vitro and in vivo , we were interested to identify the mechanistic basis of these findings . We hypothesized that there was an interruption or inhibition of an aspect of chemokine signalling that regulates cell migration . In our initial studies , we found no differences in G-protein coupled receptor ‘upstream’ phenomena , such as cAMP levels , Ca2+ mobilisation or β-arrestin recruitment as a consequence of apoA1 pre-treatment ( Figure 6—figure supplement 1 ) . Hence , we turned our attention to other known downstream signalling events that are essential for cytoskeletal reorganization , including PI3K and Rho/Rac ( Vorotnikov , 2011 ) . Human monocytes were pre-treated with apoA1 ( 40 µg/ml ) for 60 min , prior to C5a stimulation . Pre-treatment with apoA1 reduced Akt phosphorylation by 54% , as determined by western blotting and densitometry ( Figure 6A ) . A similar significant reduction in Akt signalling was also observed following apoA1 and CyD pre-treatment of biogel elicited murine macrophages ( reduction of 32% and 63% respectively , Figure 6B–C ) . In addition to Akt phosphorylation , we also measured the effects of apoA1 on its upstream activator , PI3K . Pre-treatment of RAW macrophages with apoA1 for 1 hr prior to stimulation with C5a led to a reduction in PI3K activity by 36% ( Figure 6D ) , and PI3K inhibitor studies revealed significant reduction of macrophage chemotaxis in vitro ( Figure 6E–F ) . Taken together , these results provide evidence that apoA1 modulates PI3K/Akt signalling resulting in a significant impairment of monocyte and macrophage chemokine-induced migration . 10 . 7554/eLife . 15190 . 011Figure 6 . ApoA1 suppresses Akt signalling and PI3K activity following chemokine activation of human and murine monocyte-derived cells . CD14+ selected human monocytes or biogel elicited murine macrophages ( 2 × 105 ) were pre-incubated with either vehicle ( PBS ) , apoA1 ( 40 μg/ml ) or 3 mM methyl β-cyclodextrin ( CyD ) for 60 min as described in Materials and methods . Human CD14+ monocytes ( A ) or murine macrophages ( B , C ) were then stimulated with either 10 nM C5a or CCL5 . Relative levels of phosphorylated Akt were determined by Western blotting . Representative blots from independent biological replicates are shown . Densitometry of western blots is shown in panels below representative western blots ( n > 2/group ) . Data are expressed as mean ± SEM , n = 2–6 biological replicates . Murine RAW macrophages were pre-incubated with apoA1 ( 40 μg/ml ) or vehicle for 60 min at 37°C , 5% CO2 prior to stimulation with C5a ( 10 nM ) for 10 min ( D ) . PI3K was immunoprecipitated from cell lysates and PI3K activity determined by incubating with PIP2 and measuring PIP3 production by competitive ELISA . Data are expressed as mean + SD , n = 4 independent experiments . Statistical analysis was conducted by one-way ANOVA with Dunnett’s multiple comparison post-test . p<0 . 05 , relative to C5a alone . Biogel elicited macrophages were incubated with 100 nM wortmannin for 60 min before being added to the upper chamber ( 4 × 105/well ) of a CIM-16 plate and allowed to migrate for 8 hr at 37°C , 5% CO2 towards 10 nM CCL5 . Representative traces are shown in panel ( E ) . Migration was measured by max-min analysis ( F ) . Statistical analysis was conducted by one-way ANOVA with Dunnett’s multiple comparison post-test . p<0 . 05 , *p<0 . 001 relative to CCL5 alone . DOI: http://dx . doi . org/10 . 7554/eLife . 15190 . 01110 . 7554/eLife . 15190 . 012Figure 6—figure supplement 1 . Gαi/o-signalling , β-arrestin recruitment and intracellular Ca2+ flux remain unaffected in response to apoA1 . ( A ) Intracellular cAMP levels were measured as described in Supplemental Methods in CHO-K1 cells expressing human C5aR1 . Cells were incubated as described in Materials and methods with either vehicle or apoA1 ( 40 μg/ml ) as indicated . Following treatment , vehicle or C5a was added in assay buffer containing forskolin ( FSK: 20 µM final assay concentration ) to give the indicated final concentrations . The cells were then incubated for 30 min before the addition of the detection reagents . Data are mean ± SEM , n = 3 . ( B ) Recruitment of β-Arrestin was measured as described in Supplemental Methods in CHO-K1 cells expressing murine CCR2 . Cells were incubated as described in Materials and methods with either vehicle or apoA1 as indicated . Following treatment , vehicle or CCR2 was added and cells incubated for 90 min before the addition of the detection reagents . Data are mean ± SEM , n = 3 . Biogel elicited macrophages were loaded with FURA2-AM in the presence or absence of 40 μg/ml apoA1 for 45 min . Macrophages were then stimulated with either vehicle or 10 nM C5a . ( C ) Representative tracings of intracellular Ca2+ levels following C5a stimulation in vehicle or apoA1 pre-treated cells . ( D ) Pre-treatment with apoA1 has no effect on intracellular Ca2+ levels . Data are mean ± SEM , n = 4 biological replicates . Statistical analysis was performed by two-way ANOVA with Sidak’s multiple comparisons correction . DOI: http://dx . doi . org/10 . 7554/eLife . 15190 . 012 Given that signalling in the PI3K/Akt pathway regulating cell movement involves proteins and lipids associated with membrane lipid rafts ( Zajchowski and Robbins , 2002 ) , we hypothesized that the apoA1-mediated suppression of Akt phosphorylation in monocytes and macrophages occurred due to perturbation of lipid raft content as a consequence of cholesterol efflux . To assess this , biogel-elicited macrophages ( Figure 7A–F ) or human THP-1 macrophages ( Figure 7G–J ) were treated with CyD or apoA1 or its variants , and the abundance of the raft-enriched components caveolin-1 and ganglioside GM1 assessed . Pre-treatment with apoA1 , rapoA1 , ox-rapoA1 , or CyD led to a significant reduction in total cell caveolin-1 ( Figure 7A–B ) as determined by western blotting , and ganglioside GM1 expression as determined by flow cytometry ( Figure 7C ) or confocal fluorescence microscopy ( Figure 7D–F ) . Membrane lipid rafts are enriched in cholesterol , and perturbations to this component leads to raft reorganization , which affects the activity of receptors and raft-bound signalling factors ( Simons and Toomre , 2000 ) . To determine whether apoA1 reorganized cholesterol-rich domains , sucrose density gradient ultracentrifugation of human macrophages was performed and raft markers identified ( Figure 7G ) . As previously reported ( Adorni et al . , 2007; Koseki et al . , 2007 ) , cellular free cholesterol is distributed between both low ( rafts ) and high ( non-rafts ) density fractions ( Figure 7H ) . Treatment with apoA1 ( 40 µg/ml ) , ox-rapoA1 ( 40 µg/ml ) , or CyD leads to a significant reduction in cholesterol content in both raft and non-raft fractions , demonstrating that apoA1 can significantly modulate the composition of this signalling nanodomain ( Figure 7I–J ) . 10 . 7554/eLife . 15190 . 013Figure 7 . ApoA1 modulates monocyte-derived cell lipid raft cholesterol content in cells of murine and human origin . ( A–B ) Total cell caveolin-1 was assessed by Western blotting following treatment of biogel elicited macrophages with apoA1 ( 40 μg/ml ) , ox-rapoA1 ( 40 μg/ml ) or 3 mM methyl β-cyclodextrin ( CyD ) for 1 hr . The lipid raft content of biogel elicited macrophages was assessed following treatment with apoA1 ( 40 μg/ml ) or 3 mM CyD for 1 hr , and incubation of the cells with cholera toxin B Alexa Fluor 647 conjugate to bind lipid raft ganglioside G M1 . Relative fluorescence ( compared to PBS-treated control cells ) was assessed by flow cytometry ( C ) , and cells were imaged by confocal microscopy ( D–F ) . Prior to fractionation of membrane lipid rafts , human THP-1 macrophages were treated with either vehicle , apoA1 ( 40 μg/ml ) , ox-rapoA1 ( 40 μg/ml ) , or 3 mM CyD for 1 hr prior to stimulation with 10 nM CCL5 for 5 min . Cells were then homogenized in a 0 . 2% Triton-X 100 containing buffer prior to 5–45% sucrose density gradient centrifugation . Fractions were collected from the top ( fraction 1 , lightest fraction ) to the bottom ( fraction 10 , heaviest fraction ) , and fraction proteins separated by SDS-PAGE , transferred to PVDF membranes and caveolin-1 , flotillin-1 , or transferrin receptor ( TfR ) content assessed by western blotting . Representative distribution of raft and non-raft marker proteins presented in panel ( G ) . The cholesterol content of each fraction was determined via the amplex red assay , with a representative distribution shown in panel ( H ) . The sums of the cholesterol content in lipid raft fractions ( fractions 2–4 ) , and non-raft fractions ( fractions 8–10 ) following treatment with apoA1 , ox-rapoA1 and CyD are shown in panels ( I ) and ( J ) , respectively . Western blots are representative of three independent experiments . Cholesterol contents are means + SEM of three independent experiments . Statistical analysis was conducted by one-way ANOVA with Dunnett’s multiple comparison post-test . p<0 . 05 , *p<0 . 01 relative to PBS . DOI: http://dx . doi . org/10 . 7554/eLife . 15190 . 01310 . 7554/eLife . 15190 . 014Figure 7—figure supplement 1 . ApoA1 depletes membrane lipid rafts of cholesterol in cholesterol-loaded THP-1 cells . Human THP-1 macrophages were incubated with methyl-β-cyclodextrin:cholesterol complexes ( 0 . 5 mg/mL ) for 1 hr to cholesterol-load cells prior to incubation for 1 hr with 40 µg/mL apoA1 or PBS . Cells were then homogenized in a 0 . 2% Triton-X 100 containing buffer prior to 5–45% sucrose density gradient centrifugation . Fractions were collected from the top ( fraction 1 , lightest fraction ) to the bottom ( fraction 10 , heaviest fraction ) , and cholesterol content of each fraction quantified . The sums of the cholesterol content in lipid raft fractions ( fractions 2–4 ) , and non-raft fractions ( fractions 8–10 ) , are shown in panels ( A ) and ( B ) respectively . Statistical analysis was conducted by using Student’s T test . p<0 . 05 relative to loaded cells alone . DOI: http://dx . doi . org/10 . 7554/eLife . 15190 . 014 Diseases in which chronic inflammation is a central feature of their pathology are significant contributors to mortality and serious morbidities worldwide . A major factor that sustains inflammation in these diseases is a mobilization and activation of the innate immune system ( Tabas and Glass , 2013 ) . Chronic inflammation is characterized by the continued recruitment of circulating monocytes , in part mediated by their response to chemokines , such as CCL2 and CCL5 , which are synthesized locally in response to tissue damage and resultant locally-generated inflammatory mediators ( Tabas and Glass , 2013 ) . After entering the tissue , recruited monocytes become macrophages and the degree of their activation is frequently judged in vitro by the chemotactic response to chemokines . Thus , the identification and development of agents with the ability to inhibit monocyte/macrophage chemotaxis to various chemokines is an extremely active area of investigation with the ultimate goal of deploying such agents as therapeutics to combat chronic inflammatory diseases , including arthritis and atherosclerosis ( Allen et al . , 2007; Koelink et al . , 2012; Koenen and Weber , 2011 ) . It has been previously reported that both HDL and its major protein component apoA1 can inhibit leukocyte chemotaxis to various chemokines ( Bursill et al . , 2010; Wang et al . , 2010; Ansell et al . , 2003 ) . Previous studies have been heavily weighted towards experimental conditions in which the assays were exclusively performed in vitro or the duration of treatment was prolonged ( ≥24 hr ) . Furthermore , the mechanistic basis for the effects on chemotaxis has been largely unexplored . In this present report , we have addressed these issues in both murine and human monocytes and macrophages in vitro and in models of inflammation in vivo . Our major findings are that: ( 1 ) only a short exposure ( <1 hr ) of monocytes or macrophages to apoA1 is required to reduce chemotaxis to chemoattractants in vitro and most notably , recruitment to an inflammatory site in vivo; ( 2 ) apoA1 treatment reduces both the velocity and the directionality of macrophage migration to an established chemokine gradient; ( 3 ) inhibition of monocyte-derived cell chemotaxis requires cholesterol depletion of the cells , through a mechanism independent of ABCA1 , a cholesterol transporter previously invoked to explain certain anti-inflammatory effects of apoA1 ( Murphy et al . , 2008 ) , but rather through cholesterol efflux most likely by aqueous diffusion ( Adorni et al . , 2007 ) . We propose a model in which apoA1 and CyD treatment disrupts plasma membrane lipid rafts by cholesterol depletion , which , in turn , dampens the signalling pathway required for reorganization of the actin cytoskeleton , as reflected by impaired PI3K/Akt activation . These findings will now be discussed in more detail . ApoA1 or HDL treatments of monocytes/macrophages in vitro and of mice have typically been of longer duration than used in this current report , sometimes reaching one week ( Smythies et al . , 2010; Hewing et al . , 2014 ) . This makes it difficult to discern early direct effects of apoA1 from those that are downstream and likely to include many subsequent indirect effects . For example , in Bursill et al . , a 24 hr treatment of monocytes with recombinant HDL resulted in reduced chemotaxis , and this was associated with decreased expression of CCR2 and CX3CR1 ( Bursill et al . , 2010 ) . These long-term effects are unlikely to be the basis for the observed ~80% reduction in macrophage chemotaxis following only 20 min of exposure to apoA1 ( Figure 1—figure supplement 1D–F ) . In our experiments , important “upstream” or “receptor proximal” signalling functions of chemokine G protein–coupled receptors ( GPCRs ) were not affected by acute treatment with apoA1 i . e . decreased intracellular cAMP levels ( Figure 6—figure supplement 1A ) , β-arrestin recruitment ( Figure 6—figure supplement 1B ) and Ca2+ flux ( Figure 6—figure supplement 1C–D ) . In the current studies , both murine ( Figure 1A ) , and human monocyte migration to CCL2 ( Figure 3A ) were inhibited by apoA1 treatment . Importantly , in some chemotaxis experiments , apoA1 was removed after pre-treatment of cells , yet with even short pre-treatment times ( as little as 20 min ) an ~80% inhibition of chemotaxis to CCL5 was still observed ( Figure 1—figure supplement 1D–F ) . In other words , continuous exposure to apoA1 was not required for significant inhibition of chemotaxis . Our in vivo studies recapitulated important features of assays performed on immune cells in vitro . For example , continuous and prolonged exposure of leukocytes to apoA1 and HDL , inherent to the Apoa1Tg mouse model , was associated with reduced recruitment of innate immune cells to sites of inflammation ( Figure 5 ) . As with the experiments in vitro which involve long-term incubation of cells with apoA1 or HDL , early vs . late effects cannot be discerned using this transgenic mouse model . Thus , it was especially notable that in our adoptive transfer approach , only 45 min ex-vivo exposure of monocytes to apoA1 was required to reduce their recruitment to an inflammatory site by 65% ( Figure 4 and Figure 4—figure supplement 1 ) . That the rapidity of these effects in vitro and in vivo parallels the acute loss of lipid rafts in monocytes ( Murphy et al . , 2008 ) or macrophages ( Figure 7 ) is not likely to be a coincidence . Rather , this rapid desensitization to the chemoattractant effects of chemokines likely represents the perturbation of lipid rafts , which are crucial to the common signalling machinery downstream from chemokine receptors that are required for the reorganization of the cytoskeleton and integral to chemotaxis ( Vorotnikov , 2011 ) . Consistent with this is our finding that incubation with apoA1 was effective in blunting chemotaxis to multiple chemokines and chemoattractants , yet the early events following receptor ligation were unimpaired ( Figure 6—figure supplement 1 ) . More directly , we show using independent assays ( western blotting and flow cytometry ) that lipid raft abundance was reduced after incubation of macrophages with apoA1 , as was Akt phosphorylation . Akt activation in response to phosphorylation by PI3K is required for chemotaxis and was confirmed by our experiments using the PI3K inhibitor , wortmannin ( Figure 6E–F ) . Given the importance of lipid raft organization to chemotaxis , we were interested in the mechanism by which apoA1 mediated its effects . Lipid rafts are enriched in cholesterol and its depletion is known to change the content and function of many of its components ( Simons and Toomre , 2000 ) . While it is long been appreciated that apoA1 is a major acceptor of cellular cholesterol , and therefore can affect the lipid raft structure and functions , most studies of apoA1’s anti-inflammatory effects have focused on cholesterol efflux mediated by specific membrane factors , in particular ABCA1 ( e . g . Wang et al . , 2001; Murphy et al . , 2008; Pagler et al . , 2011 ) . By multiple criteria , our results are consistent with a non-specific cholesterol depletion process ( sometimes called aqueous diffusion ( Adorni et al . , 2007 ) that was responsible for the impairment in chemotaxis , as: ( 1 ) similar effects of apoA1 incubation were observed with Abca1-/- cells ( Figure 2D–F ) ; ( 2 ) CyD mediated cholesterol efflux was equally effective as apoA1 ( Figure 2J–L ) ; and ( 3 ) a mutant form of apoA1 ( ox-rapoA1 , or 5-OH-Trp72 apoA1 ) , which is a poor acceptor of cholesterol via ABCA1 , also inhibited chemotaxis ( Figure 2A–C ) . It should be noted that under the conditions of our assays using non-cholesterol loaded cells , ~80% of cholesterol efflux is likely through the aqueous diffusion mechanism ( Adorni et al . , 2007 ) , and this dominance is consistent with our results . Furthermore , that apoA1 can promote cholesterol efflux in Abca1-/- macrophages has also been reported ( Zhu et al . , 2008 ) . In addition , the inhibitory effect of apoA1 on macrophage chemotaxis to CCL5 is not affected by cholesterol loading of the cell ( data not shown ) , consistent with apoA1’s ability to promote cholesterol removal from membrane lipid rafts of these cells ( Figure 7—figure supplement 1 ) . Continued monocyte recruitment and macrophage activation are hallmarks of sites of chronic inflammation . CC chemokines play a key role in monocyte recruitment and macrophage activation in pre-clinical models of human diseases characterised by chronic inflammation ( Charo and Ransohoff , 2006; White et al . , 2013; Tabas and Glass , 2013 ) . Extending the findings from chemokine receptor knockout mice to interventional studies in man using chemokine receptor antagonists has proven challenging with multiple drugs that target single chemokine receptors failing to progress beyond early phase randomised clinical trials ( Schall and Proudfoot , 2011; Koelink et al . , 2012 ) . As an alternative to targeting single macrophage chemoattractant GPCRs for therapeutic benefit , we and others have explored the potential of targeting multiple chemokines using chemokine binding proteins ( White et al . , 2011 ) or blocking macrophage responses to multiple chemoattractants using netrin ( van Gils et al . , 2012 ) or apoA1 ( this study ) . The effects of netrin on macrophage chemotaxis are dependent on the Unc5b receptor while our experiments show that the effects of apoA1 occur via a receptor-independent mechanism by changing lipid raft organisation leading to decreased PI3K activity and hence decreased monocyte/macrophage adhesion , rolling and chemotaxis . The present results are relevant to our recent studies of apoA1’s anti-tumorigenic activity ( Zamanian-Daryoush et al . , 2013 ) . This effect was attributed , in part , to suppression of the recruitment and accumulation of myeloid-derived suppressor cells ( MDSC ) , potent stimulators of tumor growth . Notably , MDSC recruitment to tumors was significantly reduced in both apoA1 Tg mice and in apoA1-deficient mice receiving apoA1 therapy . MDSCs are comprised of a mixture of myeloid cells with granulocytic and monocytic morphology . The ability of apoA1 to suppress the recruitment of MDSCs to tumors , therefore , is in agreement with our current results showing the ability of apoA1 to suppress myeloid cell migration to inflammatory sites . While the exact molecular mechanism behind apoA1’s anti-tumor effect is currently unknown , given the similarities between the results in the cancer and inflammation settings , it is likely that apoA1’s ability to modulate membrane lipid raft cholesterol content influences signalling pathways regulating MDSC recruitment to tumors . Using recombinant apoA1 particles , Nissen and colleagues showed regression of coronary atherosclerosis in patients given 5 weekly intravenous injections ( Nissen et al . , 2003; Kingwell et al . , 2014 ) , and it is tempting to speculate that the efficacy was due , at least in part , by the mechanisms illustrated by our studies . Our in vivo findings ( Figures 4 , 5 and Figure 4—figure supplement 1 ) suggest that apoA1 infusion may also be therapeutic in other clinical situations where reduced monocyte recruitment to sites of inflammation would be beneficial . In future studies , it will be interesting to test the effect of apoA1 or CyD infusion in pre-clinical models of ischemia reperfusion injury ( mimicking a myocardial infarction ) , septic shock or ischemic stroke , all of which are situations where limiting the initial rush of inflammatory leukocytes into sites of injury could have a significant effect on subsequent tissue injury and repair processes . All cell culture media and buffers were obtained from PAA systems ( Yeovil , UK ) or Gibco ( Life Technologies , NY ) unless otherwise specified . Laboratory chemicals were obtained from Sigma-Aldrich ( Gilligham , Dorset , UK or St . Louis , MO ) . Chemokines and other chemoattractants were purchased from Peprotech ( London , UK or Rocky Hill , NJ ) , Merck Millipore ( Feltham , UK or Billerica , MA ) and R&D Systems ( Abingdon , Oxford , UK or Minneapolis , MN ) . Inhibitors were purchased from TOCRIS ( Bristol , UK ) . Human apolipoproteinA1 ( apoA1 ) was purchased from Merck Millipore and Academy Bio-medical Co . ( Houston , Texas ) . Recombinant apoA1 ( rapoA1 ) was prepared as described previously using a genetically engineered E . coli mutant strain deficient in endotoxin synthesis ( Huang et al . , 2014 ) and 5-hydroxy Trp72-apoA1 ( ox-rapoA1 ) was prepared using an orthogonal tRNA synthetase pair that incorporates 5-hydroxy tryptophan ( Ellefson et al . , 2014 ) . UK animal studies were conducted with ethical approval from the Dunn School of Pathology Local Ethical Review Committee and in accordance with the UK Home Office regulations ( Guidance on the Operation of Animals , Scientific Procedures Act , 1986 ) . Male ( 10–14 weeks ) C57BL/6J and SJL/1 mice ( CD45 . 1 ) mice were obtained from Harlan Laboratories ( Oxfordshire , UK ) . Female Cd68-GFP mice ( 6–8 weeks ) were obtained from our in-house colony . All USA animal experiments were carried out according to the guidelines of the National Institutes of Health and approved by the New York University Institutional Animal Care and Use Committee ( Protocol 102090 ) . Male ( 10–14 weeks ) C57BL/6J , and C57BL/6-Tg ( APOA1 ) 1Rub/J ( Apoa1Tg ) mice were obtained from the Jackson Laboratory , Maine , USA . Abca1-/- mice were kindly provided by Kathryn J Moore ( NYU School of Medicine , USA ) . Human blood from anonymous healthy donors was obtained in the form of leukocyte cones from the NHS Blood and Transplant service . Leukocyte cones contain waste leukocytes isolated from individuals donating platelets via apharesis , and consist of a small volume ( ~10 ml ) of packed leukocytes with few red blood cells or platelets . For monocyte isolation , blood was diluted with 1:3 with PBS followed by separation using ficoll gradient centrifugation as previously described ( White et al . , 2014 ) . Following human peripheral blood mononuclear cell ( PBMC ) isolation and washing , a total of 1 . 25 × 108 PBMCs were labelled and positively selected with CD14 micro-beads and MACS separation ( Miltenyi Biotec , Bisley , Surrey , UK ) . Monocytes were resuspended at 4 × 106 cells/ml in chemotaxis buffer ( RPMI 1640/25 mM HEPES/0 . 5% ( w/v ) BSA ) and left on ice until required . Mice were injected i . p . with 1 ml of sterile 2% biogel P-100 fine polyacrylamide beads ( 45–90 µm; Bio-Rad Laboratories , Hertfordshire , UK or Hercules , CA ) suspended in PBS . Mice were sacrificed 4 days later and the peritoneum lavaged with 10 ml of ice-cold PBS/2 mM EDTA . Monocytes were isolated from Cd68-GFP bone marrow using EasySep Mouse Monocyte Enrichment kit ( Stem Cell Technologies , France ) as described previously ( Iqbal et al . , 2014 ) . Briefly , bone marrow cell suspensions were passed through a 70 µm cell strainer and red blood cells lyzed ( Pharmlyse , BD Bioscience , SJ ) according to the manufacturer's instructions . The bone marrow cell suspension was treated with the EasySep reagents and monocytes isolated by depletion with an EasyPlate magnet ( Stem Cell Technologies ) using a 96 well plate for 5 min , followed by a second selection for a further 2 min . The purity of the resulting populations confirmed by flow cytometry using antibodies specific for CD45 ( BD Bioscience ) , CD11b ( BD , Bioscience ) , 7/4 ( AbD Serotec , Oxford , UK ) and Ly-6G ( Biolegend , London , UK ) . Bone marrow isolations from a total of 4 femurs were typically found to yield 1–2 × 106 cells at a purity of 65–85% monocytes . Monocytes were incubated with apoA1 ( 40 µg/ml ) for 45 min at 37°C in 5% CO2 and washed twice before i . v . administration . C57BL/6J mice were administered 100 µg zymosan A in PBS ( Sigma-Aldrich ) i . p . 30 min before i . v . administration of 1 × 106 isolated monocytes ( 70–80% GFP positive , in 200 µl ) . After 16 hr mice were sacrificed and peritoneal exudates were collected by lavage with 5 ml of ice cold sterile PBS/2 mM EDTA . Total cell counts and cellular composition of peritoneal exudate were determined as previously described ( Iqbal et al . , 2014 ) . Experiments were carried out with CIM-16 plates and an xCELLigence RTCA-DP instrument ( ACEA , San Diego , USA ) as previously described ( Iqbal et al . , 2013 ) . Chemoattractants were made to desired concentrations and loaded into the lower wells of the CIM-16 plate . Upper wells were filled with chemotaxis buffer and plates equilibrated for 30 min at RT . Biogel elicited macrophages were resuspended in chemotaxis buffer and incubated with apoA1 or other treatments for 20–60 min at 37°C , 5% CO2 . Cell suspensions were placed into the wells of the upper chamber , and the assay performed over 8 hr ( 5 s data points ) . Bone marrow monocyte derived macrophages ( BMDM ) were generated as previously described ( Iqbal et al . , 2013; Marim et al . , 2010 ) . Briefly , fresh bone marrow cells from tibiae and femurs of C57BL6/J mice ( 8–12 weeks ) were flushed and cultured in DMEM supplemented with 10% heat-inactivated fetal bovine serum ( FBS ) , 2 mM l-glutamine , 100 U/ml penicillin and 15% supernatant from L929 cells as a source of macrophage colony stimulating factor ( Englen et al . , 1995 ) . BMDM were generated by culturing 4 × 106 bone marrow cells in 10 ml of medium in 100 mm non-tissue culture treated petri dishes ( Sterilin , Abergoed UK ) . On day 3 , an additional 5 ml of medium was added . Cells were harvested with PBS/2 mM EDTA on day 7 . C57BL/6J or apoA1 TG mice were administered 100 µg zymosan A ( Sigma-Aldrich ) in PBS . After 16 hr mice were sacrificed and peritoneal exudates collected by lavage with 5 ml of ice cold sterile PBS with 2 mM EDTA . Total cell counts and cellular composition of peritoneal exudate were determined as previously described ( Iqbal et al . , 2014 ) . Antibodies used to identify neutrophils ( CD45+/CD115lo/Ly6G+ ) and monocytes ( CD45+/CD115hi/Ly6Chi/lo ) were APC anti-mouse Ly-6G/Ly-6C ( Gr-1 ) , PE anti-mouse CD115 ( CSF-1R ) , and PE/Cy7 anti-mouse CD45 from Biolegend ( San Diego , CA ) . Total blood cell counts were determined with a Genesis blood counter ( Oxford Science , CT ) . Biogel elicited macrophages were filtered through a 40 microns cell strainer , to remove the biogel beads , centrifuged at 300 g for 5 min and then resuspended in chemotaxis buffer at a concentration of 3 × 106 cells/ml . Filtered biogel elicited macrophages were then incubated with either vehicle or apoA1 ( 40 µg/ml ) for 45 min at 37°C , 5% CO2 , before being seeded into ibidi Chemotaxis 3D µ–slides . Cells were then allowed to adhere for 30 min before chemoattractant or the vehicle was added to the appropriate side of the chemotaxis plate ( as per the manufacturer's instruction ) . Cells were then imaged every 5 min for 60 frames using an Olympus FV1200 inverted confocal microscope equipped with a heated and humidified live cell imaging stage . Macrophages ( 20–30 per chamber ) were then tracked using the manual tracking plugin within ImageJ software and chemotaxis analysis conducted using the chemotaxis and migration tool software ( Zengel et al . , 2011 ) . Slides VI 0 . 4 ( IB , Germany ) were coated with 0 . 5% bovine gelatin and seeded with human umbilical vein endothelial cells ( HUVEC; passage 2–3 ) such that monolayers were confluent the next day . HUVEC were stimulated with TNF-a ( 10 ng/ml , 7 hr; Sigma-Aldrich ) prior to commencing the flow assay . PBMCs were freshly isolated as previously described ( Cooper et al . , 2008 ) from healthy volunteers using a double density gradient of histopaque ( Sigma-Aldrich ) . Briefly , blood was diluted 1:2 with RPMI , layered over histopaque and centrifuged at 400 ×g for 30 min . The PBMC layer was removed and washed twice . PBMCs were pre-incubated with apoA1 for 45 min at 37°C , diluted to 1 × 106 cells/ml in PBS-BSA ( 0 . 1% ) and perfused over HUVEC at a flow rate of 1 dyne/cm2 for 5 min . Six random fields/treatment were recorded for 10 s each . The total number of interacting PBMCs was quantified as captured and further classified as rolling ( cells that moved in the direction of flow over the 10 s analysis period ) , adherent ( cells that remained stationary over the 10 s analysis period ) or transmigrated ( cells that migrated through the endothelial monolayer ) using Image-Pro Plus software ( MediaCybernetics , Buckinghamshire , UK ) . PBMCs were freshly isolated as previously described ( Cooper et al . , 2008 ) from healthy volunteers using a double density gradient of histopaque ( Sigma-Aldrich ) . PBMCs were pre-incubated with apoA1 for 45 min at 37°C , diluted to 1 × 106 cells/ml in PBS-BSA ( 5% ) and stimulated with C5a ( 10 nM ) for 5 min and then stained with active CD11b ( Clone CBRM 1/5; eBioscience ) , total CD11b ( Clone ICRF44; Biolegend ) , CD49b ( Clone P1E6-C5; Biolegend ) , CD14 ( Clone 63D3; Biolegend ) , C5aR1 ( Clone S5/1; Biolegend ) and CD62L ( Clone DREG-56; Biolegend ) and analysed by flow cytometry . Human THP-1 monocytes were cultured in RPMI containing 10% ( v/v ) FBS at 37°C in 5% CO2 . For differentiation to macrophages , THP-1 monocytes were seeded at a density of 1 . 2 × 106 cells/ml ( in 150 cm2 dish ) and incubated with phorbol 12-myristate 13-acetate ( ( PMA ) 100 ng/mL ) for 72 hr in RPMI with 10% FBS . Prior to treatments , differentiated macrophages were serum starved in RPMI containing 0 . 2% BSA overnight . THP-1 macrophages were then incubated with apoA1 ( 40 ug/ml ) , methyl-β-cyclodextrin ( 3 mM ) or vehicle ( PBS ) for 1 hr prior to stimulation with 10 nM C5a for 5 min . For cholesterol loading experiments differentiated macrophages were incubated with methyl-β-cyclodextrin:cholesterol complexes ( 0 . 5 mg/mL ) for 1 hr , prior to apoA1 incubation . Lipid rafts were prepared according to methods described previously ( Airiian and Mkrtchian , 1987; Gaus et al . , 2005 ) . Following treatment , cells were washed three times with ice-cold PBS , centrifuged ( 100 ×g , 10 min , 4°C ) and resuspended in 1 ml of ice-cold lysis buffer ( 25 mM MES , pH 6 . 5 , 150 mM NaCl , 0 . 2% Triton X-100 , 1 mM PMSF , 1 mM NaF , 0 . 1 mM Na3VO4 , 5 μg/ml leupeptin , 5 μg/ml aprotinin ) . Cells were then sonicated three times on ice for 15 s , and lysates centrifuged ( 100 ×g , 10 min , 4°C ) . The supernatant was gently mixed with an equal volume of 90% ( w/v ) sucrose in MBS ( 25 mM MES , pH 6 . 5 , 150 mM NaCl ) . Two millilitres of the mixture was overlaid with 2 ml of 35 , 30 , 25 , and 5% ( w/v ) sucrose ( all in MBS ) . The sucrose gradient was then spun at 23 , 000 ×g at 4°C in a Beckman SW41 rotor for 16–20 hr . Ten fractions of 1 ml were collected from the top and analysed for protein , and cholesterol . Free cholesterol content of each fraction was measured using the Amplex Red cholesterol assay ( Invitrogen ) , and protein content by the Pierce 660 nm protein assay . To assess the distribution of raft and non-raft marker proteins , fractions were mixed with 4x Laemmli buffer ( 250 mM Tris-HCl , pH 6 . 8 , 8% SDS , 40% glycerol , 0 . 004% bromophenol blue , 20% β-mercaptoethanol ) , heated at 95°C for 5 min , resolved by SDS-PAGE gels , and transferred to PVDF membranes . Membranes were blocked with 5% BSA in TBS-T for 1 hr and then incubated with either rabbit anti-caveolin ( 1:2000 ) ( BD Biosciences ) , mouse anti-flotillin1 ( 1:1000 ) , mouse anti-TfR ( 1:1000 ) diluted in 5% BSA/TBS-T overnight at 4°C . Membranes were then incubated with an HRP-conjugated anti-rabbit or anti-mouse secondary antibodies ( 1:20 , 000 ) for 1 hr at RT . Protein bands were visualised by incubating the membranes for 5 min with Amersham ECL prime and chemiluminescent detection . In some experiments , lipid rafts were characterized in biogel elicited macrophages ( 1 × 106 ) after incubation with PBS , apoA1 ( 40 µg/ml ) or methyl-β-cyclodextrin ( CyD , 3 mM ) for 1 hr prior to staining membrane rafts with cholera toxin subunit B conjugated with Alexa Fluor 647 ( Life Technologies ) for 5 min at 4°C . Cells were then fixed and membrane lipid raft fluorescence determined by flow cytometry , or visualized by microscopy . Intracellular cAMP levels were measured using Discoverx cAMP Hunter eXpress kits ( DiscoveRx , Birmingham , UK ) following the manufacturer’s protocol . Briefly , CHO-K1 cells overexpressing human C5aR1 receptor were plated into a ½ area 96 well plate ( 15 , 000 cells/well ) and incubated at 37°C , 5% CO2 for 24 hr . The medium was then removed and replaced with assay buffer containing a cAMP capture antibody . Cells were pre-treated for 1 hr at 37°C , 5% CO2 with either vehicle or 40 µg/ml apoA1 prior to being stimulated with either vehicle or C5a at the indicated concentration for 30 min at 37°C , 5% CO2 . Cell lysis and cAMP detection were then performed as per the manufacturer’s protocol . Luminescence measurements were taken using a PHERAstar microplate reader ( BMG Labtech , Aylesbury , UK ) . Recruitment of β-Arrestin was measured using the DiscoveRx PathHunter eXpress β-Arrestin GPCR Assay following the manufacturer’s protocol . Briefly , cells were seeded into ½ area 96 well plates and incubated at 37°C , 5% CO2 for 48 hr prior to testing . Agonist or vehicle was then added to the corresponding wells and the plate incubated at 37°C , 5% CO2 for 90 min . Cell lysis and detection reagents were subsequently added and one hour later , luminescence measurements were taken using a PHERAstar microplate reader . Biogel elicited macrophages were seeded into black walled 96 well microplates and left for 4 hr at 37°C , 5% CO2 . The medium was then removed and replaced with RPMI 1640 containing 4 µM FURA2-AM and 0 . 04% pluronic acid ( Life Technologies ) supplemented with either vehicle or 40 μg/ml apoA1 . Plates were then left for 45 min at RT in the dark . Subsequently , cells were washed twice with PBS , the medium replaced with Fluorobrite DMEM ( Life technologies ) and the plate placed into a PHERAstar microplate reader set to 37°C . Macrophages were then stimulated with either vehicle or C5a at the indicated concentration and changes in FURA-2 fluorescence were measured using excitation wavelengths of 340 nm and 380 nm and an emission wavelength of 520 nm . Biogel elicited macrophages ( 2 × 106 ) were pre-incubated with PBS , apoA1 or 3 mM methyl-β-cyclodextrin ( Sigma-Aldrich ) for 1 hr and then stimulated with 10 nM CCL5 for 5 min , washed and lysed in ice cold lysis buffer ( 150 mM NaCl , 0 . 8 mM MgCl2 , 5 mM EGTA , 50 mM HEPES , 1% IGEPAL CA-630 ) supplemented with protease and phosphatase inhibitors ( Sigma-Aldrich ) . Protein concentration was determined using a BCA protein assay kit ( Thermo Fisher Scientific , Loughborough , UK ) following the manufacturer’s protocol . Samples were then diluted 3:1 with 4x Laemmli buffer ( 250 mM Tris-HCl , pH 6 . 8 , 8% SDS , 40% glycerol , 0 . 004% bromophenol blue , 20% β-mercaptoethanol ) and heated at 95°C for 5 min . Samples ( 30 µg of protein ) were resolved on SDS-PAGE gels and transferred onto Hybond ECL nitrocellulose membranes ( GE Healthcare , Buckinghamshire , UK ) . Membranes were blocked with 5% BSA in TBS-T for 2 hr at RT or overnight at 4°C . After blocking , membranes were incubated with rabbit anti-phospho-Akt ( 1:2000 ) ( Cell Signalling Technology , MA ) , or rabbit anti-β-tubulin ( 1:2000 ) ( EMD Millipore ) diluted in 5% BSA/TBS-T for 2 hr at RT or overnight at 4°C . Membranes were then incubated with an HRP-conjugated anti-rabbit secondary antibody ( 1:20 , 000 ) for 1 hr at RT . Protein bands were visualised by incubating the membranes for 5 min with Amersham ECL prime and subsequent exposure to x-ray film over a range of exposure times . To confirm equal protein loading between samples , bound antibodies were removed by incubating the nitrocellulose membranes in stripping buffer ( 60 mM Tris-HCl pH 6 . 8 , 2% SDS , 0 . 8% β-mercaptoethanol ) for 30 min at 50°C . Membranes were blocked with 5% BSA in TBS-T for 2 hr at RT and then incubated with rabbit anti-Akt diluted in 5% BSA/TBS-T for 2 hr at RT . Protein band detection was conducted as described above . Densitometry was performed with Li-cor Image Studio Lite 4 . 0 ( Li-Cor , Cambridge , UK ) RAW 264 . 7 cells were incubated with PBS , apoA1 ( 40 µg/ml ) or methyl-β-cyclodextrin ( 3 mM ) for 1 hr prior to stimulation with C5a ( 10 nM ) . Following treatments , cells were washed three times with 137 mM NaCl , 20 mM Tris HCl pH7 . 4 , 1 mM CaCl2 , 1 mM MgCl2 , 0 . 1 mM Na3VO4 ( buffer A ) , and then lysed in 1 ml of buffer A supplemented with 1 mM phenylmethylsulphonyl fluoride , and 1% NP-40 . Lysed cells were then centrifuged , and the supernatant incubated with 5 µl anti-PI3K p85 antibody ( ABS233 , N-SH2 domain , 0 . 36 mg/mL ) ( Millipore ) and incubated overnight at 4°C . To immunoprecipitate PI3K , 60 µl of 50% slurry of Protein A-agarose beads was added , and samples incubated for 2–4 hr at 4°C . Immunoprecipitated enzyme was collected by centrifugation at 9300 ×g for 5 s . Pellets were washed three times in buffer A plus 1% NP-40 , three times in 0 . 1 M Tris-HCl , pH 7 . 4 , 5 mM LiCl , 0 . 1 mM Na3VO4 and twice with 10 mM Tris-HCl , pH 7 . 4 , 150 mM NaCl , 5 mM EDTA , 0 . 1 mM Na3VO4 . Beads were then incubated with 30 µl KBZ reaction buffer ( Echelon Biosciences , Utah , USA ) , and PI3K activity measured by incubating with PIP2 , and measuring PIP3 production by a competitive ELISA as per the manufacturer’s instructions ( PI3-Kinase Activity ELISA: Pico , Echelon Biosciences ) . All quantitative data are reported as mean ± SEM of n observations . Statistical evaluation was performed using Student’s t-test ( where two variables were analysed ) or one-way analysis of variance ( ANOVA ) ( Prism 6 GraphPad Software , San Diego , CA ) followed by Dunnett’s or Bonferroni multiple comparison posthoc test , taking a probability p<0 . 05 as statistically significant .	A molecule called cholesterol is an important component of the membranes found in cells and is also used to make some hormones and other useful molecules . However , cholesterol can also contribute to the formation of plaques in arteries , which can lead to a disease called atherosclerosis , the cause of heart attacks . Particles called high density lipoproteins ( HDL ) carry cholesterol around the body in the bloodstream and are thought to have anti-inflammatory properties . A protein called apoA1 is a major component of HDL particles and , acting as part of a HDL particle or alone , it removes cholesterol from cells . Atherosclerotic plaques form when white blood cells collect in places where the arteries are inflamed . The membranes that surround the white blood cells contain receptors that are able to detect inflammatory signals called chemokines . These receptors eventually communicate with the machinery needed for cell movement . This machinery is concentrated in parts of the membrane known as lipid rafts . Iqbal , Barrett et al . investigated whether apoA1 can block the movement of mouse and human white blood cells towards the chemokines produced during inflammation . The experiments show that apoA1 treatment strongly inhibited the movement of white blood cells towards a range of chemokines in a culture dish . The apoA1 protein removes cholesterol from lipid rafts in the membrane of the white blood cell , which changes the properties of the membrane and decreases the activity of the machinery needed for cell movement . Further experiments in mice with inflammation of the peritoneum , the thin layer of tissue that lines the inside of the abdomen , produced similar findings . The next step following on from this work would be to investigate whether apoA1 treatment can reduce the accumulation of white blood cells in mice that act as models of other inflammatory diseases , such as arthritis and atherosclerosis .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "immunology", "and", "inflammation" ]	2016	Acute exposure to apolipoprotein A1 inhibits macrophage chemotaxis in vitro and monocyte recruitment in vivo
In triploblastic animals , Par-proteins regulate cell-polarity and adherens junctions of both ectodermal and endodermal epithelia . But , in embryos of the diploblastic cnidarian Nematostella vectensis , Par-proteins are degraded in all cells in the bifunctional gastrodermal epithelium . Using immunohistochemistry , CRISPR/Cas9 mutagenesis , and mRNA overexpression , we describe the functional association between Par-proteins , ß-catenin , and snail transcription factor genes in N . vectensis embryos . We demonstrate that the aPKC/Par complex regulates the localization of ß-catenin in the ectoderm by stabilizing its role in cell-adhesion , and that endomesodermal epithelial cells are organized by a different cell-adhesion system than overlying ectoderm . We also show that ectopic expression of snail genes , which are expressed in mesodermal derivatives in bilaterians , is sufficient to downregulate Par-proteins and translocate ß-catenin from the junctions to the cytoplasm in ectodermal cells . These data provide molecular insight into the evolution of epithelial structure and distinct cell behaviors in metazoan embryos . Bilaterian animals comprise more than the 95% of the extant animals on earth and exhibit enormous body plan diversity ( Martindale and Lee , 2013 ) . One of the most important morphological features in bilaterian evolution is the emergence of the mesoderm , an embryological tissue that gives rise important cell types such as muscle , blood , cartilage , bone , and kidneys in the space between ectoderm and endoderm . The emergence of mesoderm clearly contributed to the explosion of biological diversity throughout evolution ( Martindale and Lee , 2013; Martindale , 2005 ) . Cnidarians ( e . g . sea anemones , corals , hydroids , and ‘jellyfish’ ) are the sister group to bilaterians , and despite their surprisingly complex genomes ( Putnam et al . , 2007 ) , do not possess a distinct mesodermal tissue layer . Instead , the gastrodermal lining to their gut cavity consists of a bifunctional endomesodermal epithelium with molecular characteristics of both bilaterian endodermal and myoepithelial mesodermal cells ( Martindale and Lee , 2013; Technau and Scholz , 2003; Martindale et al . , 2004; Jahnel et al . , 2014 ) . For example , brachyury and snail , among other genes , contribute to the specification of the endomesodermal fates in both bilaterian and cnidarian embryos ( Technau and Scholz , 2003; Martindale et al . , 2004; Magie et al . , 2007; Yasuoka et al . , 2016; Servetnick et al . , 2017 ) . Yet in bilaterians , mesodermal cells segregate from an embryonic endomesodermal precursor to form both endoderm and a third tissue layer ( mesoderm ) not present in the embryos of diploblastic cnidarians ( Martindale et al . , 2004; Rodaway and Patient , 2001; Davidson et al . , 2002; Maduro and Rothman , 2002; Solnica-Krezel and Sepich , 2012 ) . How mesodermal cells originally segregated from an ancestral endomesodermal epithelium during animal evolution is still unclear ( Martindale and Lee , 2013; Martindale , 2005; Technau and Scholz , 2003 ) , particularly because virtually all of the genes required for mesoderm formation are present in cnidarian genomes ( Putnam et al . , 2007; Baumgarten et al . , 2015; Chapman et al . , 2010; Shinzato et al . , 2011 ) . During the last decade , several studies have described molecular and cellular characteristics related to the segregation of mesoderm during bilaterian development ( Solnica-Krezel and Sepich , 2012; Keller et al . , 2003; Darras et al . , 2011; Schäfer et al . , 2014 ) . Here , we investigate the cellular basis of morphogenesis during embryogenesis of the ‘diploblastic’ sea anemone , Nematostella vectensis . In most bilaterian embryos described to date , after a series of synchronous and stereotyped cleavage divisions , maternal determinants induce the localization of nuclear ß-catenin to blastomeres derived from the vegetal pole ( Martindale and Lee , 2013 ) . Hence , gastrulation and the specification of endomesodermal fates is restricted to the vegetal pole . In these species , brachyury is expressed at the border of the blastopore and snail is expressed in the prospective mesodermal tissues ( Technau and Scholz , 2003 ) . The formation of mesoderm involves a variety of cellular processes including the downregulation of E-cadherin , loss of apicobasal cell polarity , and in some cases , the induction of epithelial-to-mesenchymal transition ( EMT ) ( Solnica-Krezel and Sepich , 2012; Schäfer et al . , 2014; Acloque et al . , 2009; Lim and Thiery , 2012 ) . Embryos of the cnidarian starlet sea anemone N . vectensis develop without a stereotyped cleavage pattern but cell fates become organized along the embryonic animal-vegetal axis ( Fritzenwanker et al . , 2007; Salinas-Saavedra et al . , 2015 ) . During blastula formation , embryonic cells of N . vectensis form a single hollow epithelial layer . Epithelial cells of the animal pole , characterized by the nuclear localization of Nvß-catenin prior to gastrulation ( Wikramanayake et al . , 2003; Lee et al . , 2007 ) , invaginate by apical constriction to form the endomesodermal epithelium ( Magie et al . , 2007; Tamulonis et al . , 2011 ) . The expression of Nvbrachyury around the presumptive border of the blastopore and Nvsnail genes in the presumptive endomesodermal gastrodermis of N . vectensis embryos occurs even before the morphological process of gastrulation begins ( Scholz and Technau , 2003; Röttinger et al . , 2012 ) . Interestingly , the components of the intracellular polarity Par system ( NvaPKC , NvPar-6 , NvPar-3 , NvPar-1 , and NvLgl ) , which show a highly polarized bilaterian-like subcellular distribution throughout all epithelial cells at the blastula stage in N . vectensis ( Salinas-Saavedra et al . , 2015 ) , are specifically degraded and down-regulated from the endomesoderm during the gastrulation process ( Figure 1A ) . We have previously suggested that the expression of bilaterian ‘mesodermal genes’ ( e . g . Nvsnail ) might induce the loss of apicobasal cell-polarity indicated by the absence of the components of the Par system in the endomesoderm of N . vectensis embryos ( Salinas-Saavedra et al . , 2015 ) . Recent studies in N . vectensis and bilaterians have provided information that supports this hypothesis . For example , it has been shown that snail is necessary and sufficient to downregulate Par3 in Drosophila mesoderm , inducing the disassembly of junctional complexes in these tissues ( Weng and Wieschaus , 2016 , 2017 ) . In addition , we have shown that Nvbrachyury regulates epithelial apicobasal polarity of N . vectensis embryos , suggesting some aspects of epithelial cell polarity are highly conserved ( Servetnick et al . , 2017 ) . Together , this evidence suggests a plausible cellular and molecular mechanism for the segregation of a distinct cell layer in bilaterian evolution from an ancestral bifunctional endomesodermal tissue . Thus , in this study , we describe the functional association between the components of the Par system , apical junctions , epithelial integrity , and the nuclearization of Nvß-catenin in a cnidarian embryo . In addition , we demonstrate that the endomesoderm in N . vectensis is organized by different junctional complexes that confer different functional properties to this tissue than the overlying ectoderm . And finally , we investigate the putative interactions between the components of the Par system , the canonical Wnt signaling pathway , and snail gene expression , giving insights on the evolution of the mesoderm and EMT . Components of the Par system are not present in the cells of endomesodermal epithelium of N . vectensis during gastrulation , even though the very same cells express these proteins during the blastula stage ( Salinas-Saavedra et al . , 2015 ) ( Figure 1 ) . This absence is consistent with the absence of apical Adherens Junctions ( AJs ) in the endomesoderm of N . vectensis ( Figure 1—figure supplement 1 ) and other cnidarians ( Magie et al . , 2007; Chapman et al . , 2010; Ganot et al . , 2015 ) . At polyp stages , neither ß-catenin ( an AJ-associated protein ) ( Figure 1B and C ) nor the Par proteins ( Figure 1—figure supplement 1C ) are detectable in endomesodermal cells of either the gastrodermis or the pharynx . When N . vectensis embryos are stained with antibodies to ß-catenin ( Figure 1 ) or if Nvß-catenin::GFP mRNA is expressed in uncleaved zygotes ( Figure 1—figure supplement 1B ) , clear localization of ß-catenin can be seen in the cortex of ectodermally derived epithelial cells ( Figures 1B , C , D , G and I ) , but not in endomesodermal cells ( Figure 1B and C ) . In pharyngeal cells that are located between the epidermis and gastrodermis , Nvß-catenin ( Figure 1B and D ) , NvPar-6 ( Figure 1E ) , and NvPar-1 ( Figure 1F ) expression begins to disappear , and is localized only in the most apical regions , indicating that AJs are being disassembled/degraded during the gastrulation process ( Figure 1G and H ) . During later planula stages , ß-catenin and the components of the Par system display scattered patterns in the cytoplasm of a small subset of endomesodermal cells ( Figure 1I and J ) . Even though we do not know the identity of these cells , this expression temporally coincides with the transient activation of Wnt signaling emanating from the oral pole ( Kusserow et al . , 2005; Marlow et al . , 2013 ) at those developmental stages . In bilaterians ( Acloque et al . , 2009; Lim and Thiery , 2012 ) and N . vectensis ( Kusserow et al . , 2005; Marlow et al . , 2013 ) , the later activation of Wnt signaling is also associated with neurogenesis , and may cause the observed changes in protein localization . Regardless of this scattered expression , it is clear that cells that undergo gastrulation in N . vectensis lose their polarized ectodermal cell-cell adhesion complex and components of the Par system , including ß-catenin , are downregulated from endomesodermal tissues ( Figure 1 ) . In bilaterians , the proper formation of an epithelial paracellular barrier ( essential for tissue homeostasis ) depends on the establishment of adhesive complexes between adjacent cells ( Higashi et al . , 2016; Jonusaite et al . , 2016 ) , which are regulated by the aPKC/Par complex ( Ohno et al . , 2015 ) . To test if this absence of protein expression is correlated to differential cell-cell adhesion in the endomesodermal epithelium of N . vectensis , we assessed their role in regulating paracellular movements between ectodermal and endomesodermal epithelia by using a fluorescent tracer dye penetration assay ( Figure 2A ) ( Higashi et al . , 2016 ) . For the purposes of these experiments , in order to avoid unwanted results related to tissue specification , cell proliferation , and cell movements , we used newly hatched juvenile polyps where the gastrodermis ( endomesodermally derived ) is fully differentiated . N . vectensis polyps were exposed to media containing 10 , 000 MW fluorescent dextran ( Molecular Probes , Inc . ) . When juvenile polyps are incubated in dextran for 5–10 min ( Figure 2B ) , fluorescent dextran solution moves into the gastric cavity and then spreads into the mesoglea through the gastrodermal epithelium ( Figure 2C ) but does not enter the mesoglea through the outer ectodermally-derived epidermis ( Figure 2C and D ) . These results suggest that cell-cell adhesion is differentially regulated between the epidermis and gastrodermis and the absence/disruption of AJs may compromise Septate Junctions ( SJs ) in the gastrodermis . Similar results were obtained in N . vectensis polyps when we overexpressed NvPar-3::mVenus by injection of mRNA into uncleaved eggs which is normally expressed in ectodermal but not endodermal epithelial tissue ( Figure 2D and E ) . However , in polyps expressing a dominant negative version of NvPar-3::mVenus ( dnNvPar-3; microinjected into uncleaved eggs ) dye penetrated between epithelial cells in both the gastrodermis and the outer epidermis ( Figures 2D , F and G ) , demonstrating an ancestral role of the aPKC/Par complex in the maintenance of cell-cell adhesion and the paracellular boundary ( SJs ) of epithelial cells during animal development . Our results suggest that the absence of Par proteins in the endomesoderm is associated with changes in cell-cell adhesion complexes . Pharmacological treatment of N . vectensis embryos with an aPKC activity inhibitor blocks cytokinesis but not mitosis in cleaving embryos ( Figure 3A ) . In addition , a dominant negative version of NvPar-1 ( dnNvPar-1 ) , that lacks its kinase domain , localizes only to the cortex of cell–cell contacts ( Figure 3B ) . Since NvPar-1 is phosphorylated by NvaPKC ( Figure 3—figure supplement 2 ) , we predict that , as in other systems , dnNvPar-1 could be phosphorylated by aPKC but would not phosphorylate the aPKC/Par complex ( Vaccari et al . , 2005; Böhm et al . , 1997 ) . Thus , dnNvPar-1 can localize to the cell cortex where aPKC may be inactive . These results together suggest that the formation of cell–cell contacts is regulated by the activity of the aPKC/Par complex in N . vectensis embryos ( Figure 3C ) . We further tested this hypothesis by using genome editing by CRISPR/Cas9 targeting Nvpar-6 and Nvpar-3 genes ( Figure 3D ) . We did not observe any effects on the embryo until 36 hpf at 16°C ( late blastula stage ) , indicating the activity of maternally loaded proteins up until that stage . When NvPar-6 and NvPar-3 are mutated , the ectodermal epithelium loses its integrity , presenting changes in thickness ( Figure 3—figure supplement 1B and Figure 3—figure supplement 3 ) , and interestingly , the endomesoderm ( which does not express these proteins ) generates cells with mesenchymal-like morphotypes that are never normally seen in this species ( Figure 1D and E ) . In Nvpar-6 and Nvpar-3 mutant embryos , we also observed the disruption of microtubules and actin cytoskeleton ( Figure 3—figure supplement 4B ) , and AJs ( visualized with the ß-catenin antibody in Figure 4 ) that confirms our previous observations of their role in regulating ectodermal cell polarity . Although it was difficult to dissect significant changes in the expression of germ layer markers ( e . g . Nvbra , Nvsnail , NvSix3/6 , and Nvfz10 ) from the morphological changes associated with epithelial integrity when these genes were disrupted ( Figure 3—figure supplement 4E ) , it is clear that the primary defect in NvPar3 KO were aspects of cell adhesion and not cell type specification . Similar results were obtained when we overexpressed the mRNA encoding for a dominant negative version NvPar-6 ( dnNvPar-6 ) and NvPar-3 ( dnNvPar-3 ) into N . vectensis eggs ( Figure 3—figure supplement 1 and Figure 3—figure supplements 5 and 6 ) . However , dominant negative effects on the injected embryos were observed at earlier stages ( 10–12 hpf ) than the CRISPR/Cas9 mutants ( zygotic expression ) because the mutant proteins compete with the wild type proteins ( maternally loaded ) . Hence , in these experiments , embryonic lethality ( 90% ) and cell death were higher . One surprising observation from the experiments described above show that the changes observed in the ectodermal and endodermal epithelium after disrupting NvPar-6 and NvPar-3 ( Figure 3 ) suggests some sort of trans-epithelial regulation of cell–cell adhesion ( most likely involving AJs ) because these Par genes are not expressed in the endomesoderm . The polarizing activity of the aPKC/Par complex in the ectoderm is thus necessary to maintain the integrity of both ecto- and endodermal epithelia during cellular movements associated with gastrulation . To assess whether the observed phenotypes on cell-cell adhesion are related to non-autonomous cell regulation ( trans-epithelial interactions ) , we repeated the above experiments randomly injecting single blastomeres at 3–4 hpf ( 8–16 cell-stage ) to make mutant clones in an otherwise wild type background ( Figure 4 ) . In these experiments , only the cell-lineage of the injected blastomere would be affected and would exhibit defective cell-cell adhesion in an otherwise undisturbed wild-type background . If endomesodermal cells derived from an injected blastomere display fibroblast/mesenchymal cell morphology , it would indicate that the organization of the endomesodermal epithelium is not dependent on the ectoderm but , rather , an intrinsic cell-autonomous activity of the aPKC/Par complex ( Figure 4—figure supplement 1 ) . Our results show that only ectodermal- but not endomesodermal-lineages are affected by these mutations ( Figure 4 and Figure 4—figure supplement 2 ) . Presumptive ectodermal cells derived from an injected blastomere fail to maintain AJs ( and potentially SJs ) and the resulting clone of epithelial cells loses its structural integrity inducing cell extrusion . In contrast , presumptive endomesodermal cells derived from an injected blastomere develop into a normal endomesodermal epithelium ( Figure 3F ) . Our results complement the work of ( Kirillova et al . , 2018 ) and demonstrate that the proper cell-cell adhesion of the ectodermal layer somehow regulates trans-epithelially the integrity of the endomesodermal layer . This regulation may maintain the tension between cells during invagination at gastrula stages , or , in conjunction with the extracellular matrix ( ECM ) and basal cues , it may influence signaling patterns necessary to organize epithelial layers during N . vectensis embryogenesis . N . vectensis has two snail genes , Nvsnail-A and Nvsnail-B , which are both expressed in the endomesodermal plate prior to and throughout the gastrulation process , and which define the boundary between gastrodermis and ectodermal pharynx ( Magie et al . , 2007; Röttinger et al . , 2012; Amiel et al . , 2017 ) . To determine the role of Nvsnail genes on ß-catenin nuclearization , we co-injected the mRNA of NvSnail-A::mCherry , NvSnail-B::mCherry , and Nvß-catenin::GFP into uncleaved eggs . The overexpression of both proteins NvSnail-A::mCherry and NvSnail-B::mCherry together induce the ectopic translocation of Nvß-catenin::GFP to the nuclei of ectodermal cells ( Figure 6A ) . This treatment also delocalizes NvPar-3 from the cell cortex when both NvSnail::mCherry proteins are co-expressed with NvPar-3::mVenus ( Figure 6B ) . To determine the role of Nvsnail genes on cell adhesion/epithelial polarity , we randomly injected single blastomeres at the 8–32 cell-stage with mRNA from both NvSnail-A::mCherry and NvSnail-B::mCherry together . The fluorescent dextran that was co-injected with the mRNAs could be used to detect the clones where the over-expression of the co-injected mRNAs occurred in a ‘wild-type’ background ( Figure 6D ) . Similar to the Nvpar-3 knock-out ( Figure 4 ) , the expression of Nvsnail genes is sufficient to induce the degradation of Par proteins and AJs ( ß-catenin ) from the ectoderm and disrupts its epithelial integrity; however , nuclear ß-catenin was not observed under these treatments ( Figure 6D ) . Thus , nuclear Nvß-catenin::GFP observed in vivo when we overexpressed NvSnail proteins ( Figure 6A ) is a consequence of the high cytosolic availability generated by its ectopic overexpression and release from AJs . Interestingly , not every ectodermal cell was affected by these treatments even though all of the cells expressed the injected mRNAs ( Figure 6A and E , and Figure 6—figure supplement 1 ) . This patched pattern suggests that the response to Nvsnail over-expression is spatially regulated . These results suggest that the role of Nvsnail genes on AJs and apicobasal cell polarity is constrained to the site of gastrulation in N . vectensis embryos under natural conditions , and that these genes may be required for gastrulation movements . Therefore , we predicted that ß-catenin ( AJs ) and Par proteins will be retained in the cells of the N . vectensis endomesodermal plate if both Nvsnail genes are disrupted . The snail genes temporally down-regulate E-cadherin during mesoderm segregation and EMT in bilaterian animals ( Lim and Thiery , 2012 ) . As we have shown here , as well as in previous studies ( Magie et al . , 2007; Magie and Martindale , 2008 ) , the cells comprising the endomesodermal plate lose their cell-cell adhesion during gastrulation in N . vectensis embryos . It may be possible that temporal regulation of endomesodermal patterning might act upon the AJs . Our data suggest that once gastrulation is complete and the pharynx forms , components of the Par system and the ß-catenin components of the AJs are degraded from both the cortex and cytoplasm of endomesodermal cells ( Figure 1 and Figure 1—figure supplement 1 ) . Hence , it could be possible that Nvbrachyury induces the disruption of apicobasal polarity ( Servetnick et al . , 2017 ) , remnant AJs maintain the endomesodermal-plate cells together , and Nvsnail genes degrades and prevents the reassembly of AJs in the endomesoderm of N . vectensis . To address these issues , we used CRISPR/Cas9 knock-out of Nvsnail-A and Nvsnail-B genes together to inhibit zygotic function of these genes and investigate their role on the temporal regulation of AJs and cell polarity . In CRISPR/Cas9 mutants , the endomesodermal plate forms but it does not migrate further than its first invagination during gastrulation ( Figure 7 ) . Furthermore , AJs ( labeled with ß-catenin ) and apical Par proteins ( labeled with antiNvPar-6 and antiNvaPKC ) are retained at the apical cortex of the cells of the endomesodermal plate ( Figure 7B and C and Figure 7—figure supplement 1 ) . Surprisingly , NvPar-1 and NvLgl were not detected in those cells ( Figure 7C ) , suggesting that the degradation of these basolateral proteins precede or do not depend on the activity of the Nvsnail genes . This suggests that Nvsnail regulates apical cell-polarity , AJs turnover , and the migration ( ‘zippering’ ) but not the invagination of the endomesodermal plate during gastrulation of N . vectensis embryos ( Figure 7D ) . Interestingly , the invagination of the endomesodermal plate ( controlled by the Wnt/PCP pathway ) is uncoupled from its specification in N . vectensis embryos ( Kumburegama et al . , 2011; Wijesena et al . , 2011 ) , which is consistent with our observations . The segregation of different germ layers during embryogenesis of many bilaterian animals is carried out by similar cellular mechanisms . EMT is a shared mechanism utilized by mesoderm , neural crest cell ( NCC ) , and tumorigenesis to delaminate cells in bilaterian animals ( triploblastic animals ) . During EMT , the nuclearization of ß-catenin induces the expression of ‘endomesodermal’ genes like brachyury and snail ( Acloque et al . , 2009 ) . The expression of these genes downregulates epithelial cadherins , disrupts apicobasal polarity ( mediated by the aPKC/Par complex ) , disassembles AJs , and induces changes in cytoskeleton organization ( Acloque et al . , 2009; Lim and Thiery , 2012 ) . A rearrangement of the actin-myosin cytoskeleton induces apical constriction of cells ( generating a bottle-like shape ) , which detach from the epithelial sheet , break down the basal membrane , and invade a specific tissue as mesenchymal cells ( Acloque et al . , 2009; Lim and Thiery , 2012; Ohsawa et al . , 2018 ) . Interestingly , mesoderm formation , tumorigenesis , and EMT have never been described as natural processes during N . vectensis ( a diploblastic animal ) embryogenesis . During N . vectensis gastrulation ( Magie et al . , 2007; Tamulonis et al . , 2011 ) , cells around the edge of the blastopore at the animal pole ( which expresses Nvbrachyury ) acquire a bottle-like shape by apical constriction , leading to epithelial buckling and the invagination of presumptive endomesoderm ( which expresses Nvsnail ) . However , throughout this process the endomesoderm remains as a monolayer of epithelial cells and individual mesenchymal cells never detach and invade the blastocoel . We have shown that by disrupting the aPKC/Par complex ( apicobasal cell-polarity ) in N . vectensis ( Figures 3 , 4 and 5 ) , we are able to convert cells from the endomesodermal epithelium into mesenchymal-like cells , translocate Nvß-catenin ( Figure 5A ) , and emulate EMT-like processes ( apical constriction and individual cell-detachments ) in the ectodermal epithelium of N . vectensis treated-embryos ( Figure 5C and D ) . These results demonstrate that the cnidarian N . vectensis possesses mechanisms necessary to segregate individual germ layers ( e . g . mesoderm and NCC ) described in bilaterians; however , they do not do it . Our working hypothesis is that the N . vectensis embryo is composed of two independent morphogenetic modules that are integrated and organized by the pharynx ( Figure 7D ) . The first observation is that the ectoderm , whose apicobasal polarity ( and thus , AJs and epithelial integrity ) is regulated by Nvbrachyury that promotes ectodermal epithelial morphogenesis and pharynx formation ( Servetnick et al . , 2017 ) , and the second module is generated by endomesodermal differentiation and cell-movements that are regulated by Nvsnail genes . This is supported by the expression Nvbrachyury in Nvsnail knock-out embryos ( Figure 7—figure supplement 2 ) , and Nvsnail knock-out phenotypes where ectodermal pharynx develops normally but no clear endomesoderm is formed ( Figure 7—figure supplement 1 ) . Additional work is required to elucidate any differences in function between Nvsnail-A and Nvsnail-B genes , however , both modules are specified by nuclear ß-catenin ( Röttinger et al . , 2012 ) , suggesting that the nuclear ß-catenin ( maternal ) shift from the animal pole in cnidarians to the vegetal pole in bilaterians is mechanistically plausible and sufficient to re-specify the site of gastrulation and germ-layers along the animal-vegetal axis during Metazoan evolution ( Martindale and Lee , 2013; Lee et al . , 2007 ) . Bilaterian-EMT has been a focus of study for decades as a mechanism to segregate different cell layers involved in a variety of different normal and pathological biological processes ( Ohsawa et al . , 2018; Nieto et al . , 2016 ) . This process appears to depend on the fine regulation of snail expression levels and their temporal activity . For example , during NCC migration , cells display ‘partial-EMT’ where cells remain attached to several neighboring cells but their apicobasal polarity and AJs are down-regulated , allowing collective-cell migration ( Weng and Wieschaus , 2017; Nieto et al . , 2016; Lee et al . , 2006; Theveneau and Mayor , 2013; Ribeiro and Paredes , 2014 ) . Our data suggest that ‘partial-EMT’ may be the mechanism by which the endomesodermal epithelium migrates into the blastocoel in N . vectensis embryos during normal gastrulation ( Figure 8 ) . In this scenario , upstream factors that regulate snail transcription may be critical for this process . In bilaterian animals , there are many other pathways in addition to the canonical Wnt pathway that activate snail transcription and induce the disruption of AJs and apicobasal cell polarity . For example , TGFß , BMP , NANOS , FGF , and MEK/ERK/ERG take on roles during the specification of mesoderm , NCC migration , tumorigenesis , and other EMT-related processes ( Lim and Thiery , 2012; Nieto et al . , 2016; Barrallo-Gimeno and Nieto , 2005 ) . Concordantly in N . vectensis embryos , cells of the pharyngeal and endomesodermal tissues express components of all these pathways ( Röttinger et al . , 2012; Amiel et al . , 2017; Extavour et al . , 2005; Matus et al . , 2006 , 2007; Wijesena et al . , 2017 ) that may modify their cellular characteristics . For example , one cadherin ( NvCDH2 Clarke et al . , 2016: 1g244010 ) , and kinases that modify tubulin and histones are differentially regulated between ecto- and endomesodermal epithelium ( Wijesena et al . , 2017 ) . In conclusion , N . vectensis has both up and downstream cellular and molecular mechanisms associated with EMT described in bilaterians . However , N . vectensis does not segregate a distinct mesodermal germ layer nor display EMT under natural conditions . In bilaterians , this mechanism must have evolved to segregate mesodermal cells from the endoderm to retain the tight cell-cell junctions required in endodermal epithelia . Interestingly , mesoderm segregation via EMT in Drosophila takes place after epithelial folding in response to snail expression . In these embryos , contractile myosin enhances the localization of AJs and Par-3 in the presumptive mesoderm and prevents their downregulation by Snail , thus delaying EMT ( Weng and Wieschaus , 2016 , 2017 ) . Furthermore , the overexpression of Snail in Drosophila embryos is sufficient to disassemble ectodermal-AJs , but mesodermal-AJs are maintained by actomyosin contraction that antagonize Snail effects ( Weng and Wieschaus , 2016 , 2017 ) . Our results suggest a similar mechanism since Nvsnail overexpression in endomesodermal lineages ( Figure 6—figure supplement 1 ) is not sufficient to segregate cells and the endomesoderm remains as an epithelium . However , unlike Drosophila , Par proteins and AJs are not enhanced but degraded during the gastrulation of N . vectensis ( Figure 1 ) . As it is discussed in ( Weng and Wieschaus , 2017 ) , not only the degradation but also the turnover of AJs and Par proteins in adjacent epithelia is essential for EMT-mediated germ layer segregation in different animals . The dual identity of N . vectensis endomesoderm is characterized by the continuous expression of Nvsnail genes ( Martindale et al . , 2004 ) that repress the turnover of AJs and may play a role in inhibiting EMT from occurring ( Figures 6 and 7 ) . Interestingly , components of the Wnt/PCP pathway are expressed only in the endomesoderm ( Kumburegama et al . , 2011; Wijesena et al . , 2011 ) , while components of the Par system are expressed only in the ectoderm ( Salinas-Saavedra et al . , 2015 ) . It could be that NvSnail degrades AJs and inhibits their re-assembly in the endomesoderm , but the activation of contractile myosin by the Wnt/PCP pathway maintains the endomesodermal cells together in N . vectensis embryos . Hence in bilaterians , a mechanism ( most likely downstream of Snail ) that connects the cytoskeleton with cell-polarity may have evolved to tighten cell-cell adhesion in the endoderm and allow EMT . To elucidate this , further comparative research and funding are needed to understand the cellular mechanisms that evolve to segregate mesoderm and control epithelial cell polarity at the base of the metazoan tree . For example , cnidarians , poriferans , and ctenophores present intriguing characteristics to study . In cnidarians , different modes of gastrulation have been described between species including unipolar and multipolar cell ingression and delamination ( Kirillova et al . , 2018; Byrum and Martindale , 2004; Marlow and Martindale , 2007 ) . Poriferans display EMT-like processes and cell morphologies during regeneration and trans-differentiation ( Nakanishi et al . , 2014; Coutinho et al . , 2017 ) . However , whether or not these processes involve similar molecular and cellular mechanisms are still unclear . Interestingly , ctenophores segregate a mesodermal cell population during embryogenesis but do not have the genes that encode for all cell-cell adhesion complexes and specify for mesoderm in bilaterians ( Figure 8—figure supplement 1 ) ( Ganot et al . , 2015; Ryan et al . , 2013 ) . Thus , there is much to be learned by the comparative study of cell biology to understand the evolutionary origins of EMT and germ layer formation . Spawning , gamete preparation , fertilization and embryo culturing of N . vectensis ( RRID:SCR_005153 ) embryos was performed as previously described ( Röttinger et al . , 2012; Hand and Uhlinger , 1992; Layden et al . , 2013; Wolenski et al . , 2013 ) . Adult N . vectensis were cultivated at the Whitney Laboratory for Marine Bioscience of the University of Florida ( USA ) . Males and females were kept in separate glass bowls ( 250 ml ) in 1/3x seawater ( salinity: 12pp ) reared in dark at 16°C . Animals were fed freshly hatched Artemia three times a week and macerated oyster the day before spawning . Spawning was induced by incubating the adults under an eight-hour light cycle at 25°C the night before the experiment . Distinct groups of animals were spawned once every 2 weeks . Oocytes and sperm were collected separately and fertilized in vitro by adding sperm to egg masses for 25 min . The jelly mass surrounding the fertilized eggs was removed by incubating the eggs in 4% L-Cysteine ( in 1/3x seawater; pH 7 . 4 ) for 15–17 min and then washed 3 times with 1/3x seawater . De-jellied eggs were kept in glass dishes ( to prevent sticking ) in filtered 1/3 seawater at 16°C until the desired stage . All immunohistochemistry experiments were carried out using the previous protocol for N . vectensis ( Salinas-Saavedra et al . , 2015 ) with a slight modification in the glutaraldehyde concentration to allow better antibody penetration . Embryos were fixed on a rocking platform at room temperature in two consecutive steps . Embryos of different stages were fixed for no longer than 3 min in fresh Fix-1 ( 100 mM HEPES pH 6 . 9; 0 . 05M EGTA; 5 mM MgSO4; 200 mM NaCl; 1x PBS; 3 . 7% Formaldehyde; 0 . 2% Glutaraldehyde; 0 . 5% Triton X-100; and pure water ) . Then , Fix-1 was removed and replace with fresh Fix-2 ( 100 mM HEPES pH 6 . 9; 0 . 05M EGTA; 5 mM MgSO4; 200 mM NaCl; 1x PBS; 3 . 7% Formaldehyde; 0 . 05% Glutaraldehyde; 0 . 5% Triton X-100; and pure water ) . Embryos were incubated in Fix-2 for 1 hr . Fixed embryos were rinsed at least five times in PBT ( PBS buffer plus 0 . 1% BSA and 0 . 2% Triton X-100 ) for a total period of 3 hr . PBT was replaced with 5% normal goat serum ( NGS; diluted in PBT ) and fixed embryos were blocked for 1 to 2 hr at room temperature with gentle rocking . Primary antibodies were diluted in 5% NGS to desired concentration . Blocking solution was removed and replaced with primary antibodies diluted in NGS . All antibodies incubations were conducted over night on a rocker at 4°C . After incubation of the primary antibodies , samples were washed at least five times with PBT for a total period of 3 hr . Secondary antibodies were then applied ( 1:250 in 5% NGS ) and samples were left on a rocker overnight at 4°C . Samples were then washed with PBT and left on a rocker at room temperature for an hour . To visualize F-actin , samples were incubated then for 1 . 5 hr in Phalloidin ( Invitrogen , Inc . Cat . # A12379 ) diluted 1:200 in PBT . Samples were then washed once with PBT and incubated with DAPI ( 0 . 1 µg/µl in PBT; Invitrogen , Inc . Cat . # D1306 ) for 1 hr to allow nuclear visualization . Stained samples were rinsed again in PBS two times and dehydrated quickly into isopropanol using the gradient 50 , 75 , 90 , and 100% , and then mounted in Murray’s mounting media ( MMM; 1:2 benzyl benzoate:benzyl alcohol ) for visualization . Note that MMM may wash DAPI out of your sample . For single blastomere microinjection experiments , after Phalloidin staining , samples were incubated with Texas Red Streptavidin ( 1:200 in PBT from 1 mg/ml stock solution; Vector labs , Inc . Cat . # SA-5006 . RRID:AB_2336754 ) for 1 hr to visualize the injected dextran . We scored more than 1000 embryos per each antibody staining and confocal imaged more than 50 embryos at each stage . The primary antibodies and concentrations used were: mouse anti-alpha tubulin ( 1:500; Sigma-Aldrich , Inc . Cat . # T9026 . RRID:AB_477593 ) , rabbit anti-ß-catenin ( 1:300; Sigma-Aldrich , Inc . Cat . # C2206 . RRID:AB_476831 ) , mouse anti-histone H1 ( 1:300; F152 . C25 . WJJ , Millipore , Inc . RRID:AB_10845941 ) . Rabbit anti-NvaPKC , rabbit anti-NvLgl , rabbit anti-NvPar-1 , and rabbit anti-NvPar-6 antibodies are custom made high affinity-purified peptide antibodies that were previously raised by the same company ( Bethyl Inc . ) . All these four antibodies are specific to N . vectensis proteins ( Salinas-Saavedra et al . , 2015 ) and were diluted 1:100 . Secondary antibodies are listed in Key resources table . Primary polyps were incubated and mounted in 1/3 sea water with fluorescent dextran solution ( 0 . 5 mg/ml ) . For uninjected embryos we used Dextran , Alexa Fluor 555 ( Molecular Probes , INC . Cat . # D34679 ) . For injected embryos , expressing fluorescent proteins , we used Dextran , Alexa Fluor 647 ( Molecular Probes , INC . Cat . # D22914 ) . Animals were observed within 10 min of incubation . 15 animals were recorded per treatment . For better visualization of the dextran solution inside the gastric cavity as shown in Figure 2B , we delivered additional dextran solution by microinjecting dye through the polyp’s mouth . For the rest of the experiments , we mainly focused in the ectodermal permeability and we let the polyps to eat the solution by themselves as grown babies . The coding region for each gene of interest was PCR-amplified and cloned into pSPE3-mVenus or pSPE3-mCherry using the Gateway system ( Roure et al . , 2007 ) . Eggs were injected directly after fertilization as previously described ( Salinas-Saavedra et al . , 2015; Layden et al . , 2013; DuBuc et al . , 2014 ) with the mRNA encoding one or more proteins fused in frame with reporter fluorescent protein ( N-terminal tag ) using final concentrations of 450 ng/µl for each gene . Fluorescent dextran was also co-injected to visualize the embryos . For single blastomere microinjections , we raised the embryos until 8–16 cell stages ( 3–4 hpf ) and co-injected the mRNA solution with Biotinylated Dextran Amine-Texas Red ( 10 µg/µl; Vector labs , Inc . Cat . # SP-1140 . RRID:AB_2336249 ) . Live embryos were kept at 16°C and visualized after the mRNA of the FP was translated into protein ( 2–3 hr ) . To avoid lethality , lower mRNA concentrations of the mutant proteins ( 250 ng/µl ) were used to image the specimens for Figures 2 and 4 , and Figure 3—video 1 . Live embryos were mounted in 1/3 sea water for visualization . Images were documented at different stages from 3 to 96 hr post fertilization . We injected and recorded more than 500 embryos for each injected protein and confocal imaged approximately 20 specimens for each stage for detailed analysis of phenotypes in vivo . We repeated each experiment at least five times obtaining similar results for each case . The fluorescent dextran and primers for the cloned genes are listed in Key resources table . To target our gene of interest , we used synthetic guide RNAs ( sgRNA; Synthego , Inc . ) and followed the instructions obtained from the manufacturer to form the RNP complex with Cas9 ( Cas9 plus sgRNAs ) . Target sites , off-target sites , and CFD scores were identified and sgRNA were designed using CRISPRscan ( Doench et al . , 2014; Moreno-Mateos et al . , 2015 ) . We delivered the RNP complex by microinjection as previously described ( Servetnick et al . , 2017; Wijesena et al . , 2017; Ikmi et al . , 2014 . Lyophilized Cas9 ( PNA Bio . , Inc . Cat . # CP01 ) was reconstituted in nuclease-free water with 20% glycerol to a final concentration of 2 µg/µl . Reconstituted Cas9 was aliquoted for single use and stored at −80°C . Embryos were injected , as described for mRNA microinjections , with a mixture ( 12 . 5 µl ) containing sgRNAs ( 80 ng/μl of each sgRNA ) , Cas9 ( 3 μg ) , and Alexa Fluor 488-dextran ( 0 . 2 μg/μl; Molecular Probes , Inc . Cat . # D22910 ) . Cas9 and sgRNA guides only controls were injected alongside each round of experiments . sgRNA guides controls are only shown in figures as Cas9 had no significative effects . 3 sgRNA were used to knock out Nvpar-3 , 3 sgRNA were used to knock out Nvpar-6 , 6 sgRNA were used to knock out Nvsnail-A , and 6 sgRNA were used to knock out Nvsnail-B . Single-embryo genomic DNA was analyzed as previously described ( Servetnick et al . , 2017 ) . Gene expression was confirmed by in situ hybridization . We injected and recorded more than 1000 embryos for each treatment . We repeated each experiment at least six times obtaining similar results for each case . sgRNAs’ sequences and PCR primers flanking the targeted region are listed in Key resources table . In situ hybridization was carried out following a previously published protocol for N . vectensis ( Wolenski et al . , 2013 ) . Animals were fixed in ice-cold 4% paraformaldehyde with 0 . 2% glutaraldehyde in 1/3x seawater for 2 min , followed by 4% paraformaldehyde in PBTw for 1 hr at 4°C . Digoxigenin ( DIG ) -labeled probes , previously described ( Salinas-Saavedra et al . , 2015; Röttinger et al . , 2012 ) , were hybridized at 63°C for 2 days and developed with the enzymatic reaction of NBT/BCIP as substrate for the alkaline phosphatase conjugated anti-DIG antibody ( Roche , Inc . Cat . #11093274910 . RRID:AB_514497 ) . Samples were developed until gene expression was visible as a purple precipitate . We incubated N . vectensis embryos in 20 µM of aPKC pseudosubstrate inhibitor ( Protein kinase Cζ pseudosubstrate , myristoyl trifluoroacetate salt , Sigma , Cat . #P1614 ) from 0 to 4 hpf . Controls and 1-azakenpaullone ( AZ; Sigma , Cat . #A3734 ) drug treatment of N . vectensis embryos was performed as previously described ( Röttinger et al . , 2012; Leclère et al . , 2016 ) . Embryos were developed in 5 µm AZ from 3 to 76 hpf . Controls were incubated in 0 . 08% DMSO . Images of live and fixed embryos were taken using a confocal Zeiss LSM 710 microscope using a Zeiss C-Apochromat 40x water immersion objective ( N . A . 1 . 20 ) . Pinhole settings varied between 1 . 2 and 1 . 4 A . U . according to the experiment . The same settings were used for each individual experiment to compare control and experimental conditions . Results from in situ hybridization studies were imaged using a Zeiss Imager . M2 with a Zeiss 425 HRc color digital camera run by Zeiss Zen 2012 software . Z-stack images were processed using Imaris 7 . 6 . 4 ( Bitplane Inc . ) software for three-dimensional reconstructions and FIJI for single slice and movies . Final figures were assembled using Adobe Illustrator and Adobe Photoshop . Tissue homogenization and protein extraction was performed as described in ( Salinas-Saavedra et al . , 2015; Suzuki et al . , 2001; Wang et al . , 2012 ) . Briefly , embryos were homogenized in 200 µl of ice cold lysis buffer ( 30 mM HEPES , pH to 7 . 5 , 1 mM EDTA , 150 mM NaCl , 50 mM NaF , 1 mM Na3VO4 , 1 mM Na2MoO4 , 1 mM MgCl2 , 1% NP-40 , 10% Glycerol , Protease Inhibitor cocktail ( Sigma P8340 ) and PMSF . After 15 min’ incubation on ice , crude lysate was carefully laid on top of a 200 µl sucrose cushion ( 1M sucrose , 30 mM HEPES , pH to 7 . 5 , 1 mM EDTA , 150 mM NaCl , 50 mM NaF , 1 mM Na3VO4 , 1 mM Na2MoO4 ) and yolk pelleted by centrifugation at 1000 rpm for 10 min . The top layer was transferred to a clean microcentrifuge tube and 300 µl of lysis buffer was added . Approximately , 5 mg of protein was obtained from 60 µl of embryos ( more than 15 , 000 embryos ) homogenized in 500 µl of lysis buffer . 2 mg of total protein ( in 500 µl of lysate buffer ) was incubated with Par-specific antibodies cross-linked to Pierce Protein A/G Magnetic Beads ( Pierce Biotechnology , Rockford , IL ) . ( Pre-immune ) IgG-IP pull downs were performed as a negative control for each experiment . Three antibodies , described previously ( Salinas-Saavedra et al . , 2015 ) , against three different proteins ( NvaPKC , NvPar-6 , and NvPar-1 ) were utilized . We performed co-IP experiments using early cleavage ( 2–4 hpf ) and for gastrula stages ( 24–30 hpf ) lysates . Co-IP experiments were repeated four times for each stage using fresh lysates every time . For a detailed protocol , please , go to http://www . whitney . ufl . edu/research/faculty/mark-q-martindale/mark-q-martindale-lab-protocols/ Epithelial thickness was measured using confocal images of embryos immunohistochemically labeled , processed with Imaris 7 . 6 . 4 ( Bitplane Inc . ) . For detailed and graphical explanation , please see Figure 3—figure supplement 3 and Figure 3—figure supplement 7 . For each treatment , epithelial thickness was determined by the average cell length ( µm ) along the apico-basal axis ( A-B axis ) of five individual cells . These values were made for two perpendicular axes Axis1 ( A1 ) and Axis2 ( A2 ) and normalized by the embryonic diameter ( µm ) , in order to minimize technical artifacts ( e . g . fixation and mounting ) that could have affected the shape/size of the cell . Giving a proportion ( p ) that was calculated for A1 and A2 , named p1 and p2 , respectively . The values of p1 and p2 obtained for 90 control embryos , were statistically compared with the respective values obtained for 103 mutant-embryos using the Mann–Whitney U test ( nonparametric; normality was tested using SPSS software ) with a critical p-value of 0 . 05 . The null hypothesis assumed no differences between the cell sizes of control and mutant-embryos . Values and statistics can be found in Figure 3—figure supplement 3—source data 1 .	Most animals – including birds , fish and mammals – have symmetrical left and right sides , and are known as bilaterians . During early life , the embryos of animals in this group develop three distinct layers of cells: the ectoderm ( outer layer ) , the endoderm ( inner layer ) , and the mesoderm ( middle layer ) . These layers then go on to form the animal’s tissues and organs . The ectoderm produces external tissues , such as the skin and the nervous system; the endoderm forms internal tissues , like the gut; and the mesoderm creates all tissues in between , like muscles and blood . Another , smaller group of animals , called cnidarians , do not have left and right sides . Instead , they have a ‘radial symmetry’ , meaning they have multiple identical parts arranged in a circle . These animals – which include corals , jellyfish and sea anemones – only develop two distinct layers of cells , equivalent to the outer and inner layers of bilaterians . Cnidarians evolved before bilaterians , but their genetic material is equally complex . So why did these two groups evolve to have different layers of cells ? And how exactly do animal embryos develop these distinct layers ? To address these questions , Salinas-Saavedra et al . studied embryos of the sea anemone Nematostella vectensis . Molecules called Par-proteins play an important role in controlling how cells behave and attach to one another ( and therefore how they form layers ) . So , using a technique called immunohistochemistry to look inside cells , Salinas-Saavedra et al . examined these proteins in the two layers of cells in sea anemone embryos . The experiments found that in the sea anemones , Par-proteins are arranged differently in cells that form the ‘skin’ compared to cells that form the ‘gut’ . In other words , cells in the outer layer attach to one another in a different way than cells in the inner layer , where the Par-proteins are degraded by ‘mesodermal’ genes . The findings also show that these sea anemones have all they need to form a third middle layer of cells . Like bilaterians , they could potentially move cells in and out of sheets that line surfaces inside the body – but they do not naturally do this . Understanding how animals form different layers of cells is important for scientists studying evolution and the development of embryos . However , it also has wider applications . For instance , some cells involved in developing the mesoderm are also involved in forming tumors . Future research in this area could help scientists learn more about how cancer-like cells form in animals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "evolutionary", "biology" ]	2018	Germ layer-specific regulation of cell polarity and adhesion gives insight into the evolution of mesoderm
Tissue microarrays ( TMAs ) have been used in thousands of cancer biomarker studies . To what extent batch effects , measurement error in biomarker levels between slides , affects TMA-based studies has not been assessed systematically . We evaluated 20 protein biomarkers on 14 TMAs with prospectively collected tumor tissue from 1448 primary prostate cancers . In half of the biomarkers , more than 10% of biomarker variance was attributable to between-TMA differences ( range , 1–48% ) . We implemented different methods to mitigate batch effects ( R package batchtma ) , tested in plasmode simulation . Biomarker levels were more similar between mitigation approaches compared to uncorrected values . For some biomarkers , associations with clinical features changed substantially after addressing batch effects . Batch effects and resulting bias are not an error of an individual study but an inherent feature of TMA-based protein biomarker studies . They always need to be considered during study design and addressed analytically in studies using more than one TMA . Tissue microarrays ( TMAs ) were first developed in the 1990s as an efficient way to examine tissue-based biomarkers ( Kononen et al . , 1998 ) . Since then , TMAs have been used in thousands of studies to evaluate histologic and molecular biomarkers , mostly in cancer tissue . Individual TMAs consist of cylindrical cores from hundreds of tissue samples embedded in one recipient block ( Kononen et al . , 1998; Kallioniemi et al . , 2001 ) . Studies often include more than one TMA . Even when biomarker assays are well standardized and run conditions are diligently kept fixed , some TMA slides ( batches ) may have measurements systematically too low or too high , and some batches may have wider spread around the true values of the biomarker than others . In general , such batch effects can have a profound impact on the validity of biomarker studies , such as those using RNA microarrays ( Tworoger and Hankinson , 2006; Leek et al . , 2010 ) . Contrary to popular belief , whether such measurement error induces upward or downward bias in results is not guaranteed to follow simple heuristics ( van Smeden et al . , 2020 ) . Whether and to what extent TMAs are affected by batch effects has not been empirically assessed . TMAs pose unique challenges . For example , when tumor tissue is collected prospectively for inclusion on TMAs , tumor characteristics may differ between batches due to nonrandom assignment of cases , as well as temporal trends in tumor risk factors , screening , and diagnosis . Differences in tissue processing or storage across tissue specimens may have differential impact on biomarkers . Including calibration samples for quality control is also more challenging for TMAs than , for example , assaying of blood samples , because repeat sections from a tumor may differ due to intratumoral heterogeneity rather than only batch effects . In this study , we assess batch effects in a large set of centrally constructed TMAs from prostate cancer tissue from 1448 men in two nationwide cohort studies . We quantify the extent to which protein biomarker variation could be explained by batch effects . We probe different methods for mitigating batch effects while maintaining true , “biological , ” between-TMA variation , including in a plasmode simulation . Finally , we demonstrate the impact of handling batch effects on commonly performed biomarker analyses . To evaluate the presence of batch effects in studies using TMAs , we studied tumor tissue from 1448 men with primary prostate cancer on 14 TMAs ( labeled “A” through “N” ) , each including multiple tumor cores from 47 to 158 patients per TMA ( Figure 1 ) . Multiple cores from the same tumor ( usually 3 ) were always located on the same TMA . TMAs were used to quantify 20 protein biomarkers ( Figure 2 ) . Biomarker values showed noticeable between-TMA variation , despite immunohistochemical staining having been conducted at the same time for all 14 TMAs . We estimated that across the 20 biomarkers , between-TMA variation explained between 1% and 48% of the overall variation in biomarker levels ( intraclass correlation coefficient , ICC ) , with half of the biomarkers having ICCs greater than 10% ( Figure 2 ) . In an example biomarker , estrogen receptor alpha in nuclei of stromal cells ( Figure 3 ) , the means of the most extreme TMAs differed by 2 . 2 standard deviations in intensity of expression and variances differed up to 9 . 3-fold . Other biomarkers showed similar between-TMA variation by magnitude and by which TMAs had the most extreme values ( Figure 4A ) . Likewise , we observed that not only means , but also variances of biomarker levels differed between TMAs , although patterns of heteroskedasticity appeared weaker than for means ( Figure 4—figure supplement 1 ) . In contrast , we found little evidence for more complex patterns of batch effects , such that tumors with specific grade , stage , or year of diagnosis would have been particularly affected by between-TMA differences ( Supplementary file 1a ) . Nevertheless , observations from the same TMAs tended to be clustered together when projected onto the first two principal components , capturing 27% of the variance in all biomarkers ( Figure 4B ) . Some biomarkers were stained using automated staining systems , other stains were done manually ( Figure 2 ) . Moreover , the method of scoring , including human ( eye ) scoring and computer-assisted quantification , differed between biomarkers , as did the main quantitative score , typically a measure of staining intensity , a proportion of cells above an intensity threshold , or a combination of both ( Figure 2 ) . Notably , between-TMA differences were present with any of these approaches . For example , batch effects were not only present when considering intensities of biomarker staining , as for the estrogen receptor alpha and beta example . Even when setting cutoffs for staining visible by eye and quantifying the number of stain-positive cells , 8% ( 95% confidence interval [CI] , 2–15 ) of variance in estrogen receptor alpha positivity and 27% ( 95% CI , 11–42 ) of estrogen receptor beta positivity were attributable to between-TMA variation ( Figure 2—figure supplement 1 ) . Our data do not allow distinguishing which of these approaches , if any , were less prone to batch effects . In summary , we observed a large and concerning degree of between-TMA variation for several biomarkers that were quantified using different approaches , suggesting that addressing batch effects could significantly impact scientific inference . The noticeable proportion of variance attributable to TMAs could have two possibly co-existing explanations . First , that between-TMA differences in biomarkers reflect different patient and tumor characteristics that need to be retained . Second , that between-TMA differences are artifacts due to systematic measurement error that need to be removed ( batch effects ) . In support of the first hypothesis , there were noticeable differences in patient and tumor characteristics between TMAs that are likely associated with biomarker levels ( Figure 1 ) . Along with a 14-year range between the per-TMA medians of cancer diagnosis year , there were differences in the proportion of tumors with a Gleason score of 8 or higher ( between 11% and 33% ) and rates of lethal disease ( between 2 and 16 events per 1000 person-years of follow-up ) . In support of the second hypothesis , we observed that certain TMAs had consistently higher or lower biomarker values for the majority of tested biomarkers ( Figure 4A ) . For example , the same batches that showed higher-than-average biomarker values for stathmin also had higher-than-average values for PTEN . This example is noteworthy because both markers were assayed together on the same section of each TMA using multiplex immunofluorescence , and stathmin would be expected to be expressed in more aggressive tumors with activation of the PI3K signaling pathway while PTEN expression would be expected to be low in the same tumors ( Stopsack et al . , 2020 ) . Further supporting the second hypothesis , we did not observe any meaningful reduction in ICCs when we considered tumors that had the same clinical features in terms of Gleason score and stage ( Figure 4—figure supplement 2 ) . Moreover , the association between Gleason score and biomarker levels ( Figure 2D ) was considerably lower than between TMAs and biomarker levels , as underscored by less pronounced visual separation of principal components by Gleason score ( Figure 4C ) than by TMA ( Figure 4B ) . Gleason score differences explained no more than 13% of the variance in biomarker levels ( for prostate-specific membrane antigen , PSMA; 95% CI for ICC , 0 . 02–0 . 29 ) , and 13 of the 20 biomarkers had ICCs by Gleason score of 1% or less ( Figure 4—figure supplement 3 ) . To directly disentangle both hypotheses , we further examined data on 10 tumors with a total of 53 tumor cores for which some cores were included on different TMAs ( Figure 4D ) . These were not included in the previous analyses and had estrogen receptor scoring data . This design allowed us to estimate biomarker differences directly attributable to between-TMA variability within the same tumors while controlling for the between-core variability expected due to intratumoral heterogeneity . Of the total variance in estrogen receptor alpha levels , 30% ( 95% CI , 0–67 ) was explained by between-TMA variation; for estrogen receptor beta , 24% ( 95% CI , 0–60 ) was explained by between-TMA variation . For comparison , between-tumor variation explained 37% ( 95% CI , 4–68 ) of the variance of estrogen receptor alpha levels and 26% ( 95% CI , 0–57 ) of the variance of estrogen receptor beta levels . Collectively , while these observations highlighted moderate differences in clinical and pathological characteristics between TMAs , they suggested that between-TMA differences were largely due to batch effects . We implemented six different approaches for batch effects mitigation and compared these to the uncorrected biomarker levels ( Figure 3 , Figure 3—figure supplement 1 ) . Two mitigation approaches , batch means ( approach 2 ) and quantile normalization ( approach 6 ) , assumed no true difference between TMAs is arising from patient and tumor characteristics , while all other approaches attempted to retain such differences between TMAs . It is possible that the choice of mitigation approaches may be optimized using knowledge of the source of the batch effect . This would be the case if each method “specialized” in mitigating effect from specific sources . We have not investigated this possibility here . Overall , correlations between values adjusted by different approaches were higher ( mean Pearson r , 0 . 97–1 . 00 ) than between uncorrected values and corrected values ( r , 0 . 90–0 . 95 ) , regardless of mitigation approach ( Figure 4E ) . Approaches 2–7 reduced visible separation by batch on plots of the first two principal components ( Figure 4—figure supplement 4 ) . Variance attributable to between-TMA differences decreased to ICCs of <1% for all markers ( Supplementary file 1b ) . An exception was the quantile regression-based approach 5; the ICCs after this approach remained up to 10% . This method does not explicitly address differences in means between batches but allows associations between clinical and pathological factors and biomarker levels to differ at high and low quantiles ( Figure 4—figure supplement 5 ) . The differences between uncorrected values and batch effect-corrected values were remarkably similar between the mean-based approaches using approaches 2 ( simple means ) , 3 ( standardized batch means ) , and 4 ( inverse probability-weighted batch means; Figure 4—figure supplement 6 ) . Consequently , batch effect-corrected values by approaches 2–4 were highly correlated ( Figure 4E ) . All mean-only batch effect mitigations also gave the same results when fitting outcome models stratified by batch ( Figure 4—figure supplement 7 ) . However , batch-specific results differed for approaches that targeted between-batch differences in the variance of biomarkers . To compare the performance of the different batch mitigation approaches in a time-to-event analysis , we applied plasmode simulation ( Franklin et al . , 2014 ) to fix the expected strength of the biomarker exposure–outcome relationship a priori before artificially introducing batch effects . The correlation structure between biomarker and confounders and between confounders and batches from the actual data ( Figure 5—figure supplement 1A , C ) was preserved in the plasmode-simulated data . Likewise , across a range of hazard ratios for the biomarker–outcome association , confounder–outcome associations remained unchanged ( Figure 5—figure supplement 1B , D ) . We first evaluated a setting in which we did not introduce batch effects ( Figure 5A ) . Here , the observed hazard ratios without batch effect mitigation equaled the expected . When performing ( unnecessary ) batch effect mitigation , observed hazard ratios were still comparable with the expected hazard ratios ( Figure 5D; see Supplementary file 1c for CIs ) . We then introduced batch effects by adding batch-specific mean differences to the observed biomarker levels , yet without introducing differences in variance by batch ( Figure 5B ) . Without batch effect mitigation , for a true hazard ratio of 3 . 0 , the observed hazard ratio , averaged over simulations , was 2 . 17 ( 95% CI , 1 . 86–2 . 53 ) , an underestimate by 28% ( Figure 5E; Supplementary file 1c ) . In contrast , all mitigation approaches produced CIs that covered the expected hazard ratio ( e . g . , approach 6 quantile normalization: hazard ratio , 3 . 03; 95% CI , 2 . 48–3 . 69 ) . When we introduced batch-specific differences in both means and in variances ( Figure 5C ) , the observed hazard ratio without batch effect mitigation decreased to 1 . 90 ( 95% CI , 1 . 66–2 . 16 ) compared to the expected hazard ratio of 3 . 0 ( Figure 5F; Supplementary file 1c ) . Batch effect mitigation methods that only focus on means ( approaches 2–4 ) reduced but did not fully eliminate bias , with hazard ratios ranging between 2 . 67 and 2 . 70 . Methods that address differences in both mean and variance resulted in less bias , with an observed hazard ratio of 3 . 11 ( 95% CI , 2 . 54–3 . 81 ) for approach 6 ( quantile normalization ) . We also included two stratification-based approaches . Fitting survival models separately by batch , followed by inverse-variance pooling ( approach 8 ) , resulted in approximately unbiased estimates but was less efficient than other approaches , comes with a risk of sparse-data bias , and resulted in considerably wider CIs in our simulation . Including batch as a stratification variable in a single Cox model ( approach 9 ) was unbiased and efficient . A drawback of both stratification-based approaches is that they do not explicitly estimate batch effect-adjusted biomarker values that could be visualized directly . Scenarios evaluated thus far were based on the actual , modest imbalance of confounders between batches and at most weak associations between the biomarker and confounders , resulting in weak confounding overall . We additionally introduced both modest and strong associations between biomarker and confounders and created more severe imbalance between batches ( Figure 5—figure supplement 2 ) . In all scenarios , the ranking of mitigation methods was preserved ( Figure 5—figure supplement 3 , Supplementary file 1c–d ) , with the least bias obtained through quantile normalization ( approach 6 ) . Bias occurred when using uncorrected biomarker levels in the presence of any batch effects , except if there was no association between biomarker and outcome ( i . e . , a hazard ratio of 1 ) , and with mean-only approaches 2–4 if variance was also affected by batch effects . In no situation , except possibly with the quantile regression-based approach 5 , were estimates after batch effect mitigation farther from the expected values than results based on uncorrected biomarker levels . To illustrate how batch effect mitigations alter the results of commonly conducted tumor biomarker analyses , we estimated how uncorrected and corrected biomarker levels were associated with Gleason score and with rates of lethal disease . For markers with little between-TMA variability ( low ICCs ) such as beta-catenin , there were no noticeable differences in associations between using unadjusted and adjusted biomarker levels irrespective of adjustment model , as expected from plasmode simulation . However , for markers with higher between-TMA variability ( higher ICC ) and stronger associations with the outcome , adjustment approaches led to noticeable differences ( Figure 6 ) . For example , uncorrected stathmin expression levels were not associated Gleason score ( difference , 0 . 00 standard deviations per one grade-group increase; 95% CI , –0 . 05 to 0 . 05 ) , while the difference in levels corrected according to approach 6 ( quantile normalization ) was 0 . 04 ( 95% CI , 0 . 00 to 0 . 07 ) , suggesting a potentially qualitatively different interpretation ( Figure 6A; Supplementary file 1e ) . In models for lethal disease ( Figure 6B ) , the otherwise unadjusted hazard ratio for the highest quartile of the vitamin D receptor , compared to the lowest quartile , was 0 . 44 ( 95% CI , 0 . 23–0 . 86 ) ; after mitigation using approach 6 , the hazard ratio was 0 . 19 ( 95% CI , 0 . 09–0 . 40 ) , suggesting that unadjusted biomarker levels could underestimate the prognostic association by 2 . 3-fold ( Supplementary file 1f–g ) . The key strength of using TMAs is their utility in parallelizing the assessment of biomarkers on a large number of tissue specimens ( Kononen et al . , 1998 ) . Similar to other high-throughput platforms , batch effects have to be considered in every TMA biomarker study . As we demonstrated , for some of the biomarkers , batch effects can be of substantial magnitude . We show that batch effect mitigation is possible and can enhance study findings . In our study of prostate tumor specimens , between-TMA differences explained 10% or more of the variance in biomarker levels for half of the included biomarkers , considerably more than one of the strongest pathological features in prostate cancer , Gleason grade . All analytical mitigation approaches to reduce batch effects , whether they attempted to retain real differences between tumors from different TMAs or not , led to corrected biomarker levels that were more similar to each other than they were , in general , to the uncorrected biomarker levels . In drawing from a large set of protein tumors biomarkers in prostate cancer , we show how appropriately mitigating batch effects strengthens results and their validity for biomarkers affected by batch effects . Ideally , batch effects between TMAs are minimized when designing a study . Standardizing how tumor samples are obtained , stored , processed , and assayed is critical , as are stratified or random allocation of tumor samples to different TMAs ( Tworoger and Hankinson , 2006 ) when possible . However , the batch effects that we observed occurred despite all feasible standardization efforts . Moreover , samples will be collected sequentially , and TMAs may be constructed sequentially in large-scale prospective studies over time . There were modest differences in the clinical and pathological characteristics between our TMAs , an issue that may be inevitable in larger-scale biobank studies . Allocation schemes of tumors to TMAs that appear ideal retrospectively , for example by matching “cases” of lethal tumors with “controls” of non-lethal tumors , may not be feasible prospectively . Likewise , in few of the thousands of studies using TMAs will it be possible to reallocate tumors to different TMAs and repeat all pathology work merely to reduce the implications of batch effects . An additional challenge in the design phase is that tissue samples are inherently heterogeneous and cannot simply be diluted , like blood samples . “Quality control” tumor samples that could serve as a quantitative calibration series suitable for all future biomarkers do not exist . One potential strategy is to include cell lines that have been formalin-fixed and paraffin-embedded on each TMA . While cell lines address issues of heterogeneity , the cell lines are often genomically unique and as such may not be relevant for all biomarkers . Another potential approach is to include samples from the same tumor case across TMAs , which would allow for direct estimation of batch effects . For these reasons , a principled approach that anticipates batch effects and addresses them analytically is critical . Beyond efforts to prevent batch effects during the study design phase , we suggest the following best practices when undertaking TMA-based tissue biomarker studies ( Figure 7 ) . First , the extent of potential batch effects should be explored and reported in any study of cancer tissue using TMAs . Inspecting TMA slides and plots ( Figure 3; Manimaran et al . , 2016 ) is important . Between-TMA variation should be quantified , for example by calculating ICCs , that is , to contrast variation of biomarker levels between TMAs compared to that between or within tumors ( Nakagawa and Schielzeth , 2010 ) . In our study , for half of the biomarkers , ICCs for between-TMA variation were low , at less than 10% , although the proportion of tolerable batch variation should be chosen based on the context . Whether TMAs differ in terms of average biomarker levels , low levels ( possibly reflective of background ) , or variability between tumors will also inform what impact of between-TMA differences to expect . Second , the source of between-TMA differences should be elucidated . Ideally , including multiple cores from the same tumors in more than one TMA will help estimating , again using ICCs , how biomarker levels vary between TMAs , between tumors , and within tumors . Alternatively , ICCs between TMAs can be estimated by restricting to or adjusting for tumor features associated with differences in the biomarker , if known . In our study , both approaches indicated that the largest share of between-TMA differences was likely due to batch effects rather than due to true differences between tumors on different TMAs . However , one should not simply assume this to be the case in other settings , and also explore between-tumor differences as one source of between-TMA differences . In multidisciplinary team discussions ( Marrone et al . , 2019 ) , it may be possible to directly pinpoint the source of batch effects and eliminate its cause . All study steps , including the pre-analytic , analytic , and post-analytic phases , should be considered . If sources of batch effects can be identified , it is preferable that they be addressed directly during the pre-analytical or analytical phase , rather than applying the post-analytical methods that we have described here and that may not adequately incorporate knowledge on the source of batch effects . For example , if immunohistochemical staining was performed separately for each TMA , then immunohistochemistry and quantification should be repeated using new sections from all TMAs at once . Imaging of pathology slides can also be a source of batch effects ( Kothari et al . , 2014 ) , as could be image analysis . In other cases , particularly if such obvious reasons for batch effects were avoided through standardized processing , as in our examples , it may remain elusive whether batch effects were induced through subtle differences in how tumors were cored and embedded during TMA construction , how long they had been stored , how they were sectioned , how well the staining process was standardized , or how successfully background signal was eliminated during software-based quantification . Yet even biomarkers scored by manual quantification were not free from batch effects . Third , if a biomarker is affected by batch effects and no “physical” remediation is possible , then post-analytical approaches should be used to reduce bias in results ( Tworoger and Hankinson , 2006; Leek et al . , 2010 ) . We demonstrate that in all plausible or exaggerated real-world scenarios , estimates after applying batch effect mitigations were consistently closer to the true underlying values than they were without . If batches do not only differ in terms of mean values , but also in terms of their variances , then methods that focus solely on means may be insufficient . A simple quantile-normalization-based approach was successful in reducing bias in real-world scenarios and could be preferred for its simplicity . It is important to note that any method tested in this study is preferable over not addressing batch effects , and thus the choice between methods should be secondary to the choice to address batch effects altogether . Only results for biomarkers that are affected by batch effects and that are associated with the outcome of interest will show large changes in estimates , as the vitamin D receptor in our example . In contrast , for the majority of our example biomarkers , results did not change appreciably because batch effects were low , associations with the outcome were close to null , or both ( Figure 6 ) . We recommend that researchers openly address batch effects in their TMA-based studies: they are not an error of an individual study , but an inherent feature of TMA-based studies . Batch effects have long been recognized in studies of the transcriptome using microarrays and next-generation sequencing , where batch effect mitigations are a component of standard workflows ( Leek et al . , 2010; Leek et al . , 2012 ) . Our data strongly suggest that protein biomarker studies using multiple TMAs are at risk of batch effects just like any other biomarker study . The extent of batch effects is difficult to predict , and empirical evaluation is necessary each time . Future studies should quantify between-TMA differences and , if they deem batch effect mitigations to be unnecessary , provide evidence for absence of batch effects , rather than merely assuming their absence . The methods that we provide facilitate the appropriate migration of batch effects between TMAs and help strengthen scientific inference . It may be prudent to extend this approach to in-situ tissue biomarkers other than proteins , such as RNA in-situ hybridization , even if our study only demonstrated batch effects for proteins . Having mitigated batch effects will allow researchers to focus on increasing study validity by addressing other sources of measurement error ( van Smeden et al . , 2020 ) , selection bias ( e . g . , from tumor biospecimen availability ) ( Liu et al . , 2018 ) , and confounding . Tumor tissue in this study was from men who were diagnosed with primary prostate cancer during prospective follow-up of two nationwide cohort studies . The Health Professionals Follow-up Study is an ongoing cohort study that enrolled 51 , 529 male health professionals across the United States in 1986 . The Physicians’ Health Study 1 and 2 were randomized-controlled trials of aspirin and dietary supplements , starting in 1982 with 22 , 071 male physicians . Participants were diagnosed with and treated for prostate cancer at local health care providers across the United States . The study team collected formalin-fixed paraffin-embedded tissue specimens from radical prostatectomy and transurethral resection of the prostate ( TURP ) , and study genitourinary pathologists performed central re-review , including standardized Gleason grading of full hematoxylin–eosin-stained slides from the tumor blocks ( Stark et al . , 2009 ) . Written informed consent was obtained from all participants , and the study protocol was approved by the institutional review boards of the Brigham and Women’s Hospital and Harvard T . H . Chan School of Public Health ( IRB19-1430 ) , and those of participating registries as required . TMAs were constructed using 0 . 6 mm tissue cores of the primary nodule or the nodule with the highest Gleason score ( Pettersson et al . , 2012 ) , including three or more cores of tumor tissue per participant ( tumor ) . For a subset of tumors , additional cores of tumor-adjacent , histologically normal-appearing prostate tissue were included . TMAs were constructed at the same laboratory across a 10-year period , as tissue from cohort participants became available , without matching on patient or tumor characteristics and without randomization . Cores from the same participant were generally included on the same TMA , with exceptions noted below , and summarized as the mean . We include information from 14 prostate tumor TMAs . Immunostaining was generally performed separately for individual biomarkers yet always for all TMAs at the same time . Detailed immunohistochemistry staining and quantification procedures for each marker have been published ( Rider et al . , 2015; Flavin et al . , 2014; Fiorentino et al . , 2008; Ahearn et al . , 2016; Ding et al . , 2011; Nguyen et al . , 2010; Pettersson et al . , 2018; Kasperzyk et al . , 2013; Dhillon et al . , 2009; Hendrickson et al . , 2011; Zu et al . , 2013; Stopsack et al . , 2020 ) or are in preparation for estrogen receptor alpha ( antibody SP1; Thermo Fisher Scientific , Waltham , MA ) and an antibody ( PPG5/10; Bio-Rad Laboratories , Hercules , CA ) widely used to measure estrogen receptor beta . If batch effect mitigation approaches had been applied in the original studies , the uncorrected levels were retrieved . Right-skewed biomarker scores ( Ki-67 , pS6 , TUNEL ) were loge transformed . All biomarkers were scaled to mean 0 and standard deviation 1 solely to facilitate comparisons of batch effects across markers; batch effect mitigation does not necessitate scaling and preserves absolute biomarker values . To visualize the extent of biomarker variation between TMAs , we plotted uncorrected biomarker values by tumor , biomarker , and TMA . We summarized biomarker variation using the first two principal components ( Lê et al . , 2008 ) . We calculated between-TMA mean differences and ratios of variances versus the first TMA . We tested if tumors with different clinical/pathological characteristics had higher biomarker levels in TMAs with higher means ( i . e . , multiplicative effect modification ) . For each biomarker and each clinical/pathological feature ( ordinal Gleason score , ordinal stage , or calendar year of diagnosis ) , let Zij be the within-TMA z-score ( mean 0 , standard deviation 1 ) for tumor i from TMA j; Ai , the clinical/pathological feature of tumor i; Bj , the TMA-specific biomarker mean , rj , the TMA-specific random effect , and eij , residual error . In the regression model Zij=β0+β1Ai+β2Bj+β3AiBj+rj+eij , we evaluated the β3 term to assess for multiplicative effect measure modification . We calculated the proportion of variation in biomarker levels attributable to TMA using intra-class correlations ( ICCs , also “repeatability”; Nakagawa and Schielzeth , 2010 ) based on one-way random effects linear mixed models with an independent variance–covariance structure ( Nakagawa and Schielzeth , 2010; Hankinson et al . , 1995 ) for Yij , the biomarker level per tumor i and TMA j; where β0 is the biomarker mean; rj , the random effect for TMA j; and eij , the residual error: Yij=β0+rj+eij . The ICC was defined as the proportion of between-TMA variance in the total variance: ICC=var ( r ) var ( r ) +var ( e ) . 95% CIs for ICCs were obtained using parametric bootstrapping using 500 repeats ( Stoffel et al . , 2017 ) . To directly distinguish between-TMA variation caused by batch effects from variation caused by differences in patient and tumor characteristics , we compared ICCs per biomarker overall to ICCs per biomarker when restricting analyses to a subset of tumors with the same clinical features . We also leveraged a small subset of tumors that had cores included on more than one TMA . Here , we used two-way random effects linear mixed models with independent variance–covariance structure ( Bates et al . , 2015 ) to separate between-TMA variation from between-core variation ( i . e . , intratumoral heterogeneity ) and residual modeling error: Yijk=β0+rj+si+eijk . Compared to the model described earlier , this model additionally includes tumor-specific random effects si , and thus ICC=var ( r ) var ( r ) +var ( s ) +var ( e ) . In addition to ( 1 ) using uncorrected values , we implemented eight different approaches to handle between-TMA batch effects: ( 2 ) Simple means . This approach assumes that all TMAs , if not affected by batch effects , would have the same mean biomarker value . Differences in mean biomarker values per batch are corrected by estimating batch-specific mean effects ( differences from the overall mean level ) using a linear regression model with uncorrected biomarker values as the outcome and batch indicators as predictors . Corrected biomarker values are then obtained by subtracting batch-specific effects from the uncorrected biomarker values . Mean differences can either indicate the difference of each batch mean to the overall mean , as implemented here , or be defined by comparison to a reference batch . ( 3 ) Standardized means . This approach estimates marginal means per batch using model-based standardization ( in the epidemiologic sense ) . It assumes that batches with similar characteristics have the same means if not affected by batch effects . A linear regression model for a specific biomarker is fit , adjusting for tumor variables that differ in distribution between TMAs , similar to an approach described in the epidemiology literature by Rosner et al . , 2008 . Let Yij indicate the biomarker value for tumor i on TMA j; Bj , TMA j; C1 to Cm , the m covariates to be retained; and eij , the residuals . Then Yij=β0+βjBi+γ1C1+…+γnCn+eij . Batch effect-corrected biomarker values can be obtained by subtracting batch-specific effects βj predicted from the model above from uncorrected biomarker values . We included the following clinical and pathologic variables as plausible sources of real between-TMA differences that should be retained in this approach , as well approaches 4–7: calendar year of diagnosis ( linear ) , Gleason score ( categorical: 5–6; 3+4; 4+3; 8; 9–10 ) , and pathologic tumor stage ( categorical: pT1/T2 , pT3/T3a , pT3b/T4/N1 , missing/tissue from transurethral resection of the prostate ) . ( 4 ) Inverse-probability weighted batch means . Like the preceding approach , this approach assumes that batches with similar characteristics have the same means if not affected by batch effects . While the preceding approach assumes a constant association between covariates and biomarker levels across batches , this approach allows for associations to differ between batches . We first used inverse probability weighting for marginal standardization of the distribution of clinical and pathological features per batch to the distribution in the entire study population . Stabilized weights ( Cole and Hernán , 2008 ) , truncated at the 2 . 5th and 97 . 5th percentile , were obtained through multinomial regression models , modeling the probability of assignment to a specific batch based on same clinical and pathological variables as in approach 3 . In the weighted pseudopopulation , we then used a marginal linear model to estimate batch-specific mean differences , which were further used as in approaches 2 and 3 . ( 5 ) Quantile regression . This approach assumes that batches with similar characteristics have the same values for a selected set of batch-specific quantiles , in this application the upper and lower quartile . The lower quartile may be particularly affected by background noise , while the upper quartile may more likely reflect differences in batches due to covariates . A corollary of separately modeling the two differently is that clinical and pathological variables are allowed to have different effects on these quartiles ( Bann et al . , 2020 ) . These assumptions contrast with approaches 2–4 that focus on mean levels only . We used quantile regression with the Frisch-Newton approach ( Portnoy and Koenker , 1997 ) separately for the first and third quartile of biomarker values with batch indicators to predict adjusted batch-specific quantile values with the same confounders as above . We then used the batch-specific 25th percentiles ( yτ=0 . 25 ) as the offset and the interquartile range between the 25th and 75th percentiles ( yτ=0 . 75 ) as the scaling factor when batch-correcting biomarker levels . Let yij indicate the batch effect-corrected biomarker level for tumor i on TMA j; yij , the uncorrected biomarker level for tumor i on TMA j; y^iτ=x , xth quantile of y for batch j ( predicted value for yj from unadjusted quantile regression ) ; y^jτ=x , ∗ is y^jτ=x with adjustment for confounders ( predicted value for yj from adjusted quantile regression ) ; and y´τ=x , the xth quantile of y overall . Then the corrected biomarker level isyij∗= ( yij−y^jτ=0 . 25 ) ( y¯τ=0 . 75−y¯τ=0 . 25 ) ( y^jτ=0 . 75 , ∗−y^jτ=0 . 25 , ∗ ) +y¯τ=0 . 25−y^jτ=0 . 25 , ∗+y^jτ=0 . 25 ( 6 ) Quantile normalization . This approach assumes that samples on all batches , if not affected by batch effects , would not only have the same mean and variance but also the same distribution of individual biomarker values . Uncorrected biomarker values are ranked within each batch and then ranks are replaced by the mean of values with the same rank across batches . We implemented quantile normalization using limma ( Ritchie et al . , 2015; Bolstad et al . , 2003 ) . A conceptually related approach , for example , employed in molecular epidemiology ( Tworoger and Hankinson , 2006; Marrone et al . , 2019 ) , would be to use within-batch ranks as the batch-corrected biomarker , often grouped into data-driven categories such as batch-specific quartiles . We did not further consider these derivatives because they do not retain absolute biomarker levels and can distort rank distances . ( 7 ) ComBat . For comparison , we additionally included the ComBat algorithm , which like approach 4 attempts to retain differences in batch means due to covariate differences; it is frequently applied together with approach 6 . ComBat and its derivatives ( Leek et al . , 2012; Johnson et al . , 2007; Zhang et al . , 2018 ) were initially designed for microarray studies of gene expression , which include considerably more than one biomarker per sample . This property would typically limit their use for a protein biomarker quantified on a TMA unless a large number of biomarkers is available , as in our study . Mitigation depends on values of other biomarkers on the same batches . Even if multiple protein biomarkers were available , the non-randomly selected set of concomitantly available biomarkers may influence how batch effects are corrected . ComBat scales means and ( optionally ) variances while ( optionally ) retaining adjustment variables . ComBat is implemented using an empirical Bayes approach to achieve more favorable properties for small batches . The underlying model is similar to the regression above and has been emulated by a two-way analysis of variance ( Nygaard et al . , 2016 ) . In using ComBat , we scaled both means and variances , adjusting for the same clinical and pathological variables as before . Because ComBat cannot handle biomarkers if they are missing on entire batches , we ran ComBat separately for groups of biomarkers measured on 8 , 9 , 10 , or 14 TMAs . ( 8 ) Stratification with inverse-variance pooling . An alternative approach to treating batch effects is to estimate outcome regression models separately by batch . This approach can be applied for a variety of regression models but does not result in corrected values . We pooled estimates with inverse variance-weighting to obtain summary estimates . ( 9 ) Stratification in Cox proportional hazards regression . In a special case of stratification for time-to-event outcomes , Cox proportional hazards models allow for nonparametric batch effect mitigation by including batch as a stratification factor in the model specification . Comparisons are performed within batches . Unlike approach 8 , only batch-specific baseline hazard functions but no batch-specific effects are estimated . For approaches 1–7 , we calculated Pearson correlation coefficients between uncorrected and corrected biomarker levels . Additionally , we repeated ICC and principal components analyses with corrected levels , and we estimated associations between Gleason score and biomarker levels after batch effect mitigation , stratifying by batch using approach 8 . Approaches 2–6 , which result in batch effect-adjusted biomarker levels , are implemented in the R package batchtma , available at https://stopsack . github . io/batchtma and the Comprehensive R Archive Network ( CRAN ) . We evaluated the impact of batch effect mitigation approaches on known , investigator-determined biomarker–outcome associations using plasmode simulation , an approach used , for example , for evaluating confounding control for binary exposures in pharmacoepidemiology ( Franklin et al . , 2014 ) . We used observed data from all tumors included on the 14 TMAs to determine covariates ( Gleason grade , pathological stage ) and outcome ( lethal disease ) , preserving the observed correlation structure ( e . g . , joint distribution of clinical characteristics across TMAs ) . The only simulated elements were the biomarker levels and the strengths of biomarker–outcome associations ( hazard ratios ranging from 1/3 to 3 ) that we fixed by simulating event times with flexible parametric survival models ( Crowther and Lambert , 2013 ) . Models used a baseline hazard function consisting of cubic splines with three knots ( Jackson , 2016 ) . Group differences were based on proportional hazards for the observed confounder–outcome coefficients in the real data and the fixed biomarker ( exposure ) –outcome hazard ratios . First , we used plasmode simulation to generate the fixed associations of the biomarker levels with the outcome , which are unknowable outside simulation studies , generating 200 plasmode data sets for each association . Second , we introduced batch effects . Batch effects were either only for the mean or for both mean and variance , using the actual standardized between-TMA differences in means and variances for the estrogen receptor-alpha protein , a biomarker with high ICCs . We also added batch effects for mean and variance with effect modification , making mean and variance changes due to batch effects strongly correlated with Gleason scores . Third , we calculated batch effect-adjusted biomarker levels using approaches 2–6 . Finally , we compared the expected hazard ratios for the biomarker–outcome association , fixed during simulations , with the estimated hazard ratios from Cox regression ( with normality-based 95% CIs ) before and after batch effect mitigation approaches 2–6 and using the two stratification-based approaches 8 and 9 . In sensitivity analyses , we simulated “moderate” associations between the biomarker and confounders ( 0 . 2 standard deviations difference in biomarker levels per Gleason grade group , 0 . 1 per stage category ) , “strong” associations ( differences of 0 . 4 and 0 . 2 standard deviations , respectively; stronger than observed for any biomarker in our study ) , as well as “strong” associations and additional imbalance in Gleason grade and stage between TMAs ( by excluding tumors with low grades from TMAs with higher-than-average means and excluding tumors with high stage from TMAs with low-than-average means ) , all before the four steps described above . To quantify the impact of different approaches to batch-effect handling on scientific inference , we focused on two commonly employed types of analyses in biomarker research in prostate cancer: first , a cross-sectional analysis of Gleason score and biomarker levels , using linear regression models; second , a time-to-event analysis of biomarker levels and rates of lethal disease , using Cox proportional hazards regression . Gleason scores were modeled as ordinal variables and biomarkers as linear variables to obtain one single estimate per model . We also categorized biomarkers into four quartiles and compared hazard ratios for lethal disease of the extreme quartiles . Models were designed only for investigating issues of batch effects and not for subject matter inference on specific biomarkers . The batchtma R package is available at https://stopsack . github . io/batchtma and the Comprehensive R Archive Network ( CRAN ) . Code used to produce results this manuscript is at https://github . com/stopsack/batchtma_manuscript ( copy archived at swh:1:rev:a588f10906f8685b055e5a6f0a487f5f850d13bc , Stopsack , 2022 ) . Data are available for analysis on the Harvard FAS computing cluster through a project proposal for the Health Professionals Follow-up Study ( https://sites . sph . harvard . edu/hpfs/for-collaborators ) .	To understand cancer , researchers need to know which molecules tumor cells use . These so-called ‘biomarkers’ tag cancer cells as being different from healthy cells , and can be used to predict how aggressive a tumor may be , or how well it might respond to treatment . A popular technique for assessing biomarkers across multiple tumors is to use tissue microarrays . This involves taking samples from different tumors and embedding them in a block of wax , which is then cut into micro-thin slices and stained with reagents that can detect specific biomarkers , such as proteins . Each block contains hundreds of samples , which all experience the same conditions . So , any patterns detected in the staining are likely to represent real variations in the biomarkers present . Many cancer studies , however , often compare samples from multiple tissue microarrays , which may increase the risk of technical artifacts: for example , staining may look stronger in one batch of tissue samples than another , even though the amount of biomarker present in these different arrays is roughly the same . These ‘batch effects’ could potentially bias the results of the experiment and lead to the identification of misleading patterns . To evaluate how batch effects impact tissue microarray studies , Stopsack et al . examined 14 wax blocks which contained tumor samples from 1 , 448 men with prostate cancer . This revealed that for some biomarkers , but not others , there were noticeable differences between tissue microarrays that were clearly the result of batch effects . Stopsack et al . then tested six different ways of fixing these discrepancies using statistical methods . All six approaches were successful , even if the arrays included tumors with different characteristics , such as tumors that had been diagnosed more or less recently . This work highlights the importance of considering batch effects when using tissue microarrays to study cancer . Stopsack et al . have used their statistical approaches to develop freely available software which can reduce the biases that sometimes arise from these technical artifacts . This could help researchers avoid misleading patterns in their data and make it easier to detect real variations in the biomarkers present between tumor samples .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2021	Extent, impact, and mitigation of batch effects in tumor biomarker studies using tissue microarrays
Metabolic co-regulation between biosynthetic pathways for secondary metabolites is common in microbes and can play an important role in microbial interactions . Here , we describe a novel mechanism of metabolic co-regulation in which an intermediate in one pathway is converted into signals that activate a second pathway . Our study focused on the co-regulation of 2 , 4-diacetylphloroglucinol ( DAPG ) and pyoluteorin , two antimicrobial metabolites produced by the soil bacterium Pseudomonas protegens . We show that an intermediate in DAPG biosynthesis , phloroglucinol , is transformed by a halogenase encoded in the pyoluteorin gene cluster into mono- and di-chlorinated phloroglucinols . The chlorinated phloroglucinols function as intra- and inter-cellular signals that induce the expression of pyoluteorin biosynthetic genes , pyoluteorin production , and pyoluteorin-mediated inhibition of the plant-pathogenic bacterium Erwinia amylovora . This metabolic co-regulation provides a strategy for P . protegens to optimize the deployment of secondary metabolites with distinct roles in cooperative and competitive microbial interactions . Many microorganisms produce multiple secondary metabolites with diverse chemical properties and ecological functions that contribute to the fitness of the producing organism in natural environments ( Haas and Keel , 2003; O’Brien and Wright 2011 , Traxler and Kolter , 2015 ) . The biosynthesis of different secondary metabolites by a microorganism is often coordinated ( Bosello et al . , 2011; Tsunematsu et al . , 2013; Hidalgo et al . , 2014; Vingadassalon et al . , 2015; Cano-Prieto et al . , 2015; Cui et al . , 2016 ) . Co-regulation of different secondary metabolites is thought to be selected during microbial evolution because it confers competitive advantages to the producing organism in microbial interactions ( Challis and Hopwood , 2003 ) . Among the secondary metabolites , antibiotics are particularly intriguing because of their roles in both cooperative and competitive interactions among microorganisms in natural habitats ( Yim et al . , 2007; Raaijmakers and Mazzola , 2012 ) . This study focuses on 2 , 4-diacetylphloroglucinol ( DAPG ) and pyoluteorin , two broad-spectrum antibiotics with toxicity against fungi , bacteria , oomycetes and plants ( Ohmori et al . , 1978; Keel et al . , 1992 ) . Certain strains of Pseudomonas spp . produce DAPG but not pyoluteorin ( Loper et al . , 2012 ) and some strains of Pseudomonas aeruginosa produce pyoluteorin but not DAPG ( Ohmori et al . , 1978 ) . To date , only a subset of strains of the species Pseudomonas protegens are known to produce both compounds ( Loper et al . , 2012 ) . Among these DAPG- and pyoluteorin-producing strains is the soil bacterium P . protegens Pf-5 . In the genome of Pf-5 , genes for DAPG and pyoluteorin biosynthesis are present in two different gene clusters ( Figure 1A and C ) separated by 3 . 7-megabases ( Paulsen et al . , 2005 ) . The biosynthetic substrates and intermediates of the two pathways are distinct ( Figure 1B and D ) . DAPG biosynthesis begins with synthesis of phloroglucinol ( PG ) from three molecules of malonyl-CoA by the type III polyketide synthase PhlD ( Achkar et al . , 2005 ) . PG is then acetylated through the action of PhlABC to form DAPG ( Bangera and Thomashow , 1999 ) . Pyoluteorin biosynthesis begins with activation of L-proline by the L-prolyl-AMP ligase PltF , and attachment to the peptidyl-carrier protein PltL ( Thomas et al . , 2002 ) . The Pro-S-PltL intermediate is oxidized by PltE and chlorinated by PltA to yield 4 , 5-dichloropyrrolyl-S-PltL ( Thomas et al . , 2002; Dorrestein et al . , 2005 ) . The dichloropyrrole is extended by the type I polyketide synthase PltBC followed by release and formation of the resorcinol , presumably catalyzed by PltG to yield pyoluteorin ( Nowak-Thompson et al . , 1997 , 1999 ) . 10 . 7554/eLife . 22835 . 003Figure 1 . Biosynthetic gene clusters for 2 , 4-diacetylphloroglucinol ( DAPG ) ( A ) , and pyoluteorin ( C ) , the proposed biosynthetic pathways of DAPG ( B ) and pyoluteorin ( D ) of P . protegens Pf-5 , and the GFP reporter constructs ( E , F , G ) used in this work . Arrows represent gene location and orientation in the biosynthetic gene clusters , and are colored according to their functions . The open arrow in ( E ) indicates that pltR is not included in the pL-gfp construct . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 003 Despite the independent biochemical and genetic determinants for their biosynthesis , production of DAPG and pyoluteorin by P . protegens is tightly coordinated . Mutations in pyoluteorin biosynthetic genes result in loss of pyoluteorin production along with overproduction of DAPG ( Brodhagen et al . , 2004; Quecine et al . , 2016 ) , and addition of pyoluteorin to bacterial cultures represses the expression of certain DAPG biosynthetic genes and the production of DAPG ( Schnider-Keel et al . , 2000; Baehler et al . , 2005 ) . Conversely , PG , an intermediate in DAPG biosynthesis , has a concentration-dependent influence on expression of pyoluteorin biosynthetic genes and production of pyoluteorin: nanomolar concentrations of PG are required for pyoluteorin production but micromolar concentrations of PG repress pyoluteorin production ( Kidarsa et al . , 2011; Clifford et al . , 2016 ) . The essential role of PG in pyoluteorin production became evident when we discovered that a phlD mutant , which lacks the type III polyketide synthase PhlD responsible for PG biosynthesis , produces neither pyoluteorin nor DAPG . The phlD mutant can be complemented for pyoluteorin production by addition of nanomolar concentrations of PG to the culture medium ( Kidarsa et al . , 2011 ) . Furthermore , addition of nanomolar concentrations of PG induces expression of pyoluteorin biosynthetic genes ( Clifford et al . , 2016 ) , indicating that the influence of PG on pyoluteorin production is mediated through gene regulation . Pyoluteorin itself also induces expression of pyoluteorin biosynthetic genes ( Brodhagen et al . , 2004; Li et al . , 2012 ) . The pyoluteorin biosynthetic genes are expressed under the positive control of PltR ( Brodhagen et al . , 2004 ) ( Figure 1 ) , a transcriptional regulator belonging to the LysR family . Both PG and pyoluteorin are known to induce PltR-mediated transcription of pyoluteorin biosynthetic genes , but the relative importance of the two small molecules in inducing gene expression has not been evaluated . The metabolic co-regulation between DAPG and PLT biosynthetic pathways differs from known examples of metabolic co-regulation between secondary metabolic pathways , which can be classified into four types based on the underlying mechanisms . The first type of co-regulation involves one or more intermediates that are shared by different biosynthetic pathways ( Vingadassalon et al . , 2015; Cano-Prieto et al . , 2015 ) . The second type results when a single enzyme is shared by different pathways ( Lazos et al . , 2010; Tsunematsu et al . , 2013 ) . The third type of co-regulation is mediated through functionally redundant enzymes present in different pathways ( Lautru et al . , 2007 ) . The fourth type occurs when the biosynthetic genes for two pathways are controlled by the same ‘pathway-specific’ regulator ( Pérez-Llarena et al . , 1997; Bergmann et al . , 2010 ) . Overall these mechanisms of metabolic co-regulation involve a shared biosynthetic intermediate , enzymatic activity , or regulator , but none of these features is shared by the DAPG and pyoluteorin pathways ( Figure 1 ) . Here , we describe a mechanism of metabolic co-regulation that is distinct from all described previously . Our results show that PG , an intermediate in DAPG biosynthesis , is transformed by a halogenase encoded in the pyoluteorin gene cluster into chlorinated derivatives that function as cell-cell communication signals inducing expression of pyoluteorin biosynthetic genes . Our work reveals a new mechanism of co-regulation between two secondary metabolic pathways , which is mediated by an intermediate in one pathway that is converted into signals that activate the second pathway . To determine the differences between the regulatory roles of PG and pyoluteorin in the production of pyoluteorin , we evaluated the effects of both metabolites on expression of the pyoluteorin biosynthetic gene pltL using pL-gfp ( Figure 1E ) , a reporter construct containing a transcriptional fusion of the promoter of pltL to a promoterless gfp ( Yan et al . , 2016 ) . GFP expressed from pL-gfp was monitored in a ΔpltAΔphlD double mutant grown in a medium amended with each of the two metabolites alone or in combination ( Figure 2 ) . In the non-amended medium , and relative to wild-type Pf-5 , GFP expression was significantly lower in the ΔpltAΔphlD mutant ( p=2 . 76E-4 ) , indicating only background expression of pltL . This result was expected because of the deletion of pltA and phlD , which encode enzymes for biosynthesis of pyoluteorin and PG , respectively ( Figure 1 ) ; both pyoluteorin and PG are known to induce expression of pyoluteorin biosynthetic genes ( Brodhagen et al . , 2004; Clifford et al . , 2016 ) . In the medium amended with 10 nM PG , GFP expression of the ΔpltAΔphlD mutant increased nearly tenfold relative to expression levels in non-amended medium , indicating induction of pltL . In contrast , addition of 100 nM pyoluteorin failed to alter GFP expression by the ΔpltAΔphlD mutant . However , when both pyoluteorin and PG were added , the GFP expression was significantly higher than under treatment with PG alone ( p=1 . 44E-4 ) . Therefore , pyoluteorin was not required for pltL expression but further induced pltL expression in the presence of PG . As expected based on the necessity of PltR for expression of the pyoluteorin biosynthetic genes ( Brodhagen et al . , 2004; Li et al . , 2012 ) , pltR was required for expression of pltL and the presence of PG could not restore pltL expression in a ΔpltAΔphlDΔpltR mutant ( Figure 2 ) . These results indicated that PG , not pyoluteorin , is required for PltR-mediated activation of pltL . 10 . 7554/eLife . 22835 . 004Figure 2 . Effect of phloroglucinol ( PG ) and pyoluteorin on expression of pltL::gfp in P . protegens Pf-5 . Pf-5 wild-type and derivatives containing pL-gfp were treated with PG ( 10 nM ) and/or pyoluteorin ( 100 nM ) , as indicated . Expression levels of pltL::gfp were measured and recorded as relative GFP ( fluorescence of GFP divided by OD600 ) . Letters above columns indicate treatments significantly different from one another , as determined by ANOVA analysis ( p<0 . 05 ) . Data are means of at least three biological replicates from a representative experiment repeated three times with similar results , and error bars represent the standard deviation of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 00410 . 7554/eLife . 22835 . 005Figure 2—source data 1 . Expression of pltL::gfp by Pf-5 wild-type ( WT ) and its derivatives in response to PG and/or pyoluteorin . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 005 To test whether PG directly or indirectly regulates expression of the pyoluteorin biosynthetic genes , we assessed the effect of PG on expression of pltL in P . fluorescens SBW25 . Like P . protegens Pf-5 , strain SBW25 is a member of the P . fluorescens group but , unlike Pf-5 , SBW25 has neither the phl gene cluster nor the plt gene cluster ( Silby et al . , 2009 ) . Because SBW25 does not have pltR , which is required for pltL expression , we made the transcriptional reporter construct pRL-gfp , which contains pltR and the intergenic region between pltR and pltL that includes the promoter of pltL fused to a promoterless gfp ( Figure 1F ) . Although PG was sufficient to induce pltL expression in Pf-5 ( Figure 2 ) , PG failed to induce GFP expression from pRL-gfp in P . fluorescens SBW25 ( Figure 3A ) and is therefore not likely to function as a signal directly . One possible explanation for these results is that Pf-5 processes PG to another compound , which then directly induces pyoluteorin gene expression . To test this possibility , we amended SBW25 carrying pRL-gfp with culture supernatants from Pf-5 or the ΔphlD mutant and measured pltL expression . Consistent with our hypothesis , culture supernatants from Pf-5 , but not the ΔphlD mutant , could induce GFP expression of SBW25 ( Figure 3A ) . Therefore , we concluded that PG does not directly influence pyoluteorin gene expression . Rather , PG could be converted to a compound ( s ) that regulates pyoluteorin gene expression . 10 . 7554/eLife . 22835 . 006Figure 3 . Effect of phloroglucinol ( PG ) and Pf-5 culture supernatants on expression of pltL::gfp in P . fluorescens SBW25 ( A , C ) , and P . protegens Pf-5 ( B ) . ( A ) Expression of pltL::gfp in SBW25 containing pRL-gfp . ( B ) Expression of pltL::gfp in wild-type ( WT ) Pf-5 and the ΔphlM mutant containing pL-gfp . ( C ) Expression of pltL::gfp in SBW25 containing pMRL-gfp . In each experiment , the concentration of PG is 10 nM , and the same volume of methanol was added to the bacterial cultures as a control . The supernatants of bacterial cultures were mixed with the fresh bacterial cultures at a 1:1 vol ratio to test their effects on pltL::gfp expression; WT-Sup , the culture supernatants of wild-type Pf-5; ΔphlD-Sup , the culture supernatants of a ΔphlD mutant . The expression levels of pltL::gfp were measured and recorded as relative GFP ( fluorescence of GFP divided by OD600 ) . Letters above columns indicate treatments significantly different from one another , as determined by ANOVA analysis ( p<0 . 05 ) . Data are means of at least three biological replicates from a representative experiment repeated three times with similar results , and error bars represent the standard deviation of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 00610 . 7554/eLife . 22835 . 007Figure 3—source data 1 . Expression of pltL::gfp by SBW25 and Pf-5 in response to phloroglucinol ( PG ) and Pf-5 culture supernatants . ( A ) Expression of pltL::gfp by SBW25 containing pRL-gfp . ( B ) Expression of pltL::gfp by Pf-5 wild-type ( WT ) and the ΔpltM mutant that contains pL-gfp . ( C ) Expression of pltL::gfp by SBW25 containing pMRL-gfp . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 00710 . 7554/eLife . 22835 . 008Figure 3—figure supplement 1 . Transcriptional profiles of pltM , pltR and the other genes in the pyoluteorin gene cluster of Pf-5 . Data were generated from our previous transcriptomic study ( Clifford et al . , 2016 ) . The height of the black peaks indicates the expression levels of genes in the pyoluteorin gene cluster of the wild-type Pf-5 ( Pf-5 combined coverage ) , the ΔphlAΔphlDΔphlG three-fold mutant ( phlADG combined coverage ) , and the ΔphlAΔphlDΔphlG three-fold mutant grown in a medium containing 100 nM PG ( phlADG_PG combined coverage ) . For the three strains/treatments , expression levels of genes are indicated in different scales to show the general expression profile of the whole gene cluster . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 00810 . 7554/eLife . 22835 . 009Figure 3—figure supplement 2 . Production of pyoluteorin , DAPG and MAPG by P . protegens Pf-5 wild-type ( WT ) , a ΔpltM mutant , and a pltM+ complemented strain . Secondary metabolites were extracted from bacterial cultures incubated for 24 hr in NBGly . Data are means of three biological replicates from a representative experiment repeated twice with similar results , and error bars represent the standard deviation of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 00910 . 7554/eLife . 22835 . 010Figure 3—figure supplement 2—source data 1 . Concentration of pyoluteorin , MAPG and DAPG ( µM ) in cultures of Pf-5 wild-type ( WT ) and its derivatives . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 010 We then attempted to identify a mechanism by which Pf-5 could convert PG to the compound ( s ) that directly induces pyoluteorin gene expression . Within the plt gene cluster of Pf-5 , pltM attracted our attention because of its proximity to pltR ( Figure 1C ) . pltM and pltR are predicted to form a bicistronic operon because of a four-base overlap between the 5’ end of pltM and the 3’ terminus of pltR ( Paulsen et al . , 2005 ) . Moreover , pltM and pltR share similar transcriptional profiles , which differ from those of other genes in the plt gene cluster ( Figure 3—figure supplement 1; Clifford et al . , 2016 ) . PltM is a putative FADH2-dependent halogenase and is required for pyoluteorin production: a pltM insertion mutant is deficient in the production of pyoluteorin ( Nowak-Thompson et al . , 1999 ) . The reason why PltM is required is unclear because the plt gene cluster encodes another FADH2-dependent halogenase PltA that catalyzes the addition of both chlorines to the pyrrole moiety to form pyoluteorin ( Dorrestein et al . , 2005 ) . Therefore , we hypothesized that PltM is responsible for conversion of PG into the compound ( s ) that induces the expression of pyoluteorin biosynthetic genes . To avoid any concerns regarding potential polar effects of the insertion in the previously derived pltM mutant ( Nowak-Thompson et al . , 1999 ) , we generated a mutant with an in-frame deletion of pltM and confirmed that the ΔpltM mutant does not produce pyoluteorin ( Figure 3—figure supplement 2 ) . Pyoluteorin production by the ΔpltM mutant could be complemented by a plasmid-borne copy of pltM , confirming that pltM has an essential role in pyoluteorin production . We introduced pL-gfp into the ΔpltM mutant and found that the mutant expressed only background levels of GFP , indicating that expression of pltL was not induced in this mutant ( Figure 3B ) . The GFP expression remained at background levels even when cells were grown in the presence of 10 nM PG . However , GFP expression of the ΔpltM mutant was induced by culture supernatants of Pf-5 ( Figure 3B ) . As expected , culture supernatants of the ΔphlD mutant failed to induce GFP expression of the ΔpltM mutant containing pL-gfp . These results supported our hypothesis and indicated that wild-type Pf-5 , but not the ΔpltM mutant , produces the compound ( s ) that induces the pltL expression . To test whether pltM is sufficient for signaling in SBW25 , we made a transcriptional fusion construct pMRL-gfp , which contains pltM , pltR , and the intergenic region between pltR and pltL that includes the promoter of pltL fused with the promoterless gfp ( Figure 1G ) . Like SBW25 containing pRL-gfp , SBW25 containing pMRL-gfp had only low , background GFP expression in the non-amended medium , and higher GFP expression level in the presence of culture-supernatants from wild-type Pf-5 ( Figure 3A and C ) . While PG failed to induce GFP expression of SBW25 containing pRL-gfp ( Figure 3A ) , it induced GFP expression of SBW25 containing pMRL-gfp ( Figure 3C ) . These results indicated that pltM is sufficient for signaling in SBW25 in the presence of PG and the regulatory gene pltR . To identify the compound ( s ) that induce ( s ) expression of pltL , we extracted bacterial metabolites from cultures of the wild-type Pf-5 and the ΔpltM mutant , and analyzed the extracts by liquid chromatography–mass spectrometry ( LCMS ) . This analysis revealed the presence of 2-chlorobenzene-1 , 3 , 5-triol ( PG-Cl ) in culture extracts of the wild-type Pf-5 but not the ΔpltM mutant ( Figure 4—figure supplement 1 ) . We also separated the extracts of the wild-type into fractions using semi-preparatory HPLC , and assayed the fractions for signaling activity . This analysis revealed two peaks that induced pltL expression in SBW25 ( pPL-gfp ) ( Figure 4C ) , suggesting that Pf-5 produces at least two signaling compounds . 10 . 7554/eLife . 22835 . 011Figure 4 . Identification of the signaling metabolites from Pf-5 that induce expression of pltL::gfp in P . fluorescens SBW25 ( pRL-gfp ) . Metabolites were extracted from wild-type ( WT ) Pf-5 , and a phloroglucinol ( PG ) -fed ΔgacA mutant of Pf-5 carrying a plasmid that constitutively expresses pltM ( pME6010-pltM ) . The extracts were fractioned into 48 fractions using semi-preparatory HPLC ( no observable signal activity was detected from the last 12 fractions , data not shown ) . The signal activity of the fractions from both strains was tested using strain P . fluorescens SBW25 containing pRL-gfp ( C ) . Data are means of three biological replicates , and error bars represent the standard deviation of the mean . ( A ) Mass spectra of compound eluting at 3 . 3 min in Fraction 21 ( left ) and 9 min in Fraction 26 ( right ) from the ΔgacA ( pME6010-pltM ) culture extracts . ( B ) Comparisons between the signal compounds identified in Pf-5 cultures to synthesized PG-Cl and PG-Cl2 , which elute at 3 . 3 and 9 min , respectively , as indicated by the dashed arrows . BPC , base peak chromatogram; EIC , extracted ion chromatogram; Std , synthesized standards . The m/z used for EIC analysis of PG-Cl and PG-Cl2 are also shown . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 01110 . 7554/eLife . 22835 . 012Figure 4—figure supplement 1 . LCMS analysis of crude extracts of bacterial metabolites from cultures of wild-type Pf-5 ( A ) and ΔpltM mutant ( B ) . BPC , base peak chromatogram ( top ) ; EIC , extracted ion chromatogram ( 158 . 8–159 . 0 m/z , middle ) ; Std , PG-Cl standard ( bottom ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 012 To obtain sufficient levels of signaling molecules for chemical analysis , we constructed a derivative strain of Pf-5 that overproduced the signaling compounds . This overproducing strain is a ΔgacA mutant of Pf-5 carrying plasmid pME6010-pltM , which expresses pltM from a constitutive kanamycin-resistance promoter ( Pk ) ( Heeb et al . , 2000 ) . The ΔgacA mutant does not produce many of the secondary metabolites produced by wild-type Pf-5 ( Hassan et al . , 2010 ) and thus its use helps to reduce the noise in LCMS analysis . The strain ΔgacA ( pME6010-pltM ) was cultured in a medium amended with PG , and bacterial metabolites from the cultures were extracted and assayed for signal activity using SBW25 ( pPL-gfp ) , as described for the wild-type Pf-5 above . As in the extract of wild-type Pf-5 , two peaks in the extract of the ΔgacA ( pME6010-pltM ) strain , spread across six fractions ( fractions 19–21 , 24–26 ) , showed signal activity ( Figure 4C ) . LCMS analysis revealed compounds with protonated ions consistent with PG-Cl in fraction 21 ( obs . 158 . 9847 , calc . 158 . 9854 , 4 . 4 ppm error ) and 2 , 4-dichlorobenzene-1 , 3 , 5-triol ( PG-Cl2 ) in fraction 26 ( obs . 192 . 9474 , calc . 192 . 9465 , 4 . 7 ppm error ) ( Figure 4A ) . The identities of PG-Cl and PG-Cl2 partially purified from the culture extracts were confirmed by comparing them with the synthesized authentic compounds ( Figure 4B ) . PG-Cl , a molecule with previously undescribed biological activity , has been isolated from the red alga Rhabdoina verticillata ( Blackman and Matthews , 1982 ) . To our knowledge , PG-Cl2 has not been previously reported . The structures of these two chlorinated phloroglucinol derivatives are consistent with the expected products of a reaction catalyzed by PltM , an FADH2-dependent halogenase . To determine the catalytic activity of PltM , we characterized the products generated in vitro from PG by PltM . As a putative FADH2-dependent halogenase , PltM is predicted to require a flavin reductase for catalysis ( Nowak-Thompson et al . , 1999 ) , but no putative flavin reductase is encoded in the plt gene cluster ( Figure 1C ) . Therefore , the previously characterized NAD ( P ) H-dependent flavin reductase SsuE from E . coli ( Keller et al . , 2000 ) was purified and used here . The results showed that , in the presence of co-factors and flavin reductase , PltM converted PG into a compound ( s ) that activated pltL expression , as determined from GFP expression of SBW25 containing pRL-gfp ( Figure 5 ) . Furthermore , LCMS analysis showed that both PG-Cl and PG-Cl2 were present in the in vitro reaction containing PltM , but not in the reaction without PltM , based on comparisons with the authentic standards ( Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 22835 . 013Figure 5 . Effect of extracts from in vitro catalyzed reactions on expression of pltL::gfp in P . fluorescens SBW25 . PltM was purified from E . coli and incubated with a mixture of flavin reductase SsuE , substrate PG , and co-factors FAD and NADPH . The effect of extracted compounds from the reactions was tested using P . fluorescens SBW25 containing pRL-gfp . "+” and "–” indicate the presence or absence of the component in the reactions , respectively . The expression level of pltL::gfp was measured and recorded as relative GFP ( fluorescence of GFP divided by OD600 ) . Letters above columns indicate treatments significantly different from one another , as determined by ANOVA analysis ( p<0 . 05 ) . Data are means of at least three biological replicates from a representative experiment repeated twice with similar results , and error bars represent the standard deviation of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 01310 . 7554/eLife . 22835 . 014Figure 5—source data 1 . Expression of pltL::gfp by SBW25 containing pRL-gfp in response to extracts from in vitro reactions . EM-1# to −6# refers to extracts from each in vitro reaction with the components shown . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 01410 . 7554/eLife . 22835 . 015Figure 5—figure supplement 1 . LCMS analysis of PltM reaction mixtures . ( A ) Base peak chromatogram ( BPC ) for the complete reaction mixture containing PG , FAD , NADH , SsuE , and PltM in Tris buffer ( top ) ; BPC for the reaction mixture lacking PltM ( containing PG , FADH , NADH , and SsuE in Tris buffer ) ( middle ) ; BPC for chlorinated phloroglucinol standards , PG-Cl , PG-Cl2 , and PG-Cl3 elute at 12 . 5 , 17 . 2 , and 18 . 8 min , respectively . A peak observed only in the complete reaction mixture is highlighted in yellow and has an identical retention time to that of the PG-Cl standard . ( B ) Extracted ion chromatograms ( EIC ) for the complete reaction mixture ( red ) or the reaction mixture lacking PltM ( black ) . Top panel , EIC 158 . 8–159 . 0 ( PG-Cl ) ; second panel , EIC 192 . 8–193 . 0 ( PG-Cl2 ) ; third panel , EIC 226 . 8–227 . 0 ( PG-Cl3 ) . Bottom panel , BPC for PG-Cl , PG-Cl2 , and PG-Cl3 standards , which elute at 12 . 5 , 17 . 2 , and 18 . 8 min , respectively . ( C ) Mass spectrum generated by the peak observed in the EIC 158 . 8–159 . 0 of the complete PltM reaction eluting at 12 . 5 min ( PG-Cl calc . [M-H]− 158 . 9854 , 1 . 3 ppm error ) . ( D ) Mass spectrum generated by the peak observed in the EIC 192 . 8–193 . 0 of the complete PltM reaction eluting at 17 . 2 min ( PG-Cl2 calc . [M-H]− 192 . 9465 , 2 . 6 ppm error ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 01510 . 7554/eLife . 22835 . 016Figure 5—figure supplement 2 . Effect of different compounds that have structures similar to phloroglucinol ( PG ) on the expression of pltL::gfp in P . fluorescens SBW25 . Resorcinol , triclosan , and 2 , 4-D ( 2 , 4-Dichlorophenoxyacetic acid ) , which have structures similar to PG , were used to test the substrate specificity of PltM . Different concentrations of these compounds were added to cultures of SBW25 containing reporter plasmid pMRL-gfp in NBGly broth . PG was used as a positive control in this experiment . The same volume of methanol was added to the cultures ( Me ) as a negative control . The expression level of pltL::gfp was monitored and recorded as relative GFP , which was calculated by GFP divided by OD600 . Differing letters above the columns indicate treatments significantly different from one another , as determined by ANOVA analysis ( p<0 . 05 ) . Compared with PG , no comparable signaling effects of the tested compounds were observed . Data are means of three biological replicates from a representative experiment repeated two times with similar results , and error bars represent the standard deviation of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 01610 . 7554/eLife . 22835 . 017Figure 5—figure supplement 2—source data 1 . Expression of pltL::gfp by SBW25 containing pMRL-gfp in response to PG and compounds with similar structures to PG . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 017 By evaluating a range of compounds related to PG as possible precursors , we also demonstrated that PltM exhibits a high degree of substrate specificity ( Figure 5—figure supplement 2 ) . These in vitro data , combined with the genetic data above , demonstrate that pltM is necessary and sufficient for converting PG into PG-Cl and PG-Cl2 , which induce pltL expression . We then set out to test the function of PG-Cl or PG-Cl2 in pyoluteorin biosynthesis of Pf-5 . Addition of synthesized PG-Cl or PG-Cl2 to cultures induced , in a concentration-dependent manner , expression of pltL ( Figure 6A and B ) and pyoluteorin production ( Figure 6C and D ) by the ΔpltM mutant of Pf-5 . The minimum concentration of PG-Cl2 that induced the expression of pltL or production of pyoluteorin was 0 . 01 μM , which is 100-fold less than that observed for PG-Cl . These results suggest that PG-Cl2 is a more potent signaling compound than PG-Cl . Additionally , there is evidence of toxicity: above 0 . 1 µM PG-Cl2 , growth of P . protegens Pf-5 decreased ( Figure 6B ) . PG-Cl was also toxic to Pf-5 , but only at concentrations of 10 μM or higher . Similarly , PG-Cl and PG-Cl2 showed inhibitory effects on the growth of P . fluorescens strain SBW25 ( Figure 6—figure supplement 1 ) . Induction of pltL expression and pyoluteorin production by both PG-Cl and PG-Cl2 required pltR ( Figure 6—figure supplements 2 and 3 ) . This is as expected because PltR is known to be the transcriptional inducer of pyoluteorin biosynthetic genes ( Brodhagen et al . , 2004; Li et al . , 2012 ) . 10 . 7554/eLife . 22835 . 018Figure 6 . Influence of PG-Cl ( left ) and PG-Cl2 ( right ) on expression of pltL::gfp ( A , B ) and production of pyoluteorin ( C , D ) in the ΔpltM mutant of P . protegens Pf-5 . The ΔpltM mutant was cultured in NBGly amended with increasing concentrations of PG-Cl and PG-Cl2 . To test the regulatory effects of PG-Cl and PG-Cl2 on expression of pyoluteorin biosynthetic genes , GFP activity was determined from the ΔpltM mutant containing pL-gfp 20 hr after inoculation ( A , B ) . The OD600 value of bacterial cultures ( 24 hr after inoculation ) was measured to show the influence of PG-Cl and PG-Cl2 on the bacterial growth ( A , B ) . The bacterial cultures were extracted 24 hr after inoculation and production of pyoluteorin , MAPG and DAPG were quantified ( C , D ) . Letters above the symbols indicate treatments significantly different from one another , as determined by ANOVA analysis ( p<0 . 05 ) . Data are means of at least three biological replicates from a representative experiment repeated twice with similar results , and error bars represent the standard deviation of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 01810 . 7554/eLife . 22835 . 019Figure 6—source data 1 . Expression of pltL::gfp and production of pyoluteorin by the ΔpltM mutant of Pf-5 in response to chlorinated phloroglucinols . The production of MAPG and DAPG is also shown . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 01910 . 7554/eLife . 22835 . 020Figure 6—figure supplement 1 . Effect of PG-Cl and PG-Cl2 on growth of P . fluorescens SBW25 . Strain SBW25 was cultured in NBGly amended with PG-Cl and PG-Cl2 . The final concentrations of the compounds used in the cultures are indicated . Bacterial growth was recorded by measuring OD600 value of the cultures at 20 hr . Data are means of three biological replicates , and error bars represent the standard deviation of the mean . The experiment was done at least twice with similar results and results from one experiment are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 02010 . 7554/eLife . 22835 . 021Figure 6—figure supplement 1—source data 1 . Toxicity of PG-Cl and PG-Cl2 to P . fluorescens SBW25 . PG-Cl and PG-Cl2 were tested at different concentrations as indicated . 0 means methanol was added to the cultures as a control . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 02110 . 7554/eLife . 22835 . 022Figure 6—figure supplement 2 . Effect of PG-Cl and PG-Cl2 on expression of pltL::gfp ( A ) and production of pyoluteorin ( B ) in the ΔpltMΔphlD mutant and the ΔpltMΔpltR mutant of P . protegens Pf-5 . The ΔpltMΔphlD mutant and the ΔpltMΔpltR mutant were cultured in NBGly broth with addition of PG-Cl and PG-Cl2 . The final concentrations of the compounds ( µM ) used in the cultures are shown . ( A ) To test the regulatory effects of PG-Cl and PG-Cl2 on expression of pyoluteorin biosynthetic genes , the mutants containing pL-gfp were used . The expression level of pltL::gfp was monitored and recorded as relative GFP , which was calculated as GFP divided by OD600 . ( B ) To test the regulatory effect of PG-Cl and PG-Cl2 on the production of pyoluteorin , secondary metabolites were extracted from the cultures and analyzed by HPLC . Data are means of three biological replicates from a representative experiment repeated two times with similar results , and error bars represent the standard deviation of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 02210 . 7554/eLife . 22835 . 023Figure 6—figure supplement 2—source data 1 . Expression of pltL::gfp and the production of pyoluteorin by Pf-5 wild-type ( WT ) and its derivatives in response to chlorinated phloroglucinols . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 02310 . 7554/eLife . 22835 . 024Figure 6—figure supplement 3 . Effect of PG-Cl and PG-Cl2 on the expression of pltL::gfp in P . fluorescens SBW25 . SBW25 strains containing reporter construct pRL-gfp ( left of dotted line ) and pL-gfp ( right of dotted line ) were cultured in NBGly amended with PG-Cl and PG-Cl2 . The final concentrations of the compounds ( µM ) used in the cultures are indicated . The same volume of methanol was added to the cultures serving as negative controls ( Methanol ) . The expression level of pltL::gfp was monitored and recorded as relative GFP , which was calculated as GFP divided by OD600 . Data are means of three biological replicates from a representative experiment repeated three times with similar results , and error bars represent the standard deviation of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 02410 . 7554/eLife . 22835 . 025Figure 6—figure supplement 3—source data 1 . Expression of pltL::gfp by SBW25 containing pL-gfp or pRL-gfp in response to chlorinated phloroglucinols . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 025 PG-Cl and PG-Cl2 induced pyoluteorin production concurrently with decreased monoacetylphloroglucinol ( MAPG ) and DAPG production by the ΔpltM mutant ( Figure 6C and D ) . This result is consistent with the known co-regulation between the pyoluteorin and DAPG biosynthetic pathways ( Brodhagen et al . , 2004 ) . DAPG and pyoluteorin are known signals produced by distinct biosynthetic pathways with described roles in intercellular communication of P . protegens ( Schnider-Keel et al . , 2000; Brodhagen et al . , 2004 ) . Several observations suggested that PG-Cl and PG-Cl2 also function as cell-cell communication signals: ( 1 ) they are released by the bacterial cell ( Figure 3 and Figure 4 ) ; ( 2 ) they can be taken up by the bacterial cell ( Figure 6 ) ; ( 3 ) they regulate gene expression at the transcriptional level and affect the output ( antibiotic production ) of the cell ( Figure 6 ) ; ( 4 ) their production varies with the growth stage of the bacterial cultures ( Figure 7—figure supplement 1 ) ; and ( 5 ) their regulatory activities rely on a transcriptional regulator , PltR ( Figure 6—figure supplements 2 and 3 ) . These results , combined with our previous finding that PG can function as a chemical messenger between cells of P . protegens ( Clifford et al . , 2016 ) , collectively support the model depicted in Figure 7A , in which PG-Cl and PG-Cl2 can be synthesized from PG by bacterial cells , released into the environment and taken up by neighboring cells to induce expression of pyoluteorin biosynthetic genes and produce pyoluteorin . 10 . 7554/eLife . 22835 . 026Figure 7 . PG-Cl and PG-Cl2 function as cell-cell communication signals of P . protegens Pf-5 . ( A ) Proposed cell-cell communication model . Cell [1] ( ΔphlAΔpltM ) produces phloroglucinol ( PG ) and releases it into the environment . PG is taken up by cell [2] ( ΔphlAΔphlDΔpltA ) and converted into PG-Cl and PG-Cl2 , which are secreted into the environment . PG-Cl and PG-Cl2 are taken up by cell [1] and induce expression of pyoluteorin biosynthetic genes and production of pyoluteorin by cell [1] . The plt gene cluster is shown . Red font indicates that the gene was deleted . ( B ) GFP expression from the pure- and co-culture of the ΔphlAΔpltM mutant containing pL-gfp and the ΔphlAΔphlDΔpltA mutant containing pL-gfp on NAGly plate . ( C ) Antibiosis of the pure- and co-culture of the ΔphlAΔpltM mutant and ΔphlAΔphlDΔpltA mutant against growth of Erwinia amylovora on a NAGly plate . The co-culture was prepared by mixing the two mutants 1:1 and spotting on the plate . The experiment was repeated at least twice , with similar results . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 02610 . 7554/eLife . 22835 . 027Figure 7—figure supplement 1 . Levels of signaling compounds in cultures of P . protegens Pf-5 of different ages . Relative levels of signaling compounds that induce pltL expression were determined from extracts of Pf-5 culture supernatants by assessing GFP expressed by P . fluorescens SBW25 containing the reporter construct pRL-gfp . Cultures of wild-type Pf-5 were grown in 5 ml NBGly for different times , as indicated . Data are means of three biological replicates from a representative experiment repeated three times with similar results , and error bars represent the standard deviation of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 02710 . 7554/eLife . 22835 . 028Figure 7—figure supplement 1—source data 1 . Expression of pltL::gfp by SBW25 in response to extracts of wild-type Pf-5 cultures . Pf-5 wild-type was cultured for different times , as indicated . The culture extracts were tested by SBW25 containing pRL-gfp . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 02810 . 7554/eLife . 22835 . 029Figure 7—figure supplement 2 . Toxicity of pyoluteorin ( A ) and DAPG ( B ) to the plant pathogen Erwinia amylovora . Erwinia amylovora was cultured in NBGly broth for 24 hr in a 96-well plate with or without the tested compounds . The final concentrations of the tested compounds are indicated . The growth of the bacteria was monitored by measuring the value of OD600 of the bacterial cultures . Me , negative control composed of a culture amended with the same volume of methanol used in cultures amended with pyoluteorin or DAPG . Data are means of three biological replicates from a representative experiment repeated two times with similar results , and error bars represent the standard deviation of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 02910 . 7554/eLife . 22835 . 030Figure 7—figure supplement 2—source data 1 . Growth of Erwinia amylovora in response to pyoluteorin and DAPG . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 030 We used ΔphlAΔpltM and ΔphlAΔphlDΔpltA mutants of Pf-5 to test the model . Both mutants expressed the pyoluteorin biosynthetic gene pltL at low levels as indicated by low GFP expression from pL-gfp ( Table 1 , Figure 7B ) . The ΔphlAΔpltM mutant can produce PG but cannot convert it into PG-Cl or PG-Cl2 , because of the deletion of pltM . Adding PG-Cl or PG-Cl2 to the ΔphlAΔpltM mutant culture led to induced expression of pltL ( Table 1 ) , indicating that this mutant could respond to the signals . The ΔphlAΔphlDΔpltA mutant could not produce PG because of the deletion of phlD , but still has the capability to convert PG into PG-Cl and PG-Cl2: adding PG to the culture led to increased expression of pltL ( Table 1 ) . Neither mutant could produce pyoluteorin when grown in pure culture , because of the phlD and pltA mutations in the ΔphlAΔphlDΔpltA mutant and the pltM mutation in the ΔphlAΔpltM mutant . In contrast , pyoluteorin was produced when the two mutants were grown in co-culture ( Table 1 ) . Consistently , both mutants showed increased expression of pltL in the co-culture compared with the mutants grown in pure culture ( Table 1 , Figure 7B ) . These results support our model ( Figure 7A ) that PG produced by the ΔphlAΔpltM mutant could be converted into PG-Cl and PG-Cl2 by co-cultured cells of the ΔphlAΔphlDΔpltA mutant; PG-Cl and PG-Cl2 could then be used by the ΔphlAΔpltM mutant to induce production of pyoluteorin . As expected , the co-culture , but not the pure cultures , of the two mutants inhibited the growth of E . amylovora ( Figure 7C ) , a plant pathogenic bacterium that is sensitive to pyoluteorin ( Figure 7—figure supplement 2A ) . It is worthy to note that neither mutant produces DAPG ( Table 1 ) because of the deletion of phlA , a structural gene required for DAPG biosynthesis ( Figure 1A and B ) , so DAPG production did not obscure the influence of pyoluteorin production on antibiosis . 10 . 7554/eLife . 22835 . 031Table 1 . Concentrations of DAPG , pyoluteorin , and expression of pltL::gfp in pure cultures and co-culture of the ΔphlAΔpltM mutant and the ΔphlAΔphlDΔpltA mutant of P . protegens Pf-5 . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 031Product/expressionAmendment†Bacterial cultures‡ΔphlAΔpltMΔphlAΔpltM + ΔphlAΔphlDΔpltAΔphlAΔphlDΔpltAPyoluteorinNoneBD6 . 2 ± 1 . 8 nmol/gBDDAPGNoneBDBDBDRelative GFPNone34 . 6 ± 2 . 3137 . 4 ± 3 . 4§; 214 . 7 ± 11 . 2¶27 . 6 ± 1 . 9PG ( 10 nM ) 37 . 6 ± 3 . 1NT133 . 3 ± 6 . 4PG-Cl ( 1 µM ) 261 . 0 ± 13 . 3NT308 . 6 ± 21 . 8PG-Cl2 ( 10 nM ) 128 . 2 ± 10 . 7NT268 . 4 ± 16 . 2The mutants were cultured for 3 d on a NAGly plate to determine production of secondary metabolites , and for 20 hr in NBGly broth to determine expression of pltL::gfp from the reporter construct pL-gfp ( Figure 1E ) . The co-culture was prepared by mixing the two mutants at a 1:1 ratio . †PG , PG-Cl or PG-Cl2 was amended in the mutant cultures to test their influences on the expression of pltL::gfp . The concentrations of the tested compounds are shown in parenthesis . ‡Secondary metabolites were extracted from the agar plate and the concentrations of DAPG and pyoluteorin were determined by HPLC . Expression of pltL::gfp was monitored and recorded as relative GFP ( fluorescence of GFP divided by OD600 ) . §The ΔphlAΔpltM mutant , but not the ΔphlAΔphlDΔpltA mutant , contains pL-gfp in the co-culture . ¶The ΔphlAΔphlDΔpltA mutant , but not the ΔphlAΔpltM mutant , contains pL-gfp in the co-culture . Data are presented as mean ± standard derivation . Mean values are calculated from three biological replicates . BD = below detection . NT = not tested . Collectively , our results support the hypothesis that PG-Cl and PG-Cl2 function in cell-cell communication of P . protegens and verify that cells of this bacterium can detect and respond to the chlorinated phloroglucinols produced by neighboring bacterial cells . The primary contribution of this work was the discovery of a novel mechanism of co-regulation between secondary metabolite biosynthetic pathways ( Figure 8 ) . This co-regulatory mechanism involves a metabolic intermediate that is transformed through enzymatic activity into a signal regulating gene expression . In contrast to previously described co-regulatory mechanisms involving a metabolic intermediate ( Vingadassalon et al . , 2015; Cano-Prieto et al . , 2015 ) , the metabolic intermediate responsible for coordination of DAPG and pyoluteorin production is not shared by the two biosynthetic pathways . Instead , PG is an intermediate in DAPG biosynthesis and a precursor for formation of signaling molecules that activate expression of pyoluteorin biosynthetic genes ( Figure 8 ) . 10 . 7554/eLife . 22835 . 032Figure 8 . Proposed model of metabolic co-regulation between DAPG and pyoluteorin biosynthetic pathways in P . protegens . Phloroglucinol ( PG ) is synthesized by the type III polyketide synthase PhlD in the DAPG biosynthetic pathway , and processed into PG-Cl and PG-Cl2 by the halogenase PltM encoded in the pyoluteorin gene cluster . The PG-Cl and PG-Cl2 putatively bind to regulator PltR and induce transcription of pyoluteorin biosynthetic genes and prompt production of pyoluteorin . In these two pathways , the final products ( DAPG and pyoluteorin ) , the biosynthesis intermediate ( PG ) and its derivatives ( PG-CL , PG-Cl2 ) can be released and used by bacteria and function as cell-cell communication signals . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 032 The role of antibiotics in cell-cell signaling is now widely recognized ( Yim et al . , 2007 ) , but the results of this study show that intermediates of antibiotic biosynthesis pathways can also function as , or be converted to , signaling molecules . DAPG and pyoluteorin are known to function as intra- and inter-cellular signals ( Brodhagen et al . , 2004; Maurhofer et al . , 2004; Powers et al . , 2015 ) . Here we found that the chlorinated phloroglucinols PG-Cl and PG-Cl2 function as intra- and inter-cellular signaling compounds . When added to the culture medium , these compounds can induce expression of pyoluteorin biosynthetic genes in P . protegens Pf-5 ( Figure 6 , Figure 6—figure supplement 2 ) . By co-culturing mutants of Pf-5 with deficiencies in phlD and pltM , we also showed that PG , PG-Cl and PG-Cl2 could be released from the producing cells and used by neighboring cells to induce the production of pyoluteorin ( Table 1 , Figure 7 ) . These results support and extend those from a previous report showing that PG can serve as an intra- and inter-cellular chemical messenger , influencing the expression of pyoluteorin biosynthetic genes as well as hundreds of other genes with diverse functions in P . protegens Pf-5 ( Clifford et al . , 2016 ) . Therefore , it appears that the final products DAPG and pyoluteorin , the biosynthetic intermediate PG , and its derivatives PG-Cl and PG-Cl2 all serve as signals influencing gene expression , the production of antibiotics , and the antimicrobial activity of P . protegens ( Figure 8 ) . The signaling roles of small molecules released as intermediates , derivatives , or end products of antibiotic biosynthesis pathways are likely to be more consequential than previously recognized . Activation of pyoluteorin biosynthetic genes requires PltR ( Brodhagen et al . , 2004; Li et al . , 2012 ) ( Figure 1 ) , the pyoluteorin pathway-specific positive regulator belonging to the LysR family . LysR regulators typically require a signal for activity , and the signal is commonly an intermediate or product of the regulated gene cluster ( Maddocks and Oyston , 2008 ) . Pyoluteorin has been proposed to be the signal of PltR ( Li et al . , 2012 ) . However , our data indicate that the signaling effect of pyoluteorin relies on the presence of PG ( Figure 2 ) , which is converted to PG-Cl and PG-Cl2 ( Figure 3 and Figure 5—figure supplement 1 ) . Despite our exhaustive attempts , we were not able to purify the PltR protein to evaluate whether or not PG-Cl or PG-Cl2 binds to PltR in vitro ( data not shown ) . Nevertheless , based on our data that PG-Cl and PG-Cl2 are required for induction of the pyoluteorin biosynthetic genes ( Figure 6 ) , and pltR is required for the signaling action of PG-Cl and PG-Cl2 in both P . protegens Pf-5 ( Figure 6—figure supplement 2 ) and P . fluorescens SBW25 ( Figure 6—figure supplement 2 ) , we speculate that that PG-Cl and PG-Cl2 , rather than pyoluteorin , are signals required for the activity of PltR . An interesting question is why DAPG and pyoluteorin production are so intricately enmeshed ? P . protegens can occupy different environments , including the soil , plant surfaces , and insects , and interact with different organisms including bacteria , fungi , protozoa , and insects ( Brodhagen et al . , 2004; Jousset et al . , 2009; Kidarsa et al . , 2013; Henkels et al . , 2014; Loper et al . , 2016 ) . Both DAPG and pyoluteorin contribute to the antagonism of P . protegens against a range of bacteria , fungi , and oomycetes ( Ohmori et al . , 1978; Keel et al . , 1992 ) and both facilitate the bacterium’s establishment in fungal tissues , such as basidiocarps of the button mushroom Agaricus spp . ( Henkels et al . , 2014 ) . However , the sensitivity of a specific organism to these two antibiotics can be different . For example , pyoluteorin is twenty times more toxic than DAPG to the growth of E . amylovora ( Figure 7—figure supplement 2 ) . Furthermore , these two compounds also have distinct ecological roles . For example , DAPG but not pyoluteorin exhibits toxicity to the amoeba Acanthamoeba castellanii ( Jousset and Bonkowski , 2010 ) . Therefore , natural situations may exist in which it could be more advantageous to produce one compound versus the other . Additionally , antibiotic biosynthesis is associated with metabolic costs . The coordinated production of pyoluteorin and DAPG may be a mechanism for the bacterium to balance metabolic cost against the benefits conferred by production of the two antibiotics . In short , metabolic co-regulation may enable bacteria to activate the production of one antibiotic to combat certain competitors or predators or to occupy specific habitats , while reducing the metabolic burden on the cell by repressing the production of a second compound . In summary , this study demonstrated that PG , an intermediate in the DAPG biosynthetic pathway , is converted into PG-Cl and PG-Cl2 , which serve as signals activating expression of pyoluteorin biosynthetic genes . We also showed that PG-Cl and PG-Cl2 can be released from and sensed by bacteria to regulate antibiosis against a bacterial pathogen , indicating that these metabolites function as cell-cell communication signals . Many questions remain related to the co-regulation of the DAPG and pyoluteorin biosynthesis pathways . Membrane receptors recognizing the various signaling compounds ( DAPG , pyoluteorin , PG , PG-Cl and PG-Cl2 ) have not been identified , but are likely needed to facilitate uptake of these molecules across the bacterial membrane . PG-Cl and PG-Cl2 are now known to be responsible for the previously described activation of the pyoluteorin biosynthesis by nanomolar concentrations of PG , but these compounds do not appear to mediate the repression of pyoluteorin production by higher ( micromolar ) concentrations of PG ( Kidarsa et al . , 2011 ) . The mechanism ( s ) involved in inhibition of pyoluteorin production by PG remain unknown . Here , we report a novel mechanism of co-regulation that is responsible for only one aspect of the complex and intricate co-regulation of two secondary metabolites in P . protegens . Our results highlight the diversity of mechanisms involved in metabolic co-regulation in bacteria and provide an unprecedented example of co-regulation between separate biosynthetic pathways in which the intermediate of one pathway functions as a precursor of the signal ( s ) activating the biosynthesis of a second pathway . The bacterial strains and plasmids used in this study are listed in Table 2 . The sequences of oligonucleotides used in this study are listed in Table 3 . 10 . 7554/eLife . 22835 . 033Table 2 . Bacterial strains and plasmids used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 033Strains and plasmidsGenotype and relevant characteristicsReference or sourceStrainsP . protegensLK099Wild-type Pf-5 ( Howell and Stipanovic , 1979 ) JL4928ΔpltAΔphlDThis studyLK269ΔpltAΔphlDΔpltRThis studyLK270ΔpltMThis studyLK024ΔphlD ( Kidarsa et al . , 2011 ) LK293ΔpltMΔpltRThis studyLK397ΔphlDΔpltMThis studyLK096ΔpltAΔphlA ( Henkels et al . , 2014 ) LK403ΔphlAΔpltMThis studyLK413ΔphlAΔphlDΔpltAThis studyP . fluorescens SBW25SBW25 is a member of the P . fluorescens group but does not contain the phl or plt gene clusters ( Silby et al . , 2009 ) E . coliBL21 ( DE3 ) ompT hsdSB ( rB−mB− ) gal dcmNEBS17-1recA pro hsdR-M+ RP4 2-Tc::Mu-Km::Tn7 Smr Tpr ( Simon et al . , 1983 ) PlasmidspEX18TcGene replacement vector with MCS from pUC18 , sacB+ Tcr ( Hoang et al . , 1998 ) p18Tc-ΔpltMpEX18Tc containing pltM with an internal deletion of 1345 bpThis studypPROBE-NTpBBR1 containing promoterless gfp , Kmr ( Miller et al . , 2000 ) pRL-gfppPROBE-NT containing pltR as well as the intergenic region between pltR and pltL including the promoter of pltL fused with a promoterless gfpThis studypMRL-gfppPROBE-NT containing pltM and pltR as well as the intergenic region between pltR and pltL including the promoter of pltL fused with a promoterless gfpThis studypL-gfpCalled ppltL-gfp previously . Contains the intergenic region between pltR and pltL including the promoter of pltL fused with a promoterless gfp ( Yan et al . , 2016 ) pME6010pACYC177-pVS1 shuttle vector , Tcr ( Heeb et al . , 2000 ) pME6010-pltMpME6010 containing pltM , which is expressed from the constitutive promoter PkThis studypET28a ( + ) Protein overexpression vectorNovagenpET28a-PltMpET28a containing pltM , used for overexpression of the PltM proteinThis studypET28a-SsuEpET28a containing ssuE , used for overexpression of the SsuE proteinThis studySmr , streptomycin resistance; Tpr , trimethoprim resistance; Tcr , tetracycline resistance; Kmr , kanamycin resistance . 10 . 7554/eLife . 22835 . 034Table 3 . Sequences of oligonucleotides used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 22835 . 034OligonucleotidesDNA sequencespltR UpF-XbaGAGAGGTCTAGAAGTGGAGTCTGGTCATCAAGpltR UpRGCTCTTGTCTTGTAGTCTTCCTGTTTCGGAGAApltR DnFAACAGGAAGACTACAAGACAAGAGCCAGACATCpltR DnR-XbaGGTGTGTCTAGAGTGAAGAATGAGCAGGTGTCpltM-F1ATGGTACCCTGCATTTCGATACCGCpltM-R1ATAGAATTCGGCGGACGGGTGGATCpltMRL-F1AGCGATTTCGATCTTCATCCCCpltM-up-ovlpGAAGCGGCGACACCGGATCCAATCGTTTCATCCpltM-dn-ovlpGGATGAAACGATTGGATCCGGTGTCGCCGCTTCpltRfATGGTACCAAGGATTTAGGAATGAAGGCpltM 5'primerGTCATACATATGAATCAGTACGACGTCATTATCpltM 3' primerCAGTGCCTCGAGTCAGACTTTGAGGATGAAACGssuE 5' primerGGAGAGCATATGCGTGTCATCACCssuE 3' primerGTAAAGCTTTTACGCATGGGCATTpltM 3' primer3ATAGGATCCGGCCCCGGGCATCACTCAGpltRrTATAAGCTTTTCAGCCCGGACTTCGCGAGGgfp-plt-r1ATGGTACCATAGACGTACGCTCCTGCRestriction sites used for cloning are underlined in oligonucleotides . P . protegens Pf-5 and derivatives , P . fluorescens SBW25 , and Erwinia amylovora were cultured at 27°C on King’s Medium B agar , Nutrient Agar ( Becton , Dickinson and Company , Sparks , MD ) supplemented with 1% glycerol ( NAGly ) or Nutrient Broth ( Becton , Dickenson and Company ) supplemented with 1% glycerol ( NBGly ) . Liquid cultures were grown with shaking at 200 r . p . m . PG was purchased from Sigma-Aldrich ( St . Louis , MO , USA ) , MAPG was purchased from Sigma Aldrich Chemie GmbH ( Schnelldorf , Germany ) , and DAPG was purchased from Toronto Research Chemicals ( North York , Canada ) . Chlorinated phloroglucinols PG-Cl , PG-Cl2 and PG-Cl3 were purchased from Synthon-Lab ( St . Petersburg , Russia ) . All compounds were >99% pure , as determined by HPLC , and were solubilized in methanol . The ΔpltAΔphlD mutant was made by sequentially deleting phlD and pltA genes from the chromosome of wild-type P . protegens Pf-5 . To delete phlD from the chromosome of Pf-5 , a phlD deletion construct made previously ( Kidarsa et al . , 2011 ) was transferred into strain E . coli S17-1 and used to delete 888 bp from the phlD gene in the chromosome of Pf-5 using a previously described protocol ( Kidarsa et al . , 2011 ) . To further delete pltA , a pltA deletion construct made previously ( Henkels et al . , 2014 ) was used to delete 275 bp from the pltA gene in the chromosome of Pf-5 . The deletion of phlD and pltA was confirmed by PCR and subsequent DNA sequencing . To delete pltR from the chromosome of Pf-5 , two DNA fragments flanking the pltR gene were PCR amplified using the oligonucleotide pair pltR UpF-Xba/pltR UpR and pltR DnR-Xba/pltR DnF ( Table 3 ) . These two fragments were fused together by PCR and digested using XbaI to generate a 1325 bp DNA fragment containing pltR with a 925 bp internal deletion . This DNA fragment was ligated to pEX18Tc to create construct p18Tc-ΔpltR . This deletion construct was introduced into the ΔpltAΔphlD mutant to make the ΔpltAΔphlDΔpltR triple mutant . The deletion of pltR was confirmed by PCR and subsequent DNA sequencing . To delete pltM from the chromosome of Pf-5 , two DNA fragments flanking the pltM gene were PCR amplified using the oligonucleotide pair pltMRL-F1/pltM-up-ovlp and pltM-dn-ovlp/pltRf . These two fragments were fused together by PCR and digested using EcoRI and KpnI to generate a 1120 bp DNA fragment containing pltM with a 1345 bp internal deletion . This DNA fragment was ligated to pEX18Tc to create construct p18Tc-ΔpltM . The construct p18Tc-ΔpltM was transferred into strain E . coli S17-1 and used to delete 1345 bp from the pltM gene in the chromosome of Pf-5 . The deletion of pltM from Pf-5 was confirmed by PCR and subsequent DNA sequencing . To complement the ΔpltM mutant , a 1661 bp DNA fragment containing pltM was PCR amplified using oligonucleotides pltM-F1 and pltM-R1 . This PCR product was digested by KpnI and EcoRI and ligated downstream of the constitutively expressing promoter ( Pk ) of the kanamycin resistance gene in pME6010 to make the complementation construct pME6010-pltM . The plasmid pME6010-pltM was transferred into Pf-5 derivatives by bi-parental mating ( Kidarsa et al . , 2011 ) . For heterologous expression in Escherichia coli , the pltM coding sequence was PCR amplified using oligonucleotides pltM 5’primer and pltM 3’primer . The PCR product was digested with NdeI and XhoI , and ligated to the expression vector pET28a ( + ) to create the expression construct pET28a-PltM , in which pltM was expressed with a N-terminal fused 6X His Tag . Similarly , for SsuE , the ssuE coding sequence was PCR amplified using E . coli K12 genomic DNA as a template , and oligonucleotides ssuE 5’primer and ssuE 3’primer . The PCR product was digested with NdeI and HindIII and ligated to the expression vector pET28a ( + ) to create the expression construct pET28a-SsuE , in which SsuE was expressed with a N-terminal fused 6X His Tag . For reporter constructs , three variants of pltL::GFP were constructed . pL-gfp contains the promoter of pltL fused to a promoterless gfp ( Yan et al . , 2016 ) . The construct pRL-gfp includes the entire pltR coding sequence and the predicted promoter sequence of pltL . To make this construct , a 1600 bp DNA fragment was PCR amplified using oligonucleotides pltRr and gfp-plt-r1 . The resultant PCR fragment was digested with HindIII and KpnI and ligated to pPROBE-NT to create the construct pRL-gfp . The construct pMRL-gfp contains pltM and pltR and the promoter region of pltL fused to a promoterless gfp . To make this construct , a 3050 bp DNA fragment was PCR amplified using pltM 3'primer3 and gfp-plt-r1 . The resultant PCR fragment was digested with BamHI and KpnI and ligated to pPROBE-NT to create the construct pMRL-gfp . Strains P . protegens Pf-5 and P . fluorescens SBW25 containing gfp-based reporter plasmids were cultured overnight in NBGly at 27°C with shaking . The cells were washed once with fresh NBGly and used to inoculate 5 ml NBGly to an optical density of 600 nm ( OD600 ) of 0 . 01 . Five microliters of purified compounds at various concentrations ( pyoluteorin , PG , and chlorinated derivatives of PG ) or extracts ( described below ) were added to the 5 ml cultures . The cultures were incubated for 20 hr at 27°C with shaking . The OD600 was measured to monitor growth of the bacteria . The green fluorescence of bacteria was monitored using a 96-well plate reader ( Tecan Infinite 200Pro , Männedorf , Switzerland ) by measuring emission at 535 nm with an excitation at 485 nm . For each measurement , the green fluorescence value was divided by the corresponding OD600 to determine the relative GFP level . The green fluorescence emitted by cells of wild-type strains without gfp reporter constructs was determined and used for background correction . The expression plasmids pET28a-PltM and pET28a-SsuE were transformed into E . coli BL21 ( DE3 ) . Cells were grown in LB broth , supplemented with 50 mg/ml kanamycin , at 37°C to an OD600 of 0 . 5 , induced with 0 . 4 mM IPTG , and grown for an additional 12 hr at 20°C . The cells were harvested and suspended in Tris-NaCl buffer ( 20 mM Tris pH 7 . 4/200 mM NaCl ) , and lysed by sonication . The His-tagged proteins were purified using the Ni-NTA Purification System ( Invitrogen ) under native purification conditions . Purified PltM and SsuE were dialyzed in Tris-NaCl-Glycerol buffer [20 mM Tris pH 7 . 4/200 mM NaCl/10% ( vol/vol ) glycerol] . The protein concentrations were determined using a DC Protein Assay ( Bio-Rad ) and proteins were stored at −80°C . To test the catalytic activity of PltM , 20 μl of 56 µM PltM , 50 μl of 21 µM SsuE , 4 μl of 10 mM FAD , 10 μl of 100 mM NADPH , and 10 μl of 10 mM PG were added to 106 μl of Tris-NaCl buffer ( 20 mM Tris pH 7 . 4/200 mM NaCl ) . Five additional reactions were also set up , with one of the five components replaced by an equivalent volume of the Tris-NaCl buffer in each reaction . The mixtures were incubated for 3 hr at 25°C and extracted three times with ethyl acetate . The ethyl acetate extracts were dried under vacuum at room temperature and dissolved in 200 μl methanol . A volume of 5 μl from each methanol solution was added to P . fluorescens SBW25 containing the reporter plasmid pRL-gfp to determine the signaling activity of each reaction product . For LCMS analysis , the reactions were concentrated to dryness and dissolved in 100 μl methanol and 10 μl were analyzed as described below . Pf-5 and derivative strains were inoculated at a starting OD600 of 0 . 01 in 5 ml NBGly and cultured for 24 hr . Four milliliters of the culture supernatant were extracted twice with ethyl acetate , then subsequently dried under vacuum , and suspended in methanol . The culture extracts were analyzed by high-performance liquid chromatography ( HPLC ) to quantify production of MAPG , DAPG , and pyoluteorin as described below . To quantify the production of MAPG , DAPG and pyoluteorin by Pf-5 and derivatives , HPLC analyses were accomplished using an Agilent 1100 HPLC instrument , which consisted of a quaternary pump , vacuum degasser , autosampler , column thermostat ( set to 30°C ) , and diode array detector . Separation was achieved using a Luna C18 column ( 4 . 6 × 150 mm , 5 μm , Phenomenex , Torrance , CA ) with a flow rate of 1 ml/min where line A was water + 0 . 1% ( vol/vol ) formic acid , and line B was acetonitrile + 0 . 1% ( vol/vol ) formic acid with the following program . The column was pre-equilibrated in 90% A/10% B , and upon injection this composition was held for 2 min . The composition of the mobile phase was then changed to 0% A/100% B over 28 min using a linear gradient . This composition was held for 6 min then changed to 90% A/10% B over 2 min . The column was equilibrated in 90% A/10% B for 6 min prior to the next injection . Under these chromatographic conditions , MAPG eluted at 12 . 1 min , pyoluteorin eluted at 15 . 1 min , and DAPG eluted at 18 . 1 min . The HPLC was operated with and data viewed withChemStation ( version B . 04 . 03 , Agilent , Santa Clara , CA ) . Quantification was performed by integrating the area under the curve at 300 nm and comparing with a standard curve prepared by injection of purified pyoluteorin , MAPG , and DAPG . Data were processed with GraphPad Prism ( GraphPad Software , San Diego , CA ) . For LCMS analysis of crude culture extracts of the wild-type Pf-5 and the ΔpltM mutant , these two strains were grown for 20 hr in 500 ml NBGly at 27°C with shaking . The supernatants of the cultures were extracted twice with ethyl acetate ( 250 ml ) . The extracts were concentrated in vacuo to dryness . The dried extracts were dissolved in 50 µl methanol and then diluted to 400 µl with water . The large amount of precipitate formed was removed by centrifugation ( 15 , 000 x g , 60 min , 10°C ) , and 10 µl of the clear supernatant were analyzed by LCMS as described below . LCMS analysis was performed using a Agilent 1260 HPLC ( consisting of degasser , quaternary pump , autosampler , and diode array detector ) upstream of a 6230 ESI-TOF operated in negative ionization mode with the following parameters: Mass range , 100–3200 m/z in profile mode; Gas temperature , 350°C; Drying gas , 10 L/min; Nebulizer , 40 psig; Capillary voltage , 3500 V; Fragmentor , 100 V; Skimmer , 65 V; OCT 1 RF Vpp , 750 V; Acquisition rate , five spectra/sec; Time , 200 ms/spectrum; Transients/spectrum , 1902 ) . The synthesized authentic compounds PG-Cl , PG-Cl2 and PG-Cl3 were used as standards for comparisons in the analysis . Separation was achieved with an Extend C18 column ( 2 × 100 mm , 3 µm , Agilent ) at a flow rate of 0 . 2 ml/min . The column was pre-equilibrated in 98% A/2% B , where A was water + 0 . 1% ( vol/vol ) formic acid and B was acetonitrile + 0 . 1% ( vol/vol ) formic acid . This composition was held for 1 min and then changed to 100% B over 30 min using a linear gradient . After that , the composition was held for an additional 12 min and the composition was returned to 98% A/2% B over 2 min . The column was equilibrated at 98% A/2% B for 12 min prior to injection of the next sample . The LCMS was operated with and data were viewed using MassHunter ( version B . 04 . 03 , Agilent , Santa Clara , CA ) . Culture extracts from the wild-type Pf-5 were also fractioned by semi-preparatory HPLC and signal activity of the fractions was tested by the reporter strain P . fluorescens SBW25 containing pRL-gfp . The wild-type Pf-5 was cultured for 20 hr in 500 ml NBGly at 27°C with shaking before being extracted twice with ethyl acetate ( 250 ml ) . The dried extracts were dissolved in 1 ml methanol and mixed with 1 ml water . The concentrated extracts were separated by semi-preparatory HPLC using a PuriFlash Pf450 ( Interchim Inc . , Los Angeles , CA ) . Separation was achieved using a Purisphere C18 column ( 10 × 150 mm , 10 µm , Interchim Inc . ) at a flow rate of 2 ml/min . The column was pre-equilibrated in 90% A/10% B , where A was water and B was methanol . This composition was held for 2 min and then increased to 100% B over 40 min using a linear gradient and pure methanol was flown over the column for an additional 9 min . The effluent from the pump was collected directly into a 96-deep well plate in 1 min time slices after discarding the first 3 min to waste . Five microliters from each well were tested using SBW25 ( pRL-gfp ) and fluorescence was observed as described above . To obtain sufficient levels of signaling molecules for LCMS analysis , the ΔgacA mutant of Pf-5 containing pME6010-pltM was cultured in three flasks , each containing 1000 ml NBGly amended with 12 . 5 µg/ml PG . The cultures were incubated for 20 hr at 27°C with shaking before being extracted twice with ethyl acetate ( 300 ml ) . Extracts were concentrated in vacuo to dryness . The dried extracts from the three replicate cultures were combined and dissolved in 1 . 5 ml methanol . A portion ( 200 µl ) was diluted with 1 . 8 ml of water . The concentrated extracts were fractioned by semi-preparatory HPLC and the signal activity of the fractions were tested using SBW25 ( pRL-gfp ) as described above . The fractions with the highest signal activity were concentrated to dryness separately and analyzed by LCMS along with synthesized standards including PG-Cl , PG-Cl2 and PG-Cl3 as described above . Antibiosis of Pf-5 and derivatives against the plant pathogen E . amylovora was determined on NAGly . E . amylovora was cultured overnight in 5 ml NBGly at 27°C with shaking , and 80 μl of the cultures were inoculated into 8 ml melted NAGly ( 45°C ) , and poured over solidified NAGly in a Petri plate . This two-layer plate was air-dried with the lid open for 1 hr to solidify the agar and remove extra moisture . The ΔphlAΔpltM mutant and the ΔphlAΔphlDΔpltA mutant were cultured overnight in 5 ml NBGly at 27°C with shaking . The cells were washed once with fresh NBGly and used to make a bacterial suspension with an OD600 of 0 . 05 . For co-culture , cells of the two mutants were combined in a 1:1 ratio for a total OD600 of 0 . 05 . Two microliters of the bacterial suspensions were used to inoculate three two-layer plates and incubated for 3 d at 27°C . The agar medium with the Pf-5 strains in the center was cut from the plate and concentrations of DAPG and pyoluteorin in the agar were quantified as follows . Bacterial cells were carefully removed from the agar using Kimwipe . The agar was weighed , sliced into small pieces and immersed for 30 min in sterilized water before extracting twice with ethyl acetate . The extracts were dried and the residue was dissolved in 20 μl methanol and analyzed by HPLC as described above . Statistical analyses were performed by ANOVA and Tukey multiple comparison using SPSS statistics 20 ( SPSS Inc . , Chicago , IL ) . All replicates in this study were biological replicates . Sample sizes are indicated in the figure legends . All data are presented as mean ± standard derivation .	Bacteria live almost everywhere on Earth and often compete with one another for limited resources , like space or nutrients . Certain bacteria produce molecules that are toxic to other microorganisms to give themselves a competitive advantage . These toxic molecules are more commonly referred as antibiotics , and are perhaps best known for their importance in medicine . Yet , antibiotics benefit the bacteria that produce them in other ways too . Some bacteria , for example , use antibiotics as chemical signals to communicate with one another and coordinate their activities . Some bacteria produce many antibiotics with different toxic and signaling activities . These bacteria often coordinate the production of different antibiotics such that the production of one antibiotic shuts down the production of another . This kind of coordination would allow the bacterium to focus its energy on producing only the antibiotic that gives it a competitive advantage at that time . Yet , in most cases , it was not known how the bacterial cell coordinates the production of two different antibiotics . Pseudomonas protegens is a species of bacteria that lives in soil , and produces many antibiotics that are toxic to other bacteria or fungi . The antibiotics are made via distinct pathways of chemical reactions that are catalyzed by different enzymes . However , the production of two antibiotics , called 2 , 4-diacetylphloroglucinol and pyoluteorin , is tightly coordinated in some strains of P . protegens . Now , Yan et al . have discovered how P . protegens coordinates the production of these two antibiotics . It turns out that the bacterium produces an enzyme that adds chlorine atoms onto one of the intermediate building blocks used to make 2 , 4-diacetylphloroglucinol . These “chlorinated derivatives” then activate the genes required to make the second antibiotic , pyoluteorin . The derivatives also signal to other P . protegens cells and trigger them to produce pyoluteorin too . Lastly , Yan et al . confirmed that pyoluteorin could inhibit the growth of another species of bacteria called Erwinia amylovora . These new findings highlight an important role played by chemicals that might have previously been considered as merely stepping stones in other biochemical reactions . An important challenge for the future will be to evaluate if other microbes use chemical intermediates in similar ways . Understanding the natural role of more antibiotics and their intermediates should help us to more wisely use existing antibiotics , and might eventually lead to new treatments for infections in humans and other animals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "microbiology", "and", "infectious", "disease" ]	2017	Novel mechanism of metabolic co-regulation coordinates the biosynthesis of secondary metabolites in Pseudomonas protegens
Secretory cargo that cannot fold properly in the ER are selectively targeted for removal by a well-studied ER-associated degradation pathway , or ERAD . In contrast , very little is known about post-ER quality control mechanisms for damaged or misfolded integral membrane proteins . Here we describe a quality control function of the Rsp5-ART ubiquitin ligase adaptor network that functions to protect plasma membrane ( PM ) integrity . Failure to mediate this protective response during heat stress leads to toxic accumulation of misfolded integral membrane proteins at the cell surface , which causes loss of PM integrity and cell death . Thus , the Rsp5-ART network comprises a PM quality control system that works together with sequential quality control pathways in the ER and Golgi to ( i ) target the degradation of proteins that have exceeded their functional lifetime due to damage and/or misfolding and ( ii ) limit the toxic accumulation of specific proteins at the cell surface during proteotoxic stress . Cells have evolved elaborate protein quality control mechanisms that assist proteins in achieving their desired native structure and also help target the degradation of proteins that have become misfolded , damaged , or otherwise aberrant . This system , often referred to as the cellular proteostasis network , consists of an elaborate network of proteins dedicated to preventing the toxic accumulation of unfolded or misfolded proteins . Newly translated proteins interact with chaperones that direct their folding into the native active form . Proteins that do not fold correctly or older proteins that exceed their functional lifetime are recognized by cellular quality control mechanisms that target them for degradation . Often , the cell employs compartment-specific mechanisms that ensure protein quality control . For example , misfolded protein domains that are soluble and accessible to the cytosol can be targeted for degradation by CHIP , an E3 ligase that interacts with various chaperones to target ubiquitination of misfolded substrates ( Meacham et al . , 2001 ) . Similarly , unfolded soluble protein domains in the lumen of the ER are clients of the Hsp70 family member BiP/Kar2 , and extended association with BiP/Kar2 can lead to targeting for ubiquitination by the ERAD machinery and subsequent proteasomal degradation ( Claessen et al . , 2011; Walter and Ron , 2011 ) . In these examples , targeting of misfolded proteins for ubiquitination and subsequent degradation is mediated by prolonged interaction with chaperones , which serve as adaptors that recruit E3 ubiquitin ligases to modify misfolded clients . In both cases , ubiquitinated substrates are ultimately degraded by the proteasome . Quality control of integral plasma membrane proteins plays a critical role in many human diseases because it affects the quantity of functional proteins—ion channels , nutrient transporters , and signaling receptors—at the cell surface . In contrast to proteins in the cytosol or the ER , which are degraded by the proteasome , most post-ER integral membrane proteins are degraded via lysosomal trafficking . This process is typically ubiquitin-dependent and involves: ( i ) selective recognition of cargo for ubiquitination , ( ii ) capture of ubiquitinated cargo for trafficking to the endosome , ( iii ) endosomal sorting by ESCRT complexes , which package ubiquitinated cargo into vesicles that bud into the lumen of the endosome ( intralumenal vesicles , or ILVs ) , and ( iv ) fusion of these multivesicular endosomes with lysosomal/vacuolar compartments , resulting in cargo degradation ( Henne et al . , 2011; MacGurn et al . , 2012 ) . While the sorting and trafficking mechanisms in this elaborate downregulation process have been studied extensively , the initial step of cargo recognition for ubiquitination , particularly with respect to the recognition of misfolded proteins in post-ER compartments , is still poorly understood . Recently , a peripheral membrane protein quality control system was identified for the detection , ubiquitination , and downregulation of misfolded CFTR from the plasma membrane . This mechanism involves recognition of the misfolded cytosolic domain of CFTR by molecular chaperones ( Hsc70 and Hsp90 ) which then recruit the E3 ubiquitin ligase CHIP to ubiquitinate misfolded CFTR ( Okiyoneda et al . , 2010 ) . Importantly , a similar mechanism was shown to function in the endocytic downregulation of artificial PM cargo fused to cytosolic domains that undergo temperature-induced misfolding ( Apaja et al . , 2010 ) . These studies provide an elegant mechanism for recognition and clearance of membrane proteins with extensive soluble cytosolic domains but do not address how aberrant conformations of membrane spanning domains are detected and degraded . Evidence for other quality control mechanisms at the PM has been reported using many different PM cargoes . For example , early studies of Pma1 , the yeast plasma membrane H+-ATPase , report its rapid degradation following heat stress ( Piper , 1995 ) . More recent studies have proposed that an unstable mutant allele of Pma1 which is rapidly ubiquitinated , endocytosed , and degraded in the yeast vacuole may represent a misfolded form of the protein ( Liu and Chang , 2006 ) . Other studies of the yeast general amino acid transporter Gap1 demonstrate that loss of sphingolipids alters the activity and stability of Gap1 at the cell surface , suggesting that proteins may become misfolded when PM lipid composition is altered ( Lauwers et al . , 2007 ) . However , mechanisms for the recognition and targeted ubiquitination of misfolded integral membrane proteins have remained elusive . Furthermore , the physiological consequences of the accumulation of misfolded proteins at the cell surface remain unclear . Previously , we identified and characterized a network of Rsp5 ubiquitin ligase adaptors called ARTS , or arrestin-related trafficking adaptors . The ART modular adaptor network mediates the recognition and ubiquitination of cell surface proteins , thereby directing the endocytic remodeling of PM protein composition . Thus , ART proteins serve as a key point of regulation during changes in nutrient availability and environmental stress . Here , we report that various proteotoxic stresses trigger extensive cell surface remodeling characterized by the endocytic downregulation of various different PM cargo proteins . Specifically , we show that heat stress results in misfolding of certain thermolabile PM cargo molecules which are targeted for endocytosis and vacuolar degradation by the Rsp5-ART adaptor network . Defects in this endocytic targeting system result in accumulation of cell surface proteins that is toxic during heat stress . This toxicity is associated with loss of PM integrity , which we propose is caused by conformational flux or misfolding of integral membrane proteins at the cell surface . The ART-Rsp5 quality surveillance mechanism at the PM functions in parallel with both ER and Golgi QC systems to control both the quality and quantity of integral membrane proteins in the cell . Previously , we found that certain environmental changes elicit highly-specific surface remodeling programs , such as the methionine-induced endocytic down-regulation of the methionine transporter Mup1 ( Lin et al . , 2008 ) . In contrast , other environmental changes can stimulate global surface remodeling programs , such as the TORC1-mediated tuning of PM protein composition by activation of the ART adaptor network ( MacGurn et al . , 2011 ) . Given the well-characterized effect of high temperature on protein denaturation , we reasoned that the conformational flux facilitated by heat stress might trigger partial denaturation or misfolding of integral membrane proteins , providing a simple assay for the investigation of PM quality control mechanisms . To characterize the effect of heat stress on integral PM protein stability in yeast , we analyzed the trafficking and degradation of various PM cargoes in cells grown at 26°C and shifted to 38°C for 60 min . In general , most cargoes were observed to undergo endocytosis followed by vacuolar trafficking ( Figure 1A and Figure 1—figure supplements 1 and 2 ) and degradation ( Figure 1B and Figure 1—figure supplements 3–5 ) , although a few cargoes appeared to be stable ( Pdr5 , Figure 1B and Figure 1—figure supplement 5 ) or induced during heat stress ( Figure 1—figure supplement 6 ) . This heat-induced endocytic downregulation was observed with varying kinetics and efficiency for many diverse cargoes including amino acid transporters ( Mup1 , Lyp1 , Dip5 ) , hexose transporters ( Hxt3 ) , G-protein coupled receptors ( Ste3 ) and proton pumps ( Pma1 ) ( Figure 1B ) . Similar to heat stress , we found that other known proteotoxic stresses ( growth in 10% ethanol , 5 mM DTT , or 2 . 5 mM diamide ) triggered the endocytosis and vacuolar trafficking of different PM cargoes , suggesting that endocytic clearance may be associated with protein damage or misfolding . For example , ethanol stress triggered the endocytic downregulation of the arginine transporter Can1 and the methionine transporter Mup1 , but had little effect on the surface stability of Lyp1 ( Figure 1—figure supplement 7 ) . In contrast , oxidative stress triggered the endocytic downregulation of Lyp1 and Can1 but had little effect on the surface stability of Mup1 ( Figure 1—figure supplement 7 ) . Given its dramatic affect on surface protein stability , and given the ability of Saccharomyces cerevisiae to tolerate a wide range of temperatures for growth ( Steinmetz et al . , 2002 ) , we decided to further investigate the molecular basis for heat-induced endocytic downregulation . 10 . 7554/eLife . 00459 . 003Figure 1 . Heat stress triggers endocytic downregulation . ( A ) Fluorescence distribution of GFP-tagged endocytic cargoes ( green ) was analyzed in wildtype yeast cells expressing the vacuolar marker Vph1-mCherry ( red ) . Cells were grown to mid-log phase at 26°C ( left panels ) and then shifted to 38°C for 2 hr ( right panels ) . Plasma membrane ( ‘PM’ ) and vacuole ( ‘vac’ ) localization are indicated . Note that the GFP moiety of cargo fusions resists hydrolysis and thus fluorescence signal accumulates in the vacuole even as the protein appears to be degraded as monitored by immunoblot . ( B ) Stability of affinity-tagged cargoes was analyzed following temperature shift from 26°C ( left lane ) to 40°C . The number beneath each lane indicates quantification of protein abundance ( relative to 26°C , t = 0 and normalized to G6PDH as a loading control ) determined using the Li-Cor system . ( C ) Analysis of heat-induced degradation of Lyp1 in the presence of glycerol , a chemical chaperone . ( D ) Detergent solubility of a thermostable cargo ( Pdr5 ) and a thermolabile cargo ( Lyp1 ) was analyzed at low and high temperature . A schematic representation for the experimental design is shown at the left . ( E ) Detergent soluble ( S40 ) and insoluble ( P40 ) fractions from membranes incubated at 40°C for 30 min were analyzed for mobility on sucrose step gradients . Top ( T ) fractions were immobile on the gradient , while bottom ( B ) fractions migrated through the gradient . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 00310 . 7554/eLife . 00459 . 004Figure 1—figure supplement 1 . Cell surface fluorescence intensity was measured for Mup1-GFP and Can1-GFP following growth at 26°C or following a shift to 38°C for 2 hr . Top panels illustrate how PM fluorescence intensity was measured and calculated and bottom panels depict fluorescence intensity measurements over many cells ( n > 30 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 00410 . 7554/eLife . 00459 . 005Figure 1—figure supplement 2 . Mup1-pHluorin was used to quantify surface abundance of Mup1 for a population of yeast cells at 26°C or following a shift to 38°C for 2 hr ( bottom panel ) . Mup1 pHluorin signal originates almost exclusively from the PM ( top panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 00510 . 7554/eLife . 00459 . 006Figure 1—figure supplement 3 . Stability of affinity-tagged Mup1 and Pma1 was analyzed following temperature shift from 26°C to indicated temperatures . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 00610 . 7554/eLife . 00459 . 007Figure 1—figure supplement 4 . Stability of Lyp1 was analyzed following temperature shift from 26°C to indicated temperatures in the presence or absence of 3% glycerol . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 00710 . 7554/eLife . 00459 . 008Figure 1—figure supplement 5 . Stability of affinity-tagged Aqr1 and Pdr5 was analyzed following temperature shift from 26°C to indicated temperatures . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 00810 . 7554/eLife . 00459 . 009Figure 1—figure supplement 6 . Identification of integral PM proteins induced by shifting cells to 40°C . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 00910 . 7554/eLife . 00459 . 010Figure 1—figure supplement 7 . Various proteotoxic stresses trigger endocytic downregulation . Fluorescence distribution of GFP-tagged endocytic cargoes ( green ) was analyzed in wildtype yeast cells expressing the vacuolar marker Vph1-mCherry ( red ) . Cells were grown to mid-log at 26°C and then subject to the following types of proteotoxic stress: 10% ethanol for 2 hr , 5 mM DTT for 2 hr , or 2 . 5 mM diamide for 3 hr . Trafficking of Can1-GFP ( left ) , Mup1-GFP ( middle ) , and Lyp1-GFP ( right ) was analyzed . Plasma membrane ( ‘PM’ ) and vacuole ( ‘vac’ ) localization are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 01010 . 7554/eLife . 00459 . 011Figure 1—figure supplement 8 . Analysis of cargo thermostability . For each heat-induced degradation timecourse experiment , the half-life of each cargo was estimated at each temperature using linear regression . Cargo half-life is shown plotted as a function of temperature for each cargo , revealing different cargoes exhibit different thermostability profiles . The graph on the left shows the thermostability of Lyp1 , Mup1 , and Aqr1 . The graph on the right includes the same data but is scaled to also include thermostability data for Pdr5 . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 01110 . 7554/eLife . 00459 . 012Figure 1—figure supplement 9 . Analysis of Lyp1 thermostability in the absence or presence of 3% glycerol . See also Figure 1—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 01210 . 7554/eLife . 00459 . 013Figure 1—figure supplement 10 . Analysis of the kinetics of Mup1 lysosomal degradation following the addition of methionine at 26°C in the absence or presence of 3% glycerol . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 013 To further characterize the thermostability of specific cargoes , we grew yeast cells at 26°C and measured the kinetics of cargo degradation following shift to different temperatures ( 26°C , 34°C , 38°C , 40°C , 42°C ) all within the normal growth range for S . cerevisiae . We found that different cargoes exhibited a range of thermostabilities ( Table 1 and Figure 1—figure supplement 8 ) suggesting heat stress triggers thermo-instability that is cargo-intrinsic . Since conformational flux and protein misfolding both increase as a function of temperature , we reasoned that heat-induced instability of PM proteins may relate to misfolding or conformational instability caused by increased temperature . To test this idea , we performed thermostability analysis of the highly thermo-labile lysine transporter Lyp1 in the presence of glycerol , which is known to promote/stabilize protein folding and function as a chemical chaperone ( Bernier et al . , 2004 ) . Importantly , glycerol enhanced the thermostability of Lyp1 ( Figure 1C and Figure 1—figure supplements 4 and 9 ) but did not stabilize the methionine-induced endocytic downregulation of the methionine transporter Mup1 ( Figure 1—figure supplement 10 ) . These data demonstrate that glycerol can stabilize certain integral membrane proteins during heat stress but has no effect on substrate-induced endocytosis . Together , these data are consistent with a mechanism whereby heat stress induces different degrees of misfolding in distinct PM proteins and this conformational instability of integral PM proteins results in their selective recognition and subsequent targeting for endocytic downregulation . 10 . 7554/eLife . 00459 . 014Table 1 . Cargo thermostabilityDOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 014Cargo34°C38°C40°C42°CLyp11605025–Pma1Stable9545–Mup1Stable10065–Aqr1–Stable10070Pdr5–Stable400200Kinetic analysis of heat-induced cargo degradation ( Figure 1—figure supplements 3 and 4 ) was used to estimate half-lives of each cargo at each temperature tested . Half-lives are indicated in minutes . ‘–’ indicates half-life not determined . We decided to further explore the possibility that the thermoinstability observed for certain cargo at the PM is a direct result of misfolding during heat stress . First , we tested for changes in the physico-chemical properties of a relatively stable cargo ( Pdr5 ) and a comparatively heat-labile cargo ( Lyp1 ) using a detergent solubilization assay . Specifically , we isolated yeast plasma membranes ( P13 fraction ) , incubated them at 26°C or 40°C , and analyzed the percentage of cargo that is detergent soluble . While Pdr5 did not exhibit a significant change in detergent solubility at high temperature ( Figure 1D , left ) , Lyp1 detergent solubility was significantly reduced during heat stress ( Figure 1D , right ) indicating that Lyp1 , but not Pdr5 , may be misfolding or aggregating at high temperature . To further explore this possibility , we analyzed S40 ( detergent soluble ) and P40 ( detergent insoluble ) fractions from membranes subject to heat stress ( 40°C for 30 min ) by scoring mobility on a sucrose gradient ( Figure 1E ) . Importantly , while detergent-soluble cargo in the S40 floated on the sucrose gradient , detergent-insoluble Lyp1 in the P40 fraction migrated to the bottom of the sucrose gradient ( Figure 1E ) , suggesting Lyp1 aggregation during heat stress . These data demonstrate that a thermolabile cargo ( Lyp1 ) tends to aggregate at high temperature and that the rapid degradation of such cargo in response to heat stress may protect cells from the accumulation of aggregated proteins at the surface . To identify the molecular mechanism that targets misfolded cargo for endocytosis during heat stress , we first tested if this stress response requires Rsp5 , a yeast Nedd4 family ubiquitin ligase that targets many PM cargo for ubiquitination and endocytosis ( MacGurn et al . , 2012 ) . To do this , we analyzed heat-induced cargo trafficking in yeast strains expressing various mutant alleles of Rsp5 . Like all Nedd4 family members , Rsp5 contains a C-terminal HECT ubiquitin ligase domain and multiple WW domains required for substrate targeting . These WW domains direct substrate targeting for ubiquitination by binding to a specific peptide sequence called a PY motif ( PPXY ) present in adaptor proteins ( ARTs , ARRDCs ) and some substrates ( ENaC ) . While the WW1 domain of Rsp5 was dispensable for heat-induced cargo degradation , mutations in either WW2 or WW3 significantly abrogated the endocytic response ( Figure 2A–C ) . Since these mutations result in significant accumulation of cargo at the PM during heat stress , we tested if mutant alleles of RSP5 exhibited defects in thermotolerance by scoring for growth at elevated temperatures . While rsp5-ww1 mutants did not exhibit any observable temperature sensitivity defect , rsp5-ww2 and rsp5-ww3 mutants were extremely temperature sensitive and failed to grow at 38°C ( Figure 2D ) . 10 . 7554/eLife . 00459 . 015Figure 2 . Rsp5 mediates the heat-induced endocytic response . ( A ) Fluorescence distribution of GFP-tagged endocytic cargoes ( green ) was analyzed in wildtype ( left panels ) , rsp5-ww2 ( middle panels ) , or rsp5-ww3 ( right panels ) yeast cells expressing the vacuolar marker Vph1-mCherry ( red ) . Cells were grown to mid-log at 26°C and then shifted to 38°C for 2 hr . Plasma membrane ( ‘PM’ ) and vacuole ( ‘vac’ ) localization are indicated . ( B ) and ( C ) Stability of affinity-tagged cargoes was analyzed following temperature shift from 26°C ( left lane ) to 40°C . The number beneath each lane indicates quantification of protein abundance ( relative to 26°C , t = 0 and normalized to G6PDH as a loading control ) determined using the Li-Cor system . ( D ) Heat-sensitivity analysis of wildtype , rsp5-ww2 , or rsp5-ww3 yeast cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 015 To better understand the basis of this temperature sensitive phenotype , we utilized a polar fluorescent vital dye , propidium iodide ( PI , MW = 668 . 4 Da ) , which binds to nucleic acids but is membrane impermeant . Thus , PI staining can be used to score the number of cells in a population that have lost cellular integrity . For example , wildtype yeast cells grown at 26°C exhibit negligible ( <1% ) PI-positive cells within the population ( Figure 3A ) . However , if these same yeast cells are heat-shocked ( 65°C for 10 min ) or treated with the drug nystatin , which binds to ergosterol and forms pores in the yeast plasma membrane , all yeast cells ( 100% ) stain positive for PI ( Figure 3A ) , demonstrating the utility of PI as a marker for PM integrity . Next , we used flow cytometry to analyze PI staining in wildtype and mutant cells following heat stress . After three hours at 40°C , wildtype and rsp5-ww1 yeast cells did not exhibit any significant PI staining ( Figure 3B , C ) . In contrast , rsp5-ww2 and rsp5-ww3 mutant cells exhibited a significant fraction of the population that stained PI-positive in response to heat stress ( Figure 3B , C ) . These results suggest that the rapid loss of PM integrity of rsp5-ww2 and rsp5-ww3 mutant cells during heat stress is either the basis of thermosensitivity or a secondary effect resulting from cell death . 10 . 7554/eLife . 00459 . 016Figure 3 . Rsp5-mediated endocytic clearance protects PM integrity during heat stress . ( A ) Yeast cells that were either grown at 26°C to mid-log phase ( left ) , heated to 65°C for 10 min ( middle ) , or treated with nystatin ( right ) were stained with propidium iodide ( PI ) and analyzed by fluorescence microscopy ( top ) and flow cytometry ( bottom ) . ( B ) Wildtype ( left ) , rsp5-ww2 ( middle ) , or rsp5-ww3 ( right ) yeast cells were grown to mid-log at 26°C , shifted to 42°C for 2 hr , stained with PI , and analyzed by fluorescence microscopy ( top ) and flow cytometry ( bottom ) . ( C ) Flow cytometry was used to analyze PI staining of the indicated strains at 26°C ( blue bars ) or following growth at 40°C for 3 hr ( red bars ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 016 Heat stress broadly induces protein misfolding and denaturation across all cellular compartments . We next wanted to determine if the observed loss of PM integrity in rsp5 mutants during heat stress is linked to the accumulation of misfolded cargo specifically at the cell surface . To test this , we analyzed PI staining following heat stress for mutants that affect different stages of the endocytic pathway ( Figure 3C ) . While mutations that abrogate endocytosis ( Δend3 , Δrvs167 ) resulted in significant PI staining in response to heat stress , mutant cells defective for downstream trafficking events such as endosomal sorting and multivesicular body biogenesis ( Δvps23 ) , endosomal-vacuolar fusion ( Δvam3 ) , and vacuolar degradation ( Δpep4 ) exhibited negligible PI staining in response to heat stress ( Figure 3C ) . These results indicate that plasma membrane clearance is the critical event protecting cells from heat-induced loss of PM integrity , while downstream sorting and trafficking events are dispensable . Indeed , the misfolding of integral PM proteins , many of which are channels and transporters designed to facilitate import of nutrients into the cell , could contribute to loss of critical ion gradients if not targeted for rapid clearance from the PM . Thus , PM quality control is an important line of defense for maintaining the integrity of the cell . In many cases , Rsp5 ubiquitin ligase activity is directed by adaptor proteins which target specific cargoes at the PM in response to various stimuli . To test if any Rsp5 adaptor proteins function in targeting cargoes for endocytic downregulation during heat stress , we screened a set of yeast knockout strains to identify Rsp5 adaptors required for thermotolerance . While most ART ( arrestin-related trafficking adaptor ) family proteins were found to be dispensable for growth at high temperatures , Δart1 yeast cells were temperature sensitive for growth at 38°C similar to rsp5-ww2 and rsp5-ww3 mutant cells ( Figure 4A and Figure 4—figure supplement 1 ) . Importantly , our analysis revealed that heat-induced endocytosis and vacuolar degradation of most cargoes is Art1-independent , with the exception of the lysine transporter Lyp1 ( Figure 4B ) . In contrast to other cargoes , Lyp1 endocytosis and vacuolar degradation is abrogated in the absence of Art1 ( Figure 4B , C ) , underscoring the cargo-specific role Art1 plays in the heat-induced endocytic response . Consistent with its role in the heat-induced endocytic response , we found that Art1-GFP translocates to the PM in response to heat stress ( Figure 4—figure supplement 2 ) . Thus , while rsp5 mutant strains exhibit broad defects in the heat-induced endocytic response ( Figure 2 ) , Δart1 mutant cells are defective for the turnover of only a subset of PM proteins . 10 . 7554/eLife . 00459 . 017Figure 4 . ARTs protect plasma membrane integrity during heat stress . ( A ) Heat-sensitivity analysis of wildtype and art mutant yeast cells . See Figure 4—figure supplement 1 for additional characterization of art mutant yeast cells . ( B ) Fluorescence distribution of GFP-tagged endocytic cargoes ( green ) was analyzed in wildtype ( left panels ) and Δart1 ( right panels ) yeast cells expressing the vacuolar marker Vph1-mCherry ( red ) . Cells were grown to mid-log at 26°C and then shifted to 38°C for 2 hr . Plasma membrane ( ‘PM’ ) and vacuole ( ‘vac’ ) localization are indicated . ( C ) Stability of affinity-tagged Lyp1 was analyzed following temperature shift from 26°C ( left lane ) to 40°C . The number beneath each lane indicates quantification of protein abundance ( relative to 26°C , t = 0 ) determined using the Li-Cor system . ( D ) Heat-sensitivity analysis of wildtype , Δart1 , Δart2 , and Δart1Δart2 mutant yeast cells . ( E ) Flow cytometry was used to analyze PI staining of the indicated strains at 26°C ( blue bars ) or following growth at 40°C for 3 hr ( red bars ) . ( F ) Kinetic analysis of PI staining for a 40°C heat stress timecourse . ( G ) Fluorescence distribution of GFP-tagged Mup1 ( green ) was analyzed in wildtype ( left panels ) and Δart1Δart2Δart3Δart4Δart5Δart6Δart9 ( right panels ) yeast cells during heat stress . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 01710 . 7554/eLife . 00459 . 018Figure 4—figure supplement 1 . Heat-sensitivity analysis of wildtype and art mutant yeast cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 01810 . 7554/eLife . 00459 . 019Figure 4—figure supplement 2 . Fluorescence distribution of GFP-tagged Art1 ( green ) was analyzed at 26°C and 38°C in yeast cells expressing the vacuolar marker Vph1-mCherry ( red ) . Cells were grown to mid-log at 26°C and then shifted to 38°C for 2 hr . Plasma membrane ( ‘PM’ ) and Golgi localization are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 01910 . 7554/eLife . 00459 . 020Figure 4—figure supplement 3 . Results of a screen to identify multicopy bypass suppressors of the Δart1 ts phenotype . Overexpression of ART2 can suppress the ts phenotype but not the canavanine hypersensitivity phenotype of Δart1 mutant cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 020 Our results indicate that the temperature sensitive growth phenotype exhibited by rsp5-ww2 , rsp5-ww3 , and Δart1 mutant cells likely results from accumulation of misfolded proteins in the plasma membrane , which ultimately triggers loss of PM integrity and cell death . Based on this rationale , we performed a screen to identify bypass suppressors of the Δart1 mutant temperature sensitive phenotype in order to uncover other components of the ART-Rsp5 network or new pathways for PM protein quality control . All multicopy overexpression plasmids that suppressed the Δart1 temperature sensitive growth defect ( 84 total suppressors isolated ) were found to encode either ART1 ( 35 individual isolates ) or ART2 ( 49 individual isolates ) ( Figure 4—figure supplement 3 ) . The identification of ART2 as a suppressor of the Δart1 temperature sensitivity phenotype indicated that ( i ) multiple ART family proteins may protect cells during heat stress and ( ii ) different ART proteins may have overlapping cargo specificities which create redundancy in the system . Although Δart2 mutant cells did not exhibit any temperature sensitive phenotypes , we found that deletion of ART2 in Δart1 mutant cells resulted in an enhanced temperature sensitivity phenotype ( Figure 4D ) that correlated with increased Lyp1 accumulation in the Δart1Δart2 double mutant ( Figure 4C ) . To quantitatively measure the protective role of the ART adaptor network , we combined flow cytometry analysis and the PI staining assay to score a panel of ART deletion mutants for loss of PM integrity during heat stress . By limiting the time of exposure to heat stress ( 40°C for only 3 hr ) we aimed to detect the onset of PM permeability defects associated with accumulation of cargo at the surface . Although Δart1 mutant cells exhibited a slight loss of PM integrity ( ∼18% of the cells are PI-positive following 40°C for 3 hr ) , the other adaptor mutants tested exhibited no defects in PM integrity in response to heat stress ( Figure 4E ) . However , striking synthetic defects were observed when Δart1 was combined with deletion of other ART genes ( Figure 4E , F ) . Furthermore , while ART1 and ART2 are dispensable for Mup1 trafficking to the vacuole in response to heat stress ( Figure 4B and data not shown ) , the loss of multiple ART genes effectively blocks Mup1 turnover in response to heat stress ( Figure 4G ) . These results underscore the critical role of the ART adaptor network in protecting PM integrity during heat stress . Our observations that Rsp5 and the ART adaptor network are critical mediators of the heat-induced endocytic response suggested that these proteins may function as a critical quality control mechanism at the PM . Given that loss of this response during heat stress results in ( i ) cargo accumulation at the PM , ( ii ) loss of PM integrity as defined by the permeabilization of the cell to a large ( 668 . 4 Da ) polar dye ( PI ) and ( iii ) loss of cell viability , we hypothesized that heat stress triggers conformational instability or misfolding of integral membrane proteins at the cell surface and that failure to recognize , remove and degrade these proteins threatens the integrity of the plasma membrane . Based on our observation that Δart1 mutant cells , which are temperature sensitive , exhibit specific accumulation of the lysine transporter Lyp1 during heat stress , we decided to test if Lyp1 accumulation is toxic at high temperature . To do this , we analyzed temperature sensitivity of Δart1Δlyp1 mutant cells and , to our surprise , found that loss of Lyp1 partially suppressed the temperature sensitive phenotype of Δart1 mutant cells ( Figure 5A ) . This result suggests that Lyp1 accumulation at the PM during heat stress is toxic and significantly contributes to the temperature sensitive phenotype of Δart1 mutant cells . 10 . 7554/eLife . 00459 . 021Figure 5 . Cargo accumulation at the PM is toxic during heat stress . ( A ) heat-sensitivity analysis of wildtype , Δart1 , Δlyp1 , and Δart1Δlyp1 mutant yeast cells . ( B ) Empty vector or plasmids encoding Lyp1-FLAG expressed from different promoters ( pCPY , pLYP1 , pADH1 ) were transformed into Δart1Δlyp1 mutant yeast cells and the affect on heat tolerance was scored by growth at the indicated temperatures . Lyp1 expression level for each strain was analyzed by quantitative Western blot ( Li-Cor ) . ( C ) Empty vector or plasmids encoding Lyp1-K10XR-FLAG ( which lacks N-terminal lysine residues ) expressed from different promoters ( pCPY , pLYP1 , pADH1 , pTDH3 ) were transformed into wildtype yeast cells and the affect on heat tolerance was scored by growth at the indicated temperatures . ( D ) Flow cytometry was used to analyze PI staining of Δart1Δlyp1 mutant yeast cells expressing Lyp1 from different promoters ( pCPY , pLYP1 , pADH1 ) following growth at 40°C for 3 hr . ( E ) Flow cytometry was used to analyze PI staining of wildtype yeast cells coordinately expressing Lyp1 from the TDH3 promoter and Art1 from different promoters ( pART1 , pCPY , pADH1 , pTDH3 ) following growth at 40°C for 3 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 02110 . 7554/eLife . 00459 . 022Figure 5—figure supplement 1 . Heat-sensitivity analysis of wildtype yeast cells expressing titrated levels of Lyp1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 02210 . 7554/eLife . 00459 . 023Figure 5—figure supplement 2 . Stability of wildtype Lyp1 or the K10XR mutant , which cannot be ubiquitinated on its N-terminal cytosolic tail . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 02310 . 7554/eLife . 00459 . 024Figure 5—figure supplement 3 . Fluorescence distribution of GFP-tagged wildtype Lyp1 ( green , only in cells expressing Vph1-mCherry ) or the K10XR mutant ( green , only in cells without Vph1-mCherry ) was analyzed following growth at 38°C for 2 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 024 To explore this idea , we titrated Lyp1 expression using several different promoters to determine if Lyp1 accumulation in Δart1Δlyp1 cells is toxic during heat stress . Importantly , we found that Lyp1 expression correlated with temperature sensitivity in Δart1Δlyp1 cells , where the expressed Lyp1 could not be efficiently internalized in response to heat stress ( Figure 5B ) . This same Lyp1 titration failed to cause temperature sensitivity in wildtype cells where Lyp1 undergoes heat-induced endocytosis , but higher levels of Lyp1 expression using the TDH3 promoter did confer temperature sensitivity ( Figure 5—figure supplement 1 ) . To further explore the toxicity of Lyp1 at high temperatures , we engineered a mutant of Lyp1 lacking cytosolic lysine residues ( Lyp1-K10XR ) that could be targets for Art1-Rsp5 ubiquitination . Importantly , the Lyp1-K10XR protein is efficiently delivered to the PM but cannot be internalized or degraded by the heat-induced endocytic response ( Figure 5—figure supplements 2 and 3 ) . High expression levels of Lyp1-K10XR resulted in increased temperature sensitivity even in wildtype cells ( Figure 5C ) , demonstrating that accumulation of Lyp1 at the PM is toxic to the cell during heat stress . Importantly , the increased temperature sensitivity observed when Lyp1 accumulates at the PM correlates with increased PI staining during heat shock ( Figure 5D ) . However , protection against this loss of PM integrity could be conferred upon increasing expression of Art1 ( Figure 5E ) , which prevents integral membrane protein accumulation at the surface ( Lin et al . , 2008; MacGurn et al . , 2011 ) . Thus , our results indicate that accumulation of Lyp1 at the cell surface during heat stress is associated with loss of PM integrity and decreased cell viability . Under normal circumstances , heat stress triggers the recognition of Lyp1 by Art1 , which targets its ubiquitination and removal from the PM , protecting the cell from toxic Lyp1 accumulation at the surface . Our results indicate that Rsp5 and the ART adaptor network function as part of a PM quality surveillance system , detecting integral membrane proteins at the cell surface as they become conformationally unstable and targeting their internalization and degradation . If the Rsp5-ART system is functioning as a bona fide quality control pathway at the PM , we would expect to observe genetic interactions with other quality control systems for integral membrane proteins along the secretory pathway ( Figure 6A ) . To test this , we constructed a genetic array consisting of pairwise deletions known to affect protein quality control in the cytosol , nucleus , ER , Golgi complex , PM , endosome , and vacuole . We systematically scored each double deletion strain for PI staining following 3 hr of growth at 40°C to identify synthetic genetic interactions between quality control pathways ( Figure 6B ) . In general , mutations that affected quality control of soluble cytosolic proteins did not interact genetically with mutations that affected quality control of integral membrane proteins . In contrast , we found that mutations in sequential quality control pathways for integral membrane proteins significantly enhanced the PM integrity defect observed for Δart1 mutant cells following heat stress . For example , abrogation of either Golgi quality control ( Δvps10 , Δgga1 , Δtul1 ) or ER quality control ( Δdoa10 , Δhrd1 , Δire1 ) systems significantly enhanced the PM integrity defect observed for Δart1 mutant cells during heat stress ( Figure 6C , D and Figure 6—figure supplement 1 ) . Furthermore , cells with defects in all three integral membrane quality control systems ( ERQC , GQC , and PM quality control; Δart1Δdoa10Δgga1 triple mutant cells ) exhibited dramatic loss of PM integrity ( ∼55% of total cell population ) following 3 hr at 40°C ( Figure 6D ) . Importantly , mutations that abrogate either ERQC or GQC , or abrogate both ERQC and GQC , do not result in temperature sensitive growth or significant PI-positive staining of cells following heat stress ( Figure 6D ) , suggesting that PM quality control can compensate for defects in quality control along the secretory pathway . These genetic interactions underscore the critical protective function of sequential quality control systems along the secretory pathway and demonstrate how all three integral membrane quality control systems—ERQC , GQC , and PMQC—cooperate to protect the integrity of the PM during proteotoxic stress ( Figure 7 ) . 10 . 7554/eLife . 00459 . 025Figure 6 . Systems level coordination of integral membrane protein quality control . ( A ) Model illustrating the major quality control mechanisms for integral membrane proteins: ERAD ( proteasomal degradation ) , Golgi quality control ( GQC; vacuolar/lysosomal degradation ) , and plasma membrane quality control ( PMQC; vacuolar/lysosomal degradation ) . ( B ) Genetic array constructed for known quality control systems in the cell . Red color indicates the percentage of yeast cells staining PI positive following growth at 40°C for 3 hr ( see key inset ) . Gray box indicates a strain which could not be obtained due to synthetic lethality in the SEY6210 background . ( C ) Node network illustrating all genetic interactions observed for protein quality control systems ( cytosolic quality control , CQC; ER quality control , ERQC; Golgi quality control , GQC; PM quality control , PMQC ) in Figure 6B . ( D ) Flow cytometry was used to analyze PI staining of the indicated strains following growth at 40°C for 3 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 02510 . 7554/eLife . 00459 . 026Figure 6—figure supplement 1 . Heat-sensitivity analysis of wildtype and mutant yeast cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 02610 . 7554/eLife . 00459 . 027Figure 7 . Cellular mechanisms of integral membrane protein quality control . Integral membrane proteins are subject to sequential quality control mechanisms including ERAD in the ER ( I ) , GQC in the Golgi ( II ) , and ART-Rsp5 mediated PMQC at the PM ( III ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00459 . 027 An incredible yet poorly-understood property of protein quality control systems is the ability to preferentially recognize misfolded or aberrant proteins and target them either for re-folding or degradation . The potential number of disordered conformer states occupied by a protein as it misfolds is seemingly infinite , which poses a challenge for quality control systems to specifically recognize misfolded species with a broad range of substrates . In this study , we observed that many PM cargo proteins are turned over following heat stress , which is known to induce broad misfolding of proteins in the cell ( Fang et al . , 2011; Theodoraki et al . , 2012 ) . One thermolabile protein , the lysine transporter Lyp1 , was shown to aggregate in isolated plasma membranes subject to heat stress , suggesting that heat stress triggers broad misfolding of cargo in the PM . We show that the ART-Rsp5 network ( i ) is required for the endocytic downregulation of cargo during proteotoxic stress and ( ii ) protects cells from the loss of PM integrity associated with accumulation of misfolded cargo at the surface . The recognition of misfolded proteins by ARTs is consistent with the finding that glycerol , a chemical chaperone that protects against protein misfolding , stabilized Lyp1 during heat stress without broadly affecting endocytosis ( Figure 1C and Figure 1—figure supplement 10 ) . Furthermore , we demonstrate that different cargo substrates targeted by the ART-Rsp5 network exhibit unique thermostability profiles and are recognized by specific ART-Rsp5 complexes , underscoring the cargo-intrinsic nature of substrate selection . Since each specific protein is expected to experience unique conformational dynamics in response to different proteotoxic stresses , these results are consistent with the targeting of proteins as they become conformationally disordered or misfolded ( Claessen et al . , 2011 ) and underscores the role of specific ART proteins in recognizing distinct misfolded signatures . Although it is still poorly understood how quality control machinery discriminates between folded and unfolded clients , recent studies have established two general mechanisms for targeting misfolded proteins . One mechanism involves targeting of misfolded proteins by binding of E3 ubiquitin ligases to molecular chaperones . For example , the E3 ubiquitin ligase CHIP binds to Hsp70 , which effectively targets chronically misfolded proteins for ubiquitination and subsequent proteasomal degradation ( Meacham et al . , 2001; Okiyoneda et al . , 2010 ) . While it is possible that the ART-Rsp5 is targeted to misfolded integral membrane proteins via chaperones , candidate chaperones for the detection of misfolded integral membrane proteins are not obvious . A second quality control strategy involves the recognition of misfolded proteins directly by E3 ubiquitin ligases via intrinsically disordered regions with the potential to target exposed hydrophobic domains ( Fredrickson et al . , 2011 ) . For example , the nuclear E3 ubquitin ligase San1 is capable of targeting the degradation of a broad range of soluble nuclear proteins by a mechanism that requires intrinsically disordered segments of the San1 protein ( Rosenbaum et al . , 2011 ) . Indeed , while San1 is predicted to contain 12 disordered regions , the complete network of ART adaptors is predicted to contain 173 disordered regions ( some greater than 100 amino acids in length ) demonstrating the potential capacity for this system to directly detect misfolded proteins . Ultimately , recognition of integral membrane protein misfolding may require mechanisms distinct from those employed in the recognition of misfolded soluble proteins . For example , it is possible that misfolding stress may trigger the aggregation of membrane spanning domains and the ART-Rsp5 network may recognize biochemical features of this type of aggregation . Future studies will need to address structural mechanisms for the recognition of misfolded PM proteins by the ART-Rsp5 network . Ultimately , the most critical feature of protein quality control systems is that they are required to protect cells during proteotoxic stress and prevent possible cell death . We found in the course of our investigation that abrogation of the ART-Rsp5 network results in hypersensitivity to heat stress that is associated with accumulation of proteins at the cell surface and loss of PM integrity . Thus , the ART-Rsp5 network plays a critical role in protecting the cell—and in particular , protecting the plasma membrane—during conditions of proteotoxic stress . It is unclear exactly how accumulation of misfolded proteins at the cell surface could result in loss of PM integrity , but we speculate that nutrient transporters and ion channels—whose primary function is to permit selective transport of small molecules through the PM—are a particular liability given their potential to allow non-specific leakage if not properly folded . Indeed , the function of many PM channels and transporters requires transition between multiple conformational states in order to facilitate transfer of molecules between the extracellular space and the cytosol . These conformational transitions must be carefully controlled to prevent loss of critical ion gradients , especially during conditions of proteotoxic stress where excessive conformational flux may present a significant liability to PM integrity . Alternatively , the accumulation of misfolded integral membrane proteins could result in aggregation which broadly disrupts the function of proteins at the surface . Understanding not only the mechanism for recognition and removal of misfolded integral PM proteins but also the basis of their toxicity will contribute to our overall understanding of proteostasis networks and help shed light on the cytotoxicity associated with protein misfolding disorders . In many cases , protein quality control systems are induced or activated during conditions of proteotoxic stress . For example , the heat shock response activates a transcriptional program that increases expression of chaperones , ubiquitination machinery , and other proteins dedicated to dealing with protein folding stress . Similarly , the unfolded protein response is induced following ER stress , resulting in the induction of ER chaperones and ERAD machinery ( Walter and Ron , 2011 ) . While the ART-Rsp5 network does not appear to be induced transcriptionally , we find that it does appear to be activated in response to heat stress . Specifically , we find that heat stress triggers PM translocation of Art1 ( Figure 4—figure supplement 2 ) . Our previous studies revealed that TORC1 signaling induced Art1 dephosphorylation which triggered PM translocation , however we find no evidence that heat stress affects the phosphorylation status of Art1 ( unpublished results ) . Thus , the mechanistic basis of Art1 translocation to the PM in response to heat stress remains to be elucidated . Although removal of misfolded integral PM proteins is critical to protect PM integrity during proteotoxic stress , we speculate that similar quality control mechanisms may also serve important housekeeping functions during normal cell growth . Indeed , all integral PM proteins have a functional lifetime and must ultimately be removed and degraded . Even in the absence of proteotoxic stress , accumulation of these proteins could be detrimental to the cell . Indeed , many human channelopathies , which include many neurological or cardiac disorders , are very likely caused by mutations that affect both the quality and quantity of specific ion channels at the cell surface ( Kullmann , 2010; Rougier et al . , 2010 ) . Our results indicate that quality control systems for integral membrane proteins—including ERQC , GQC , and PMQC—cooperate to limit the accumulation of misfolded proteins at the PM . Although ERAD is the best understood quality control system that targets integral membrane proteins for degradation , we found that defects in ERAD did not significantly affect thermotolerance in the presence of Art1 but did exacerbate the loss of PM integrity in the absence of Art1 . GQC , while still poorly understood , has also recently been shown to protect the cell from proteotoxic stress by capturing misfolded proteins at the Golgi and diverting them to the endosomal system for degradation in the lysosome ( Wang et al . , 2011 ) . We found that defects in all three sequential pathways—ERQC , GQC , and PMQC—resulted in severe PM integrity defects when subject to misfolding stress , suggesting they function as spatially distinct yet mechanistically overlapping protein quality control systems . We cannot exclude the possibility that the synthetic genetic interactions observed between the ART-Rsp5 network , GQC , and PMQC may result not from defects in sequential quality control systems but instead from nonspecific effects of impairing quality control at multiple organelles in the cell . Nevertheless , we propose that the ART-Rsp5 network in yeast functions as the major pathway for clearance of PM proteins that exceed their functional lifetime or misfolded proteins that traffic to the surface by escaping ER and Golgi QC . Understanding the coordination of compartment-specific quality control systems along the secretory and endocytic pathways has the potential to contribute to improved therapeutic strategies for protein misfolding diseases like CFTR and other channelopathies ( Kullmann , 2010; Rougier et al . , 2010 ) . The earliest studies reporting the existence of post-ER quality control of integral membrane proteins provided evidence that misfolding of mutant membrane proteins in yeast is triggered by either high temperature ( Jenness et al . , 1997; Li et al . , 1999; Liu et al . , 2006 ) or changes in PM lipid composition ( Wang and Chang , 2002; Lauwers et al . , 2007 ) . Recent studies have provided evidence of a PM quality control system in mammalian cells that contributes to the ubiquitination and lysosomal trafficking of misfolded ΔF506 CFTR ( Okiyoneda et al . , 2010 ) as well as thermolabile synthetic cargoes ( Apaja et al . , 2010 ) . Both studies converged on a common mechanism that involves recognition of misfolded soluble cytosolic domains by chaperones ( Hsp70/Hsc70/Hsp90 ) which recruit CHIP , an E3 ubiquitin ligase . While the yeast genome does not encode a homolog of mammalian CHIP , the analogous proteins , Ubr1 and Ubr2 , are also required for the degradation of misfolded proteins ( Khosrow-Khavar et al . , 2012 ) . However , Ubr1 and Ubr2 are dispensable for the heat-induced endocytic response associated with the turnover of integral PM proteins ( Figure 6B and data not shown ) . Instead , our results indicate that the recognition of misfolded integral membrane proteins in yeast is mediated by ART adaptor proteins . It is tempting to speculate that a similar peripheral quality control function may be mediated by the ARRDC family of proteins in mammalian cells , which encode proteins similar to ARTs with an N-terminal arrestin-domain and a C-terminal domain containing multiple PY motifs capable of binding to members of the Nedd4 family of ubiquitin ligases ( Patwari and Lee , 2012 ) . In the context of a multicellular organism , these membrane quality control systems may prevent the affects of accumulated protein damage over time and thus ultimately protect cells from premature aging and cell death . Yeast cells expressing fluorescent fusion proteins were grown to mid-log phase in synthetic media . Microscopy was performed using a fluorescence microscope ( DeltaVison RT; Applied Precision , Issaquah , WA ) equipped with FITC and rhodamine filters . Images were captured with a digital camera ( Cool Snap HQ; Photometrics , Tucson , AZ ) and deconvolved using softWoRx 3 . 5 . 0 software ( Applied Precision , Issaquah , WA ) . Yeast cells expressing epitope-tagged protein were grown to mid-log phase in synthetic media . 5 OD600 equivalents of mid-log cells pretreated at the indicated temperatures were harvested by precipitation in 10% trichloroacetic acid ( TCA ) . Precipitates were washed in acetone , aspirated , resuspended in lysis buffer ( 150 mM NaCl , 50 mM Tris pH7 . 5 , 1 mM EDTA , 1% SDS ) , and mechanically lysed with glass beads . Protein sample buffer ( 150 mM Tris pH 6 . 8 , 6M Urea , 6% SDS , 10% beta-mercaptoethanol , 20% Glycerol ) was added and extracts were analyzed by SDS-PAGE and immunoblotting with anti-FLAG ( Sigma , St . Louis , MO ) antibody . Yeast cells were disrupted by bead-beating in lysis buffer ( PBS + 1 mM EDTA + protease inhibitors ) . Lysates were centrifuged at 500×g for 5 min to remove unbroken cells and nuclei . Cleared lysates were then spun at 13 , 000×g for 15 min to isolate P13 fractions . P13 fractions were washed once , resuspended in lysis buffer and incubated for 30 min at either 26°C or 40°C then placed on ice . Detergent ( n-dodecyl-β-D-maltopyranoside; Affymetrix , Santa Clara , CA ) was added to 1% and samples were incubated for 30 min at 4°C . Samples were then spun at 40 , 000×g for 30 min to isolate S40 and P40 fractions . To assay protein aggregation , S40 and P40 fractions were loaded onto 20%/40% sucrose step gradients and spun at 55 , 000×g for 60 min . Top and bottom fractions of the sucrose gradient were isolated , precipitated in 10% TCA and analyzed by Western blot . Yeast strains were grown to early-log phase ( around 0 . 2 OD600 ) in synthetic media , shifted to 40°C and cultured for an additional 3 hr . 1 OD600 equivalent of cells was pelleted and resuspended in PBST ( 0 . 01% Tween 20 ) and cells were stained with propidium iodide ( Sigma ) for 20 min . Cells were then washed twice with ddH20 and analyzed by flow cytometry .	Cells have evolved elaborate mechanisms for the detection of misfolded or damaged proteins , and for targeting their degradation . Since the accumulation of misfolded proteins is toxic to the cell , these protein quality control systems are critical for the maintenance of normal cellular function over the lifetime of an organism . The breakdown of this quality control correlates with the progression of neurodegenerative disorders including Alzheimer's , Huntington's and Parkinson's disease . Normal function of the protein quality control machinery can also cause disease: this is the case with channelopathies such as cystic fibrosis , in which mutant ion channels are targeted for degradation and therefore cannot function correctly at the cell surface . Understanding how protein quality control systems recognize misfolded proteins and target their degradation , and designing ways to stabilize or destabilize specific targets , particularly at the cell surface , could thus lead to the development of new therapeutic strategies . While protein quality control mechanisms in the cytosol and endoplasmic reticulum ( ER ) have been studied extensively , much less is known about quality control of integral membrane proteins after they exit the ER . Maintaining the quality of cell surface proteins impacts many critical biological functions including nutrient uptake , signaling and the functioning of specialized surface structures such as cell junctions . Here , Zhao et al . describe a new quality control mechanism that prevents misfolded proteins from accumulating in the plasma membrane . Building upon earlier work describing a network of adaptor proteins ( called ARTs ) for the Rsp5 ubiquitin ligase , Zhao et al . show that subjecting cells to proteotoxic stress , particularly thermal stress , triggers ART-Rsp5-mediated clearance of misfolded plasma membrane proteins . When ART-Rsp5-mediated clearance is abrogated , misfolded proteins accumulate at the cell surface , resulting in a rapid loss of cellular integrity . In the brain , such proteotoxicity can lead to cell death and neurodegeneration , thereby highlighting the importance of this plasma membrane quality control system .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2013	The ART-Rsp5 ubiquitin ligase network comprises a plasma membrane quality control system that protects yeast cells from proteotoxic stress
We describe the genome-wide distributions and temporal dynamics of nucleosomes and post-translational histone modifications throughout the maternal-to-zygotic transition in embryos of Drosophila melanogaster . At mitotic cycle 8 , when few zygotic genes are being transcribed , embryonic chromatin is in a relatively simple state: there are few nucleosome free regions , undetectable levels of the histone methylation marks characteristic of mature chromatin , and low levels of histone acetylation at a relatively small number of loci . Histone acetylation increases by cycle 12 , but it is not until cycle 14 that nucleosome free regions and domains of histone methylation become widespread . Early histone acetylation is strongly associated with regions that we have previously shown to be bound in early embryos by the maternally deposited transcription factor Zelda , suggesting that Zelda triggers a cascade of events , including the accumulation of specific histone modifications , that plays a role in the subsequent activation of these sequences . In most animals , the first phase of embryonic development depends solely on maternally deposited proteins and RNAs and is often accompanied by very low or undetectable transcription ( Newport and Kirschner , 1982a , 1982b; Tadros and Lipshitz , 2009 ) . After several hours to several days , depending on the species , zygotic transcription initiates , marking the beginning of a process known as the maternal-to-zygotic transition ( MZT ) during which maternally deposited RNAs are degraded and the zygotic genome assumes control of its own mRNA production . In Drosophila melanogaster , sustained zygotic transcription begins around mitotic cycle 7 , about an hour into development , although there is growing evidence that very low levels of transcription occur even earlier ( Ali-Murthy et al . , 2013; ten Bosch et al . , 2006 ) . Zygotic transcription gradually increases with each subsequent mitotic cycle , but it is not until the end of mitotic cycle 13 that widespread zygotic transcription is observed ( Pritchard and Schubiger , 1996; Lécuyer et al . , 2007; Lott et al . , 2011; McKnight and Miller , 1976 ) . This zygotic genome activation , along with the elongation of mitotic cycle , and cellularization of the syncytial nuclei defines the mid-blastula transition ( MBT ) . Approximately 3000 genes are transcribed in the cellular blastoderm ( De Renzis et al . , 2007; Lécuyer et al . , 2007; Lott et al . , 2011 ) . Of these , roughly 1000 are expressed in spatially restricted patterns ( Tomancak et al . , 2007; Combs and Eisen , 2013 ) , a result of the differential binding by around 50 spatially patterned transcription factors to several thousands known and putative patterning transcriptional enhancers ( Li et al . , 2008; MacArthur et al . , 2009 ) . In the cellular blastoderm , active genomic regions are biochemically distinct from the rest of the genome: they have relatively low nucleosome densities; are bound by transcription factors , polymerases and other proteins that mediate their activity; and have characteristic histone modifications ( Li et al . , 2011; Nègre et al . , 2011 ) . This high level of activity and relatively complex landscape of genome organization is remarkable given that an hour earlier the genome was being continuously replicated and doing little else . Although the Drosophila cellular blastoderm is among the most well-characterized animal tissues , the transition from quiescent to active state that precedes the formation of this tissue remains poorly understood , despite increasing evidence of its importance ( Liang et al . , 2008; Blythe et al . , 2010; Harrison et al . , 2011; Nien et al . , 2011; Liang et al . , 2012; Lee et al . , 2013; Leichsenring et al . , 2013 ) . We have previously shown that a single seven base-pair DNA motif is found in the vast majority of patterning enhancers active in the cellular blastoderm ( Li et al . , 2008 ) , and that the maternally deposited transcription factor Zelda ( ZLD ) ( Liang et al . , 2008 ) , which binds to this sequence , is present at these enhancers by mitotic cycle 8 , in the early phase of the MZT ( Harrison et al . , 2011 ) . ZLD binding sites were also found in the promoters of most genes activated in this early phase ( ten Bosch et al . , 2006 ) , suggesting that ZLD may play a broad role in early embryonic genome activation and suggesting that ZLD might play a role analogous to the pioneer transcription factors that choreograph the reorganization of genome activity during differentiation , as reviewed in Zaret and Carroll , ( 2011 ) . Although ZLD mutants alter the expression of a large number of cellular blastoderm genes ( Liang et al . , 2008 ) , and affect transcription factor binding in the cellular blastoderm ( Yanez-Cuna et al . , 2012; Foo et al . , 2014; Xu et al . , 2014 ) , little is known about its molecular function or when its activity is required . We hypothesized that ZLD might affect transcription factor binding and enhancer activity indirectly through interactions with chromatin . To further explore this possibility , and to better situate ZLD action in the broader context of early embryogenesis , we decided to characterize the chromatin landscape of D . melanogaster embryos throughout the MZT . To define the chromatin landscape before , during and after the maternal-to-zygotic transition , we collected D . melanogaster ( Oregon-R ) embryos from population cages at 25°C for 30 min , and aged them for 55 , 85 , 120 and 160 min to target mitotic cycles 8 , 12 , 14a and 14c respectively , prior to fixing them with formaldehyde ( Figure 1A ) . 10 . 7554/eLife . 03737 . 003Figure 1 . Hand sorting based on morphology results in tightly staged embryos . ( A ) Experimental scheme . D . melanogaster embryos were collected and allowed to develop before being fixed with formaldehyde . Fixed embryos were hand sorted to obtain pools of embryos within a relatively narrow age distributions between mitotic cycle 8 and the end of cycle 14 . To serve as carrier and normalization standard , chromatin from fixed stage 5 ( cycle 14 ) D . pseudoobscura embryos was prepared and added to the chromatin from the sorted embryos prior to chromatin immunoprecipitation . In ChIP-seq data analysis , the sequencing reads for D . pseudoobscura were used to normalize the D . melanogaster ChIP-seq signals . ( B ) Embryo collection and sorting . The timeline of the early embryogenesis is depicted on top with the relative lengths of each mitotic cycle approximated by the size of the box . The developmental stages ( from 1–5 ) are indicated by different colors . The earliest sustained transcription is detected is at cycle 7 , and the mid-blastula transition ( MBT ) occurs when a large number of genes are transcriptionally activated at approximately the end of cycle 13 . We generated four pools of sorted embryos with developmental stages centered around cycles 8 , 12 , 14a , or 14c as shown by differential interference contrast ( DIC ) and DAPI . ( C ) We determined the distribution of the developmental cycle of the embryos in each pool as shown by counting the number of nuclei in DAPI-stained embryos or by examining the extent of membrane envagination during cycle 14 . DOI: http://dx . doi . org/10 . 7554/eLife . 03737 . 00310 . 7554/eLife . 03737 . 004Figure 1—Figure supplement 1 . Normalization using D . pseudoobscura . Analysis of the D . pseudoobscura chromatin used to normalize ChIP signals between different stages . Here we show the variance among the four replicates of D . pseudoobscura used to normalize the four D . melanogaster datasets ( for cycles 8 , 12 , 14a and 14c ) . ( A ) Number of D . pseudoobscura peaks , as in Figure 2B ( for D . melanogaster reads ) . ( B ) Comparison of each D . pse replicate ( Y-axis ) to their average ( X-axis ) . Dpse . = D . pseudoobscura . DOI: http://dx . doi . org/10 . 7554/eLife . 03737 . 004 As D . melanogaster females often retain eggs post-fertilization , leading to unacceptable levels of contaminating older embryos in embryo pools ( Harrison et al . , 2011 ) , we manually removed embryos of incorrect stages by inspection under a light microscope , as previously described ( Harrison et al . , 2011 ) . The purity of the resulting embryo pools was confirmed by examining the density of nuclei in 4' , 6-diamidino-2-phenylindole ( DAPI ) stained samples from each pool ( Figure 1B , C ) . We carried out chromatin immunoprecipitation and DNA sequencing ( ChIP-seq ) using commercial antibodies against nine post-translation modifications ( acetylation at H3K9 , H3K18 , H3K27 , H4K5 and H4K8 , mono-methylation at H3K4 , and tri-methylation at H3K4 , H3K27 and K3K36 ) , as well as histone H3 ( Table 1 ) . 10 . 7554/eLife . 03737 . 005Table 1 . Antibodies used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 03737 . 005MarkAB sourceAB catalog #H3Abcamab1791H4K5acMillipore07-327H4K8acAbcamab15823H3K18acAbcamab1191H3K27acAbcamab4729H3K4me1Abcamab8895H3K4me3Abcamab8580H3K9acActiveMotif39 , 138H3K27me3Millipore07-449H3K36me3Abcamab9050 As we sought to compare not just the genomic distribution of marks but also their relative levels across the MZT , we developed a strategy to normalize across time-points for the same antibody . Briefly , we prepared chromatin from stage 5 Drosophila pseudoobscura embryos ( mitotic cycle 14 ) , and ‘spiked in’ a fixed amount of this common reference to each D . melanogaster chromatin sample prior to ChIP and sequencing . We chose D . pseudoobscura since it is sufficiently diverged from D . melanogaster that there is very little ambiguity in the assignment of reads to the correct species ( Paris et al . , 2013 ) . The D . pseudoobscura chromatin served as an internal standard . Since the D . pseudoobscura chromatin in each sample was identical , we expected it to be identically immunoprecipitated ( within experimental error ) . Indeed , we found that both the number of peaks ( Figure 1—Figure supplement 1A ) and the peak-by-peak signal ( Figure 1—Figure supplement 1B ) for the D . pseudoobscura fraction across time-points were fairly stable . We therefore used the relative recovery of D . melanogaster compared to D . pseudoobscura in each time-point as a measure of the relative abundance of the corresponding mark in that time-point . We used three measures of genome-wide recovery of each histone mark to examine their dynamics: the total normalized number of D . melanogaster reads ( Figure 2A ) , the number of regions scored by MACS as enriched ( Figure 2B ) and the average ChIP signal among all enriched regions ( Figure 2C ) . These all gave qualitatively similar results except for H4K5ac , which had anomalously few peaks at early stages despite being found at uniformly high levels across the genome . 10 . 7554/eLife . 03737 . 006Figure 2 . Global levels of histone marks change over early development . ( A ) The number of aligned reads ( after normalization to D . pseudoobscura ) for the four developmental time points are indicated for each histone mark and histone H3 . ( B ) The number of peaks detected using the peak calling program , MACS ( Zhang Y et al . , 2008 ) , for each histone mark at each stage are shown . ( C ) Box plots show the trend of average ChIP-seq signals over ±500 bp around the peaks detected across all stages for each histone mark . The dark line in the middle of the plot represents the median , the edges of the box represent the first and third quartiles . DOI: http://dx . doi . org/10 . 7554/eLife . 03737 . 006 As expected , global levels of histone H3 were relatively stable , although we observed a gradual increase of approximately 1 . 4-fold over time , possibly reflecting an overall increase in nucleosome density and chromatin compaction in cycle 14 relative to cycle 8 . The replication associated mark H4K5ac ( Sobel et al . , 1994 ) , found ubiquitously across the genome , declined rapidly from cycle 8 onwards , consistent with the elongation of cell cycles duration over time and the decreasing fraction of nuclei caught in S phase . The remaining marks all showed dramatic increases over the MZT . H4K8ac , H3K18ac , and H3K27ac were enriched at hundreds of loci at cycle 8 and steadily increased through cycle 14 . The remaining marks , H3K9ac , H3K4me1 , H3K4me3 , H3K36me3 and H3K27me3 , were effectively absent at cycles 8 and 12 , but showed sharp increases at cycle 14a . This distinction between these two groups of marks is evident when examining levels of histone modification at individual loci ( Figure 3 ) . 10 . 7554/eLife . 03737 . 007Figure 3 . Dynamics of H3 and histone marks around selected genes . The normalized ChIP-seq signal profiles , for histone H3 and nine different histone marks at four development time points at selected genomic loci . Shown are the early onset genes , sna and amos , and late onset genes , btsz and lea . The peak regions of histone acetylation marks detectable prior to MBT are highlighted with cyan-colored boxes . The peak regions for histone marks detected only after the MZT are highlighted by yellow-colored boxes . Below are the ZLD ChIP-seq profile ( Harrison et al . , 2011 ) from c8 , 13 , and 14 embryos , as well as RNA-seq signals ( Lott et al . , 2011 ) at c11 , c13 , c14b . DOI: http://dx . doi . org/10 . 7554/eLife . 03737 . 007 The transcription of several thousand genes is initiated during the period covered by our analyses , and we were interested in the relationship between the timing of the onset of transcription at individual loci and their chromatin dynamics . We used high-temporal resolution expression data previously collected by our lab ( Lott et al . , 2011 ) to identify genes that were exclusively zygotically transcribed , genes whose mRNAs were deposited into the egg maternally , and genes that are not transcribed in the early embryo . We divided the exclusively zygotic genes into four temporal groups according to their onset times ( Figure 4 ) , and used RNA polymerase II binding data from ( Chen et al . , 2013 ) to divide maternally deposited genes into those transcribed in the early embryo ( maternal-zygotic genes ) and those that are not . We then examined patterns of nucleosome enrichment and histone modifications around the transcription start sites , and in the gene body , of genes in each of these classes ( Figure 5 ) . 10 . 7554/eLife . 03737 . 008Figure 4 . Classification of genes based on timing of transcriptional initiation during early embryogenesis . Using single-embryo RNA-seq data from our group ( Lott et al . , 2011 ) , we identified three broad classes of genes: those at high levels in the earliest embryos ( ‘maternal’ genes ) , those not present in the earliest embryos , but transcribed prior to or during mitotic cycle 14 ( ‘zygotic’ ) , and those not present through cycle 14 ( ‘silent’ ) . We further divided the zygotic genes into four different groups based on their onset of zygotic expression—‘Early’ genes with onset of expression around mitotic cycles 10–11 , ‘Mid’ genes at cycles 12–13 , ‘Late’ genes at early cycle 14 , and ‘Later’ zygotic genes whose onset of expression was during late cycle 14 . Post-MBT polII ChIP data ( Chen et al . , 2013 ) was used to define two maternal groups of genes—those bound by polII in the embryo ( ‘Mat/Zyg’ genes ) , and those that are strictly maternally deposited ( ‘Mat-only’ genes ) . ( A ) Heatmap showing the expression levels for all groups at 8 timepoints ( from cycle 10 through 14D ) across the MZT ( Lott et al . , 2011 ) . ( B ) Heatmaps showing RNA polymerase II ChIP-seq signals ( Chen et al . , 2013 ) around the transcription start sites ( ±1 . 5 kb ) of the genes in each category for three developmental time points , pre-MBT ( left ) , MBT ( middle ) , and post-MBT ( right ) embryos . Genes within each group were ordered based on cycle 14 RNA polymerase II signals ( genes with the highest signal are on top ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03737 . 00810 . 7554/eLife . 03737 . 009Figure 5 . Relationship between H3 depletion , histone modifications and transcription dynamics . Heatmaps show ChIP-seq signals for histone H3 and different histone modification marks at each stage centered around the transcription start sites ( ±1 . 5 kb ) . Genes are groups and ordered as described in Figure 4 . For each histone mark and for histone H3 the same color scaling was used for heatmaps across all four developmental time points . TSS = transcriptional start site . DOI: http://dx . doi . org/10 . 7554/eLife . 03737 . 009 Nucleosome free regions ( NFRs; areas of relatively low histone H3 recovery ) emerged around the transcription start sites of zygotically transcribed genes at roughly the same time that their transcripts were evident in our transcription data ( Figure 5 ) . Several histone modifications appeared along with transcription: H4K8ac , H3K18ac and H3K27ac ( Figure 5 ) . In contrast , H3K9ac and the four histone methylation marks examined here were absent until cycle 14a , when widespread transcription begins . In cycle 14 embryos , regions both upstream and downstream of the promoters of zygotically expressed genes were enriched for the Polycomb-associated mark H3K27me3 ( Figure 5 ) , while the mark was almost completely absent from maternal genes , consistent with the known role of Polycomb group proteins in cell-type specific silencing of developmental genes ( Boyer et al . , 2006; Lee et al . , 2006; Schwartz et al . , 2006 ) . We also observed that maternally deposited genes , both those transcribed in the early embryo and those that are not , had fairly strong NFRs upstream of the promoter at all time points ( Figure 5 ) . A strong transcription-independent NFR in maternally deposited genes was been previously described by Gaertner et al . , ( 2012 ) , who also showed that , based on DNA sequence alone , these genes also have a strong predicted NFR upstream of the promoter . Our data extend this observation , showing that this NFR is developmentally stable . It has been previously observed that maternal-zygotic genes have different promoter motifs than zygotic genes ( Gaertner et al . , 2012; Chen et al . , 2013 ) and this may be at least partially responsible for the difference . Many of the genes transcribed by cycle 14 are expressed in clear spatial patterns ( Lécuyer et al . , 2007; Tomancak et al . , 2007; Combs and Eisen , 2013 ) driven by the action of distinct transcriptional enhancers . Although several catalogs of blastoderm enhancers exist ( Gallo et al . , 2011 ) , they are limited in scope . To generate a larger set of likely enhancers , we took advantage of the strong correlation between the binding of transcription factors known to regulate blastoderm expression and enhancer activity ( MacArthur et al . , 2009; Fisher et al . , 2012 ) . We calculated the cumulative in vivo binding landscape of 16 early developmental transcription factors , including the anteroposterior regulators Bicoid , Caudal , Hunchback , Giant , Krüppel , Knirps , Huckebein , Tailless , and Dichaete; and the dorsoventral regulators Dorsal , Snail , Twist , Daughterless , Mothers against dpp , Medea , and Schnurri ( MacArthur et al . , 2009 ) . We then identified a set of 784 regions showing the strongest overall binding , excluded peaks overlapping promoters and coding regions , and obtained a stringent set of 588 likely blastoderm enhancers in introns and intergenic regions . The chromatin state we observed associated with these putative enhancers during cycle 14 , when blastoderm enhancers are active and bound by multiple transcription factors were as expected based on previous studies of transcriptionally active mammalian and insect cells and tissues ( Heintzman et al . , 2007 , 2009; Wang et al . , 2008; Creyghton et al . , 2010; Kharchenko et al . , 2011; Nègre et al . , 2011; Rada-Iglesias et al . , 2011 ) . As shown in Figure 6 , enhancers at cycle 14 exhibited strong nucleosome depletion ( Kaplan et al . , 2011; Li et al . , 2011 ) . Flanking nucleosomes were enriched with H3K4me1 , H3K27ac , and H3K18ac , marks previously shown to be enriched at active enhancers ( Heintzman et al . , 2007; Wang et al . , 2008; Creyghton et al . , 2010; Kharchenko et al . , 2011; Rada-Iglesias et al . , 2011 ) and depleted for H3K4me3 and H3K36me3 . As many early developmental genes are located in broad domains of Polycomb-associated H3K27me3 ( Nègre et al . , 2011 ) , many of our putative enhancers are found within large regions containing high levels of this repressive mark . 10 . 7554/eLife . 03737 . 010Figure 6 . Dynamics of histone H3 depletion and histone modifications around blastoderm embryo enhancers . Heatmaps show ChIP-seq signals for histone H3 and different histone modification marks at each stage centered around putative enhancers ( as described in text ) . Enhancers are ordered by chromatin accessibility , as measured by DNaseI–seq signals from cycle 14 embryos ( Thomas et al . , 2011 ) from high ( top ) to low ( bottom ) . On the right , the heatmaps show the ChIP-seq signals for ZLD binding around these enhancers at c8 , c13 , and c14 ( Harrison et al . , 2011 ) . Line plots at the bottom show the average ChIP-seq for histone H3 , histone modifications , and ZLD at each stage around the enhancers . Enh . = enhancer . DOI: http://dx . doi . org/10 . 7554/eLife . 03737 . 010 While the chromatin status of enhancers at cycle 14 has been intensively investigated , their status earlier during the MZT has received much less attention . Our set of likely blastoderm enhancers had , as a class , relatively high nucleosome densities at mitotic cycle 8 ( Figure 6; top left ) . At this early time point , flanking nucleosomes were weakly enriched for the three early appearing histone acetyl marks , especially H3K18ac , with these marks becoming more strongly enriched by cycle 12 . The process of nucleosome depletion was initially evident at cycle 12 , but was much stronger at cycle 14a , when these enhancers begin to be active . The enhancer-associated mark H3K4me1 appeared on flanking nucleosomes by cycle 14a , but the repressive H3K27me3 did not appear in surrounding regions until cycle 14c . This raises the possibility that early events , reflected by the appearance of these enhancer-associated acetylation and methylation marks , play an important role in keeping these regions active once broader domains of inactivity are established . As expected from our previous study ( Harrison et al . , 2011 ) , almost all enhancer sequences described above are also strongly associated with early binding of the transcription factor ZLD ( Figure 6 ) . This suggests that at least at the enhancers , ZLD binding is likely to play a major role in directing the deposition of the histone acetylation marks in early embryos . To investigate this further and to identify other factors that may play a major role in determining overall histone acetylation patterns , not just the enhancers in early embryos , we carried out k-mer enrichment analysis and used the motif search tool MEME ( Bailey and Elkan , 1994 ) to identify sequence motifs associated with different histone mark peaks identified at each stage . We found that the motif most strongly correlated with the early appearing marks , H3K27ac , H3K18ac and H4K8ac , was ZLD's CAGGTAG binding ( Figure 7A ) . A small number of other motifs also showed modest enrichment using these two methods , but they failed to show substantial enrichment when the enrichment is plotted around the histone mark peaks . These analyses thus suggested a close connection between ZLD binding and early histone acetylation in general , which is further highlighted by the extremely high degree of overlap between early ( cycle 8 ) ZLD-bound peaks and early ( cycles 8 or 12 ) peaks for H3K27ac , H3K18ac and H4K8ac ( Figure 7B ) . The relationship is quantitative , with higher levels of ZLD binding coupled to increased levels of the same three marks in cycle 8 and cycle 12 embryos ( Figure 7C ) . The relationship between ZLD binding and these histone marks decays over time ( Figure 7D ) , likely reflecting the increasingly complex transcriptional profile of the genome . However , the strength of this association in early stages of the MZT suggests that ZLD is indeed a dominant factor shaping the early chromatin landscape . 10 . 7554/eLife . 03737 . 011Figure 7 . Relationship between histone occupancy , histone modification pattern and ZLD binding . ( A ) ZLD DNA binding motif enrichment around cycle 8 peaks for H4K8ac , H3K18ac , and H3K27ac . Peaks were ranked based on peak height , divided into bins of 100 , and analyzed . Heatmaps show , for each location ( column ) and set of peaks ( row ) , the average number of ZLD putative sites at each position/set . ( B ) Heatmap showing the overlap between histone acetylation peaks detected at cycle 8 and 12 , a . nd ZLD peaks detected at cycle 8 ( top 2000 ranked peaks ) . As a control , overlaps between histone mark peaks with random set of genomic positions that matched the number of ZLD peaks are shown . ( C ) Scatter plots showing the correlation between the signals around the peaks for the histone acetylation marks at cycle 8 and cycle 12 ( X-axis ) and the heights of the associated ZLD peaks within 1 kb of the histone mark peaks ( Y-axis ) . The signal for each peak was the average over the ±1 kb region surrounding the peak . The correlation coefficient ( r ) for each plot is shown . ( D ) Heatmaps showing Pearson correlation coefficients between ChIP signal of top 5000 ZLD peaks and histone marks at same locations . ChIP-seq signals for histone H3 was averaged over a ±200 bp region around each ZLD peak . Histone marks were averaged over ±1 kb around ZLD peaks . The correlation coefficients were calculated individually for all the ZLD peaks ( ‘All’ ) , for intergenic and intronic ZLD peaks ( ‘INT’ ) , for promoter peaks ( ‘Promoter’ ) , and for ZLD peaks within coding sequences ( CDS ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03737 . 011 To directly analyze the role that ZLD plays in the activation of the zygotic genome during the MZT , we carried out a limited series of ChIP experiments using embryos lacking maternal zld mRNA , obtained from zld- germ-line clones ( Liang et al . , 2008 ) . These females lay significantly less than their wild-type counterparts , and thus obtaining sufficient amounts of staged chromatin was a challenge . ChIP with an anti-ZLD antibody on these nominally zld- embryos at mitotic cycle 12 to mid-14 , showed modest ZLD binding at the same set of sites bound in wild-type embryos ( Figure 8 ) . This residual ZLD activity is likely due to weak zygotic transcription of the paternal copy in female embryos ( zelda is on the X chromosome and thus male offsprings do not receive a functional zelda from their father ) . Thus these ChIP data reflect the depletion , rather than complete elimination , of ZLD . 10 . 7554/eLife . 03737 . 012Figure 8 . Effect of zld mutation on histone occupancy and modifications . Heatmaps show ChIP-seq data from WT embryos ( left ) and embryos lacking maternal zld ( right ) . ( A ) Heatmaps centered at transcription start sites ( TSS ) and ordered by cycle 14 RNA polymerase II binding . ( B ) Heatmaps centered around intergenic ZLD peaks ( Harrison et al . , 2011 ) . Shown are all the intergenic and intronic peaks , among top 1000 ZLD bound regions ( total of 656 ) , ordered by ZLD ChIP signal and aligned by peak position . TSS , transcription start site . DOI: http://dx . doi . org/10 . 7554/eLife . 03737 . 012 Intergenic ZLD-bound regions ( Figure 8B ) had a marked loss of H3K4me1 and a decrease of H3K18ac . The ZLD-associated NFR present in wild-type embryos was almost completely gone . In contrast , we saw a limited effect of ZLD depletion on the promoter histone state globally . There is still a strong NFR , and the marks we observed are present at roughly the same levels . During the first stage of embryonic development , the genome must be reprogrammed from the differentiated states associated with the egg or sperm to create a set of totipotent cells capable of generating a new organism . By combining high-resolution gene expression analysis with precise mapping of nine histone marks throughout this early stage of development , our data suggest that this reprogramming , at least in Drosophila melanogaster , occurs by transitioning through a naïve state in which many histone marks commonly present in somatic cells are absent or at comparably low levels . We further demonstrate that histone acetylation of H3K18 , H3K27 , and H4K8 precedes most histone methylation . Thus we suggest that the establishment of the totipotent chromatin architecture proceeds in an ordered process with acetyl marks being deposited prior to methyl marks . Studies in other organisms have similarly suggested that this early reprogramming is characterized by a transition through a relatively unmodified chromatin state . There is a loss and then reestablishment of DNA methylation following fertilization in mouse embryos ( Santos et al . , 2005 ) . Immunostaining in mouse and bovine embryos has demonstrated that some histone methylation marks are removed following fertilization ( Burton and Torres-Padilla , 2010 ) . Additionally , studies in zebrafish have demonstrated widespread changes in chromatin marks as the embryo progresses through the MZT . While the extent and location of specific histone modifications in zebrafish is not consistent between recent studies ( Vastenhouw et al . , 2010; Lindeman et al . , 2011 ) , a general widespread increase in histone methylation ( H3K4me3 , H3K27me3 , H3K36me3 , and H3K9me3 ) is evident at the MZT . Thus in most , if not all , organisms studied to date there is a dramatic increase in the abundance of histone modifications at the MZT , coinciding with zygotic genome activation . One important lingering question is how this naïve state is established; whether there is a specific system that removes gametic marks associated with sperm and egg at fertilization , or whether the rapid replication cycles of early development are simply incompatible with active and differentiated chromatin . The lack of some histone marks in developing mouse and bovine embryos , which do not undergo rapid cell cycles early in development , suggests that while the mechanisms may vary between species , the removal of parental histone modifications may be a general feature of reprogramming . In the future , it will be important to understand how this transition is regulated to allow for the generation of a totipotent cell population . The presence of several post-translational modifications ( e . g . H3K4me3 and H3K36me3 ) is correlated with transcription and with different states of gene expression in a variety of eukaryotic cells and tissues ( c . f . ( Mikkelsen et al . , 2007 ) ) . By comparing high temporal-resolution transcription data to the quantitative histone modification data , we find that , at least in the early embryo , transcription often occurs in the near absence of these marks . These data confirm and extend the previous observation that H3K4me3 levels are first detectable at the MBT ( Chen et al . , 2013 ) . Interestingly global H3K36me3 levels are also below the level of detection in the mouse embryo as it undergoes the first wave of zygotic genome activation ( Boskovic et al . , 2012 ) . Together these data suggest that marks canonically associated with gene activation at later stages of development are not associated with transcriptional activity in the very early zygote . Moreover , recent data has demonstrated that these marks may not be required for transcription later in development . For example , in the Drosophila wing disc , methylation of H3K4me3 is not required for transcriptional activity ( Hodl and Basler , 2012 ) . In Drosophila embryonic tissue culture cells transcriptionally active euchromatin can be divided into two classes and only one of these is enriched for H3K36me3 ( Filion et al . , 2010 ) . Therefore transcription in the absence of H3K4me3 and H3K36me3 is likely not a distinctive feature of early embryonic development . Additionally , data from Saccharomyces cerevisiae show robust transcriptional activation of a gene localized to heterochromatin in the presence of minimal amounts of H3K36me3 ( Zhang et al . , 2014 ) . By contrast , in zebrafish H3K4me3 was identified at promoters of genes prior to their occupancy by RNA polymerase ( Vastenhouw et al . , 2010; Lindeman et al . , 2011 ) . Because those genes marked by H3K4me3 early are more likely than most genes to be activated at the MZT , it has been proposed that this mark is preparing genes for activation in the early zebrafish embryo . Thus , while histone marks can be associated with specific transcriptional outputs it appears that they are neither necessary nor sufficient for predicting gene expression . While the data presented here are far from complete , together with our previously published high-resolution transcriptional analysis , they suggest a model in which regions of genomic activity in the cellular blastoderm are established by events that transpire earlier in development . In particular , they indicate that the binding of ZLD to target sites across the genome prior to the MZT may trigger a cascade of events—reflected in early histone depletion , the appearance of several histone acetylation marks , and the subsequent appearance of functional class-specific methylation marks—that may act to counter the establishment of Polycomb-mediated repression in many loci ( Figure 9 ) . 10 . 7554/eLife . 03737 . 013Figure 9 . Model for ZLD function during zygotic genome activation . ZLD binds to enhancers in pre-MBT embryos at as early as cycle 8 . This leads to histone acetylation and nucleosome remodeling around ZLD binding sites , which facilitates binding by other transcription factors , and in many other cases leads to additional deposition of histone marks including H3K4me1 while at the same time prevents local deposition of repressive histone mark H3K27me3 and presumably formation of repressive higher order chromatin structure . DOI: http://dx . doi . org/10 . 7554/eLife . 03737 . 013 Indeed it has been shown in D . melanogaster S2 cells that acetylation of H3K27 by Nejire inhibits Polycomb silencing and the establishment of H3K27 trimethylation ( Tie et al . , 2009 ) , and ZLD may play a role in the process by directing H3K27 acetylation to enhancers . One possibility is that ZLD directly recruits histone acetyltransferases , several of which including Nejire and Diskette are maternally deposited , and that these modifications play a direct or indirect role in genome activation . Alternatively , ZLD may simply act as a kind of steric impediment to subsequent chromatin compaction and silencing—with the observed histone acetylation an indirect byproduct of early ZLD binding . We have previously observed that , while ZLD binding is fairly stable across the MZT , some of the regions it binds at cycle 8 are unbound at cycle 14 ( Harrison et al . , 2011 ) . This may reflect the need for other factors to work in conjunction with ZLD while more restrictive chromatin is established . Indeed , while ZLD protein levels remain high through the MZT , the increasing number of nuclei means that absolute ZLD levels are dropping in each nucleus and may reach a point at which ZLD binding alone is insufficient to keep regions active or resist silencing . Work from Ken Zaret and others over the past decades has identified a class of transcription factors , known as ‘pioneer’ factors , that bind early to enhancers during differentiation and thereby promote the binding of other factors to the enhancer . Zaret attributes two characteristics to pioneer factors: 1 ) they bind to DNA prior to activation and prior to the binding of other factors , and 2 ) they bind their target sites in nucleosomes and in condensed chromatin ( Zaret and Carroll , 2011 ) . ZLD clearly has the first characteristic . But it is not clear that is has the second . Our data suggest that there is essentially no condensed chromatin in the early embryo , as nucleosome density is relatively low , nucleosomes are relatively evenly distributed across the genome , and hallmarks of repressed chromatin are absent . This is consistent with the unusually broad binding of ZLD to its target sequences: ZLD binds to more than fifty percent of its target sites in the genome , far more than what is typical for other factors later in development ( Harrison et al . , 2011 ) . We and others have shown that the restricted binding of other factors is largely due to the occlusion of most of their sites by condensed chromatin . Perhaps ZLD binds to a large fraction of its sites because there simply is no condensed chromatin in the early embryo . If so , ZLD would not require , and therefore would likely not possess , the ability to bind its sites in condensed chromatin . Nonetheless , it is clear , if our model is correct , that ZLD is fulfilling the same general role that pioneer factors carry out—getting to the genome first and facilitating the subsequent binding of other factors . While there are no clear ZLD homologs outside of insects , it has recently been shown that in zebrafish the transcription factor Pou5f1 ( Oct4 ) , in combination with Nanog and SoxB1 , drives zygotic genome activation and may share with ZLD a pioneer-like activity ( Lee et al . , 2013; Leichsenring et al . , 2013 ) . Together these data suggest that pioneer transcription factors may generally be required to prepare the embryonic genome for widespread transcriptional activation at the MZT . Interestingly , Pou5f1 is homologous to the canonical pluripotency factor Oct4 , which along with Nanog and Sox2 , are transcription factors expressed to generate induced pluripotent stem cells . Together these data make an explicit connection between the role of pioneer-like factors in zygotic genome activation and the establishment of a totipotent state . Another attractive feature of the model presented above is that it would explain an important and unexplained question about transcriptional enhancers: given that essentially every enhancer sized stretch of the Drosophila genome contains a large number of binding sites for the factors active in the cellular blastoderm ( or any other stage of development ) ( Berman et al . , 2002 ) , why is it that only a small fraction of the genome functions as an enhancer ? It has long been thought that the difference between enhancers and the remainder of the genome is that enhancers do not simply contain binding sites , but rather have these sites in a particular configuration that leads to activation . However , the arrangement of binding sites within Drosophila enhancers is highly flexible ( Ludwig and Kreitman , 1995; Ludwig et al . , 1998 , 2005; Hare et al . , 2008 ) , and we have struggled to find any evidence for strong ‘grammatical’ effects in enhancer organization . The data presented here and elsewhere on ZLD binding and activity support the alternative explanation that the specification of enhancer location and output are distinct processes carried out by specific sets of factors: pioneer factors like ZLD—that determine where an enhancer will be , by influencing the maturation of genomic chromatin , and more classical patterning factors that determine what the transcriptional output of the enhancer will be . The antibodies for histone H3 , and various histone modifications were purchased from commercial sources as listed in Table 1 . The zld294 mutant and the ovoD1 mutant lines used to obtain zld maternal mutant embryos using the FLP-DFS technique ( Chou and Perrimon , 1996 ) have been described previously ( Liang et al . , 2008 ) and were obtained from the laboratory of Christine Rushlow at New York University . D . melanogaster flies were maintained in large population cages in an incubator set at standard conditions ( 25°C ) . Embryos were collected for 30 min , and then allowed to develop for 55 , 85 , 120 or 160 additional minutes before being harvested and fixed with formaldehyde . The fixed embryos were hand sorted in small batches using an inverted microscope to remove embryos younger or older than the targeted age range based on morphology of the embryos as previous described ( Harrison et al . , 2011 ) . After sorting , embryos were stored at −80°C . After all collections were completed , the sorted embryos of each stage were pooled , and a sample of each pool were stained with DAPI . The ages of the embryos and their distribution in the two younger embryo pools ( c7–9 , and c11–13 ) were determined based on nuclei density of the stained embryos . The ages of embryos between c14a and c14c , both which were distinct from c13 based nuclei density , were determined based on morphology . 7 . 5 , 0 . 7 , 0 . 4 , and 0 . 3 g of embryos at four different stages respectively , were used to prepare chromatin for immunoprecipitation following the CsCl2 gradient ultracentrifugation protocol as previously described ( Harrison et al . , 2011 ) . The chromatin obtained was fragmented to sizes ranging from 100 to 300 bp using a Bioruptor ( Diagenode , Inc . , Seraing , Belgium ) for a total of processing time of 140 min ( 15 s on , 45 s off ) , with power setting at ‘H’ . Prior to carrying out chromatin immunoprecipitation , we mixed the chromatin from each sample with a roughly equivalent amount of chromatin isolated from stage 5 ( mitotic cycle 14 ) D . pseudobscura embryos , and used about 2 µg of total chromatin ( 1 µg each of the D . melanogaster and D . pseudobscura chromatin ) for each chromatin immunoprecipitation . The chromatin immunoprecipitation reactions were carried out as described previously ( Harrison et al . , 2011 ) with 0 . 5 μg anti-H4K5ac ( 07-327; Millipore , Billerica , MA ) , 0 . 5 μg of anti-H3K4me3 ( ab8580; Abcam , Cambridge , United Kingdom ) , 0 . 5 μg of anti-H3K27ac ( ab4729; Abcam ) , 1 μg of anti-H3 ( ab1791; Abcam ) , 0 . 75 μg anti-H3K4me1 ( ab8895; Abcam ) , 0 . 75 μg anti-H4K8ac ( ab15823; Abcam ) , 1 . 5 μl of anti-H3K9ac ( 39 , 138; Activemotif ) , 0 . 75 μg anti-H3K18ac ( ab1191; Abcam ) , 3 μg anti-H3K27me3 ( 07-449; Millipore ) , or 0 . 75 μg anti-H3K36me3 ( ab9050; Abcam ) . The sequencing libraries were prepared from the ChIP and Input DNA samples using the Illumina ( San Diego , CA ) TruSeq DNA Sample Preparation kit following the manufacturer's instructions , and DNA was subjected to ultra-high throughput sequencing on a Illumina HiSeq 2000 DNA sequencers . Sequenced reads were mapped jointly to the April 2006 assembly of the D . melanogaster genome [Flybase Release 5] and the November 2004 assembly of the D . pseudoobscura genome [Flybase Release 1 . 0] using Bowtie ( Langmead , 2010 ) with the command-line options ‘-q −5 5 -3 5 -l 70 -n 2 -a -m 1 –best -strata’ , thereby trimming 5 bases from each end of the 100 base single reads , and keeping only tags that mapped uniquely to the genomes with at most two mismatches . Each read was extended to 130 bp based on its orientation to generate the ChIP profiles . We called peaks for each experiment using MACS ( Zhang et al . , 2008 ) v1 . 4 . 2 with the options ‘-- nomodel--shiftsize = 130’ , and used Input as controls . The addition of D . pseudoobscura chromatin prior to the chromatin immunoprecipitation provided us with a means to normalize the ChIP signals for each histone mark and for H3 between different stages . To normalize , we first determined the scaling factor needed to normalize the number of reads for D . pseudoobscura to 10 million , and scaled the signals of D . melanogaster ChIP profile in each sample using this factor . We then multiplied the scaled D . melanogaster signals by the ratio of D . pseudoobscura reads to D . melanogaster reads in the Input sample , which represents the relative amounts of chromatin of the D . melanogaster and the D . pseudoobscura in the starting chromatin samples used for the chromatin immunoprecipitation reactions . Starting with peaks called by MACS as described above , we identified subpeaks by peaksplitter [http://www . ebi . ac . uk/research/bertone/software] , and generated a consolidated list of subpeaks for each histone mark for all stages by joining each group of subpeaks that are within 200 bp into a single peak . We calculated the ChIP signal for each subpeak at each stage by summing the ChIP signal around a 500 bp window center around of each peak position in the normalized ChIP profile generated as described above . To show the overall trend of each histone mark , the range of the ChIP signal among all the subpeaks at each stage is shown as box plot . Using our previous single-embryo RNA-seq data from Lott et al . ( Lott et al . , 2011 ) , genes were classified as zygotic or maternal . We further divided the zygotic genes into four different groups based on their onset of zygotic expression ( first time point with FPKM>1 ) . This includes 107 genes whose onset of expression was around mitotic cycles 10–11 ( ‘Early’ group ) , 99 genes at cycles 12–13 ( ‘Mid’ ) , 143 genes at early cycle 14 ( ‘Late’ ) , and 99 genes during late cycle 14 ( ‘Later’ ) . The maternal group of genes was then compared against post-MBT polII ChIP ( Chen et al . , 2013 ) and split into Maternal/Zygotic genes that show in vivo promoter binding of polII ( ‘Mat/Zyg’ ) and a group of genes that show no polII binding ( ‘Mat–only’ ) . In addition , we used RNA-seq data ( Lott et al . , 2011 ) to define another class of non-expressed genes , showing no transcription from mitotic cycles 10 through the end of cycle14 ( ‘Silent’ ) . A set of putative enhancers was defined based on the in vivo binding locations for early transcription , as measured previously by us using ChIP–chip ( MacArthur et al . , 2009 ) . Here , we summed the raw ChIP–chip signal for 16 factors , including the A/P ( Bicoid , Caudal , Hunchback , Giant , Krüppel , Knirps , Huckebein , Tailless , and Dichaete ) and D/V ( Dorsal , Snail , Twist , Daughterless , Mad , Medea , and Schnurri ) regulators . We then identified all regions with cumulative signal over 20 . This yielded 784 genomic regions , with an average length of 488 bp . These putative enhancers were then classified based on their position with regard to nearby genes , retaining only a set of 588 intergenic and intronic putative enhancers . Two methods were used to investigate the DNA motifs enriched around the peaks identified by MACS . First , 7mers enriched in the 2 kb sequences around the peaks for each experiment were identified by comparing the frequency of each 7mer to the 7mer distribution in randomly selected 2 kb sequences throughout the genome . The selection of the random sequences was restricted to the major chromosome arms excluding the heterochromatic sequences , and the distribution of the number of random sequences were set to match the distribution of peaks among different chromosome arms . The enrichment of the 7mers was ranked based on Z scores . In parallel , the motif enrichment analysis was also carried out using MEME ( Bailey and Elkan , 1994 ) with motif length set at 6–10 and maximum number of motifs to be found at 10 . In this case the sequences located in the 250–650 bp surrounding the maximum of the histone mark peak were used , and random sequences selected using the same criteria as the kmer enrichment analysis were used as negative control . The search was limited to the 150 top ranked peaks for each histone mark . After the candidate enriched motif was identified from these two methods , the motifs were used to map the enrichment around all the peaks by patser ( Hertz and Stormo , 1999 ) using a ln ( p-value ) cutoff of −7 . 5 , and with Alphabet set at ‘a:t 0 . 3 c:g 0 . 2’ . To obtain embryos depleted of maternal zld RNA , the FLP-DFS technique was used . Briefly , zld294 , FRT19A/FM7 ( Liang et al . , 2008 ) virgin females were crossed with ovoD1 , hsFLP112 , FRT19A/Y ( Liang et al . , 2008 ) males . The larvae developed from embryos laid by females from these crosses were heatshocked twice , each for 2 hr at 37°C , when they were between 24–48 hr , and between 48–72 hr old . Collection of the mutant embryos from the resulting female progeny , as well as the aging and fixation of the embryos was carried out following standard protocol as described above except that the collection period is 3 hr followed by 1 hr aging . The embryos were sorted to remove deformed post cycle 14 embryos . As a control , wild-type embryos were collected , treated in parallel and sorted to remove embryos older than stage 5 . The ChIP-seq was carried out with the chromatin from the mutant and wild-type embryos using anti-H3 , anti-H3K18ac , anti-H3K4me1 , and anti-ZLD antibodies as described above . All raw data are available at the GEO database under the accession number GSE58935 . A genome browser with tracks from the data generated and analyzed here is available at the UCSC genome browser .	For a fertilized egg to develop into an embryo , many genes must be switched on and off at specific times . A fertilized egg ( or zygote ) contains genetic material from both parents; and the life of the fruit fly Drosophila melanogaster begins with the nuclei that contain this genetic material repeatedly dividing for the first 2 hr . These nuclear divisions are initially controlled by molecules that the mother deposits into the egg cell . However , as these molecules degrade , the zygote's genome is activated and its own genes take control of embryonic development , in a process referred to as the ‘maternal-to-zygotic transition’ . In the fruit fly zygote , this burst of regulated gene activation is likely to be accompanied by changes to the way that the DNA is packed inside the nuclei . Most DNA in a cell is packaged into a structure called chromatin , which can be marked at specific sites by chemical modifications . For example , chromatin can be acetylated or methylated , which alters its physical structure , helping the underlying genes to be either activated or repressed . In the fruit fly , the first genes to be switched on ( as well as many early developmental genes ) have a DNA motif that is recognized , and is bound by , a protein called Zelda . The Zelda protein plays a major role in activating the genome of the early fruit fly embryo , by marking thousands of genes and regulatory regions for activation . This is somewhat similar to the activity of so-called ‘pioneer’ factors that alter chromatin structure to allow particular genes to be switched on or off , and to trigger the formation and development of specific tissues . Here , Li et al . have investigated whether the Zelda protein—like known pioneer factors—also affects chromatin during the maternal-to-zygotic transition . Different chromatin modifications across the whole fruit fly genome were characterized at specific time-points during the maternal-to-zygotic transition , and the information gathered was then analyzed along with previous data on gene activity . In the early stages of the maternal-to-zygotic transition , Li et al . found very few of the chromatin features that characterize more mature cells . This indicates that the chromatin is in a so-called ‘naïve’ state . As the transition progresses , Li et al . observed that the chromatin becomes acetylated before it is methylated , and that marks associated with activation appear before those associated with repression . Chromatin acetylation was strongly associated with the early binding of the Zelda protein to its target genes . Li et al . 's findings show when , and in what order , the different features of mature chromatin appear in Drosophila zygotes . A future challenge will be to identify whether Zelda directly recruits the proteins that cause chromatin acetylation , or whether it blocks the changes to chromatin that repress gene expression .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "genetics", "and", "genomics" ]	2014	Establishment of regions of genomic activity during the Drosophila maternal to zygotic transition
The prevailing view of motor cortex holds that motor cortical neural activity represents muscle or movement parameters . However , recent studies in non-human primates have shown that neural activity does not simply represent muscle or movement parameters; instead , its temporal structure is well-described by a dynamical system where activity during movement evolves lawfully from an initial pre-movement state . In this study , we analyze neuronal ensemble activity in motor cortex in two clinical trial participants diagnosed with Amyotrophic Lateral Sclerosis ( ALS ) . We find that activity in human motor cortex has similar dynamical structure to that of non-human primates , indicating that human motor cortex contains a similar underlying dynamical system for movement generation . Clinical trial registration: NCT00912041 . Neurons in motor cortex exhibit complex firing patterns during movement ( Fetz , 1992; Churchland and Shenoy , 2007; Churchland et al . , 2010 ) . Though motor cortex has been extensively studied over the past century ( Lemon , 2008 ) , the complexity of these patterns remains poorly understood . Early work demonstrated that neural firing correlates with external variables such as movement direction ( Georgopoulos et al . , 1982 ) , and an ongoing debate centers on whether the firing patterns represent muscle or movement parameters ( e . g . , joint position , intended velocity , reach endpoint , muscle forces , and so on; reviewed in Kalaska , 2009 ) . A recent approach has been to ask , instead , whether these patterns of activity reflect an internal dynamical system across the population ( Fetz , 1992; Todorov and Jordan , 2002; Churchland and Shenoy , 2007; Scott , 2008; Churchland et al . , 2010 , 2012; Graziano , 2011; Shenoy et al . , 2013; Kaufman et al . , 2014 ) . In this view , the system's initial state is set by preparatory activity , and consistent rules govern how firing rates evolve over time across movement conditions , regardless of movement direction or other externally changing variables . This dynamical view is supported by recent studies in rhesus macaques ( Churchland et al . , 2010 , 2012; Shenoy et al . , 2013; Kaufman et al . , 2014 ) . At the level of single neurons , responses during movement show brief but strong oscillatory components , even for straight point-to-point reaches ( Churchland and Shenoy , 2007; Churchland et al . , 2010 ) . Furthermore , at the population level , the responses are well-described by a simple dynamical model in which the population response ( the neural state ) rotates with time ( Figure 1A ) ( Churchland et al . , 2012 ) . Importantly , the presence of a strong rotational component in these dynamics is not predicted by the prevailing view of motor cortical activity ( i . e . , cosine direction tuning , linear speed scaling ) that , instead , predicts solely expansive and contractive dynamics ( detailed in Churchland et al . , 2012 , Figure 4 and associated text ) . 10 . 7554/eLife . 07436 . 003Figure 1 . Neural population responses show rotational activity during movement epochs . ( A ) Projections of the neural population response onto the first jPCA plane for a monkey during an arm-reaching task ( monkey N , 108 conditions; adapted from Churchland et al . , 2012 ) . Each trace plots the first 200 ms of activity during the movement epoch for a given condition . Traces are colored based on the preparatory state projection onto jPC1 . a . u . , arbitrary units . ( B ) Projections for participant T6 during an 8-target center-out task controlled by index finger movements on a computer touchpad . Each trace plots the 250 ms of activity during the movement epoch ( ‘Materials and methods’ ) for a given condition . ( C ) Same as ( B ) , for participant T7 . Video 1 shows the evolution of the neural state over time for each participant . DOI: http://dx . doi . org/10 . 7554/eLife . 07436 . 003 In humans , studies in research participants with tetraplegia have demonstrated that motor cortical action potential ( AP ) responses contain information about intended movement kinematics ( Truccolo et al . , 2008 ) . Here , we investigated whether these AP responses also contain rotational dynamical structure at the ensemble level . We analyzed multi-neuron AP activity from 2 people with tetraplegia enrolled in the BrainGate2 pilot clinical trial . This ongoing , multi-site , pilot clinical trial is performed under an FDA Investigational Device Exemption and has been approved by local institutional review boards at all study sites . The study participants ( T6 , T7 ) had differing degrees of motor impairment due to Amyotrophic Lateral Sclerosis ( ALS ) . T6 retained the ability to make several dexterous movements ( especially of the fingers and wrist ) , while T7 retained limited finger movements . Neural signals were recorded using 4 mm × 4 mm , 96-channel silicon microelectrode arrays , which were implanted in the hand area of dominant M1 . APs were recorded as participants performed visual target acquisition tasks . Participants attempted to move a cursor from the center of a computer screen to one of eight peripheral targets , with the cursor's position controlled by index finger movements on a computer touchpad ( see ‘Materials and methods’ ) . We first tested whether an underlying dynamical structure existed in the neural activity . If present , a population-level analysis should reveal orderly rotational structure that is consistent across conditions ( i . e . , independent of the direction of movement ) . To search for rotational patterns in the neural population state , we used a three-step procedure: first , for each condition ( i . e . , for a given target ) , we averaged the activity on each electrode across all trials . Next , we performed principal components analysis ( PCA ) on the high-dimensional population data . We restricted the data to the top 6 PCs , that is , we only preserved the six response patterns most strongly present in the data . Finally , for this reduced-dimensional data set , we applied the jPCA method ( Churchland et al . , 2012 ) , which searches the data for 2-dimensional planes that capture the strongest rotational tendencies . Restricting the jPCA analysis to the dimensionality-reduced data ( 6-D ) ensured that any rotational structure revealed by the analysis was present in the most prominent response patterns in the data . For both participants , the population activity exhibited strong rotational dynamics ( Figure 1B , C ) . Each trace shows the population activity in the top jPC plane for a single condition . 250 ms of data are shown , beginning with the rapid change in neural activity that precedes movement onset ( the evolution of the neural state over time for each participant is shown in Video 1 ) . Rotations proceeded in the same direction across conditions , following from the initial pre-movement state . The top jPCA plane captured 61% ( T6 ) and 27% ( T7 ) of the variance of the high-dimensional neural data ( for comparison , the macaque study reported 28% for the top plane on average ) . 10 . 7554/eLife . 07436 . 004Video 1 . Neural population responses show rotational activity . Video shows the evolution of the neural state over time in the first jPCA plane for participants T6 and T7 . Low-dimensional projections were calculated as in Figure 1 . Each colored trace represents one of 8 conditions . All times are relative to target onset ( 0 ms ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07436 . 004 One potential concern is that the jPCA method might be powerful enough to find rotatory patterns in state space for any set of responses that contains complex , multiphasic patterns . To test for this possibility , we performed three control analyses , following Churchland et al . , 2012 . In these controls , the data were shuffled to disrupt underlying rotational structure across response patterns , while preserving the complexity of the individual response patterns . If the previously found rotational structure were simply a by-product of the analysis technique , then the shuffled data sets should still show prominent rotations in the top jPCA planes . This was not the case . Rotations were no longer qualitatively seen in the projected responses after shuffling ( Figure 2 , top row ) . We next measured the fraction of variance of the changes in neural state ( 6-D ) that could be explained by rotational activity alone ( see ‘Materials and methods’ ) and found that this greatly decreased after shuffling ( Figure 2 , bottom row ) . ( Two additional shuffle control analyses are presented in Figure 2—figure supplement 1 and Figure 2—figure supplement 2 ) . 10 . 7554/eLife . 07436 . 005Figure 2 . Rotational dynamics are not a by-product of the jPCA analysis method . For each data set , neural responses were shuffled in a manner that preserved the complexity of individual response patterns on each electrode , but disrupted the structure of the data across electrodes . For each channel , the pattern of activity during the movement epoch was inverted for half the conditions ( chosen at random ) . The inversion was performed around the initial time point , so that continuity with pre-movement activity was preserved . Performing jPCA on the shuffled responses did not reveal consistent rotational structure . ( A , top ) Projection of the population responses onto the first jPCA plane for a single shuffled trial ( participant T6 ) . ( bottom ) Fraction of variance of the change in neural state ( 6-D ) explained by rotational activity for the original data set ( brown ) vs the shuffled data sets ( blue ) . Error bar represents the standard deviation across 300 shuffle trials . ( B ) Same as ( A ) , for participant T7 . Two additional shuffle control analyses are presented in Figure 2—figure supplement 1 and Figure 2—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 07436 . 00510 . 7554/eLife . 07436 . 006Figure 2—figure supplement 1 . Results of the second shuffle control analysis . This shuffle control was similar to the first , but the patterns of activity during the movement epoch were inverted for all conditions . ( A ) Projections of the neural population state for each data set after shuffling are shown . ( B ) The fraction of variance captured by the rotational model before and after shuffling . This manipulation would not be expected to remove all rotational structure , as the structure should merely be sign-inverted . However , this manipulation should remove any consistent relationship between the pre-movement activity and the phase of subsequent oscillations . Thus , this control is expected to disrupt the relationship between the rotational phase and its initial state . DOI: http://dx . doi . org/10 . 7554/eLife . 07436 . 00610 . 7554/eLife . 07436 . 007Figure 2—figure supplement 2 . Results of the third shuffle control analysis . In this shuffle control , activity during the movement epoch from one condition was randomly reassigned onto the pre-movement activity from another . The firing rate at the beginning of the movement epoch was appended to the firing rate at the end of the pre-movement period , to preserve continuity . For a given shuffle trial , the same reassignment was performed for all neurons . ( A ) Projections of the neural population state for each data set after a single shuffle trial . ( B ) The fraction of variance captured by the rotational model for the original data sets and 300 shuffle trials is shown ( each shuffle trial tests a different random set of pairings of pre-movement and movement epoch activity ) . Error bars represent the standard deviation across shuffle trials . As with the second shuffle control , this manipulation is not expected to remove all rotational structure . While the relationship between rotational phase and the initial state is disrupted ( as each condition's movement epoch activity is assigned to another condition's pre-movement activity ) , there is some likelihood that a given condition will be assigned an initial state similar to its own , allowing some rotational structure to be preserved by this manipulation . DOI: http://dx . doi . org/10 . 7554/eLife . 07436 . 007 To quantify the consistency of the rotatory activity , we measured the angle from the neural state in the jPCA plane , x , to its derivative , x˙ , for each time point across all conditions ( Figure 3 ) . Angles near pi/2 indicate rotational dynamics . As shown , the distribution of measured angles peaked near pi/2 , similar to previously reported macaque data . 10 . 7554/eLife . 07436 . 008Figure 3 . Consistency of the rotational dynamics across conditions . Traces represent histograms of the angle , q , between the neural state , x , and its derivative , x˙ , for each time step . The angle was measured as illustrated schematically ( inset ) after projecting the data into the first jPCA plane . Purely rotatory activity results in angles near pi/2 , while pure scaling/expansion results in angles near 0 or pi . Y-axis denotes scale for participant data ( colored traces ) . For comparison , histograms for the example shuffle control data ( Figure 2 ) and monkey composite data are shown in gray and black , respectively . These traces are normalized to match the participant data range ( monkey data reproduced from Churchland et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07436 . 008 These results demonstrate prominent rotations of the neural population state in human motor cortex during movements . As with macaques , rotations were consistent across conditions and followed naturally from an initial pre-movement state . As mentioned above , and in contrast to the macaque study , both participants in this study had a diagnosis of ALS with resultant motor impairment . Participant T7's movements in particular were very limited and occurred with a long latency following target onset ( detailed in ‘Materials and methods’ ) . Both participants likely had substantial changes in motor cortex due to their disease progression; therefore , these results do not conclusively show that dynamical activity is present in healthy human motor cortex . However , given the finding that dynamical activity is prominent in motor cortex in people with abnormal motor function , and given the similarity of this activity to that of healthy non-human primates , the results strongly suggest the presence of dynamical activity in human motor cortex in general and may also hint at which aspects of motor cortical function are preserved despite the progression of a severe motor neuron disease . Given the findings of dynamical activity during overt movements , a potentially exciting open question is whether motor cortex exhibits dynamical activity during purely imagined movements . Recent studies ( Feldman et al . , Society for Neuroscience , 2011 , 2012; Pandarinath et al . , Society for Neuroscience , 2013 ) have demonstrated that motor cortex is active during both overt and imagined movements . If the functional role of rotational dynamics is to serve as an oscillatory basis set for generating muscle activation patterns ( e . g . electromyogram activity ) ( Churchland et al . , 2012; Sussillo et al . , 2015 ) , then one might expect these dynamics to be absent during imagined movements , when the subject is specifically trying to avoid generating motor output . However , if activity during imagined movements serves as a mechanism for the ‘covert rehearsal’ of activity during overt movements , then dynamics might still be present ( but might be , for example , orthogonal to patterns of activity that drive motor output ) . Future studies will attempt to directly address this question . The current study’s findings also suggest a promising avenue to improve the performance of Brain–Machine Interfaces ( BMIs ) for persons with tetraplegia , as previous work with macaques ( Kao et al . , 2015 ) demonstrated that incorporating neural population dynamics into BMI control algorithms may lead to performance improvements . Finally , as with macaques , the presence of these rotations calls into question the prevailing model of motor cortical activity ( i . e . , that motor cortical firing patterns represent muscle or movement parameters ) in favor of a dynamical systems perspective in humans as well . Participant T6 is a right-handed woman , 51 year old at the time of this study , with tetraplegia due to ALS . On December 7 , 2012 , a 96-channel intracortical silicon microelectrode array ( 1 . 0-mm electrode length , Blackrock Microsystems , Salt Lake City , UT ) was implanted in the hand area of dominant motor cortex as previously described ( Hochberg et al . , 2006; Simeral et al . , 2011 ) . T6 retained dexterous movements of the fingers and wrist . Data reported in this study are from T6's trial days 95–213 . Participant T7 is a right-handed man , 54 year old at the time of this study , with tetraplegia due to ALS . T7 had two 96-channel intracortical silicon microelectrode arrays ( 1 . 5-mm electrode length , Blackrock Microsystems , Salt Lake City , UT ) implanted in the hand area of dominant motor cortex on July 30 , 2013 . Data reported are from T7's trial days 231 and 245 . At that time , T7 retained very limited , but consistent , index finger movements . Subsequent to these days , further motor impairment precluded tasks that relied on finger movements . Neural data were recorded during ‘center-out’ target acquisition tasks . The data were originally collected for neural prosthetic decoder calibration , as part of research testing algorithms for closed-loop neural cursor control ( Gilja et al . , Society for Neuroscience , 2013 ) . In the ‘center-out’ task , participants controlled the position of a cursor on a computer screen by making physical movements with their fingers on a wireless touchpad ( Magic Trackpad; Apple , Cupertino , CA ) . The cursor began in the center of the screen , and targets would appear in one of 8 locations on the periphery . Participants then acquired the targets by moving the cursor over the target and holding it over the target for 500 ms . In contrast to the prior macaque study ( Churchland et al . , 2012 ) , the target acquisition task used in this work did not include a delay period . Therefore , participants were free to move as soon as the target appeared . Participant T6 was not limited in her ability to span the workspace of the touchpad . Participant T7's limited movements spanned a small region on the touchpad , approximately 1/8″–1/4″ wide . Data were aggregated over multiple sessions ( T6: 8 sessions , T7: 2 sessions ) . Trial counts varied between sessions and across participants ( T6: 75–220 trials per session , T7: 128 and 78 trials per session ) . The primary data analyzed were multi-neuron APs , which were taken as time points when a given channel's voltage exceeded a fixed threshold . Choice of threshold was dependent on the array ( T6: −60 μV , T7 , Lateral array: −80 μV , Medial array: −95 μV ) . Analyses were restricted to electrodes known to have significant modulation during attempted movements ( T6: 39 electrodes , T7: 78 electrodes ) . Firing rates per electrode were then averaged across trials and filtered with a Gaussian kernel with standard deviation of 25 ms ( T6 ) or 30 ms ( T7 ) . jPCA analyses were performed as described previously ( Churchland et al . , 2012 ) . Analyses were restricted to a 250-ms time period beginning with the rapid changes in neural activity that occur preceding movement onset . This period began approximately 180 ms and 300 ms after target onset for T6 and T7 , respectively , corresponding to a difference in reaction times between participants . Overt movement was detectable at approximately 230 ms and 600 ms after target onset for T6 and T7 , respectively . The delayed movements in T7 relative to T6 likely reflect a difference in disease progression between the two participants . Pre-processing steps ( ‘soft’ normalization , mean-centering , and initial dimensionality reduction using PCA ) were performed following Churchland et al . , 2012 . The initial dimensionality reduction step restricted the data to the top 6 PCs . jPCA is a method for finding projections that capture rotational structure in a data set . ( The method is described in detail in Churchland et al . , 2012; here , we summarize its key features . ) The method is based on comparing the neural state at a given point in time with its derivative . The initial dimensionality-reduction step ( performed on the trial-averaged firing rates ) reduces the data to the matrix Xred , which has dimensions d × ct ( where d is the number of PCs kept , c is the number of conditions , and t is the number of time points ) . We computed X˙red , of size d × c ( t−1 ) , by taking the difference in state between adjacent time points ( the final time point of Xred was subsequently removed to equalize the sizes of Xred and X˙red ) . We then fit the neural state transition matrices , M and Mskew:X˙red=MXred , andX˙red=MskewXred , where M is unconstrained ( fit via linear regression ) , and Mskew is constrained to be a skew-symmetric matrix ( i . e . , Mskew=−MskewT , which has purely imaginary eigenvalues ) , and thus , captures only rotational dynamics . The first jPCA plane is then constructed from the eigenvectors of Mskew associated with the largest eigenvalues . For a given jPCA plane , the basis vectors jPC1 and jPC2 are selected such that the pre-movement activity ( the initial state ) is maximally spread along jPC1 , and that the net rotation in the plane is anticlockwise . To measure the fraction of variance of the changes in neural state ( i . e . , the dimensionality-reduced data ) that could be explained by rotational activity ( Figure 3 , bottom panels ) , the data were modeled as a linear dynamical system , where Mskew perfectly captured the dynamics between time t and t−1 , that is , only purely rotational dynamics were allowed . Fraction of state variance explained for real and shuffled data was estimated using the top 6 PCs . One potential concern is that the jPCA method might be powerful enough to find rotatory patterns in state-space for any set of responses that contains complex , multiphasic patterns . The likelihood of finding such spurious rotatory patterns , which are common across conditions , increases as the number of conditions decreases . Thus , this is a larger concern in the current work ( 8 conditions per participant ) than previous work with macaques ( 27–108 conditions per monkey ) ( Churchland et al . , 2012 ) . To test for this possibility , we performed the three control analyses performed in Churchland et al . , 2012 ( described in their Supplementary Figures 2 and 3 ) . Each of the three ‘shuffle’ controls preserves the diversity and complexity of responses , but perturbs the structure of the responses at the population level . Specifically , the dynamical model assumes that neural activity during the movement epoch follows in an orderly fashion from its pre-movement state . To test this assumption , the three shuffle control analyses disrupt the relationship between the pre-movement activity and the movement epoch for each channel . If these shuffled data sets still showed prominent rotatory activity , it would indicate that rotations might be found by the jPCA method even when not truly present .	Every conscious movement a person makes , whether lifting a pencil or playing a violin , begins in the brain . To be more specific , neurons in a part of the brain called the motor cortex send signals to muscles to cause them to move . But many of the details about how messages from the motor cortex produce movements remain unclear . Some scientists believe that individual neurons in motor cortex send direct messages that tell the muscles which direction to move in , how fast , and how forcefully . But other scientists suggest that this is not the case . Instead , they propose that neurons in motor cortex work together as part of a dynamic system to create rhythmic patterns of activity for movement . These rhythmic patterns then sum together to create the signals that muscles need to carry out the movements . Studies in monkeys have supported the idea that the neurons in the motor cortex are not just direct messengers . These studies showed what appears to be a rotating set of patterns in neuronal activity in the motor cortex during movement . Now , Pandarinath et al . have shown that a similar rotation of neuronal activity patterns occurs during movement in two human volunteers . The participants both had a disease called amyotrophic lateral sclerosis ( or ALS for short ) . This disease had nearly paralyzed their arms and legs because it causes a progressive loss of muscle control . In the experiments , the volunteers used their index fingers to try to move a computer cursor to a target using a touch pad . Pandarinath et al . recorded activity in the volunteers' motor cortices while they completed the tasks . The experiments uncovered a predictable series of patterns that started when the individual first thought about moving . These patterns progressed in a rotation as the movement was carried out . The rotation was not tied to the direction of the movements and would be completely unexpected if the individual neurons were simply acting as direct messengers . It is hoped that these findings will help efforts to create prosthetic devices ( such as robotic arms ) that can better respond to an individual's thoughts . But further experiments are also needed in people without ALS to verify that the patterns observed weren't specifically related to this disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "medicine", "neuroscience" ]	2015	Neural population dynamics in human motor cortex during movements in people with ALS
Transposable elements ( TEs ) allow rewiring of regulatory networks , and the recent amplification of the ISX element dispersed 77 functional but suboptimal binding sites for the dosage compensation complex to a newly formed X chromosome in Drosophila . Here we identify two linked refining mutations within ISX that interact epistatically to increase binding affinity to the dosage compensation complex . Selection has increased the frequency of this derived haplotype in the population , which is fixed at 30% of ISX insertions and polymorphic among another 41% . Sharing of this haplotype indicates that high levels of gene conversion among ISX elements allow them to ‘crowd-source’ refining mutations , and a refining mutation that occurs at any single ISX element can spread in two dimensions: horizontally across insertion sites by non-allelic gene conversion , and vertically through the population by natural selection . These results describe a novel route by which fully functional regulatory elements can arise rapidly from TEs and implicate non-allelic gene conversion as having an important role in accelerating the evolutionary fine-tuning of regulatory networks . A substantial portion of animal genomes is composed of repetitive sequences , including gene duplicates , satellite DNA , and transposable elements . Gene conversion is a major force shaping the evolution of repetitive regions , and interlocus or non-allelic gene conversion between sequence duplicates has been studied extensively for its role in concerted evolution ( Chen et al . , 2007; Ohta , 2010 ) . Non-allelic gene conversion also affects selection operating in gene families . Compared to single-copy genes , a family of gene duplicates presents a larger mutational target , and a mutation arising in any gene copy can be spread among copies by non-allelic gene conversion , thereby increasing the efficiency of both positive and purifying selection ( Mano and Innan , 2008 ) . Non-allelic gene conversion homogenizes the arrays of ribosomal DNA gene copies present in the genomes of most organisms ( Eickbush and Eickbush , 2007 ) , has generated allelic diversity within the human leukocyte antigen gene family ( Zangenberg et al . , 1995 ) , and has allowed palindromic genes on the human Y chromosome to escape degeneration ( Rozen et al . , 2003 ) . Transposable elements give rise to families of duplicate sequences . A propensity for some TEs to carry regulatory motifs and to insert adjacent to coding sequence gives them the potential for being potent modulators of gene regulatory networks ( Feschotte , 2008; Cowley and Oakey , 2013 ) . The regulatory elements provided by these TEs , however , may be suboptimal in function , and subject to subsequent fine-tuning ( Polavarapu et al . , 2008 ) . Unlike regulatory elements where short binding motifs ( 10 basepairs on average for transcription factors; Stewart et al . , 2012 ) evolve de novo via point mutation or microsatellite expansion , binding sites that evolve from TEs are initially almost identical in sequence and are nested within a larger repeat unit ( hundreds or thousands of basepairs in size ) , and may thus be subject to non-allelic gene conversion . Re-wiring of the dosage compensation network in Drosophila miranda was driven by TE-mediated amplification of a functional but suboptimal binding motif ( Ellison and Bachtrog , 2013 ) . Here we show that non-allelic gene conversion is catalyzing the rapid fine-tuning of these suboptimal motifs by allowing sequence variants that optimize binding affinity to spread among elements . Dosage compensation in Drosophila is mediated by a male-specific ribonucleoprotein complex ( the male-specific lethal or MSL complex ) that binds to a GA-rich sequence motif ( the MSL recognition motif ) at a number of chromatin entry sites on the X chromosome ( Alekseyenko et al . , 2008; Straub et al . , 2008 ) . We previously studied the acquisition of novel chromatin entry sites on newly formed X chromosomes in D . miranda , a species where two independent sex chromosome/autosome fusions resulted in a karyotype composed of three X chromosome arms , each of a different age ( Alekseyenko et al . , 2013; Zhou et al . , 2013 ) . XL is homologous to the X chromosome of Drosophila melanogaster and has been a sex chromosome for at least 60 million years ( Richards et al . , 2005 ) ; chromosome XR formed roughly 15 million years ago when an autosome ( Muller element D ) fused to XL ( Carvalho and Clark , 2005 ) , and the neo-X/neo-Y chromosome pair originated around 1 . 5 million years ago when the Y fused to another autosome ( Muller element C ) ( Bachtrog and Charlesworth , 2002 ) . Dosage compensation evolved on both XR and the neo-X shortly after their emergence , through acquisition of novel chromatin entry sites and co-option of the MSL regulatory network ( Bone and Kuroda , 1996; Marin et al . , 1996 ) . Interestingly , we discovered that the acquisition of dosage compensation on both XR and the neo-X chromosome was in part mediated by the independent domestication of helitron transposable elements that contained MSL recognition motifs , which we have termed ISXR and ISX , respectively ( Ellison and Bachtrog , 2013 ) . ISX is highly enriched on the neo-X chromosome of D . miranda and is derived from the abundant ISY element . Compared to ISY , ISX contains a 10 basepair deletion that creates a MSL recognition motif , thereby allowing it to act as a chromatin entry site ( Ellison and Bachtrog , 2013 ) . Our previous study showed that while amplification of ISX about 1 million years ago provided dozens of functional chromatin entry sites on the neo-X chromosome of D . miranda , the motif dispersed by ISX is distinct from the canonical motif that is enriched within chromatin entry sites on XL and XR , and shows significantly lower affinity to the MSL complex compared to motifs on XL and XR ( Ellison and Bachtrog , 2013 ) . For these reasons , we postulated that the ISX binding motif is suboptimal , and predicted that refining mutations should accumulate within each MSL recognition motif until the neo-X chromosome becomes fully dosage compensated ( Ellison and Bachtrog , 2013 ) . To identify potential refining mutations that optimize MSL-binding at chromatin entry sites derived from the ISX element , we characterized sequence variation within the MSL recognition motifs and flanking sequence regions for all 77 insertions of the ISX element on the neo-X chromosome in the sequenced reference strain MSH22 ( Figure 1A ) . Because we have previously demonstrated that ISX contains a functional MSL recognition motif but the closely related ISY element does not ( Ellison and Bachtrog , 2013 ) , we sought to identify sequence variants that were present in multiple ISX elements but rare or absent in ISY elements from the same chromosome . 10 . 7554/eLife . 05899 . 003Figure 1 . TE-derived MSL recognition element ( MRE ) motifs from the neo-X chromosome of Drosophila miranda . ( A ) The MSL recognition motif ( MRE ) plus 20 basepairs of flanking sequence were extracted from all 77 ISX transposable elements located on the neo-X chromosome in the MSH22 reference genome assembly . The multiple sequence alignment of these 77 sequence regions ( arranged from top-to-bottom in the order in which they are found on the chromosome ) shows that there is sequence variation among elements both within and adjacent to the 21 basepair MRE motif . Each variant has been classified as ancestral or derived based on its frequency in the ISX progenitor element , ISY . The derived allele frequency for each variant in this region is shown for ISX as well as 139 ISY elements from the neo-X chromosome ( see Figure 1—figure supplement 1 for ISY alignment ) . Red arrows point to the derived TT haplotype that is common among ISX elements but rare in ISY . ( B ) Barplot showing the frequencies of all haplotypes at the GA/TT sites , for ISY and ISX elements separately . Two haplotypes are present within ISX elements ( GA and TT ) and the two alleles within each haplotype are in perfect linkage disequilibrium . In contrast , the majority of ISY elements harbor the GA haplotype , but these two alleles are not in perfect linkage disequilibrium among ISY elements . Rather , five additional allelic combinations are present at low frequencies in this location among ISY , but not ISX elements . DOI: http://dx . doi . org/10 . 7554/eLife . 05899 . 00310 . 7554/eLife . 05899 . 004Figure 1—figure supplement 1 . Alignment of ISY elements from the D . miranda MSH22 genome assembly . 139 ISY elements from the MSH22 neo-X chromosome were identified and 200 basepairs from their 5′ flanks were aligned . The black arrows point to the sites where the derived ‘T’ alleles are common among ISX elements . In contrast , only a single ISY element from the neo-X chromosome harbors the TT haplotype . DOI: http://dx . doi . org/10 . 7554/eLife . 05899 . 004 Using these criteria , we identified a sequence haplotype adjacent to the MSL recognition motif that is common among MSH22 ISX insertions and rare among ISY elements: 57% of ISX elements carry this haplotype vs 0 . 7% of neo-X ISY insertions , an asymmetry significantly different from that expected by chance ( Fisher's Exact Test; p < 2 . 2e-16 ) . The haplotype consists of two mutations ( G → T and A → T ) , separated by two basepairs , which are in perfect linkage disequilibrium among ISX but not ISY elements ( Figure 1 and Figure 1—figure supplement 1 ) . Because ISX is descended from ISY and the TT alleles are rare among ISY elements , they are likely to be derived . We hereafter refer to these mutations as the TT haplotype . To determine if the TT haplotype affects binding affinity of the MSL complex , we used published ChIP-seq data of MSL3 ( a component of the MSL complex ) from D . miranda strain MSH22 ( Alekseyenko et al . , 2013 ) . We compared in vivo MSL complex binding levels for the 44 MSH22 ISX insertions carrying the TT haplotype to the 33 insertions with the ancestral GA haplotype . The insertions with the TT alleles had significantly higher levels of MSL complex binding compared to those with the GA alleles ( Wilcoxon test p = 0 . 01; Figure 2A ) . 10 . 7554/eLife . 05899 . 005Figure 2 . The TT haplotype increases MSL binding affinity . ( A ) MSL3 ChIP-seq data from D . miranda strain MSH22 shows that the ISX insertions carrying the TT haplotype recruit significantly higher levels of MSL complex compared to those with the GA haplotype ( Wilcoxon test p = 0 . 01 ) . ( B ) Engineered ISX elements that differ only with respect to the TT haplotype bind different levels of MSL complex . There is an epistatic interaction between the two ‘T’ alleles such that separately , they decrease MSL complex binding relative to the ancestral allele , but together in the TT haplotype , they increase MSL complex binding ( Wilcoxon Test p = 0 . 028 for both comparisons [GT vs TT and TA vs TT] ) . The rectangles and error bars show the average and standard deviation of values from four biological replicates for each condition . DOI: http://dx . doi . org/10 . 7554/eLife . 05899 . 005 We previously demonstrated that insertion of an ISX element in the D . melanogaster genome results in recruitment of the MSL complex to an ectopic autosomal location ( Ellison and Bachtrog , 2013 ) . We used this same system to dissect the relationship between the TT alleles and MSL complex binding affinity . Starting with a cloned ISX element ( Ellison and Bachtrog , 2013 ) , we used site-directed mutagenesis to create variants of this element that differ only with respect to the TT haplotype . Each of the four possible haplotypes ( GA , GT , TA , and TT ) was engineered and inserted onto D . melanogaster chromosome 2L at cytosite 38F1 using recombinase mediated cassette exchange ( RMCE ) ( Bateman et al . , 2006 ) . We then measured the effect of each of the derived variants by quantifying allele-specific binding levels of the MSL complex in F1 hybrids between the ancestral haplotype ( GA ) and each of the derived haplotypes ( GT , TA , and TT ) . Interestingly , each T allele , when assayed separately , has a negative effect on MSL binding levels compared to the ancestral G or A allele ( Figure 2B ) . However , when combined , the TT haplotype results in significantly increased levels of MSL complex binding , relative to the ancestral GA haplotype ( Wilcoxon Test p = 0 . 0289; Figure 2B ) . These results suggest that there is sign epistasis between the two alleles and that the high frequency TT haplotype represents a refining/fine-tuning adaptation , since recruitment of MSL complex to the adjacent MSL recognition motif is increased . It is unlikely that the TT haplotype arose multiple times by parallel mutation , and there are two possibilities that could explain its prevalence among MSH22 ISX insertions . First , this double mutation may have occurred early during the process of ISX amplification , thus giving rise to two lineages of ISX: one that carries the ancestral GA haplotype , and the other with the TT haplotype . The TT-harboring elements in MSH22 would then all be descendants from the latter ISX lineage . Alternatively , this mutation may have occurred only after the GA-containing ISX element was fixed in the population at all 77 neo-X insertion sites , at which point it was spread among independent ISX elements via non-allelic gene conversion . We can distinguish between these possibilities by examining patterns of sequence polymorphism for each ISX insertion across multiple strains of D . miranda . A canonical signature of non-allelic gene conversion is the presence of shared polymorphisms across sequence duplicates ( Arguello et al . , 2006; Mansai and Innan , 2010 ) . If gene conversion is spreading the TT haplotype among ISX insertions , we expect it to be polymorphic among individuals at several ISX insertion sites , whereas we do not expect the TT haplotype to be polymorphic at individual ISX insertions under the alternative scenario . To genotype multiple wild-derived individuals at each of the MSH22 ISX insertions , we used paired-end Illumina genomic resequencing data from 23 inbred lines of D . miranda , including MSH22 . We aligned all reads to the MSH22 reference genome and identified mate-pairs where one mate was anchored in unique sequence flanking an ISX insertion . We then assembled these reads to generate a contig spanning the 5′ flank of the ISX element insertion , which contains the MSL recognition motif , for each inbred line . Using this approach we generated population data for 66 insertions out of the 77 total ISX insertions present in the MSH22 reference genome assembly . Uneven sequence coverage between insertions and individuals meant that not all insertions could be assembled for each individual . However , the majority of individuals are represented in the majority of datasets: each insertion dataset contained ∼20 lines on average ( see Dataset S1 in Dryad: Ellison and Bachtrog , 2015 ) . Almost all ISX insertions are fixed among strains ( 65 of 66 ) and insertion sites are identical between lines , suggesting that independent parallel insertions are unlikely to be present within our dataset . We performed PCR and Sanger sequencing on a subset of these regions and estimate the base-calling error rate of our Illumina contigs to be ∼0 . 1% . Consistent with non-allelic gene conversion spreading the TT haplotype , we observe a strong signal of allele-sharing within the sequence region flanking the MSL recognition motif among ISX insertions ( Figure 3 ) . On average , 68 . 9% of polymorphisms observed within a given insertion are shared among other insertions ( though most polymorphisms are shared only between a few elements ) . The TT haplotype is especially striking in this regard as it is polymorphic in 41% of insertions ( Figure 3 , Figure 4 and Figure 3—figure supplement 1 ) . If population subdivision contributes to this excess of allele sharing , we would expect individuals to cluster by allele state at the TT locus , across all polymorphic ISX insertions . Instead , we find that different individuals contribute to the TT polymorphism at each of these ISX insertions ( Figure 4—figure supplement 1 ) , suggesting that abundant non-allelic gene conversion is the most likely explanation for this observation . Interestingly , the population frequency of the TT haplotype is similar among insertions that are near each other on the chromosome ( permutation test p = 0 . 018; Figure 4 ) . This is consistent with higher gene conversion rates between more closely linked ISX elements generating correlated population frequencies among adjacent elements ( Sasaki et al . , 2010 ) . 10 . 7554/eLife . 05899 . 006Figure 3 . ISX variation among wild lines of D . miranda . For each ISX insertion identified within the D . miranda MSH22 reference genome assembly ( alignment shown at left , see also Figure 1 ) , we characterized sequence variation across D . miranda individuals . The TT haplotype ( magenta lines ) was fixed across individuals at 30% of insertions ( see example alignment , top right ) , polymorphic at 41% of insertions ( example shown middle right ) , and absent at 29% of insertions ( bottom right ) . Allele sharing between insertions occurs at sites other than the TT haplotype , but these sites tend to be shared across fewer insertions ( see heatmap , bottom right ) . Figure 3—figure supplement 1 shows the population alignment across all ISX insertions on the neo-X . DOI: http://dx . doi . org/10 . 7554/eLife . 05899 . 00610 . 7554/eLife . 05899 . 007Figure 3—figure supplement 1 . Shared polymorphism across sixty-nine ISX insertions . The 5′ 200 basepairs of the ISX element was assembled for an average of 20 individuals , for each of 69 ISX insertions . Each stripe corresponds to the population data for a given insertion and nucleotides are colored as in Figure 1 . Solid lines point to columns of the alignment containing polymorphisms that are shared between multiple ISX insertions . For these columns , the heatmap is shaded to reflect the degree of allele-sharing , which ranges from 3% of insertions to 41% of insertions . The ‘T’ letters under the heatmap mark the location of the TT haplotype shown in Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05899 . 00710 . 7554/eLife . 05899 . 008Figure 4 . Population frequency of TT haplotype across ISX insertions . The location of all ISX elements on the D . miranda neo-X chromosome , as inferred from the MSH22 reference genome assembly , is shown by vertical green bars . The derived TT haplotype ( frequency shown in red ) , is polymorphic at 27 of 66 ISX insertions , a pattern consistent with non-allelic gene conversion . DOI: http://dx . doi . org/10 . 7554/eLife . 05899 . 00810 . 7554/eLife . 05899 . 009Figure 4—figure supplement 1 . ISX genotype across insertions and individuals . Heatmap showing each of the 27 ISX insertions ( rows ) where the TT haplotype is polymorphic among individuals . Columns show the genotype of each individual , for each of these insertions . Each individual has a mixture of TT and GA ISX insertions , suggesting that TT polymorphism among lines is not due to population subdivision . DOI: http://dx . doi . org/10 . 7554/eLife . 05899 . 009 To test if selection has acted to increase the frequency of the TT haplotype in the population , we examined patterns of polymorphisms at GA- and TT-containing ISX elements . The TT haplotype harbors significantly less linked variation than the ancestral GA haplotype , across insertion sites and individuals ( haplotype diversity = 0 . 53 vs 0 . 81; resampling p < 0 . 001; Figure 5A ) . In addition , ISX insertions where TT is fixed have significantly lower nucleotide diversity compared to the insertions where GA is fixed ( one-sided Wilcoxon test p = 0 . 035; Figure 5B ) . Finally , the frequency spectrum at the TT haplotype also shows an excess of high frequency derived alleles , compared to the frequency spectrum at the GA haplotype ( resampling p = 0 . 027; Figure 5C ) . All of these patterns are expected if natural selection acting on the TT haplotype is driving its spread through the population . 10 . 7554/eLife . 05899 . 010Figure 5 . Selection shapes patterns of variation at the TT haplotype . ( A ) Haplotype diversity across all ISX sequences . Assembled ISX contigs were combined for all insertions and individuals . The 25 basepairs flanking each side of the TT region were extracted from a total of 1291 sequences and split into two groups based on whether they contained the TT or GA haplotype . Haplotype diversity was then calculated for each group . The difference between groups is significantly larger than expected by chance ( resampling p < 0 . 001 ) , with the sequences containing the TT haplotype having less haplotype diversity compared to those containing the GA haplotype . ( B ) Nucleotide diversity across all ISX sequences . We compared nucleotide diversity for ISX insertions where all individuals carried the ancestral GA haplotype to those where the derived TT haplotype was fixed . ISX insertions that are fixed for the TT haplotype have significantly reduced nucleotide diversity compared to insertions fixed for the GA haplotype ( one-sided Wilcoxon test p = 0 . 035 ) . ( C ) Allele-frequency spectrum across ISX sequences . The allele frequency spectrum was calculated separately for TT and GA-carrying ISX elements , across all insertions and individuals , using the first 200 basepairs of ISX sequence . Consistent with incomplete hitchhiking under positive selection , the TT frequency spectrum shows an excess of high frequency derived alleles , compared to the GA spectrum ( resampling p = 0 . 027 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05899 . 010 Recent work in a variety of eukaryotes suggests that transposable elements may be major drivers of regulatory evolution ( Feschotte , 2008; Cowley and Oakey , 2013 ) . Their high transposition rate and ability to supply ready-to use regulatory elements across the genome implies that they may rapidly wire new genes into regulatory networks ( Feschotte , 2008 ) . We recently showed that domesticated TEs contribute to rewiring of the dosage compensation network in D . miranda , but appear to supply only suboptimal binding sites for the MSL complex ( Ellison and Bachtrog , 2013 ) . Here , we identify a derived haplotype with two mutations that interact epistatically to increase binding affinity for the MSL complex . We show that these fine-tuning mutations spread among independent ISX insertions by non-allelic gene conversion , and through the population by natural selection ( Figure 6 ) . Relative to regulatory elements that evolve in isolation , a family of regulatory motifs dispersed by TEs presents a larger mutational target , and a mutation arising in any element contained within a larger repeat unit ( the TE ) can spread among copies by non-allelic gene conversion . Consequently , the rate of evolutionary fine-tuning at such regulatory elements can be greatly accelerated by increasing their effective population size ( Mano and Innan , 2008 ) . Thus , transposable elements can ‘crowd-source’ beneficial mutations to rapidly fine-tune regulatory networks . 10 . 7554/eLife . 05899 . 011Figure 6 . Non-allelic gene conversion spreads refining mutations among TE-derived MSL recognition motifs . Shared polymorphism of the TT haplotype among ISX insertions suggests a model where a mutation that refines regulatory activity arose once at a single TE-derived regulatory element , and spread across elements via non-allelic gene conversion . Over evolutionary time , such a mutation spreads in two dimensions: horizontally among TE-derived regulatory elements and vertically through the population , until it is fixed across elements and across individuals . The TT haplotype is at the midpoint of this process . Across ISX insertions , it is fixed , absent , and polymorphic , in approximately equal proportions . DOI: http://dx . doi . org/10 . 7554/eLife . 05899 . 011 Our transgenic experiments show that each individual T allele actually decreases the binding affinity for the MSL complex relative to the ancestral GA haplotype . Thus , TA or GT haplotypes should be selected against in the population if present on a functional ISX element . Consistent with the deleterious effect of individual T alleles , the TA and GT haplotypes are present on some ISY elements but completely absent from ISX , that is , we find the two T mutations to be in perfect linkage disequilibrium among ISX elements but not ISY ( Figure 1B ) . While most ISY elements carry the ancestral GA haplotype , a small fraction ( 0 . 7% of neo-X ISY insertions ) instead carry the derived TT haplotype . It is therefore possible that the TT haplotype was introduced onto the ISX background by non-allelic gene conversion from ISY . Under this scenario , the large family of ISY elements in the D . miranda genome could be acting as a reservoir of natural variation , where complex mutations can accumulate in the absence of epistasis . Non-allelic gene conversion could then transfer these haplotypes to related repetitive elements ( such as ISX ) . While many of these haplotypes are likely to be neutral or deleterious , some may be beneficial , as in the case of the TT haplotype . Such a scenario avoids the waiting time for a double mutation , as well as the fitness valley that would have to be traversed if the two mutations were to occur sequentially on the ISX background . To conclude , our findings suggest that TE-dispersed binding motifs follow an evolutionary trajectory that is fundamentally different from those that arise by other means . The complementary roles of TEs in dispersing regulatory motifs , and gene conversion in spreading subsequent refining mutations , combine to allow for the rapid rewiring and fine-tuning of gene regulatory networks . This process adds a new layer of complexity onto how TEs influence regulatory innovation , as well as a new context in which gene conversion affects genome evolution . Isofemale lines were established from individuals collected in Northern California and inbred for several generations . DNA was extracted from 1–8 females per line using the Qiagen PureGene kit ( Netherlands ) and fragmented by nebulization . Paired-end Illumina libraries were constructed using standard protocols ( Bentley et al . , 2008 ) and sequenced on an Illumina Genome Analyzer II machine ( San Diego , CA ) . Resequencing data were mapped to version 2 . 2 of the D . miranda MSH22 reference assembly ( GenBank: AJMI00000000 . 2 ) using bowtie2 ( Langmead and Salzberg , 2012 ) . ISX locations were identified in Ellison and Bachtrog ( 2013 ) . Paired-end read alignments were evaluated within 2 kilobase windows flanking each ISX insertion and reads with mapping quality of 20 or greater were extracted along with their mate . The extracted mate pairs were then assembled using IDBA-UD , for each line separately ( Peng et al . , 2012 ) . Contigs were aligned using FSA ( Bradley et al . , 2009 ) and visualized with Jalview ( Waterhouse et al . , 2009 ) . A custom Perl script ( available at https://github . com/chris-ellison/MSAvariants ) was used to identify sequence variants within the alignments . We also PCR amplified eight of the ISX insertions where the TT haplotype was polymorphic . We confirmed that this polymorphism was present at each of these insertions and estimated the base-calling accuracy of the assemblies by sequencing the PCR products using Sanger technology . We used the QuikChange Lightning site-directed mutagenesis kit from Agilent Technologies ( Santa Clara , CA ) and the ISX element cloned in Ellison and Bachtrog ( 2013 ) to engineer four ISX variants that differed only with respect to the TT haplotype: ISX-GA , ISX-GT , ISX-TA , and ISX-TT . Each construct was injected by BestGene Inc . ( Chino Hills , CA ) into D . melanogaster embryos carrying a RMCE landing site at cytosite 38F1 on chromosome 2L ( Bloomington Drosophila Stock Center strain #27388 ) . Transformants were verified by PCR and Sanger sequencing . Male third instar larvae ( ∼250 mg ) were collected from F1 hybrids between ISX-GA and each of the other three engineered lines: ISX-GT , ISX-TA , and ISX-TT . Chromatin immunoprecipitation was performed for four biological replicates of each of these lines using the MSL2 d-300 primary antibody from Santa Cruz Biotechnology Inc . ( Santa Cruz , CA ) and the protocol described in Alekseyenko et al . ( 2013 ) . Primers flanking the ISX MRE region were used to generate heterozygous amplicons from the MSL2 IP and input control . Sanger chromatograms were used in conjunction with polySNP software ( Hall and Little , 2007 ) to calculate relative abundance of ISX alleles within the IP and input control amplicons . Abundance of the ‘T’ alleles in the IP amplicons relative to the ancestral G/A alleles was calculated and normalized by the same values from the input control . To determine if the TT frequency among neighboring ISX elements was correlated , we clustered elements within 100 kb of each other and calculated the standard deviation in TT allele frequency within clusters . We then compared these values to 1000 permutations where TT allele frequency was randomly shuffled between ISX locations . The haplotype diversity and allele frequency spectrum resampling tests were performed by drawing , without replacement , two groups of size 617 and 674 , respectively , from the pool of 1291 ISX sequences . The intergroup difference in haplotype diversity , as well as the number of derived variants with frequency of 0 . 75 or greater , was calculated for each of 1000 replicates and compared to the difference between the TT and GA groups .	Mutations change genes and provide the raw material for evolution . Genes are sections of DNA that contain the instructions for making proteins or other molecules , and so determine the physical characteristics of each organism . Genetic mutations that increase an organism's number of offspring and chances of survival are more likely to be passed on to future generations . Changes to when or where a gene is switched on ( so-called regulatory mutations ) can also provide fitness benefits and can therefore be selected for during evolution . Transposable elements are sequences of DNA that are also called ‘jumping genes’ because they can make copies of themselves and these copies of the transposable element can move to other locations in the genome . Some transposable elements contain sequences that switch on nearby genes . If different copies of a transposable element that contains such a regulatory sequence insert themselves in more than one place , it can result in a network of genes that can all be controlled in the same way . The regulatory sequences contained within transposable elements are not always optimal , but they can be fine-tuned through evolution . A fruit fly called Drosophila miranda has a transposable element called ISX that has , over time , placed up to 77 regulatory sequences around one of this species' sex chromosomes . Just as in humans , female flies are XX and males are XY; but having only one copy of the X chromosome means that male flies need to increase the expression of certain genes to produce a full-dose of the molecules made by the genes . This process is called dosage compensation and in 2013 the 77 ISX regulatory sequences on the fruit fly's X chromosome were shown to help recruit the molecular machinery that carries out dosage compensation to nearby genes , albeit inefficiently . Now Ellison and Bachtrog—who also conducted the 2013 study—report how these transposable elements have been fine-tuned to make them more effective for dosage compensation . Ellison and Bachtrog uncovered two mutations that make the ISX transposable element better at recruiting the dosage compensation molecular machinery . ISX spread around different locations along the fly's X chromosome before these mutations arose; this means that initially none of the 77 insertions carried the two mutations , but now 30% of the 77 elements have the mutations in all flies , and 41% have them in only some flies . The same mutations have spread between the different ISX elements because transposable elements with the mutations have been used to directly convert other ISX elements without them . These mutations have also become more common in the fruit fly population by being passed on to offspring and increasing their survival . These two routes have accelerated the fine-tuning of these transposable elements for use in gene regulation . This implies that regulatory sequences derived from transposable elements evolve in a way that is fundamentally different from those that arise by other means , as the direct conversion between these insertions allows fine-tuning mutations to spread more rapidly .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "genetics", "and", "genomics" ]	2015	Non-allelic gene conversion enables rapid evolutionary change at multiple regulatory sites encoded by transposable elements
Barriers to microbial migrations can lead adaptive radiations and increased endemism . We propose that extreme unbalanced nutrient stoichiometry of essential nutrients can be a barrier to microbial immigration over geological timescales . At the oasis in the Cuatro Ciénegas Basin in Mexico , nutrient stoichiometric proportions are skewed given the low phosphorus availability in the ecosystem . We show that this endangered oasis can be a model for a lost world . The ancient niche of extreme unbalanced nutrient stoichiometry favoured survival of ancestral microorganisms . This extreme nutrient imbalance persisted due to environmental stability and low extinction rates , generating a diverse and unique bacterial community . Several endemic clades of Bacillus invaded the Cuatro Cienegas region in two geological times , the late Precambrian and the Jurassic . Other lineages of Bacillus , Clostridium and Bacteroidetes migrated into the basin in isolated events . Cuatro Ciénegas Basin conservation is vital to the understanding of early evolutionary and ecological processes . A ‘lost world’ is both a poetic metaphor and a scientific idea; in both cases , the term pertains to the conservation or re-creation of the deep past in a particular place . Scientists have looked for analogs of such worlds in environments possessing living microbial mats and stromatolites , since these organized forms of life were dominant for billions of years during the Proterozoic ( Nutman et al . , 2017 ) . Nevertheless , in most cases these communities represent more a physical metaphor of the past than an actual lost world , since they contain mostly contemporary microbial lineages ( White et al . , 2015; Wong et al . , 2015 ) . The exception seems to be the abundant and morphologically diverse stromatolites and microbial mats found in the endangered oasis of Cuatro Ciénegas Basin ( CCB ) in Northern Mexico . In this extremely diverse wetland ( Minckley , 1969 ) , the recycling of the deep aquifer by magmatic heat replicates many conditions of ancient oceans ( Wolaver et al . , 2013 ) , including its extremely unbalanced nutrient stoichiometry ( Elser et al . , 2006 ) and sulphur and magnesium minerals that replicate marine osmolarity ( Wolaver et al . , 2013; De Anda et al . , 2017; Rebollar et al . , 2012 ) , despite being low in NaCl . Moreover , isotopic analysis suggests that the deep aquifer maintained the ancestral marine conditions in the wetland by dissolving the existing minerals from its sediments ( Wolaver et al . , 2013 ) . These specific conditions along with an extreme unbalanced nutrient stoichiometry between nitrogen ( N ) and phosphorus ( P ) , created a unique niche that has persisted ( Elser et al . , 2006 ) . Ecological analyses have revealed that a 16:1 nitrogen to phosphorus ( the Redfield ratio ) is common to most life on Earth ( Elser et al . , 2006 ) . However , at the oasis of CCB , such proportions are skewed given the low phosphorus in the ecosystem . We believe that these niche variables can explain the survival in this oasis of ancient marine bacteria and hydrothermal vent-associated sulphur microbes ( Souza et al . , 2006 ) . Our hypothesis is that such marine microbes have stayed there for hundreds of millions of years ( Torsvik , 2003 ) , since this site was on the coasts of Laurentia for a very long time . This changed 35 mya , with the uplifts of the Sierras that isolated CCB from the Western Seaway ( Souza et al . , 2006 ) and the aridification of the Chihuahuan desert in the last 7 million years . In this 'lost world' , even the most dynamic part of the community , the viruses , have maintained a marine signature , as viral metagenomics revealed substantial divergence of viruses from continental waters and a strong similarity with those of marine habitats ( Desnues et al . , 2008; Taboada et al . , 2018 ) . There is a very high ratio of nitrogen to phosphorus ( 167:1 ) in the sediment of the Churince hydrological system , where most of the Bacillus of this study were sampled ( Lee et al . , 2017 ) . We see an imprint of this evolutionary history in the extreme imbalance at the bacterial intracellular level in many lineages ( the most extreme being nitrogen to phosphorus ratio of 965:1 in a strain of CCB Bacillus cereus group ) ( Valdivia-Anistro et al . , 2015 ) and in the capability of some CCB Bacillus species to synthesize membrane sulfolipids , in what appears to be an ancestral adaptation to limited phosphorus availability acquired a long time ago from cyanobacteria by horizontal gene transfer ( Alcaraz et al . , 2008 ) . Extreme unbalanced nutrient stoichiometry , as well as rich sulphur conditions , are niche characteristics of the Precambrian ocean , that ended abruptly at the onset of the Phanerozoic Eon 542 mya with the weathering of continental apatite as consequence of several glaciation events ( Planavsky et al . , 2010 ) . Moreover , using conservative time estimates based on geological events , molecular clock studies have suggested that some strains of culturable cyanobacteria ( Domínguez-Escobar et al . , 2011 ) as well as of Bacillus ( Moreno-Letelier et al . , 2012 ) from CCB diverged also from their close relatives in the late Precambrian . Hence , we propose that CCB is a microbial lost world , not just as a poetic metaphor , but as a real geographical site: a nutrient-unbalanced multidimensional niche isolated from the human environment . What would make a lost world more than a metaphor ? Typically , there are three models of diversification ( Figure 1 ) . The null model for microbes is ‘everything is everywhere and the environment selects’ . We can observe this to apply for cosmopolitan bacteria , such as Escherichia coli or Bacillus subtilis , since these microbes have an enormous population size and considerable migration rates ( Figure 1a ) . The standard model of biogeography is isolation by distance , a pattern similar to the one observed for most macro-organisms . This isolation has been observed for the thermophile Crenarchaeota Sulfolobus islandicus ( Reno et al . , 2009 ) ( Figure 1b ) . The third is the island model of localized adaptation and rare migration ( Figure 1c ) . This pattern also occurs in microbes , as is the case for those in the lakes in the Pyrenees , explained by the island-like nature of each lake ( Casamayor , 2017 ) . Here , we suggest that in order to explain CCB singularity , we need a 4th model , the lost world model ( Figure 1d ) . A lost world would be both a physical space and a refugia , where communities survive in relictual conditions . One signature for a lost world would be the presence of very deep phylogenetic branches , given that the time since isolation would be expected to be very long and extinctions rare . The other would be extreme niche conservatism , in this case by strong environmental filtering given the extremely unbalanced nutrient stoichiometry . In order to explore which of these four models fits our study site and its microbiota , we will first describe our site and the total microbial diversity we found in it . Churince is a closed hydrological system ( and the most endangered site within CCB ) and depends on recharge by the deep aquifer contained in the Sierra San Marcos ( Wolaver et al . , 2013 ) . The system used to consist of a spring , an intermediate lagoon and a large desiccation lagoon connected by a river ( Minckley , 1969 ) . By the time we surveyed its microbial biodiversity , the desiccation lagoon had already disappeared . We obtained samples of environmental DNA from water , sediment and soil in different sampling points from the spring to the end of the intermediate lagoon , as well as from soil associated to different vegetation . We observed a vast microbial diversity within roughly a square kilometer , through PCR amplification and sequencing of 16S rRNA genes from environmental DNA ( Figure 2 ) . This diversity can be expressed in operational taxonomic units ( OTUs ) , used to classify groups of related individuals that have 16S rRNA gene sequences exhibiting at least 97% identity . The Churince’s total Bacteria and Archaea richness is represented by a total of 5 , 167 OTUs assigned to samples from the water column , aquatic sediments , and soil . These assigned OTUs represented 60 different known phyla , three of which were Archaea . Even though each site seems to have a unique taxonomic ‘fingerprint’ ( Figure 3 ) , despite their spatial closeness , we also observed general patterns that aquatic sites share , such as predominance of Proteobacteria , Actinobacteria , and Bacteroidetes . Sediments and soils are much more diverse , and have important phyla in larger proportions than water sites , such as Firmicutes ( the phyla that encompasses Bacillus and Clostridium ) and the primary producer Cyanobacteria . In sediments , Cyanobacteria are part of the microbial mats along with Chlorobi and Spirochaetes , while in the soil , Cyanobacteria are part of the microbial crusts where Acidobacteria also play an important role in nutrient cycling along with Nitrospira ( Figure 3 ) . It is important to underline that in CCB many different lineages of bacteria along with Cyanobacteria ( López-Lozano et al . , 2012 ) have an important role in the acquisition of nitrogen in the valley , contributing to the unbalanced nutrient stoichiometry . The diversity within this small scale in the Churince is immense , in particular when compared to 343 other studies based also on 16S rRNA microbiomes . Those studies comprised several contrasting environments: microbes associated to human , plant , soil , sediments , biofilm , marine biofilms , and extreme environments like Yellowstone hot springs , Guerrero Negro salt flats , and Antarctica soils , for all of which data is available in public databases ( Supplementary file 1 ) . To further compare with the other environments , we used the individual OTUs and then computed their alpha diversity using Shannon index and Simpson function . Alpha diversity measures the number of species and their proportion within each of the sampling sites . Shannon index calculates diversity and abundance , though it is a poor predictor of diversity when rare species constitute a substantial part of the diversity ( which is our case ) . Simpson's function calculates diversity based on the total number of species but does not take into account their relative abundances . The most diverse environments , according to Shannon’s diversity index ( Table 1S ) , are the aquatic sediments of different sites in the world; accordingly , in our dataset the Churince’s sediment had the highest Shannon value . This was also supported by the Simpson’s index that showed , as expected , that sediment was more diverse than water ( Table 1S ) . Even though most of the diversity at Churince had low abundance ( Figure 2 , see the size of the circles in the phylogenetic tree ) , both Shannon’s and Simpson’s diversity indices revealed a very high microbial diversity , even when compared with other microbial diversity hotspots , such as Pearl river in China , or Guerrero Negro in Mexico . We suggest that the explanation for such a large diversity within such a small place is , in part , niche stability over geological times , and in addition , the diversification process reinforced by local adaptation . To test for the lost world clade diversification , we focused on the diversity of bacteria from a single well-known genus , Bacillus , that are easily cultured . From our collection of approximately 2500 cultured Bacillus spp . from CCB , 16S gene sequences were obtained and compared to sequences in databases . We obtained 265 unique sequences selected at 97% identity , a very conservative estimate for Bacillus . In a global tree ( Figure 4 ) with 1019 other OTUs reported for Bacillus spp . from around the world , we can observe the overall distribution and genetic distance of these CCB 16S sequences in relation to all known Bacillus lineages . We noticed that CCB strains formed multiple endemic ( only found in CCB ) lineages most of them with very deep branches , and that our sample increased by nearly 21% the number of previously known Bacillus . Within the Bacillus spp . from CCB , we can distinguish two diverse sets of endemic lineages: one from sediment and another one closely related to marine Bacillus spp . ( Figure 4 ) . CCB Bacillus spp . from sediments are significantly older than the marine related CCB lineages , and according to our analyses calibrated using the divergence between the genus Bacillus and Geobacillus ( Moreno-Letelier et al . , 2012; Battistuzzi and Hedges , 2009 ) ( Figure 5 ) may date back to the Ediacaran ( 635–541 mya , at the end of the Precambrian ) . The Ediacaran period , marks the start of the oxygenation of the ocean allowing not only the first animals to evolve ( Planavsky et al . , 2010 ) , but also the first aerobic Bacilli to diversify . Unlike the sediment lineage , CCB Bacillus species from water , did not form a monophyletic group , which suggests independent synchronized origins dating to the late Jurassic ( Figure 6 ) . A mixed representation of many lineages appears to have entered the CCB ‘multidimensional niche bubble’ simultaneously and did not become extinct . The presence of endemic and early divergent clades , with deep branches , of sediment and marine related CCB lineages , as well as the low extinction rates observed in these lineages ( Figure 7 ) , provides strong evidence for the lost world scenario ( Figure 1D ) . Why do we observe only two pulses of marine migration at CCB ? We think that the first pulse can be explained by the abrupt change of nutrient stoichiometry balance at the end of the Ediacaran ( Elser et al . , 2006; Planavsky et al . , 2010; Stüeken and Kipp , 2017 ) . However , the second pulse is more puzzling since it implies the migration of many independent lineages into the CCB shore . It is possible that the reason is tectonics , as CCB is the point where the birth of the Tethys sea occurred in the western point of Pangea breakage ( Wolaver et al . , 2013; Souza et al . , 2006 ) . Even though none of the other lineages of Bacillus endemic to CCB have been reported at any other site , they seem to constitute punctuated events of arrival to the ‘island-like’ niche and to have subsequently diversified locally . In contrast , in a study of Bacillus spp . from diverse environments in India many cosmopolitan Bacillus spp . were collected that had short branches to sister species ( Yadav et al . , 2015 ) . When we compared the Bacillus sample with a much smaller sample of anaerobic Clostridiales and Bacteroidetes strains , the later were fewer and did not form monophyletic groups endemic to CCB , even though some of their branches were early divergent ( Figure 4—figure supplement 1 ) . Environmental conditions are known barriers to dispersal . The extreme unbalanced nutrient stoichiometries at CCB could certainly constrain immigration from phosphorus demanding populations that require a proportion of nitrogen to phosphorus closer to 16:1 ( Elser et al . , 2006 ) for their survival . This is the case of Patagonia’s isolated and oligotrophic lakes with unique microbial communities with low diversity ( Aguayo et al . , 2017 ) . CCB microbial populations are not only unique , they are also very diverse , despite CCB having an extremely unbalanced nutrient stoichiometry . Our data showed that the small and endangered Churince system in CCB contains 57 out of the 86 known Bacteria phyla , which is 66 . 3% of the recorded bacterial diversity at phylum level ( data from 342 analysed microbiomes , Table 2S ) . This diversity is only comparable to Pearl River in China ( Wang et al . , 2012 ) , where 48 microbial phyla were found . The Pearl River is a highly productive environment that receives inputs from multiple sources in the 2 , 400 km extension of China’s third largest river . In contrast , Churince is a small hydrological system fed by a spring and extends 1 km at most . The local scale makes the bacterial diversity of CCB even more interesting . Red Queen evolutionary processes have been shown to cause highly localized adaptation in numerous systems , including microbial ones . The Red Queen process proposes that organisms continuously adapt to changing conditions , particularly those involving antagonistic interactions within and between species , causing an increase in localized adaptation and coevolution ( Lawrence et al . , 2012 ) . Experiments of competition between strains of Bacillus from different sites within Churince showed marked antagonisms against different strains even from sites a few meters away ( Pérez-Gutiérrez et al . , 2013 ) , consistent with the Red Queen model . We have also observed extreme antagonism in the case of CCB Actinobacteria with non-CCB bacteria ( Arocha-Garza et al . , 2017 ) , which may prevent migration from bacteria outside the basin . What circumstances can keep the marine signal intact ? This requires prevention of both genes and populations from migrating ( Souza et al . , 2008; Souza et al . , 2012 ) . Indeed , we have observed that most of the bacterial lineages are local and clonal ( Cerritos et al . , 2011; Avitia et al . , 2014 ) . Genomes are small and have few imprints of horizontal gene transfer ( HGT ) ( Alcaraz et al . , 2008; Alcaraz et al . , 2010; Gómez-Lunar et al . , 2016 ) . The only exception is Vibrio from CCB , a lineage that thrives in perturbed , higher nutrient environments within the basin , and that display recombination rates similar to the ones observed in marine Vibrio . The caveats being , that such recombination is mostly homologous , maintaining local adaptation and genetic isolation ( Gómez-Lunar et al . , 2018 ) . Even though most CCB lineages are clonal , the rare HGT events allow linkage disequilibrium to break . Hence , we do not see small populations sizes or genetic sweeps as expected in completely clonal lineages ( Cohan , 2016 ) . Moreover , extinction rates are low , while diversification rates are high in some lineages at certain times ( Figure 7 ) . We believe that selective sweeps have not purged the genetic diversity that would drive lineages to extinction , and involves Black Queen dynamics , in addition to the Red Queen processes discussed above . Tolerance and cooperation between strains are observed . For instance , strains of endemic lineages of Bacillus from Churince require cross feeding and cooperation to obtain even amino-acids ( Rodríguez-Torres et al . , 2017 ) . These cross-feeding observations fit a ‘Black Queen’ model ( Morris et al . , 2012 ) where adaptation to severely limited resources lead to genomic streamlining and metabolic co-dependency ( Alcaraz et al . , 2008; Gómez-Lunar et al . , 2016 ) . Hence , our potential explanation for the long-term survival of lost world Bacillus species in CCB involves multiple eco-evolutionary feedbacks . Migrants are suppressed by antagonistic coevolution and community cohesion that is maintained by co-dependent metabolic interactions . Moreover , although Bacillus spp . can form spores , the ultimate strategy to survive stressful conditions , we have shown that CCB Bacillus spp . are competing actively in the microbial communities ( Pérez-Gutiérrez et al . , 2013 ) . Antagonisms and cooperation occur simultaneously in microbial mats and stromatolites ( Anda et al . , 2018 ) , and it is possible that the same dynamics occurred originally in the South-Western shores of Laurentia ( Kershaw , 2017 ) . All our results suggest that extreme unbalanced nutrient stoichiometry , along with community cohesion function like a ‘semipermeable’ barrier to migration , where effective migration is possible , but rare . Fossil evidence shows that stromatolites were still abundant between the Permian and Triassic boundary , in the site where the Tethys sea opened in the South-Western shores of Laurentia ( Kershaw , 2017 ) where CCB was located at the onset of the Mesozoic ( Wolaver et al . , 2013; Souza et al . , 2006 ) . However , during the massive extinction event that marked the end of the Permian , stromatolites became rare , except on the western shores of the Tethys sea ( Kershaw , 2017 ) . Microbial mats and stromatolites can still be found in other sites of the planet , but at CCB , aside from giving testimony to the past , microbial lineages have been safeguarded , bringing evidence for a lost world . The extreme stoichiometric imbalance in the Churince can be explained in part due to the very old weathered rocks that have lost their phosphorous and , from the entrance of nitrogen into the system mainly through nitrogen fixation by many members of the community ( Lee et al . , 2017; López-Lozano et al . , 2012 ) . Even though CCB microbial communities have survived for an extended period of time , their particular niche conditions are being destroyed in the Anthropocene . This impact is even more poignant because CCB wetland has shrunk 90% in the last 50 years , and its deep aquifer has been devastated by the use of fossil water in local agricultural practices . This deep niche change has already destroyed many of the microbial complex communities in Churince . However , we believe that this change can be reversed if the channels that drain the wetland are closed and the wetland recovers its water cycle . The CCB has a population of 14 , 000 people , as well as flourishing tourism , which represent an input of large amounts of nitrogen and phosphorus . Fortunately , most of the human activities and nutrient inputs occur at least 20 km away from the oasis and their astounding microbial communities . We believe that calcium carbonate rocks have worked as buffer between the human activities and the turquoise blue ponds ( Wolaver et al . , 2013 ) . Nevertheless , our experimental evidence shows that an increment in nutrients results in algal blooms and a reduction or disappearance of endemic lineages ( Lee et al . , 2017 ) . Hence , if such mineral buffer gets saturated or the wetland disappears , the environmental singularity that makes CCB unique can change , erasing the biological evidence of this ‘lost world’ . We hope that awareness of this problem will push for proper measures for a change in agricultural practices and sewage management by the county and State authorities . Conservation of the unique niche in CCB and similar sites is paramount for our understanding of the deep past as well as to predict and protect the future of our planet . We sampled ten sites during May 2011 in the Churince system of CCB in a 300 m long lagoon plus two more sites in the spring-head , ca . 1 km away ( latitude: 26° 50’ 53 . 19’ N , longitude: 102° 8’ 29 . 98’ W ) . For each site , permission to sample was obtained from the federal government in Mexico ( SEMARNAT , dirección de vida silvestre FAUT0230 ) . In each sample site , we took 50 g of sediment and a gallon of water as well as a sample of both for biogeochemical variables , nutrients and minerals: C , N , P , Ca , Mg . We also sampled four types of vegetation from an established gradient and obtained composite soil samples . We extracted DNA from each sample using the same methodology ( López-Lozano et al . , 2012 ) . Metagenomic DNAs were sent to JCVI ( San Diego , CA , USA ) for 16S rRNA amplicon gene library ( 341 F-926R primers ) 454 pyrosequencing ( Roche , Brandford , Ct , USA ) . A total of 950 , 000 reads were sequenced; we required a minimum of 50 , 000 reads per site , with a minimum 500 bp length after Quality Control check . Not all samples produced the same amount of sequences , probably due to the natural low yield of DNA extraction in CCB water and sediments . Nevertheless , even at 97% , diversity is high , encompassing all the know phyla of Bacteria but a very low diversity and abundance of Archaea and mostly none of the cosmopolitan human related microbial taxa . The 16S rRNA gene analysis was done as previously reported ( Avitia et al . , 2014; Alcaraz et al . , 2016 ) . Briefly sequencing quality was processed and filtered using FASTQ and Fastx-toolkit , we filtered out any sequence with Phred < 30 , length <500 bp . Operational taxonomic units ( OTUs ) were clustered using cd-hit-est ( Huang et al . , 2010 ) with a 97% identity threshold cut-off . The OTUs were parsed into QIIME pipeline and the taxonomic assignments were done against Greengenes DB ( v 13 . 8 ( DeSantis et al . , 2006 ) ) . Chimeras were removed after taxonomic assignments and detected by ChimeraSlayer ( Human Microbiome Consortium et al . , 2011 ) . Data management , diversity statistic , and plots were done using R phyloseq package ( McMurdie and Holmes , 2014 ) and ggplot2 and RColorBreweer R libraries . Pplacer ( RRID:SCR_004737 ) was used to place the diversity into a reference tree ( Figure 2 ) ( Matsen et al . , 2010 ) . We are using diversity indexes , rather than OTU comparisons because of differences in sequencing technologies , primers used for 16S rRNA gene , coverage depth , and other factors that could affect an overall OTU comparison among different studies . Compared datasets ( 342 ) were retrieved from public available databases like NCBI’s SRA , MG-RAST , and HMP ( Human microbiome project ) websites . Detailed information about accessions used is available as supplementary material in Table S1 . The sequence identity clustering of all 16S rRNA gene sequences from the genus Bacillus spp . and sister genera Anoxybacillus and Geobacillus were retrieved from online databases Ribosomal Database Project ( RRID:SCR_006633 ) and Genbank ( RRID:SCR_002760 ) , 1019 of them at 97% sequence identity , plus 648 sequences of cultivated Bacillus spp . from CCB ( accession numbers in Supplementary file 2 ) selected with the same criterion out of more than 2500 cultivates strains; sequence clustering was done with cd-Hit ( http://weizhongli-lab . org/cd-hit/ , Huang et al . , 2010 , RRID:SCR_007105 ) . These sequences were further aligned with the 16 s rRNA sequences from CCB with the MUSCLE ( RRID:SCR_011812 ) plugin in Geneious ( Kearse et al . , 2012 ) ( RRID:SCR_010519 ) . Neighbour-joining trees were constructed using genetic distances , with the ape and seqinr R package . The sequences were also aligned with the 16 s rRNA sequences from CCB with the MUSCLE plugin in Geneious 5 . 4 . 6 ( Kearse et al . , 2012 ) . In order to have a control , a subset of OTUs from Clostridiales ( n = 131; 18 unique from CCB ) and Bacteroidetes ( n = 189; 12 unique from CCB , Genebank numbers in Table S2 ) was then used to construct Bayesian phylogenies including CCB cultivated anaerobic strains while for Bacillus , we only focused on a mayor clade which had a better representation from CCB samples , henceforth called sediment and soil Bacillus ( n = 311 ) and another with predominantly marine Bacillus ( n = 115; Genebank numbers in Table S2 ) . Phylogenies were reconstructed using BEAST v . 1 . 8 . 2 ( Drummond et al . , 2012 ) ( RRID:SCR_011812 ) , with a Birth-Death speciation model , relaxed lognormal clock models and the following substitution models Bacteroidetes ( HKY + I + G ) , Clostridiales ( HKY + I + G ) , and both Bacillus clades ( GTR + I + G ) . All substitution models were chosen using a Bayesian Information Criterion on likelihoods estimated with jModeltest 2 . 1 . 7 ( Darriba et al . , 2012 ) ( RRID:SCR_015244 ) . Three separate runs were performed for each dataset of 50 million chains each and then combined using LogCombiner v1 . 8 . 2 . Parameter convergence was evaluated using Tracer v . 1 . 6 . 0 . Ultrametric trees were obtained with relative node ages , which were later scaled to produce ultrametric trees with absolute ages to be used in the diversification rate analyses using the R package phytools ( Revell , 2012 ) . The calibration points to date all trees were obtained from literature . The calibration point of Clostridiales was set at 3 , 500 Ma ( Battistuzzi and Hedges , 2009 ) with a normal distribution at the root of the tree , for Bacteroidetes was set to 2 , 500 Ma ( Battistuzzi and Hedges , 2009 ) , also with a normal distribution . The node heights of the sediment Bacillus lineage were obtained by estimating the divergence dates within the genus Bacillus using a smaller phylogenetic sampling . The analysis was conducted using BEAST v . 1 . 8 . 2 ( Drummond et al . , 2012 ) and the calibration points were set at the divergence of the genus Bacillus from Geobacillus at a conservative 1 , 144 . 7 Ma ( sd = 164 ) ( Moreno-Letelier et al . , 2012 ) following the great oxidation event , set to a normal distribution . Another calibration point was set in the diversification of Bacillus at 1047 Ma ( sd = 159 ) , also set to a normal distribution , as it is the recommendation when using node ages estimated by molecular dating ( see BEAST v . 1 . 8 . 2 documentation ) . The clock model was a log normal relaxed clock and the analysis was run in BEAST v . 1 . 8 . 2 ( Drummond et al . , 2012 ) . Changes in diversification rates were estimated with a Bayesian framework using BAMM 2 . 5 . 0 ( Rabosky , 2014 ) . This method estimates the speciation rates , identifies shifts along the phylogeny and estimates the confidence intervals of the various shift configurations detected using a Markov Chain Monte Carlo to explore the universe of candidate models ( Rabosky , 2014 ) . The analyses were carried out using the scaled ultrametric trees of sediment Bacillus , Bacteroidetes and Clostridiales , with 10 million generations for all cases except Clostridiales , which required 20 million generations to reach convergence and adequate effective sampling sizes . Priors were estimated using the function setBAMMpriors implemented by the R package BAMMtools . Results were analysed using BAMMtools on R ( R Development Core Team , 2014; RRID:SCR_001905 ) to obtain the best shift configuration , Bayes factors of number of shifts and posterior probabilities of shifts distributions . Finally , we compared the distribution of rates along the tree of all lineages in order to assess the relative diversification rate differences in all lineages .	Water is a rare sight in a barren land , but there are many more reasons that make the Cuatro Cienegas Basin , an oasis in the North Mexican desert , a puzzling environment . With little phosphorous and nutrients but plenty of sulphur and magnesium , the conditions in the turquoise blue lagoons of the Basin mimic the ones found in the ancient seas of the end of the Precambrian . In fact , Cuatro Cienegas is one of the rare sites where we can still find live stromatolites , a bacterial form of life that once dominated the oceans . Many bacteria of marine origin exist alongside these living fossils , prompting scientists to wonder if the Basin could be a true lost world , a safe haven where ancient microorganisms found refuge and have kept evolving until this day . But to confirm whether this is the case would require scientists to hunt for clues within the genetic information of local bacteria . Souza , Moreno-Letelier et al . came across these hints after sampling for bacteria in a small ( about 1km2 ) lagoon named Churince , and analysing the DNA collected . The results yielded an astonishing amount of biodiversity: 5 , 167 species representing at least two-third of all known major groups of bacteria were identified , nearly as much as what was found in over 2 , 000 kilometres in the Pearl River in China . This is unusual , as most other extreme environments with little nutrients have low levels of diversity . Closer investigation into the genomes of 2 , 500 species of Bacillus bacteria revealed that the sample increased by nearly 21% the number of previously known species in the group . Most of these bacteria were only found in the Basin . These native or ‘endemic’ species have evolved from ancestors that came to the area in two waves . The oldest colonization event happened 680 million years ago , as the first animal forms just started to emerge . The most recent one took place while dinosaurs roamed the Earth about 160 million years ago , when geological events opened again the Basin to the ancient Pacific Ocean . Previous experiments have shown that different species of bacteria in the Churince have evolved to form a close-knit community which ferociously competes with microbes from the outside world . Paired with the extreme conditions found in the lagoon , this may have prevented other microorganisms from proliferating in the environment and replacing the ancient lineages . The days of this lost world may now be numbered . Drained by local farming , the wetlands of the Basin have shrunk by 90% over the past five decades . The Churince lagoon , the most diverse and fragile site where the samples were collected , is now completely dry . Human activities also disrupt the delicate and unique balance of nutrients in the oasis . But all may not be lost – yet . Local high school students have become involved in the research effort to describe and protect these unique microbial communities , and to change agricultural traditions in the area . Closing the canals that export spring water out of the Basin could give the site a chance to recover , and the microbes that are now seeking refuge in underground waters could re-emerge . Maybe there will still be time to celebrate , rather than mourn , the unique life forms of the Cuatro Cienegas Basin .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "ecology" ]	2018	The lost world of Cuatro Ciénegas Basin, a relictual bacterial niche in a desert oasis
Outside of the neurogenic niches of the brain , postmitotic neurons have not been found to undergo efficient regeneration . We demonstrate that mouse Purkinje cells ( PCs ) , which are born at midgestation and are crucial for development and function of cerebellar circuits , are rapidly and fully regenerated following their ablation at birth . New PCs are produced from immature FOXP2+ Purkinje cell precursors ( iPCs ) that are able to enter the cell cycle and support normal cerebellum development . The number of iPCs and their regenerative capacity , however , diminish soon after birth and consequently PCs are poorly replenished when ablated at postnatal day five . Nevertheless , the PC-depleted cerebella reach a normal size by increasing cell size , but scaling of neuron types is disrupted and cerebellar function is impaired . Our findings provide a new paradigm in the field of neuron regeneration by identifying a population of immature neurons that buffers against perinatal brain injury in a stage-dependent process . Most neurons in the brain are generated at specific developmental time points , and once a neuron becomes postmitotic regeneration following injury is limited , except for in two forebrain regions that maintain neurogenesis ( Chaker et al . , 2016 ) . In the context of injury , adult forebrain neurons undergo limited recovery that involves either reactive gliosis ( Buffo et al . , 2008; Robel et al . , 2011; Sirko et al . , 2013 ) or migration of neural stem cells from the neurogenic niches ( Benner et al . , 2013; Llorens-Bobadilla et al . , 2015; López-Juárez et al . , 2013; Martí-Fàbregas et al . , 2010 ) . The cerebellum ( CB ) of the hindbrain has a complex folded structure that houses the majority of neurons in the brain and is essential for balance and motor coordination , as well as higher order reasoning via circuits it forms throughout the forebrain ( Fatemi et al . , 2012; Steinlin , 2007; Tavano et al . , 2007; Tsai et al . , 2012; Wagner et al . , 2017 ) . For two weeks after birth , the postnatal mouse CB consists of both neurons generated in the embryo , and two neurogenic progenitor pools that produce late born neurons and glia . Interestingly , the proliferating granule cell progenitors can be replenished following injury by adaptive reprograming of the second Nestin-expressing progenitors ( Wojcinski et al . , 2017 ) . However , once a neurogenic process has ended , the degree to which post mitotic neurons can undergo regeneration is poorly understood . Purkinje cells ( PC ) are born by embryonic day ( E ) 13 . 5 in the mouse and during weeks 10–11 in humans ( Rakic and Sidman , 1970; Wang and Zoghbi , 2001 ) . After exiting the cell cycle in the ventricular zone , PCs express FOXP2 as they migrate to form a PC layer ( PCL ) under the cerebellar surface by E17 . 5 , and turn on Calbindin1 ( CALB1 ) in the late embryo and stop expressing FOXP2 by two weeks after birth . PCs play a central role in postnatal CB development by being the main source of sonic hedgehog ( SHH ) , which is required for proliferation of granule cell progenitors and Nestin-expressing progenitors that produce interneurons and astrocytes ( Corrales et al . , 2006; Fleming et al . , 2013; Lewis et al . , 2004 ) . PCs also are key for CB function by integrating the inputs that converge on the cerebellar cortex ( Sillitoe and Joyner , 2007 ) . Hence , PC loss is linked to cerebellar motor behavior syndromes and has also been implicated in autism ( Fatemi et al . , 2012; Tsai et al . , 2012; Wang et al . , 2014 ) . In this study we determined the regenerative potential of PCs in neonatal mice . To ablate and track PCs , the diphtheria toxin receptor ( DTR ) and a lineage tracer , TdTomato ( TdT ) , were expressed in a subpopulation of PCs using a transgenic approach ( Pcp2Cre/+; R26LSL-DTR/LSL-TdT or PC-DTR mice; LSL = lox stop-lox ) . We found that only 52 . 16 ± 21 . 84% of PCs ( n = 5 mice ) , identified by expression of CALB1 , expressed TdT and DTR at postnatal day ( P ) 1 , and surprisingly the percentage and large variation remained similar at P5 and P30 ( Figure 1—figure supplement 1 ) . Strikingly , when DT was injected at P1 into PC-DTR pups ( P1-PC-DTR ) , nearly all TdT+ PCs formed an ectopic layer below the PC layer ( PCL ) by 1 day post injection ( dpi ) ( Figure 1A–M ) . The ectopic layer was absent by P8 ( Figure 1K ) , and TdT+ cells in the ectopic layer were TUNEL positive starting at P3 with a peak at P5 . These results show that almost all DTR-expressing TdT+ cells become misplaced , die and are cleared within 5–7 dpi of DT ( Figure 1N , O ) . Unexpectedly , although the number of CALB1+ PCs in the PCL of P1-PC-DTR mice was significantly reduced at P2 compared to non-injected controls ( No DT ) , it was not significantly reduced at P3 and later stages ( Figure 1P ) . Furthermore , the total number of PCs ( ectopic layer + PCL ) was significantly greater in DT-injected cerebella than in No DT controls at P2 and P3 , and the total number of PCs was down to normal levels at P5 , overlapping with the time of clearance of the ectopic layer ( Figure 1P ) . Although the number of TdT+ cells in the PCL increased between P8 and P30 in P1-PC-DTR brains , it remained significantly lower than in No DT controls at P30 ( Figure 1Q ) . Given that there is no significant increase in the recombination induced by Pcp2Cre after birth in the control postnatal CB ( Figure 1Q , Figure 1—figure supplement 1F ) , the percentage of TdT+ cells in P1-PC-DTR brains at P30 ( ~25–30% ) matched the predicted percentage if 50% of the PCs were killed by DT and then ~50% of the regenerated PCs underwent recombination . Interestingly , and consistent with the rapid recovery of PC numbers in the PCL , no significant decrease in the sectional area of the CB was observed between P1 . 5 and P30 ( Figure 1R–T , Figure 1—figure supplement 2 ) . Furthermore , the thickness of the outer ( proliferating ) and inner ( differentiating ) external granule cell layers remained normal ( Figure 1—figure supplement 3 ) . In summary , we uncovered that the CB can rapidly recover ( within 24–48 hr ) from the loss of ~50% of PCs at P1 , by producing new PCs and resuming normal growth . In order to document the rapid production of new PCs after ablation , we tested whether PCs that had recently undergone cell division could be detected at P3 . P1-PC-DTR mice were divided into four groups; each group receiving three injections of BrdU ( 2 hr apart ) during 4–26 hr after DT-injection ( Figure 2A ) . As predicted , BrdU+ PCs ( FoxP2+ and CALB1+ ) were observed in the PCL of all groups ( Figure 2B , Figure 2—figure supplement 1 ) , with the greatest incorporation being between 10–20 hr after DT ( Figure 2C , Figure 2—figure supplement 1 ) . Importantly , in No DT mice no incorporation of BrdU was observed in PCs ( Figure 2—figure supplement 1 ) . Curiously , FOXP2 and BrdU showed non-overlapping subnuclear localization in the nuclei of PCs . In addition , when we analyzed brains of P1-PC-DTR mice at P30 that had received BrdU 10–14 hr after DT injection , we observed BrdU+ mature PCs with similar cell bodies and dendritic trees to their neighbors , showing that the newly generated PCs differentiate and survive to adulthood ( Figure 2—figure supplement 2 ) . Furthermore , a lack of BrdU incorporation in the ectopic layer at P3 shows that the labeling of PCs is not due to DNA damage induced by DT-mediated cell death ( Figure 2—figure supplement 1C ) . In order to further confirm that BrdU incorporation is not due to DNA damage following DT injection , we treated P1 pups either with 4Gy γ-irradiation or DT at P1 followed by three BrdU injections ( 2 hr apart ) at 30 min or 10 hr after treatment , respectively . The brains were then analyzed 24 hr after the last BrdU injection . In the irradiated pups we observed extensive γ-H2AX foci , including in PCs , but BrdU incorporation was not detected in any PCs . In contrast , P1-PC-DTR mice injected with DT showed BrdU incorporation in PCs without any γ-H2AX foci ( Figure 2—figure supplement 3 ) . Thus , DNA damage does not account for the BrdU incorporation into PCs following ablation of ~50% of PCs at P1 . As a second means to specifically label dividing cells that give rise to new PCs , we intracranially injected GFP-expressing retrovirus into P1-PC-DTR pups and littermate controls 12 hr after DT injection , since retroviruses can only incorporate into the DNA of dividing cells and are widely used for clonal analysis of neural stem cells and progeny ( Figure 2D ) ( Cepko , 1988; Yu et al . , 2009 ) . When the mice were analyzed at P21 , we indeed observed rare GFP-labeled CALB1+ PCs in P1-PC-DTR animals near the site of injection , and not in No DT controls . As expected , GFP+ granule cells and Bergmann glia were observed in both the No DT and P1-PC-DTR mice ( Figure 2E–I ) . These three sets of experiments thus reveal that a progenitor capable of proliferating produces the new PCs after ablation at P1 . Based on the rapid replenishment of PCs after ablation at P1 , we hypothesized that a local progenitor in the PCL must be responsible for the response . The Nestin-expressing progenitors ( NEPs ) in the PCL were a candidate , as they display plasticity upon ablation of granule cell precursors in newborn mice ( Wojcinski et al . , 2017 ) . Furthermore a putative rare Nestin+ cell in the adult CB was recently described as able to produce new neurons in response to exercise ( Ahlfeld et al . , 2017 ) . However , when we tested the contribution of NEPs to PC regeneration using a Nes-CFP reporter allele that transiently maintains CFP protein after differentiation , no CFP+ cells were found to co-express FOXP2 or CALB1 at 12 hr and 2 days post DT injection in P1-PC-DTR mice and in No DT controls ( Figure 2—figure supplement 4 ) . Furthermore , inducible fate mapping of NEPs using Nestin-FlpoER/+; R26FSF-TdT/+ ( FSF = frt stop-frt ) mice showed no TdT+ PCs at P30 in P1-PC-DTR and No DT control mice given tamoxifen at P0 ( Figure 2—figure supplement 5 ) . These results suggest that a progenitor other than NEPs mediates regeneration following PC depletion . We next examined whether a progenitor exists after birth that expresses early ( FOXP2 ) but not late ( CALB1 ) PC markers . Indeed , at P1 we identified CALB1 negative/low and FOXP2-expressing cells that could be immature PCs ( named iPC for immature Purkinje cells; Figure 3A–B , Figure 3—video 1 ) . Possibly accounting for the regeneration of PCs , iPCs were not labeled by Pcp2Cre as they were TdT and DTR negative in No DT controls , thus they escape DT-mediated cell death ( Figure 3—figure supplement 1 ) . Temporal analyses revealed a decrease in the number of iPCs from P1 ( 74 . 33 ± 5 . 69/midline sagittal section ) to P5 ( 28 . 66 ± 7 . 51/midline sagittal section , Figure 3A , Figure 3—figure supplement 1 ) , indicating the progenitors are a transient population . Interestingly , the few iPCs present at P5 were specifically enriched in the central and nodular zones of the CB , which are developmentally delayed at P5 ( Legué et al . , 2015; Sudarov and Joyner , 2007 ) ( Figure 3A ) . In order to investigate the normal fate of iPCs , we tested whether there is an increase in the number of CALB1+ PCs from P1 to P30 . In order to minimize variation across animals , we used the C57BL/6 inbred strain of mice and analyzed the entire half-vermis of each brain ( every second section ) . As expected , a significant reduction in the number of iPCs was observed between P1 and P5 , but in addition we detected a significant increase in the number of CALB1+ cells at P30 compared to P1 ( Figure 3C ) . At P5 there was a trend towards an increase in the number of CALB1+ cells . There was also a trend towards a reduction in the total number of iPCs plus CALB1+ cells at P5 and P30 compared to P1 ( Figure 3C ) , suggesting that some PCs may also undergo apoptosis during early postnatal development . These results provide evidence that iPCs are cells destined to become PCs , but normally undergo a delay in differentiation until the first week after birth . We then asked whether the number of iPCs increases after DT treatment of P1-PC-DTR mice . Quantification of iPC numbers showed a significant increase 12 hr after DT injection in P1-PC-DTR mice ( 1 . 90 ± 0 . 05 fold , Figure 3D ) , correlating with the time window of highest BrdU incorporation after injury ( Figure 2B ) . Interestingly , at P5 the number of iPCs was significantly lower in P1-PC-DTR animals than in No DT mice ( Figure 3D ) , possibly reflecting an exhaustion of the progenitor population by production of new PCs . To further show that iPCs expand in number after their neighbors are killed , we used constitutive FLP-based fate mapping in FoxP2Flpo/+; R26FSF-TdT/+ mice to transiently mark and follow PCs and iPCs . We found that all CALB1+ PCs and iPCs expressed TdT at P1 ( Figure 3—figure supplement 2 ) , and as predicted , an increase in transiently fate mapped TdT+ iPCs was seen in P1-PC-DTR mice 12 hr after DT injection at P1 compared to No DT controls ( 1 . 86 ± 0 . 46–fold , n = 3 , Figure 3—figure supplement 3 ) . Thus , iPCs expand in number after damage to neighboring PCs . To confirm that iPCs undergo proliferation upon PC depletion , we injected BrdU or EdU 10–14 hr after DT and collected cerebella 1 hr ( ~P1 . 5 ) later . Other than glial progenitors and microglia seen in No DT controls ( Figure 3—figure supplement 4 ) , all additional BrdU+ ( or EdU+ ) cells in the PCL of P1-PC-DTR mice expressed FOXP2 , and of these cells 45 . 5 ± 1 . 1% expressed CALB1 ( Figure 3E–N , Figure 3—figure supplements 5 , 6 ) . Furthermore , the total number of FOXP2+ cells in the PCL that were acutely labeled with BrdU , was similar to the number of BrdU+ cells that became PCs ( CALB1+ ) at P3 ( 38 . 7 ± 9 . 1/section vs 40 . 3 ± 19 . 0/section , n = 3 , Figure 3—figure supplement 5 ) . In addition , FOXP2+ cells that were Ki67+ ( Figure 3O–S , Figure 3—figure supplement 6C ) or pH3+ ( Figure 3T–X , Figure 3—figure supplement 6D ) and EdU+ were detected at P1 . 5 in the PCL of P1-PC-DTR pups , confirming the presence of proliferative iPCs following PC ablation . In order to further study the cell-cycle state of iPCs in uninjured cerebella , we analyzed the expression levels of the cell cycle inhibitor P27Kip1 ( Watanabe et al . , 1998 ) and KI67 in iPCs compared to CALB1 high FOXP2+ PCs at P1 . Fluorescent intensity analyses revealed that PCs have higher P27 and lower KI67 expression levels compared to iPCs ( Figure 3—figure supplement 7 ) . Collectively , our data argues that the recovery of PCs in P1-PC-DTR mice is mediated by a previously unrecognized and age-dependent progenitor population ( iPCs ) that normally transitions to a CALB1+ PC , but in response to loss of PCs proliferates and differentiates to replace the lost cells . Given that the population of iPCs is greatly reduced by P5 ( Figure 3C ) , PCs should not be efficiently replaced if ablated at P5 , when similar to at P1 Pcp2Cre induces recombination ( expression of DTR ) in 40 . 5 ± 21 . 5% of CALB1+ PCs ( Figure 1—figure supplement 1 ) . As predicted , when DT was injected at P5 ( P5-PC-DTR mice ) ( Figure 4—figure supplement 1A ) , the number of PCs was significantly reduced at P12 compared to No DT controls ( Figure 4—figure supplement 1B–I , R ) . TdT+ PCs were TUNEL+ by P8 ( Figure 4—figure supplement 1J–K ) and the majority of TdT+ cells were cleared from the PCL by P12 ( Figure 4—figure supplement 1G , P and S ) . Furthermore , in P5-PC-DTR mice at P8 and P12 the dendrites and cell bodies of the PCs were poorly organized compared to in controls ( Figure 4—figure supplement 1B–G and L–P ) and at P30 the cell bodies of some PCs were misplaced into the molecular layer ( Figure 4—figure supplement 1N–R ) . Importantly , the reduction in PC numbers observed at P12 was maintained at P30 ( Figure 4—figure supplement 1R ) , such that the number of PCs was reduced by 32 . 4 ± 6 . 5% . In summary , there is little replenishment of PCs when they are ablated at P5 ( Figure 4A ) . We next tested whether the rare iPCs at P5 ( Figure 3A ) can still proliferate upon PC depletion . In contrast to P1-PC-DTR mice , very few iPCs/PCs were BrdU+ in P5-PC-DTR cerebella injected with BrdU at 10–14 hr post DT-injection at both 1 hr ( 5 . 55 ± 0 . 51/ midline sagittal section , n = 3 ) and 1 . 5 days ( 6 . 22 ± 1 . 07/ midline sagittal section , n = 3 , Figure 4B ) post BrdU-injection . The few BrdU+ iPCs/PCs present were concentrated in the central and the nodular zones that are enriched for iPCs at P5 ( Figure 4—figure supplement 2 ) . Interestingly , compared to P1-PC-DTR mice in which 52 . 29 ± 0 . 09% ( n = 3 ) of iPCs incorporated BrdU , only 20 . 55 ± 0 . 07% ( n = 3 ) incorporated BrdU in P5-PC-DTR animals . Overall , these results demonstrate that replenishment of PCs is not efficient at P5 because with age , iPCs both diminish in number and in their ability to proliferate in response to PC depletion . We next examined whether the depletion of PCs in P5-PC-DTR mice had an effect on CB development . Indeed , the area of CB sections was significantly reduced at P12 but not P8 ( Figure 4C , Figure 4—figure supplement 3 although the thickness of the external granule cell layer was significantly reduced in P5-PC-DTR mice at P8 . By P12 the thickness of the external granule cell layer was similar in PC-ablated mice and controls ( Figure 4—figure supplement 4A–E ) . Surprisingly , despite the lack of recovery of PC numbers we found that the area of the CB was normal at P16 and P30 ( Figure 4C , Figure 4—figure supplement 3A–I ) . As a consequence , there was a reduction in PC density compared to No DT or to P1-PC-DTR mice ( Figure 4—figure supplement 3J , Figure 4—figure supplement 2N , Q ) . The density of granule cells also was lower compared to No DT and P1-PC-DTR P30 mice ( Figure 4—figure supplement 4F ) . Interestingly , PCs in P5-PC-DTR mice had a larger soma ( Figure 4D ) and longer primary and secondary dendrites ( Figure 4E ) compared to No DT or P1-PC-DTR mice , a cellular phenotype observed in some mouse mutants with PC loss ( Castagna et al . , 2016 ) . Furthermore , compared to controls , the percentage of PCs present at P30 in P5-PC-DTR animals compared to No DT controls ( ~66% ) did not match the percentage of granule cells that were produced ( ~81% of No DT controls ) . Thus , the ratio of PCs to granule cells is disrupted in P5-PC-DTR animals because granule cells are over-produced . These results reveal that independent of iPCs being stimulated to produce new PCs following their ablation , there are mechanisms of cell and organ size regulation that ensure recovery of CB size . Finally , given that the circuitry ( proportions of neurons ) is disrupted in P5-PC-DTR mice and not P1-PC-DTR mice but CB size is normal in both , we tested whether either mutant has normal motor function at P30 . Interestingly , P1-PC-DTR animals had no significant changes in their motor function compared to controls ( Figure 4F–J ) , whereas P5-PC-DTR mice showed a significant reduction in their latency to fall from the rotarod and had a shorter stride compared to both No DT and P1-PC-DTR mice ( Figure 4F–G and I–J ) but no change in grip strength ( Figure 4H ) . These results demonstrate that P5-PC-DTR mice , but not P1-PC-DTR mice , have motor behavior deficits . Thus , rapid production of new PCs by iPCs enables establishment of functional circuits following depletion of PCs at P1 . Furthermore , achieving correct cell numbers and/or proportions appears to be more important than maintaining CB size for functional recovery after injury in P5-PC-DTR mice . In summary , we discovered a new regenerative process in the developing CB involving a previously unidentified and normally dormant and immature PC progenitor ( iPC ) that is able to expand and produce additional PCs , likely to buffer against early postnatal loss of these postmitotic neurons due to injury . Proliferation of iPCs is stimulated by ablation of PCs at P1 and importantly the response is rapid ( 10–48 hr ) , ensuring other components of the developing CB that are dependent on PCs for their proliferation or differentiation are not compromised . However , iPCs decrease in number and their capacity to proliferate during the first postnatal week , and consequently PCs are poorly replenished when ablated at P5 . The cerebella of P5-PC-DTR mice nevertheless try to adapt by attaining near normal dimensions through a mechanism that includes increasing cell size ( Figure 4—figure supplement 5 ) . The CB therefore has multiple mechanisms for regulating organ size following perinatal injury that depend on the precise stage of development . Furthermore , the motor deficits seen in P5-PC-DTR mice highlight the importance of maintaining the correct number of PCs and relative neuron proportions during development , not just organ size . One possible reason for why iPCs differentiate into PCs after P1 and lose their ability to proliferate is that a critical component of the microenvironment that supports iPCs is diminished soon after birth , perhaps as a consequence of a developmental clock that the cells in the microenvironment follow . A second possibility is that the differentiation of iPCs is dictated by the timing of the establishment of the cerebellar circuitry . We speculate that efficient regeneration is possible at P1 because PCs still have an immature morphology and integration into the cerebellar circuitry , whereas at later stages the parallel fibers ( axons of granule cells ) synapse with PCs and climbing fibers ( axons of the inferior olive neurons ) refine their synapses and both cells promote PC maturation ( Good et al . , 2017; Hoxha et al . , 2017 ) . Thus , maturation and integration of a newly generated PC into the cerebellar circuitry might not be efficient or possible after P5 . By extrapolation , the replenishment process has evolved such that developmental plasticity is tightly correlated with age-dependent maturation of the neural circuit . An additional cellular process to consider for the age-dependency of regeneration is the ability of neurons to enter back into the cell-cycle . Most differentiated neurons , including PCs , when forced to proliferate undergo apoptosis ( Feddersen et al . , 1992 ) . However , previous reports have shown that following experimental manipulation or neurodegeneration , ectopic proliferation of adult retinal and pyramidal neurons can occur ( Ajioka et al . , 2007; Sage et al . , 2005; Skapek et al . , 2001; Yang et al . , 2001 ) . Our data indicate that iPCs , which lack the mature PC marker CALB1 and express the immature marker FOXP2 , show low expression of P27 and weak but higher expression of KI67 compared to CALB1+ PCs , suggesting that their cell-cycle exit may be incomplete . CALB1+ PCs , likely ones that recently began making CALB1 protein , also appear to be able to re-enter the cell-cyle . However , the increase in the number of iPCs ~12 hr after DT administration suggests that the main regenerative response is achieved by the proliferation of iPCs . The regenerative processes previously described in neuronal tissues involve adaptive reprograming of cells that are either actively proliferating or retain proliferative capacity and also have cell fate plasticity ( Benner et al . , 2013; Buffo et al . , 2008; Jinnou et al . , 2018; Lin et al . , 2017; Llorens-Bobadilla et al . , 2015; López-Juárez et al . , 2013; Martí-Fàbregas et al . , 2010; Robel et al . , 2011; Samanta et al . , 2015; Sirko et al . , 2013; Wojcinski et al . , 2017 ) . Here we identified a distinct regenerative process that involves a local and normally dormant or immature progenitor . Unlike NEPs of the CB or astrocytes and neural stem cells in the forebrain that produce neurons upon injury , iPCs do not require reprograming and cell fate plasticity as our data indicates that they normally produce additional CALB1+ PCs after birth . Thus , iPCs maintain their lineage decision , but proliferate and then mature upon injury . An important question raised by our study is whether regeneration of postmitotic neurons by age-dependent progenitors is unique to the CB where protracted development might provide a conducive milieu , or whether all brain regions retain similar progenitors for a particular time window after each neuron subtype is generated . Furthermore , understanding the mechanisms of PC regeneration in newborn mice could provide insights into how regeneration in the adult brain can be enabled . All the experiments were performed according to protocols approved by the Memorial Sloan Kettering Cancer Center’s Institutional Animal Care and Use Committee ( IACUC ) . Animals were given access to food and water ad libitum and were housed on a 12 hr light/dark cycle . The following mouse lines were used for these experiments: Pcp2Cre ( Zhang et al . , 2004 ) , Nestin-CFP ( Mignone et al . , 2004; Wojcinski et al . , 2017 ) , Nestin-FlpoER ( Wojcinski et al . , 2017 ) , FoxP2Flpo ( Bikoff et al . , 2016 ) , Rosa26LSL-DTR ( Stock no: 007900 , The Jackson Laboratories ) ( Buch et al . , 2005 ) , Rosa26LSL-TdT ( ai14 , Stock no: 007909 , The Jackson Laboratories ) ( Madisen et al . , 2010 ) , Rosa26FRT-STOP-FRT-TdT derived from Ai65 ( Stock no: 021875 , The Jackson Laboratories ) ( Madisen et al . , 2015 ) , C57BL/6J ( Stock no: 00664 , The Jackson Laboratories ) . Both sexes were used for analyses and no randomization was used . Exclusion criteria for experimental data points were sickness or death of animals during the testing period . No randomization was used and masking was used only for the behavior studies where the experimenter was blind to the genotypes . Diphtheria toxin ( 30 μg/g of mouse; List Biological Laboratories , Inc . ) was injected subcutaneously either at postnatal day ( P ) one or P5 and the brains were collected at various ages ( Figure 1a and Figure 4—figure supplement 1 ) . Mice not given DT ( No DT mice ) were Pcp2Cre/+; R26DTR/LSL-TdT littermates and injected with the same volume of vehicle ( PBS ) . BrdU or EdU ( 50 μg/g of mouse; Sigma ) was injected subcutaneously . For P5 and younger animals , brains were dissected and fixed in 4% paraformaldehyde ( PFA ) for 24–48 hr ( h ) at 4°C . Animals older than P5 were anesthetized using intraperitoneal injection of a Ketamine ( 100 mg/kg ) and Xylaxine ( 10 mg/kg ) cocktail . Once full anesthesia was achieved , animals were systemically perfused with ice-cold PBS , followed by 4% PFA . Brains were dissected and post-fixed in 4% PFA for 24–48 hr . Fixed brains were allowed to sink in 30% Sucrose in PBS solution and then embedded in OCT ( Tissue-Tek ) for cryosectioning . 14 μm-thick cryosections were obtained using a Leica cryostat ( CM3050S ) and mounted on glass slides . Frozen sections were stored at −20°C for future analysis . In order to generate the 3D renderings in Figure 3—video 1 60 μm-thick cryosections were obtained and staining was performed on free floating sections . Haematoxylene and Eosin ( H and E ) staining was performed to assess cerebellar cytoarchitecture and measure area ( size ) . For immunofluorescent ( IF ) analysis , slides were allowed to warm to room temperature ( RT ) . After washing once with PBS , slides were blocked using 5% Bovine Serum Albumin ( BSA , Sigma ) in PBS-T ( PBS with 0 . 1% Triton-X ) for 1 hr at RT . Slides were then incubated overnight at 4°C with primary antibodies diluted in blocking buffer . Figure 1—source data 1 . summarizes the primary antibodies used in this study . Upon primary antibody incubation , slides were washed with PBS-T ( 3 × 5 min ) , incubated with specific AlexaFluor-conjugated secondary antibodies ( 1:500 in blocking buffer , Invitrogen ) for 1 hr at RT and then washed again with PBS-T ( 3 × 5 min ) . Counterstaining was performed using Hoechst 33258 ( Invitrogen ) and the slides were mounted with Fluoro-Gel mounting media ( Electron Microscopy Sciences ) . EdU was detected using a commercial kit following the manufacturer’s recommendations ( Invitrogen Cat no: C10340 ) . The super folding ( sf ) -GFP-expressing VSVG-pseudotyped gamma-retrovirus ( Moloney murine leukemia virus ) was made in HEK293T ( ATCC #CRL-11268 ) cells using the pCMV-vsvg and pCMV-gp packaging plasmids and pUX-sf-GFP retrovirus vector plasmid ( cloned by inserting sf-GFP into the BglII and NotI sites of the pUX plasmid ( Gu et al . , 2011 ) as previously described ( Yu et al . , 2009; Zhao et al . , 2006 ) . 10–12 hr after DT injection , P1 P1-PC-DTR pups were anesthetized by hypothermia . 3 μL of ( sf ) -GFP-expressing retrovirus particles ( >109 Tu/mL ) were injected intracranially into P1 vermal cerebella using a stereotactic apparatus . On average 12–15 sections were analyzed that were ~50 μm apart around the injection site . 7–9 retroviral-labeled PCs per mouse were detected only in the P1-PC-DTR brains ( n = 6/ condition ) An X-RAD 225Cx ( Precision X-ray ) microirradiator in the Small Animal Imaging Core Facility at Memorial Sloan Kettering Cancer Center was used to provide a single dose of 4 Gy irradiation , as previously described ( Wojcinski et al . , 2017 ) , to P1 pups anesthetized by hypothermia . The CB was targeted using a collimator with 5 mm diameter . Images were collected either with a DM6000 Leica microscope or Zeiss LSM 880 confocal microscope and processed using ImageJ Software ( NIH ) . For each quantification , three midline parasagittal sections/brain were analyzed and data was averaged . Cells were counted using the Cell Counter plugin for ImageJ ( NIH ) . Analyses of the numbers of PCs and iPCs were performed by counting all of the PCs on a midline parasagittal section . CB area was calculated by defining a region of interest by outlining the perimeter of the outer edges of the CB , using ImageJ . EGL thickness was calculated by dividing the area of the EGL by the length of the EGL in midline sections . IGL density was calculated by counting the number of nuclei in three 40x fields from lobule eight in three midline sections and by dividing the number by the area of the region counted . In order to reduce variation and address the fate of iPCs , we used P1-30 inbred mice ( C57BL/6J ) and analyzed half of the vermis . Analysis of the number of iPCs and PCs was performed on every other section from 14 μm-thick sections to avoid counting the same cells twice due to their larger soma size . On average 25–28 sections were counted per brain . Intensity measurements for P27 and KI67 expression in iPCs compared to CALB1+ PCs at P1 were performed using ImageJ . iPC or PC nuclei were defined as the region of interest and the marker fluorescence intensity and the nuclear area were measured and reported as corrected total cell fluorescence ( CTCF ) /nuclear area . ( CTCF = Integrated Density – ( Nuclear area X mean fluorescence of background readings ) . PC soma size and dendrite length were calculated using randomly distributed TdT+ PCs from three midline sections ( >20 cells/section ) . Soma area was calculated by outlining the perimeter of the outer edges of each cell . Cells that showed primary dendrites were used for this analysis to ensure that the region where the maximum soma area observed was used for the analyses . For dendrite length quantifications , primary and secondary dendrite length was measured and summed and PCs around the base of fissures were omitted . 5 week old animals ( No DT: n = 17 , DT@P1: n = 9 and DT@P5: n = 11 ) were used to assess differences in motor behavior . The same sets of mice were used for all three tests described below . Prism ( GraphPad ) was used for all statistical analysis . Statistical comparisons used in this study were Student’s two-tailed t-test; One-way and Two-way analysis of variance ( ANOVA ) , followed by post hoc analysis with Tukey’s test for multiple comparisons . Relevant F-statistics and p values are stated in the figure legends and the p values of the relevant post hoc multiple comparisons are shown in the figures . Summary of all the statistical analysis performed can be found in Figure 1—source data 2 . The statistical significance cutoff was set at p<0 . 05 . Population statistics were represented as mean ± standard deviation ( SD ) of the mean . No statistical methods were used to predetermine the sample size , but our sample sizes are similar to those generally employed in the field . n ≥ 3 mice were used for each experiment and the numbers for each experiment are stated in the figure legends .	The cerebellum , or 'little brain' , handles movement , thought and social interaction . Unlike the rest of the brain , which primarily develops in the womb , most of its cells appear within the first year of our lives ( or first few weeks in mice ) . This makes it vulnerable to injury around the time of birth . We used to think that the brain could not replace damaged neurons , but when specific precursor cells in the cerebellum in the brains of newborn mice are removed , they are able to renew themselves . This is because specialized stem cells start to divide and produce the missing cells of the cerebellum . Another type of cells in the cerebellum , called Purkinje neurons , are already produced in the embryo . They direct the development of several other cell types in the cerebellum after birth . They are also a crucial component of the circuits within the cerebellum , and losing them can cause loss of muscle coordination . Purkinje cells do not normally divide once an animal is born , but scientists want to know if they might be able to regrow after injury at birth . Bayin et al . killed Purkinje cells in newborn mice with a toxin and used fluorescent markers to track the dying cells . Then , the remaining cells in the surrounding area were studied . This revealed that even when half of the Purkinje cells died a day after birth , the mice behaved normally . The cells regrew , and the cerebellum developed as it should . This happened because the loss of the Purkinje cells activated a population of immature Purkinje cells ( iPCs ) . These cells would normally mature into adult Purkinje cells , but in their immature state they can still divide and make copies of themselves to replace lost neurons after injury . As the mice grew older , the number of iPCs started to drop as the immature cells developed into adult Purkinje cells . When the iPCs ran out , any cells available to divide were gone and the mice could no longer replace any damaged Purkinje cells – the repair window had closed . This work raises the possibility that other types of immature cells in the brain could be set aside to help repair damage during early development . A better understanding of these cells could reveal clues about conditions such as autism , which have been linked to damages or faults in the cerebellum . It may also help to gain new insights into how to regenerate the adult brain after injury .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "short", "report", "neuroscience" ]	2018	Age-dependent dormant resident progenitors are stimulated by injury to regenerate Purkinje neurons
Thyroid hormone ( TH ) regulates diverse developmental events and can drive disparate cellular outcomes . In zebrafish , TH has opposite effects on neural crest derived pigment cells of the adult stripe pattern , limiting melanophore population expansion , yet increasing yellow/orange xanthophore numbers . To learn how TH elicits seemingly opposite responses in cells having a common embryological origin , we analyzed individual transcriptomes from thousands of neural crest-derived cells , reconstructed developmental trajectories , identified pigment cell-lineage specific responses to TH , and assessed roles for TH receptors . We show that TH promotes maturation of both cell types but in distinct ways . In melanophores , TH drives terminal differentiation , limiting final cell numbers . In xanthophores , TH promotes accumulation of orange carotenoids , making the cells visible . TH receptors act primarily to repress these programs when TH is limiting . Our findings show how a single endocrine factor integrates very different cellular activities during the generation of adult form . Mechanisms that synchronize developmental signals and integrate them across cell types and organ systems remain poorly defined but are fundamentally important to both development and evolution of adult form ( Atchley and Hall , 1991; Ebisuya and Briscoe , 2018 ) . A powerful system for elucidating how organisms coordinate fate specification and differentiation with morphogenesis is the array of cell types that arise from embryonic neural crest ( NC ) , a key innovation of vertebrates ( Gans and Northcutt , 1983 ) . NC cells disperse throughout the body , contributing peripheral neurons and glia , osteoblasts and chondrocytes , pigment cells and other derivatives . Differences in the patterning of these cells underlie much of vertebrate diversification . Thyroid hormone ( TH ) coordinates post-embryonic development of NC and other derivatives through mechanisms that are incompletely characterized ( Brent , 2012; Brown and Cai , 2007; Sachs and Buchholz , 2017; Shi , 1999 ) . During the abrupt metamorphosis of amphibians , TH drives outcomes as disparate as tail resorption and limb outgrowth . In the more protracted post-embryonic development of zebrafish—which has similarities to fetal and neonatal development of mammals ( Parichy et al . , 2009 ) —TH coordinates modifications to several traits including pigmentation . Remarkably , TH has seemingly opposite effects on two classes of NC-derived pigment cells , curtailing the population of black melanophores yet promoting development of yellow/orange xanthophores; fish lacking TH have about twice the normal number of melanophores and lack visible xanthophores ( Figure 1A ) ( McMenamin et al . , 2014 ) . We asked how a single endocrine factor can have such different effects on cells sharing a common embryonic origin . Using transcriptomic analyses of individual cell states , we comprehensively defined the context for TH activities by identifying populations and subpopulations of adult NC derivatives . We then assessed the consequences of TH status for lineage maturation across pigment cell classes . Our analyses showed that TH drives maturation of cells committed to melanophore and xanthophore fates through different mechanisms , promoting terminal differentiation and proliferative arrest in melanophores , and carotenoid-dependent repigmentation in xanthophores . These mechanisms reflect different developmental histories of melanophores and xanthophores and yield different cell-type abundances when TH is absent . Our findings provide insights into post-embryonic NC lineages , contribute resources for studying adult pigment cells and other NC-derived cell types , and illustrate how a circulating endocrine factor influences local cell behaviors to coordinate adult trait development . To explain the pigment cell imbalance of hypothyroid fish , we envisaged two models for TH activity during normal development ( Figure 1B ) . In the first model , TH influences states of specification , directing multipotent cells away from one fate and toward the other , or preventing the transdifferentiation of cells already committed to a particular fate . In the second model , TH influences cells that are already committed and remain committed , to their fates . In this scenario , discordant effects across lineages might be observed if TH promotes a cellular process in one lineage that amplifies its population , while simultaneously inhibiting the same process in the other lineage to restrain its population . To evaluate the applicability of these models to TH-dependent regulation of pigment cell populations , we sought to capture the range of intermediate states through which these cells transit during normal and hypothyroid development . Accordingly , we sequenced transcriptomes of thousands of individual NC-derived cells isolated from trunks of euthyroid and hypothyroid fish ( Figure 2—figure supplements 1 and 2 ) . Dimensionality reduction ( Becht et al . , 2018; Cao et al . , 2019 ) followed by unsupervised clustering identified melanophores , xanthophores and a third class of NC-derived pigment cells , iridescent iridophores ( Figure 2A and B; Supplementary file 1 ) . A cluster likely corresponding to multipotent pigment cell progenitors ( Budi et al . , 2011; Singh et al . , 2016 ) was marked by genes encoding pigment cell transcription factors , general markers of mesenchymal NC and factors associated with proliferation and migration but not pigment synthesis ( Figure 2—figure supplement 3; Supplementary file 2—Table 1 ) . Some cells within this cluster also expressed the zebrafish-specific embryonic NC marker crestin , which is generally down-regulated at later developmental stages but is still expressed in a subset of presumptive progenitor cells ( Budi et al . , 2011 ) . Beyond pigment cells and their presumptive precursors , other clusters were identifiable as neurons , Schwann cells , other glia , and chromaffin cells . An additional cluster expressed markers suggestive of proliferative , non-pigmentary progenitors , and one large cluster ( ‘unknown’ ) was not readily assignable to NC-derived populations described previously . Bioinformatic comparisons across all clusters revealed distinct expression profiles of genes encoding ligands and receptors , cell adhesion molecules , and products likely to have diverged in function after the teleost-specific whole-genome duplication ( Figure 2—figure supplement 4 ) . The larva-to-adult transformation of zebrafish entails changes in a variety of traits including NC derivatives ( Parichy et al . , 2009 ) . In some instances , cell types at different stages that are superficially similar ( e . g . larval vs . adult melanophores ) can be distinguished by different genetic requirements ( Budi et al . , 2008; Larson et al . , 2010; Parichy et al . , 1999 ) , raising the possibility that distinct gene expression programs regulate early larval and adult populations . If so , we predicted that NC-derived cells isolated from middle larval–juvenile stages , during development of the adult phenotype ( i . e . Figure 2a ) , should form clusters distinct from cells that developed during embryonic stages to form the embryonic–early larval ( ‘EL’ ) phenotype . To test this idea , we isolated EL NC-derived cells , which clustered in identifiable cell types similar to those of middle-larval juvenile stages ( Figure 2—figure supplement 5A ) . Combining profiles for cells at different stages failed to reveal non-contiguous , life-stage-specific clusters , although some EL cells occupied subsets of transcriptomic space relative to their broader cell type ( e . g . melanophores ) ( Figure 2—figure supplement 5B–D ) . These data do not indicate markedly different transcriptomic programs of NC-derived cell types across life stages , despite the existence of some stage-specific requirements for particular genes and pathways . Overall , our survey captured numerous NC-derived cell types , including abundant pigment cells and progenitors , and revealed substantial variation in gene expression programs among them . To understand the gene expression context in which TH impacts each pigment cell type , we compared pigment cells and progenitors , the lineages of which have been described ( Budi et al . , 2011; Mahalwar et al . , 2014; McMenamin et al . , 2014; Patterson and Parichy , 2019; Singh et al . , 2016 ) ( Figure 3A ) . These analyses revealed subsets of melanophores and xanthophores ( Figure 3B ) , consistent with differences in states of differentiation and morphogenetic behaviors ( Eom et al . , 2015; Parichy et al . , 2000b; Parichy and Spiewak , 2015 ) . For example , cells of subcluster melanophore 2 exhibited low levels of transcriptional activity and expressed fewer genes , suggesting a more advanced state of differentiation , as compared to cells of melanophore 1 ( Figure 3—figure supplement 1 ) . Likewise cells of xanthophore 1 had fewer transcripts and expressed fewer genes than cells of xanthophore 2 , suggesting they may represent undifferentiated , cryptic xanthophores and actively differentiating populations , respectively ( McMenamin et al . , 2014 ) . Additional surveys of these data revealed new markers of xanthophore and iridophore lineages ( Figure 3—figure supplements 2 and 3 ) , and cell-type-specific expression of some previously identified markers [e . g . tyrp1b , aox5 , tfec ( Lister et al . , 2011; McMenamin et al . , 2014 ) ( Figure 3C ) . Expression of other genes was broader than might be expected from mutational or other analyses ( Figure 3—figure supplement 4 ) ; for example mitfa , encoding a transcription factor required for melanophore fate specification ( Lister et al . , 1999 ) was expressed in melanophores and progenitors , but also xanthophores ( Figure 3C ) , consistent with prior reports ( Eom et al . , 2012; Parichy et al . , 2000b ) . To characterize transcriptional dynamics through lineage maturation , we pseudotemporally ordered cells ( Qiu et al . , 2017a; Qiu et al . , 2017b; Trapnell et al . , 2014 ) , yielding a differentiation trajectory with each pigment cell type arising from a common progenitor ( Figure 3D ) . This topology differed from known lineage relationships ( Figure 3A ) but was consistent with similarity of EL and mid-larval/juvenile gene expression programs ( Figure 2—figure supplement 5D ) . Branch expression analysis modeling ( BEAM ) ( Qiu et al . , 2017a ) confirmed that genes with functions in specification ( e . g . mitfa in melanophores ) were expressed early in pseudotime whereas genes associated with differentiation [e . g . dct , encoding a melanin synthesis enzyme ( Kelsh et al . , 2000b ) ] were expressed late ( Figure 3E; Figure 3—figure supplement 5A ) . These analyses revealed dynamics of dozens of genes potentially identifying discrete processes in lineage-specific maturation ( Supplementary file 2—Table 2 ) as well as broader trends . For example , transcripts per cell declined in melanophores but not iridophores , consistent with an expectation of reduced RNA abundance as melanophores—but not iridophores—exit the cell cycle with maturation ( Figure 3—figure supplement 5B ) ( Budi et al . , 2011; Darzynkiewicz et al . , 1980; McMenamin et al . , 2014; Spiewak et al . , 2018 ) . Resolution of pigment cell states through their development allowed us to test if TH functions in fate specification ( Figure 1B–i ) . If so , the excess melanophores and missing xanthophores of hypothyroid fish should reflect biases on specification of multipotent progenitors , or the transdifferentiation ( Lewis et al . , 2019; Niu , 1954 ) of initially specified cells . Such alterations should be evident in reduced-dimension transcriptomic space as strong skew in the apportionment of cells between branches or abnormal paths in the cellular trajectory , respectively . Yet , euthyroid and hypothyroid trajectories were topologically equivalent . Moreover , pigment cell progenitors were not depleted in hypothyroid fish as might occur were these cells being allocated inappropriately as melanophores ( Figure 4A–D ) . Through a second model—lineage discordance—TH could have opposite effects on cells already committed to particular fates , selectively amplifying one cell type while simultaneously repressing amplification of the other ( Figure 1B–ii ) . For example , TH could promote differentiation of xanthoblasts to xanthophores , but prevent differentiation of melanoblasts to melanophores . Alternatively , TH could be a survival factor in the xanthophore lineage but a pruning factor in the melanophore lineage . Terminal phenotypes of both hypothyroid and hyperthyroid mutant fish are consistent with such a mechanism ( McMenamin et al . , 2014 ) . If TH has discordant effects between lineages , we predicted that hypothyroid fish should exhibit a strong depletion of xanthophores from the end of their branch of the trajectory , whereas melanophores should be strongly over-represented near the tip of their branch . Yet , empirical distributions of pigment cell states in hypothyroid fish were all biased towards earlier steps in pseudotime , sometimes severely ( Figure 4E ) . Indeed , prior analyses showed that addition of exogenous TH to hypothyroid cells ex vivo can promote differentiation of unpigmented melanoblasts to melanophores ( McMenamin et al . , 2014 ) , contrary to the idea that TH specifically blocks melanophore development . Together these findings allow us to reject a model in which TH regulation of pigment cell abundance in the adult fish occurs through discordant effects on specific cellular processes across lineages . Having rejected both of our initial models ( Figure 1B ) , we considered a third possibility , that TH promotes the maturation of both lineages , but in distinct ways . For melanophores , inspection of transcriptomic states and cellular phenotypes supported a role for TH in promoting maturation of this lineage . Genes expressed during terminal differentiation of melanophores from euthyroid fish ( e . g . tfap2a , tyrp1b ) were expressed at lower levels in melanophores of hypothyroid fish , suggesting an impediment to maturation in the absence of TH ( Figure 5A ) . To test further test the idea that TH promotes the maturation of melanophores , we examined additional cellular phenotypes . Melanophores of juvenile euthyroid fish tended to be uniformly well-melanized and stellate , whereas melanophores of juvenile hypothyroid fish were variably melanized and dendritic ( Figure 5B ) , reminiscent of earlier stages of melanophore development in wild-type ( Eom et al . , 2012; Parichy and Turner , 2003 ) . Quantification of melanin content within individual cells confirmed that melanophores of euthyroid fish are more heavily melanized than those of hypothyroid fish ( Figure 5C ) . Prior analyses indicated that melanophores of euthyroid fish fail to divide whereas those of hypothyroid fish continue to do so ( McMenamin et al . , 2014 ) . These findings raised the possibility that melanophores of euthyroid fish might exhibit signs of cellular senescence or other indications of proliferative cessation not observed in melanophores of hypothyroid fish . Human nevus melanocytes , and melanophores of teleost melanoma models , exhibit senescent or senescent-like phenotypes and can be multinucleated ( Leikam et al . , 2015; Leikam et al . , 2008; Regneri et al . , 2019; Savchenko , 1988 ) . Accordingly , we asked whether similar attributes were evident for zebrafish stripe melanophores . When plated ex vivo , some stripe melanophores exhibited senescence-associated β-galactosidase ( SA-β-gal ) activity ( Figure 5—figure supplement 1A ) , although we were unable to score such staining reliably , precluding comparisons across TH conditions . SA-β-gal staining results from lysosomal β-gal activity and both β-gal activity and lysosome number increase in aging cells ( Kurz et al . , 2000; Lee et al . , 2006 ) . We therefore quantified lysosome-specific Lysotracker labeling ( Figure 5—figure supplement 1B and F ) of melanophores by fluorescence activated cell sorting tyrp1b:palm-mCherry+ melanophores . Lysosomal contents of melanophores from euthyroid fish were greater than melanophores from hypothyroid fish ( Figure 5—figure supplement 1C ) . Measurements of forward scatter ( FSC-A ) also suggested that melanophores from juvenile euthyroid fish were larger than melanophores from hypothyroid fish ( Figure 5—figure supplement 1D ) , although FSC-A can be influenced by cell-size-independent factors as well ( Tzur et al . , 2011 ) . Finally we examined multinucleation , a condition linked to increased cell survival and size ( Orr-Weaver , 2015; Usui et al . , 2018 ) . In euthyroid fish , ~20% of melanophores were binucleate near the onset of adult melanophore differentiation but >50% were binucleate by juvenile stages , confirming an overall increase in binucleation with somatic stage and melanophore age ( Figure 5—figure supplement 1E ) . In stage-matched comparisons for TH status , ~70% of melanophores from euthyroid fish were binucleated , whereas only ~25% of melanophores from hypothyroid fish were in this state ( Figure 5D and E ) . Collectively , our observations and those of McMenamin et al . ( 2014 ) suggest a model in which TH drives melanophores into a terminally differentiated state of increased melanization , larger size and lysosomal content , binucleation , and proliferative cessation . We next examined TH functions specific to the xanthophore lineage . Most adult xanthophores develop directly from EL xanthophores that lose their pigment and then reacquire it late in adult pattern formation ( Figure 3A ) ( McMenamin et al . , 2014 ) . Because xanthophores of hypothyroid fish persist , albeit in a cryptic state , we predicted that TH effects should be less pervasive in these cells than in melanophores that develop de novo from transit amplifying cells originating from multipotent progenitors . Indeed , fewer genes were expressed differentially between TH backgrounds in xanthophore than melanophore lineages ( 3 . 6% vs . 9%; Figure 6A ) . Prominent among these were several loci implicated in , or plausibly associated with , the processing of yellow/orange carotenoids ( Figure 6B and C; Figure 6—figure supplement 1 ) , dietarily derived pigments that contribute to xanthophore coloration ( Schartl et al . , 2016; Toews et al . , 2017 ) . Differences in carotenoid gene expression suggested a corresponding pigmentation deficiency in xanthophores of hypothyroid fish that we confirmed by HPLC , histology , and transmission electron microscopy ( Figure 6D; Figure 6—figure supplement 2 ) . Among carotenoid genes , scavenger receptor B1 ( scarb1 ) encodes a high-density lipoprotein receptor essential for carotenoid accumulation in birds and invertebrates ( Kiefer et al . , 2002; Toomey et al . , 2017 ) and we found it to be required in zebrafish for carotenoid deposition , although not cell persistence ( Figure 6—figure supplement 3A and B ) . scarb1 was expressed more highly in xanthophores of euthyroid than hypothyroid fish ( q = 1 . 1E-10 ) ( Figure 6B and E; Figure 6—figure supplement 1 ) and exogenous TH was sufficient to rescue both expression and carotenoid deposition ( Figure 6F; Figure 6—figure supplement 3C ) . Together these findings demonstrate an essential role for TH in carotenoid pigmentation and suggest that TH modulation of a suite of carotenoid pathway genes is required for cryptic xanthophores to re-pigment during adult pattern formation . The distinct phases of xanthophore EL and adult pigmentation ( McMenamin et al . , 2014 ) , and the TH-dependence of the latter , led us to ask whether mechanisms underlying coloration might be stage-specific . In contrast to the defect of adult xanthophore pigmentation in scarb1 mutants , we found that 5 dpf larval xanthophores were indistinguishable from wild-type ( Figure 6—figure supplement 4A ) . Conversely , mutants lacking xanthophore pigmentation at 5 dpf have normal adult xanthophores ( Lister , 2019; Odenthal et al . , 1996 ) . Because two pigment classes—carotenoids and pteridines—can contribute to xanthophore coloration , we hypothesized that visible colors at different stages depend on different pathways . Carotenoids were undetectable in euthyroid 5 dpf larvae , and carotenoid-related genes were expressed at lower levels in EL xanthophores than adult xanthophores ( Figure 6—figure supplement 4B and C ) . By contrast , pteridine pathway genes tended to be expressed similarly across stages regardless of TH status and were even moderately upregulated in hypothyroid xanthophores ( Figure 6C , Figure 6—figure supplement 4C ) . Pteridine autofluorescence and pterinosomes were also indistinguishable between euthyroid and hypothyroid fish ( Figure 6—figure supplement 4D; Figure 6—figure supplement 2B ) despite the overt difference in xanthophore color with TH status ( Figure 1A; McMenamin et al . , 2014 ) . Together , these observations imply that TH induces new , carotenoid-based pigmentation , allowing transiently cryptic xanthophores to reacquire coloration during adult pattern development . TH therefore drives maturation of both xanthophores and melanophores yet has markedly different roles in each lineage . Finally , to understand how TH effects are transduced in pigment cell lineages , we evaluated roles for TH nuclear receptors ( TRs ) that classically activate target genes when ligand ( T3 ) is present but repress gene expression when ligand is absent ( Brent , 2012; Buchholz et al . , 2003; Hörlein et al . , 1995 ) . Genes encoding each of the three zebrafish TRs ( thraa , thrab , thrb ) were expressed by melanophores and xanthophores , yet presumptive null alleles for each unexpectedly had pigment cell complements and patterns that resembled the wild type ( Figure 7A; Figure 7—figure supplement 1A–D ) . Given the absence of grossly apparent phenotypes for TR mutants , we hypothesized that instead of acting to promote maturation when T3 is present , TRs may function primarily to repress maturation when T3 is limiting . If so , we predicted that xanthophore development in hypothyroid fish should be rescued by mutation of TR . We therefore generated fish lacking TH and TRs . Loss of thrab , on its own or in conjunction with loss of thraa , partially restored the deposition of carotenoids in interstripe xanthophoes; mutation of all three receptors fully rescued the number of carotenoid-containing xanthophores ( Figure 7B and C; Figure 7—figure supplement 1E and F ) . TR receptor mutations likewise reduced the total numbers of melanophores in hypothyroid fish to levels indistinguishable from euthyroid fish ( Figure 7D ) . These findings suggest that repression by unliganded TRs contributes to pigment-associated phenotypes in hypothyroid fish , implying a function for TRs in repressing the repigmentation of xanthophores and terminal differentiation of melanophores until late stages in adult pigment pattern development . Nevertheless , roles for TRs are likely to be complex and outcomes of derepression dependent on context . For example , the simplest model of TR gating would predict that loss of TRs in euthyroid fish should result in the precocious maturation of pigment cells . Yet , we found no evidence for early pigmentation of xanthophores in euthyroid fish homozygous for thrab mutation ( Figure 7—figure supplement 1G ) , suggesting essential roles for other factors present only at later stages ( Patterson and Parichy , 2013 ) . Our study provides insights into how TH coordinates local cellular events during the development of adult form . The stripes of adult zebrafish comprise three major classes of pigment cells that develop at specific stages and from distinct NC sublineages . Perturbations that affect the times of appearance , states of differentiation or morphogenetic behaviors of these cells can dramatically alter pattern by affecting total numbers of cells and the cascade of interactions normally required for spatial organization ( Parichy and Spiewak , 2015; Patterson et al . , 2014; Watanabe and Kondo , 2015 ) . Fish lacking TH have gross defects in pigment cell numbers and pattern with ~two fold the normal complement of melanophores and the simultaneous absence of visible xanthophores ( McMenamin et al . , 2014 ) . We show that this phenotype arises not because TH normally biases cell fate specification or has discordant effects on a particular cellular behavior that amplifies one cell type while repressing the other . Rather , our findings—combining discovery-based analyses of single-cell transcriptomic states with experiments to test specific cellular hypotheses—suggest a model whereby TH promotes maturation of both melanophores and xanthophores in distinct ways that reflect the developmental histories of these cells ( Figure 8 ) . Our study provides a glimpse into the diversity of cell states among post-embryonic NC-derivatives and illustrates how a single endocrine factor coordinates diverse cellular behaviors in a complex developmental process . By sampling individual cell transcriptomes across NC-derived lineages , our study complements prior investigations of lineage relationships , morphogenetic behaviors , genetic requirements , and spatial and cell-type-specific gene expression profiles ( Eom et al . , 2015; Irion et al . , 2016; Johnson et al . , 1995; Kelsh et al . , 2017; McMenamin et al . , 2014; Parichy and Spiewak , 2015; Singh et al . , 2016; Singh et al . , 2014 ) . Multipotent progenitors that give rise to adult melanophores , some xanthophores , and iridophores are established in the embryo and reside within peripheral nerves as development progresses ( Budi et al . , 2011; Budi et al . , 2008; Camargo-Sosa et al . , 2019; Dooley et al . , 2013a; Singh et al . , 2016 ) . As the adult pattern forms , some of these cells migrate to the hypodermis where they differentiate and integrate into dark stripes or light interstripes . The peripheral-nerve association of pigment cell progenitors in zebrafish is reminiscent of nerve-associated Schwann cell precursors that contribute to melanocytes of mammals and birds ( Adameyko et al . , 2009 ) . Our collected cell-types , which include immature and mature glia , differentiating pigment cells , and presumptive progenitors of different types identify new candidate genes for promoting—and recognizing—distinct states of differentiation and morphogenetic activities , and will enable efforts to define how multipotent NC progenitors are maintained and recruited into particular lineages . That corresponding populations of embryonic and adult populations had largely overlapping transcriptomic states additionally highlights the intriguing problem of how specific pathways are deployed reiteratively across life cycle phases to achieve specific morphogenetic outcomes . Our identification of a role for TH in the adult melanophore lineage illuminates how these cells develop normally and mechanisms that likely contribute to the supernumerary melanophores of hypothyroid fish . Melanoblasts derived from peripheral-nerve associated progenitors are proliferative during adult pigment pattern formation yet this activity largely ceases as the cells differentiate ( Budi et al . , 2011; McMenamin et al . , 2014 ) . Several lines of evidence suggest that TH promotes melanophore maturation to a terminally differentiated state: in the presence of TH , melanophores were more heavily melanized , larger , had greater lysosomal contents , and were more likely to be binucleated . TH similarly promotes the melanization of melanoblasts ex vivo and a cessation of proliferative activity in vivo ( McMenamin et al . , 2014 ) . Our findings are broadly consistent with a role for TH in balancing proliferation and differentiation ( Brent , 2012 ) and may be of clinical relevance , as human melanoma is associated with hypothyroidism and recurrent TH pathway mutations ( Ellerhorst et al . , 2003; Shah et al . , 2006; Sisley et al . , 1993 ) . We suggest a model in which TH normally curtails expansion of the adult melanophore population by ensuring that cells cease to divide in a timely manner; in hypothyroid fish , the inappropriate retention of an immature state allows continued growth of the melanophore population during these post-embryonic stages . Whether TH induces a true cellular senescence and proliferative arrest , or whether cells at an apparently terminal state of differentiation remain competent to divide in specific conditions , will be interesting to learn . TH promoted the terminal differentiation of xanthophores , but in a manner distinct from melanophores . We found far fewer TH-dependent genes in xanthophores than melanophores , likely reflecting the different developmental histories of these cells . In contrast to adult melanophores that arise from a transit-amplifying progenitor , most adult xanthophores develop directly from EL xanthophores that lose their pigment and then regain color late in adult pattern formation ( McMenamin et al . , 2014; Patterson et al . , 2014 ) when TH levels are rising ( Chang et al . , 2012 ) . The yellow-orange color of xanthophores can depend on pteridine pigments , carotenoid pigments , or both ( Bagnara and Matsumoto , 2006; Granneman et al . , 2017; Lister , 2019; Odenthal et al . , 1996; Ziegler , 2003 ) . We showed that TH directly or indirectly regulates carotenoid-associated genes and carotenoid deposition , allowing cryptic xanthophores to reacquire visible pigmentation . TH did not similarly influence pteridine pathway genes . These observations suggest that TH mediates a transition from pteridine-dependent pigmentation at embryonic/early larval stages to carotenoid-dependent pigmentation of the same cells in the adult . Consistent with the notion of TH-mediated pigment-type switching , TH-dependent scarb1 was required for carotenoid accumulation during adult pattern formation , yet mutants lacked an embryonic/early larval xanthophore phenotype . Conversely , mutants with pteridine and color deficiencies in embryonic/early larval xanthophores have normally pigmented adult xanthophores ( Lister , 2019; Odenthal et al . , 1996 ) . In xanthophores at post-embryonic stages , then , TH drives a state of terminal differentiation from a developmental program that is relatively more advanced than that of progenitor-derived melanophores . That cryptic xanthophores appear poised to re-differentiate likely explains the smaller proportion of genes that were TH-dependent in these cells as compared to melanophores . Finally , our study provides clues to likely roles for TRs during adult pigment pattern formation . TR mutants lacked overt pigmentation defects yet allowed for rescues of both melanophore and xanthophore defects in hypothyroid fish , suggesting that unliganded TRs normally repress maturation of these lineages . Loss of TRs similarly allows the survival of congenitally hypothyroid mice ( Flamant et al . , 2002; Flamant and Samarut , 2003 ) . TRs may therefore prevent the inappropriate activation of gene expression programs required for lineage maturation when TH levels are low , as is thought to occur during amphibian metamorphosis ( Choi et al . , 2015; Shi , 2013 ) . Although we detected pigment cell expression of each TR locus , our analyses cannot indicate whether TH acts directly on pigment cells through TR activities that are autonomous to these lineages . A plausible alternative would be that TH acts on stromal or other cell types in which TRs might be expressed and might exert similarly repressive effects when unliganded . Indeed , stromal cells of the hypodermis ( Lang et al . , 2009 ) and also iridophores ( Frohnhöfer et al . , 2013; Patterson and Parichy , 2013 ) regulate melanophore and xanthophore numbers during adult pigment pattern formation , and we observed striking differences in iridophore maturation depending on TH status . On-going efforts seek to distinguish between these possibilities . Results of the current study , however , represent a useful first step in understanding how globally available signals can control fine-grained patterning of cells within this complex adult trait . Staging followed ( Parichy et al . , 2009 ) and fish were maintained at ~28 . 5°C under 14:10 light:dark cycles . All thyroid-ablated ( Mtz-treated ) and control ( DMSO-treated ) Tg ( tg:nVenus-v2a-nfnB ) fish were kept under TH-free conditions and were fed only Artemia , rotifers enriched with TH-free Algamac ( Aquafauna ) , and bloodworms . Fish stocks used were: wild-type ABwp or its derivative WT ( ABb ) ( Eom et al . , 2015 ) ; Tg ( tg:nVenus-v2a-nfnB ) wp . rt8 , Tg ( aox5:palmEGFP ) wp . rt22 , Tg ( tyrp1b:palm-mCherry ) wp . rt11 ( McMenamin et al . , 2014 ) ; csf1raj4blue ( Parichy et al . , 1999 ) ; Tg ( −28 . 5Sox10:Cre ) zf384 ( Kague et al . , 2012 ) ; Tg ( −3 . 5ubi:loxP-EGFP-loxP-mCherry ) cz1701 ( Mosimann et al . , 2011 ) ; tuba8l3:nEosFPvp . rt17 , thrabvp31rc1 , thraavp33rc1 , thrbvp34rc1 , scarb1vp32rc1 and tyrvp35rc1 ( this study ) . Mutants and transgenic lines were maintained in the WT ( ABb ) genetic background . Fish were anesthetized prior to imaging with MS222 and euthanized by overdose of MS222 . All procedures involving live animals followed federal , state and local guidelines for humane treatment and protocols approved by Institutional Animal Care and Use Committees of University of Virginia and University of Washington . To ablate thyroid follicles of Tg ( tg:nVenus-2a-nfnB ) , we incubated 4-day post-fertilization ( dpf ) larvae for 8 hr in 10 mM Mtz with 1% DMSO in E3 media , with control larvae incubated in 1% DMSO in E3 media . For all thyroid ablations , treated individuals were assessed for loss of nuclear-localizing Venus ( nVenus ) the following day . Ablated thyroid glands fail to regenerate ( McMenamin et al . , 2014 ) and absence of regeneration in this study was confirmed by continued absence of nVenus expression . For CRISPR/Cas9 mutagenesis , one-cell stage embryos were injected with 200 ng/μl sgRNAs and 500 ng/μl Cas9 protein ( PNA Bio ) using standard procedures ( Shah et al . , 2015 ) . Guides were tested for mutagenicity by Sanger sequencing and injected fish were reared through adult stages at which time they were crossed to Tg ( tg:nVenus-v2a-nfnB ) to generate heterozygous F1s from which single allele strains were recovered . CRISPR gRNA targets ( excluding protospacer adjacent motif ) are included in Supplementary file 2—Table 7 . Mutant alleles of scarb1 and TR loci are provided in Figure 6—figure supplement 3 and Figure 7—figure supplement 1 , respectively . The melanin free tyrvp . r34c1 allele generated for analyses of melanophore lysosomal content exhibits a four nucleotide deletion beginning at position 212 that leads to novel amino acids and a premature stop codon ( H71QEWTIESDGL* ) . For F0 thrb mutagenesis analysis in the thraa; thrab mutant background , chemically synthesized Alt-R CRISPR-Cas9 sgRNAs targeting the thrb site and Cas9 protein ( Alt-R S . p . Cas9 nuclease , v . 3 ) were obtained from Integrated DNA Technologies ( IDT ) . RNPs were prepared as recommended and ~1 nl was injected into the cytoplasm of one-cell stage embryos . To label nuclei of adult melanophores , BAC CH73-199E17 containing the puma gene tuba8l3 ( Larson et al . , 2010 ) was recombineered to contain nuclear-localizing photoconvertible fluorophore EosFP using standard methods ( Sharan et al . , 2009; Suster et al . , 2011 ) . Images were acquired on: Zeiss AxioObserver inverted microscopes equipped with Axiocam HR or Axiocam 506 color cameras; a Zeiss AxioObserver inverted microscope equipped with CSU-X1 laser spinning disk ( Yokogawa ) and Orca Flash 4 . 0 camera ( Hamamatsu Photonics ) ; or a Zeiss LSM 880 scanning laser confocal microscope with Fast Airyscan and GaAsP detectors . Images were corrected for color balance and adjusted for display levels as necessary with conditions within analyses treated identically . Melanophores and xanthophores were counted within regions defined dorsally and ventrally by the margins of the primary stripes , anteriorly by the anterior margin of the dorsal fin , and posteriorly by five myotomes from the start . Only hypodermal melanophores were included in analysis; dorsal melanophores and those in scales were excluded . Mature xanthophores were counted by the presence of autofluorescent carotenoid with associated yellow pigment . Cell counts were made using ImageJ . Individual genotypes of fish assessed were confirmed using PCR or Sanger sequencing . In situ hybridization ( ISH ) probes and tissue were prepared as described ( Quigley et al . , 2004 ) . Probes were hybridized for 24 hr at 66°C . Post-hybridization washes were performed using a BioLane HTI 16Vx ( Intavis Bioanalytical Instruments ) , with the following parameters: 2x SSCT 3 × 5 min , 11 × 10 min at 66°C; 0 . 2x SSCT 10 × 10 min; blocking solution [5% normal goat serum ( Invitrogen ) , 2 mg/mL BSA ( RPI ) in PBST] for 24 hr at 4°C; anti-Dig-AP , Fab fragments ( 1:5000 in blocking solution , Millipore-Sigma ) for 24 hr at 4°C; PBST 59 × 20 min . AP staining was performed as described ( Quigley et al . , 2004 ) . Xanthophore pigments were examined by imaging autofluorescence in eGFP and DAPI spectral ranges for carotenoids and pteridines , respectively . For imaging pteridines , fish were euthanized and treated with dilute ammonia to induce autofluorescence ( Odenthal et al . , 1996 ) . For analyses of carotenoid contents by HPLC we pooled three skin samples from each genotype and condition ( Mtz-treated or control ) into two separate samples . We homogenized the tissue in a glass dounce homogenizer with 1 ml of 0 . 9% sodium chloride and quantified the protein content of each sample with a bicinchoninic acid ( BCA ) assay ( 23250 , Thermo ) . We then extracted carotenoids by combining the homogenates with 1 ml methanol , 2 ml distilled water , and 2 ml of hexane:tert-methyl butyl ether ( 1:1 vol:vol ) , separated the fractions by centrifuging , collected the upper solvent fraction , and dried it under a stream of nitrogen . We saponified these extracts with 0 . 2 M NaOH in methanol at room temperature for 4 hr following the protocol described in Toomey and McGraw ( 2007 ) . We extracted the saponified carotenoids from this solution with 2 ml of hexane:tert-methyl butyl ether ( 1:1 vol:vol ) and dried the solvent fraction under a stream of nitrogen . We resuspended the saponified extracts in 120 μl of methanol:acetonitrile 1:1 ( vol:vol ) and injected 100 µl of this suspension into an Agilent 1100 series HPLC fitted with a YMC carotenoid 5 . 0 µm column ( 4 . 6 mm ×250 mm , YMC ) . We separated the pigments with a gradient mobile phase of acetonitrile:methanol:dichloromethane ( 44:44:12 ) ( vol:vol:vol ) through 11 min , a ramp up to acetonitrile:methanol:dichloromethane ( 35:35:30 ) for 11–21 min and isocratic conditions through 35 min . The column was warmed to 30°C , and mobile phase was pumped at a rate of 1 . 2 ml min−1 throughout the run . We monitored the samples with a photodiode array detector at 400 , 445 , and 480 nm , and carotenoids were identified and quantified by comparison to authentic standards ( a gift of DSM Nutritional Products , Heerlen , The Netherlands ) . Analyses of 5 dpf wild-type and csf1ra mutants used only larval heads where xanthophores are abundant in the wild type; other procedures were the same as for later stages . Skins of Tg ( aox5:palmEGFP ) euthyroid and hypothyroid zebrafish ( 8 . 6–10 . 4 SSL ) were dissociated and plated at low density in L-15 medium ( serum free ) on collagen-coated , glass bottom dishes ( Mattek ) for 5 hr . Cells were then fixed with freshly prepared 4% PFA for 15 m , rinsed with PBST ( 0 . 1% ) , blocked ( 5% goat serum , 1% BSA , 1X PBS ) , then incubated at 4°C overnight with rabbit anti-GFP primary antibody ( ThermoFisher ) . Stained cells were rinsed 3X with 1X PBS and fixed again with 4% PFA for 30 min . Cells were then rinsed twice with ddH2O , washed with 60% isopropanol for 5 min , and then dried completely . Cells were incubated with filtered , Oil Red O solution ( 5 mM in 60% isopropanol ) for 10 min , and rinsed 4X with ddH20 before imaging ( Koopman et al . , 2001 ) . All GFP+ cells were imaged across two plates per condition and were scored for presence or absence of red staining . For assaying senescence of melanophores ex vivo , skins from euthyroid and hypothyroid fish ( n = 3 each , 11 SSL ) were cleared of scales , dissociated and plated on glass-bottom , collagen coated dishes ( MatTek ) in L-15 medium ( Gibco ) and incubated overnight at 28°C . Cells were then rinsed with dPBS , fixed with 4% PFA and stained using a Senescence β-Galactosidase Staining Kit ( Cell Signaling Technologies , cat . #9860 ) according to manufacturer's instructions ( Ceol et al . , 2011; Dimri et al . , 1995 ) . Staining was carried out for 48 hr at pH six prior to imaging . To assay cell state as measured by lysosomal content ( Kurz et al . , 2000; Lee et al . , 2006 ) of melanophores by FACS , skins from euthyroid and hypothyroid Tg ( tyrp1b:palm-mCherry; tuba8l3:nEOS ) , tyr fish lacking melanin ( n = 12 each ) were dissociated and resuspended 1% BSA/5% FBS/dPBS . Cells were incubated for 1 hr with Lysotracker ( 75 nM ) ( ThermoFisher , L12492 ) and Vybrant DyeCycle Violet stain ( 5 μM ) ( ThermoFisher , V35003 ) shaking at 500 rpm , 28°C . Without washing , cells were FAC sorted . Single transgene controls and wild-type cells were used to adjust voltage and gating . Prior to analysis of fluorescence levels , single cells were isolated by sequentially gating cells according to their SSC-A vs . FSC-A , FSC-H vs FSC-W and SSC-H vs SSC-W profiles according to standard flow cytometry practices . Intact live cells were then isolated by excluding cells with low levels of DyeCycle violet staining ( DAPI-A ) . As expected these cells express a wide range of our tuba8l3:nlsEosFP transgene as determined by levels of green fluorescence ( FITC-A ) . Melanophores were isolated by identifying cells with high fluorescence in the FITC-A and mCherry-A channels which describe expression of the tuba8l3:nlsEosFP and tyrp1b:palm-mCherry transgenes . Lastly , lysosomal content of melanophores was determined by the median fluorescence intensity of the lysosomal marker , Lysotracker Deep Red ( APC-A ) . The data were collected on a FACS ARIA using FACSDiva version eight software ( BD Biosciences ) and analyzed using FlowJo v10 . Melanin content was measured from brightfield images in Fiji . All image quantifications were performed using the base processing and analysis functions in ImageJ . Images were aligned and centered on the horizontal myoseptum and cropped to 2500 × 1500 pixels around dorsal and ventral stripes . Images were segmented based on red channel intensity using ‘Auto Local Threshold’ with parameters ‘method = Sauvola radius = 50’ . To account for close or overlapping melanophores , particles were further segmented using watershed segmentation . Particles larger than 25 pixels and not touching an edge were used for subsequent analyses . Fish were euthanized then fixed in sodium cacodylate buffered 4% glutaraldehyde overnight at 4°C . Trunk regions were dissected then tissue stained in 2% osmium tetroxide for 30 min , washed , and then stained in 1% uranyl acetate overnight at 4°C . Samples were dehydrated with a graded ethanol series then infiltrated with a 1:1 propylene oxide:Durcupan resin for 2 hr followed by fresh Durcupan resin overnight and flat embedded prior to polymerization . Blocks were thin sectioned on a Leica EM UC7 and sections imaged on a JEOL 1230 transmission electron microscope . Trunks or skins of staged , post-embryonic zebrafish ( 7 . 2–11 . 0 SSL ) were dissected ( n = 8 per replicate ) and enzymatically dissociated with Liberase ( Sigma-Aldrich cat . 5401119001 , 0 . 25 mg/mL in dPBS ) at 25°C for 15 min followed by manual trituration with a flame polished glass pipette for 5 min . Cell suspensions were then filtered through a 70 μm Nylon cell strainer to obtain a single cell suspension . Liberated cells were re-suspended in 1% BSA/5% FBS in dPBS and DAPI ( 0 . 1 μg/mL , 15 min ) before FACS purification . All plastic and glass surfaces of cell contact were coated with 1% BSA in dPBS before to use . Prior to sorting for fluorescence levels , single cells were isolated by sequentially gating cells according to their SSC-A vs . FSC-A , FSC-H vs FSC-W and SSC-H vs SSC-W profiles according to standard flow cytometry practices . Cells with high levels of DAPI staining were excluded as dead or damaged . Cells from wild-type and Tg ( ubi:switch ) zebrafish without Cre were used as negative control to determine gates for detection of mCherry and GFP fluorescence , then cells from Tg ( sox10:Cre; ubi:switch ) zebrafish were purified according to these gates . NC-derived cells cells were isolated by identifying cells with high fluorescence in the mCherry-A channel which describes expression of the ubi:loxP-EGFP-loxP-mCherry transgene after permanent conversion to ubi:mCherry after exposure to Sox10:Cre ( see Figure 2—figure supplement 1C ) . All samples were kept on ice except during Liberase incubation , and sorted chilled . Skin tissue from stage-matched fish was dissociated as above and melanophores and xanthophores were FAC sorted for the presence aox5:palmeGFP or tyrp1b:palm-mCherry , respectively . RNA was extracted from pools of 1000 cells using the RNAqueous-Micro kit ( Thermo Fisher , cat . AM1912 ) . Full length cDNA was synthesized with Superscript III reverse transcriptase ( Thermo Fisher , cat . #18080093 ) . Amplifications were 40 cycles with Q5 DNA polymerase ( NEB , M0492 ) , 38 cycles at 94°C , 30 s; 67°C , 20 s; 72°C , 20 s . For primer sequences ( actb1 , thraa , thrab , thrb ) , see Supplementary file 2—Table 7 . Whole-trunks or skins were collected from stage-matched Tg ( tg:nVenus-2a-nfnB ) euthyroid and hypothyroid siblings , dissociated , and sox10:Cre:mCherry+ cells isolated by FACS . We replicated the experiment three times . For each replicate , we collected cells from euthyroid and hypothyroid fish at 7 . 2 SSL , 8 . 6 SSL , and 9 . 6 SSL ( mid-larval , 6–10 fish per stage , per replicate ) and sorted equal numbers of mCherry+ cells from each group into a single sample . Cells were pelleted and resuspended in 0 . 04% ultrapure BSA ( ThermoFisher Scientific ) . Representing a terminal stage of pigment pattern development , we also collected mCherry+ cells from one sample within each replicate of 11 SSL ( juvenile , five fish per condition ) euthyroid and hypothyroid fish . To capture cells representing the EL pigment pattern , we collected mCherry+ cells from five dpf larvae ( 50 fish ) . In each experiment , we ran parallel euthyroid and hypothyroid samples ( fish were siblings ) . For each sample , we targeted 2000–4000 cells for capture using the Chromium platform ( 10X Genomics ) with one lane per sample . Single-cell mRNA libraries were prepared using the single-cell 3’ solution V2 kit ( 10X Genomics ) . Quality control and quantification assays were performed using a Qubit fluorometer ( Thermo Fisher ) and a D1000 Screentape Assay ( Agilent ) . Libraries were sequenced on an Illumina NextSeq 500 using 75-cycle , high output kits ( read 1: 26 cycles , i7 Index: eight cycles , read 2: 57 cycles ) . Each sample was sequenced to an average depth of 150 million total reads . This resulted in an average read depth of ~40 , 000 reads/cell after read-depth normalization . We found that for many genes , annotated 3' UTRs in the Ensembl 93 zebrafish reference transcriptome were shorter than true UTR lengths observed empirically in pileups of reads mapped to the genome . This led to genic reads being counted as intergenic . To correct for this bias in aligning reads to the transcriptome , we extended all 3' UTR annotations by 500 bp . In rare cases , UTR extension resulted in overlap with a neighboring gene and in these instances we manually truncated the extension to avoid such overlap . We built a custom zebrafish STAR genome index using gene annotations from Ensembl GRCz11 with extended 3’ UTRs plus manually annotated entries for mCherry transcript , filtered for protein-coding genes ( with Cell Ranger mkgtf and mkref options ) . Final cellular barcodes and UMIs were determined using Cell Ranger 2 . 0 . 2 ( 10X Genomics ) and cells were filtered to include only high-quality cells . Cell Ranger defaults for selecting cell-associated barcodes versus barcodes associated with empty partitions were used . All samples were aggregated ( using 10X Cell Ranger pipeline ‘cellranger aggr’ option ) , with intermediary depth normalization to generate a gene-barcode matrix containing ~25 , 000 barcoded cells and gene expression counts . We used Uniform Manifold Approximation and Projection ( UMAP ) ( McInnes et al . , 2018 ) to project cells in two or three dimensions and performed louvain clustering ( Blondel et al . , 2008 ) using the reduceDimension and clusterCells functions in Monocle ( v . 2 . 99 . 1 ) using default parameters ( except for , reduceDimension: reduction_method = UMAP , metric = cosine , n_neighbors = 30 , mid_dist = 0 . 5; clusterCells: res = 1e-3 , k = 15 ) . We assigned clusters to cell types based on the detection of published marker genes . Cells isolated from euthyroid and hypothyroid fish were combined to maintain consistency of analysis and for comparisons between groups . Batch correction methods were not used between the two groups or across samples because we did not observe sample-specific separation or clustering in UMAP space . Cells with more than 15 , 000 UMIs were discarded as possible doublets . All genes were given as input to Principal Components Analysis ( PCA ) . The top 30 principal components ( high-loading , based on the associated scree plot ) were then used as input to UMAP for generating either 2D or 3D projections of the data . For , subclustering of pigment cell clusters ( melanophores , iridophores , xanthophores , and pigment progenitors ) , we subsetted the data set and again applied UMAP dimensionality reduction and louvain clustering . To identify genes expressed cell-type specifically , we used the principalGraphTest function in Monocle3 ( v . 2 . 99 . 1 ) with default parameters ( Cao et al . , 2019 ) . This function uses a spatial correlation analysis , the Moran’s I test , to assess spatially restricted gene expression patterns in low dimensional space . We selected markers by optimizing for high specificity , expression levels and effect sizes within clusters ( For extended list of cell-type-specific genes , see Supplementary file 2—Table 1 ) . The top 800 highly dispersed genes ( Supplementary file 2—Table 5 ) within euthyroid pigment cells ( melanophores , xanthophores , iridophores , and pigment progenitors ) were chosen as feature genes to resolve pseudotemporal trajectories using the setOrderingFilter , reduceDimension , and orderCells functions in Monocle ( v2 . 9 . 0 ) using default parameters with the exception of setting max_components = 3 and num_dim = 10 to generate the trajectory in 3D with the top 10 PCs ( high-loading based on scree plot ) during dimensionality reduction . After running trajectory analysis on pigment cells , we used the BEAM function in Monocle ( v . 2 . 9 . 0 ) with default settings ( except , branch_point = 3 ) to determine differentially expressed genes between trajectory branches . To generate the BEAM heatmap for the three pigment cell trajectory branches , we used the plot_multiple_branches_heatmap function with default settings ( except assigning branch 1 , 5 , and six to iridophores , melanophores , and xanthophores , respectively; and num_clusters = 6 ) . Genes were selected by significance levels for the three-branch BEAM analysis with additional significant genes added from the melanophore and iridophore two-branch analysis for more even distribution of genes across lineages ( q < 6 . 0E-11 for all genes , except for pax3a ( starred , q = 0 . 03 ) which is a positive indicator of early pseudotime for all lineages ) . To determine differentially expressed genes over pseudotime that were TH-dependent , we filtered the data set for genes expressed in at least five cells and performed differential expression analysis using a full model of sm . ns ( Pseudotime , df = 3 ) *condition and a reduced model of sm . ns ( Pseudotime , df = 3 ) . Gene sets for signature scores were selected using gene ontology ( terms and gene sets from zfin . org; cell-cycle , unfolded protein response , AP-1 transcription factor complex members ) or manual curation based on literature when required ( carotenoid , pteridine , melanin ) ( see Supplementary file 2—Table 4 ) . Signature scores were calculated by generating z-scores ( using scale ( ) ) of the mean of expression values ( log transformed , size factor normalized ) from genes in a given set . Parametric , non-parametric and multiple logistic regression analyses were performed using JMP 14 . 0 ( SAS Institute , Cary , NC ) or R [version 3 . 5 . 0] ( R Development Core Team , 2017 ) . For parametric analyses , residuals were assessed for normality and homoscedasticity to meet model assumptions and no transformations were found to be warranted . Data is available on GEO via accession GSE131136 . Monocle is available through GitHub ( https://github . com/cole-trapnell-lab/monocle-release . git; Trapnell , 2019 ) .	Hormones control the development of animals from embryos all the way into adulthood . For example , thyroid hormone is needed to transform a tadpole into an adult frog , and it is essential for developing the nervous system and regulating metabolism in countless other adult animals . However , it remains unclear how a single hormone can control such a diverse range of outcomes . One way to learn more about the effects of thyroid hormone during development is to study zebrafish pigmentation . Pigment cells arise from a group of stem cells in the embryo called the neural crest . Two of these pigment cells respond to thyroid hormone in different ways: it causes orange pigment cells called xanthophores to expand in number , and at the same time limits the number of black pigment cells called melanophores . To investigate how thyroid hormone effects the numbers of these pigment cells Saunders et al . mapped the active genes of individual cells derived from the neural crest . Further experiments were then performed on the fish themselves based on these gene activity maps . Comparing fish with and without thyroid hormone showed the hormone actually helps both orange and black pigment cells to mature , but in very different ways . For the orange xanthophores , thyroid hormone drives cells already poised to change into their adult form to acquire orange pigments . For the black melanophores , it causes them to mature into their final non-dividing adult state . This results in xanthophores becoming visible just as the number of melanophores is forced to curtail . Saunders et al . also found the receptor for thyroid hormone acts like a brake for both pigment cells , making sure neither cell type matures in the absence of the hormone . These experiments show how one hormone can independently regulate different cell types as they mature into their adult form . The finding that thyroid hormone limits the growth of melanocytes may explain why people who produce too little thyroid hormone are at greater risk of melanoma – a form of skin cancer that starts in the melanocytes . But more studies are needed to see if thyroid hormone has the same limiting effect on melanocytes in humans .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "genetics", "and", "genomics" ]	2019	Thyroid hormone regulates distinct paths to maturation in pigment cell lineages
Collaboration among the multitude of RNA-binding proteins ( RBPs ) is ubiquitous , yet our understanding of these key regulatory complexes has been limited to single RBPs . We investigated combinatorial translational regulation by Drosophila Pumilio ( Pum ) and Nanos ( Nos ) , which control development , fertility , and neuronal functions . Our results show how the specificity of one RBP ( Pum ) is modulated by cooperative RNA recognition with a second RBP ( Nos ) to synergistically repress mRNAs . Crystal structures of Nos-Pum-RNA complexes reveal that Nos embraces Pum and RNA , contributes sequence-specific contacts , and increases Pum RNA-binding affinity . Nos shifts the recognition sequence and promotes repression complex formation on mRNAs that are not stably bound by Pum alone , explaining the preponderance of sub-optimal Pum sites regulated in vivo . Our results illuminate the molecular mechanism of a regulatory switch controlling crucial gene expression programs , and provide a framework for understanding how the partnering of RBPs evokes changes in binding specificity that underlie regulatory network dynamics . Post-transcriptional gene regulatory mechanisms are widespread and mediated by hundreds of RNA-binding proteins ( RBPs ) that interact dynamically with target mRNAs ( Baltz et al . , 2012; Castello et al . , 2012; Gerstberger et al . , 2014 ) . Understanding how RBPs function together to control the location , timing and level of protein expression is paramount . Decades of research have established the crucial roles of two archetypal RBPs , Drosophila melanogaster Nanos ( Nos ) and Pumilio ( Pum ) , in developmental patterning , fertility , and nervous system development and function . We use this system to explore how two RBPs cooperatively define regulatory networks . Pum is a founding member of the Pum/fem-3 mRNA-binding factor ( FBF ) , or PUF family , of eukaryotic RBPs ( Wickens et al . , 2002 ) . PUF proteins share a sequence-specific RNA-binding domain known as the Pum Homology Domain ( PUM-HD ) ( Barker et al . , 1992; Macdonald , 1992; Wharton et al . , 1998; Zamore et al . , 1997; Zhang et al . , 1997 ) . Crystal structures of PUF proteins have illuminated their unique RNA recognition properties ( Edwards et al . , 2001; Miller et al . , 2008; Qiu et al . , 2012; Wang et al . , 2001 , 2002 , 2009; Wilinski et al . , 2015; Zhu et al . , 2009 ) . Pum binds with high affinity to specific sequences in target mRNAs and represses protein expression and enhances mRNA degradation ( Gerber et al . , 2006; Weidmann et al . , 2012 , 2014; Wharton et al . , 1998; Wreden et al . , 1997; Zamore et al . , 1999 , 1997 ) . A classic example of Pum activity is the establishment of embryonic body pattern through repression of maternal hunchback ( hb ) mRNA , which requires collaboration with Nos ( Lehmann and Nusslein-Volhard , 1987 , 1991; Murata and Wharton , 1995; Sonoda and Wharton , 1999; Wang and Lehmann , 1991 ) . Regulation of hb mRNA by Pum relies on the spatial distribution of Nos , a tandem CCHC Zn finger ( ZF ) RBP ( Barker et al . , 1992; Curtis et al . , 1997; Forbes and Lehmann , 1998 ) . Maternal hb mRNA and Pum protein are distributed throughout the syncytial Drosophila embryo ( Macdonald , 1992; Tautz , 1988 ) , whereas Nos protein forms a gradient emanating from the posterior end ( Wang and Lehmann , 1991 ) . Where their expression overlaps , Pum and Nos together repress hb mRNA ( Barker et al . , 1992; Lehmann and Nusslein-Volhard , 1991; Murata and Wharton , 1995; Sonoda and Wharton , 1999 ) . In the absence of Nos or Pum expression , Hb protein is produced throughout the embryo , and no abdominal segments are formed . Regulation of hb mRNA by Pum and Nos is dependent on two Nanos Response Elements ( NREs ) in the 3´UTR ( Murata and Wharton , 1995; Wharton and Struhl , 1991 ) . Each NRE contains a binding site for Pum with the RNA consensus sequence , 5´-UGUAHAUA ( where H is A , U or C ) , the Pumilio Response Element ( PRE ) . Additional nucleotides in the NRE , located 5´ of the PRE , are functionally important for hb regulation and were proposed to be recognized by Nos ( Edwards et al . , 2001; Sonoda and Wharton , 1999 ) . By itself , Nos appeared to lack RNA sequence specificity ( Curtis et al . , 1997 ) . Instead , Nos binding to hb NREs requires Pum RNA recognition ( Sonoda and Wharton , 1999 ) . Hence , combinatorial hb mRNA repression requires the sequence specificity of Pum and the spatial information provided by the Nos gradient . Nos and Pum also regulate germline and neurological processes ( Forbes and Lehmann , 1998; Mee et al . , 2004; Menon et al . , 2009 , 2004; Ye et al . , 2004 ) . Nos and Pum collaborate to repress expression of Cyclin B mRNA ( CycB ) in primordial germ cells and germline stem cells ( Asaoka-Taguchi et al . , 1999 ) and the sodium channel paralytic ( para ) in the nervous system ( Muraro et al . , 2008 ) . Like hb mRNA , CycB and para mRNAs possess NREs with PRE-like motifs . Furthermore , genome-wide analyses have identified hundreds of Pum-associated mRNAs , suggesting that Pum may play an expansive regulatory role beyond the few validated target RNAs ( Gerber et al . , 2006; Laver et al . , 2015 ) . While collaboration between Nos and Pum is firmly established , the mechanism by which they do so remains to be determined . Here , we report the crystal structures of Nos-Pum-RNA complexes , which reveal that Nos acts as a molecular clamp that embraces both Pum and RNA . The C-terminal region of Pum undergoes conformational changes to make new contacts with the RNA and Nos . We explored the hypothesis that Nos promotes repression by modulating the RNA-binding activity of Pum . We show that Nos enhances the cellular repression activity and in vitro RNA-binding affinity of Pum . Moreover , Nos contacts nucleotides upstream of the PRE . In doing so , Nos alters the specificity of the repression complex and promotes repression of RNAs that are not stably bound by Pum alone . We performed RNA target selection and high-throughput sequencing , which , together with RNA-binding and cellular repression assays , demonstrate that Nos diversifies Pum RNA regulatory networks . We established a cell-based hb reporter mRNA assay , wherein exogenous Nos robustly repressed reporter protein and RNA expression in a manner dependent on the PREs ( Figure 1 ) and Pum ( Weidmann and Goldstrohm , 2012 ) . We used D . mel-2 cells , which do not express Nos and express insufficient Pum to repress the reporter efficiently in the absence of exogenous Nos ( Weidmann and Goldstrohm , 2012 ) , and a Renilla luciferase ( RnLuc ) reporter containing the 3´UTR of hb with two NREs , each of which possesses a PRE ( Figure 1A ) . Halo-tagged , full-length Nos protein , comprising N-terminal , tandem ZF ( Z ) , and C-terminal regions ( NZC , Figure 1B ) , repressed reporter expression 75% relative to a negative control Halo-tag ( Halo ) protein alone ( Figure 1C ) . Mutation of one PRE modestly reduced repression ( Figure 1C ) . In contrast , mutation of both PREs abrogated repression . Nos also reduced the level of reporter mRNA ( Figure 1—figure supplement 1 ) , consistent with enhanced degradation of target mRNAs . Together , these data show that Nos-enhanced , Pum-mediated repression is PRE-dependent , and a single NRE is sufficient to confer regulation . 10 . 7554/eLife . 17096 . 003Figure 1 . The Zn finger and C-terminal regions of Nanos collaborate with Pumilio to repress target protein and mRNA expression . ( A ) Diagram of Renilla Luciferase ( RnLuc ) reporters including the hb 3´UTR . WT PRE1 and PRE2 sequences , located within NRE1 and NRE2 , respectively , and mutant PRE sequences ( mt1 , mt2 , and mt1 , 2 ) are shown . ( B ) Diagram of Nos protein . Amino acid residue boundaries of the N-terminal region ( N ) , central region ( Z , blue ) including ZFs , and C-terminal extension ( C ) are indicated . ( C ) Nos-enhanced repression via the hb PREs . Reporter assays were performed in D . mel-2 cells . Percent repression values are graphed for RnLuc WT and mt hb 3´UTR reporter expression with negative control Halo-tag alone ( Halo ) and full length Halo-Nos test proteins . ( D ) The Nos Z and C-terminal regions together retain efficient repression activity . Percent repression values are graphed for RnLuc hb 3´UTR WT reporter expression with negative control Halo and Halo-Nos variants are shown . For panels C and D , mean and standard error of the mean ( SEM ) values from quadruplicate experiments are shown . Expression of test proteins was visualized by tetramethylrhodamine ( TMR ) fluorescent labeling of the Halo-tag fusion proteins . Statistical analysis of the data is reported in Figure 1―source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 00310 . 7554/eLife . 17096 . 004Figure 1—source data 1 . Values and statistical analysis of luciferase reporter assays . The average value of the relative response ratio of Renilla to Firefly luciferase activities ( Rn/FF ) , average percent repression values ( %Repress ) , and Standard Errors of the Mean ( SEM ) are listed for each condition in the experiments ( from four technical replicates , n = 4 ) . The p-values ( p-val ) resulting from two-tailed t-tests between each measurement and the indicated control are represented in bold ( significant ) or italics ( not significant ) . Experimental values derived from the same experiment are outlined in boxes . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 00410 . 7554/eLife . 17096 . 005Figure 1—figure supplement 1 . Nos reduces hb 3´UTR reporter mRNA level in a PRE-dependent manner . Quantitation of Northern blot detection of the indicated RnLuc hb 3´UTR reporter mRNAs and the 7SL RNA , as a loading control , from total RNA purified from D . mel−2 cells expressing negative control Halo-tag alone ( Halo ) or Halo-Nos and the indicated reporter . For each condition , RNA was analyzed from three individually transfected cell cultures . Mean and SEM values from the triplicate samples are shown . A dotted line indicates the level of WT 3´UTR reporter mRNA in cells transfected with negative control Halo . Statistical analysis of the data is reported in Figure 1―figure supplement 1—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 00510 . 7554/eLife . 17096 . 006Figure 1—figure supplement 1—source data 1 . Values and statistical analysis of Northern blot of luciferase reporter mRNAs . Quantitation of the Northern blot in Figure 1—figure supplement 1 was performed and the average ratio of Renilla luciferase mRNA to 7SL RNA internal control ( Rn/7SL ) was calculated to normalize expression values ( Normed , n = 3 ) . Standard Error of the Mean ( SEM ) is reported for each measurment . The p-values ( p-val ) resulting from two-tailed t-tests between each measurement and the indicated control are represented in bold ( significant ) or italics ( not significant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 006 Using this approach , we performed structure-function analysis of Nos to identify which regions are required for activity . Conserved tandem ZFs are necessary for RNA binding ( Curtis et al . , 1997; Sonoda and Wharton , 1999 ) , and mutations that disrupt either ZF ( C319Y or C354Y ) completely blocked Nos repression activity ( Figure 1D ) . Similarly , a Nos variant lacking the C-terminal region ( NZ ) displayed limited repression activity , 16% repression versus 79% for full-length Nos ( NZC ) . In contrast , a Nos variant lacking the N-terminal region ( ZC ) retained 43% repression . No individual Nos region ( N , Z , or C ) was sufficient for repression . These findings indicate that multiple regions of Nos contribute to enhancing Pum-mediated repression: the ZFs and the C-terminal region are crucial , and the N-terminal region makes a minor contribution . Because the Nos ZC region retained most of the Nos-enhanced repression activity and was amenable to biochemical and structural studies , we focused on determining its mechanism of action . We determined a 3 . 7 Å crystal structure that reveals the molecular architecture of the Nos-Pum-RNA ternary complex ( Table 1 , Figure 2A , and Figure 2—figure supplement 1 ) . Strikingly , the structure illustrates how the Nos tandem ZFs envelop the RNA bases immediately upstream of the PRE and , together with the C-terminal region , embrace both the RNA and Pum ( Figure 2A and B ) . We crystalized Pum and Nos with a hb NRE2 RNA , 5´-AAAUUGUACAUAAGCC ( the core PRE sequence is underlined , and we designate the first U of the core PRE as position +1 and number the four upstream nucleotides as −1 to −4 ) . The ternary complex structure reveals critical protein-protein and protein-RNA interactions between Nos , Pum and RNA . We also determined a 1 . 14 Å crystal structure of a binary complex of Pum and a hb PRE2 RNA ( 5´-UGUACAUA ) ( Figure 2C ) , which exemplifies the modular 1 repeat:1 RNA base recognition of classical PUF proteins . For example , the UGU motif of the PRE is recognized by repeats R8 through R6 . This high-resolution structure , combined with previous crystal structures of Drosophila Pum ( Edwards et al . , 2001 ) and zebrafish Nos ZFs ( Hashimoto et al . , 2010 ) , allowed us to build and refine the ternary complex model . 10 . 7554/eLife . 17096 . 007Figure 2 . Nanos embraces Pumilio and hb RNA to stabilize the ternary complex . ( A ) Crystal structure of a Drosophila Nos-Pum-hb NRE2 RNA ternary complex . Pum is shown as a ribbon diagram ( yellow ) , Nos is shown as a surface representation ( blue ) and hb NRE2 RNA is shown as a stick model colored by atom type . N- and C-termini of proteins and 5´ and 3´ ends of the RNA are labeled . Pum repeats R1 to R8 and pseudo repeats R1´ and R8´ are indicated . ( B ) View of the Nos-Pum-hb NRE2 RNA ternary complex down the long axis of Pum . Pum ( yellow ) and Nos ( blue ) are shown as ribbon diagrams , and hb NRE2 is shown as a cartoon backbone with RNA bases . Zn atoms are shown as grey spheres . The C-terminus of Pum was truncated to L1411 to allow visualization of Nos-RNA interaction . ( C ) Ribbon diagram of a crystal structure of a Drosophila Pum binary complex with the PRE2 from hb NRE2 . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 00710 . 7554/eLife . 17096 . 008Figure 2—figure supplement 1 . Representative electron density map of the Nos-Pum-hb NRE2 complex . A 2Fo-Fc composite omit map contoured at 1 . 2 σ is shown superimposed with the Nos-Pum-hb NRE2 RNA structure at the binding pocket for the −1U . Carbon atoms are blue for Nos and grey for RNA . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 00810 . 7554/eLife . 17096 . 009Figure 2—figure supplement 2 . Nanos induces localized structural changes in Pum upon formation of the Nos-Pum-hb NRE2 RNA ternary complex . A loop between repeats R7-R8 of Pum ( R7-R8 loop ) undergoes conformational changes to promote Nos-Pum interactions . The terminal helix of Repeat 8′ ( R8′ ) also changes conformation to interact with the RNA . Crystal structures of Nos – Pum – hb NRE2 RNA ternary and Pum – hb PRE2 RNA binary complexes are shown side-by-side . Pum is shown as a ribbon diagram for both the ternary ( yellow ) and binary ( tan ) complexes . Regions that undergo structural changes are shown in magenta and C-terminal residues of Pum that are disordered in the ternary complex are colored turquoise . Nos is shown as a surface representation , colored blue . RNA is pictured as a cartoon backbone with grey PRE and brown Nanos Binding Site regions . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 00910 . 7554/eLife . 17096 . 010Figure 2—figure supplement 3 . Crystal structure of Nos – Pum – hb NRE2 RNA ternary complex highlights key Pum-RNA and Nos-Pum contacts . The C-terminal helix of Pum unfolds to promote ternary complex formation , forming additional contacts with the RNA . Important Nos and Pum residues are indicated , although electron density for most side chains is incomplete . The R7-R8 loop region of Pum , containing the indicated residue F1367 , changes conformation to contact the Nos C-terminal helix , including residue M378 . Pum is shown as a ribbon diagram ( yellow ) . Regions that undergo structural changes are shown in magenta . Nos is shown as a ribbon diagram , colored blue . RNA is pictured as a cartoon backbone with grey PRE and brown Nanos Binding Site regions . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 01010 . 7554/eLife . 17096 . 011Table 1 . Data collection and refinement statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 011Pum-RNAPum-Nos-hb RNAPum-Nos-cycB RNAPDB ID5KLA5KL15KL8Data collectionSpace groupC2P6522P6522Cell dimensionsa , b , c ( Å ) 194 . 9 , 29 . 5 , 62 . 0137 . 0 , 137 . 0 , 221 . 4135 . 1 , 135 . 1 , 220 . 4a , b , g ( ° ) 90 . 0 , 101 . 2 , 90 . 090 . 0 , 90 . 0 , 120 . 090 . 0 , 90 . 0 , 120 . 0Resolution ( Å ) 50-1 . 14 ( 1 . 16-1 . 14 ) 50-3 . 70 ( 3 . 83-3 . 70 ) 50-4 . 00 ( 4 . 12-4 . 00 ) Rsym0 . 045 ( 0 . 387 ) 0 . 128 ( 0 . 747 ) 0 . 143 ( 0 . 779 ) I / σI36 . 9 ( 2 . 7 ) 19 . 1 ( 2 . 8 ) 13 . 0 ( 3 . 6 ) Completeness ( % ) 99 . 7 ( 97 . 4 ) 99 . 3 ( 93 . 2 ) 99 . 3 ( 100 . 0 ) Redundancy4 . 2 ( 2 . 4 ) 11 . 3 ( 11 . 0 ) 8 . 9 ( 8 . 7 ) RefinementResolution ( Å ) 34 . 46 - 1 . 1438 . 3 - 3 . 7039 . 0 - 4 . 00No . reflections1270771356210715Rwork / Rfree ( % ) 16 . 0 / 17 . 426 . 4 / 30 . 028 . 3 / 31 . 2No . atomsProtein553231943021RNA253252226Water / Solvent40100B-factorsProtein29 . 0175 . 5208 . 6RNA20 . 5150 . 4183 . 4Water / Solvent34 . 9--R . m . s deviationsBond lengths ( Å ) 0 . 0070 . 0030 . 002Bond angles ( ° ) 0 . 9500 . 6050 . 508*Values in parentheses are for highest-resolution shell . Comparison of the ternary and binary complexes reveals that the addition of Nos and the upstream nucleotides induces localized conformational changes in Pum that promote Nos-Pum interaction and binding of Pum to RNA upstream of the core PRE site . While the overall structure of Pum in the ternary complex is similar to that in the binary complex ( RMSD of 1 . 2 Å over 324 Pum Cα atoms ) , the C-terminal region of Pum undergoes notable changes . Loop residues between repeats R7 and R8 rearrange in the ternary complex to promote interaction of F1367Pum with the C-terminal α helix of Nos ( Figure 2—figure supplement 2 ) . In addition , the C-terminal α helix of Pum ( helix α2 of repeat R8´ ) unfolds to promote interaction of residues with the upstream RNA backbone ( Figure 2—figure supplement 2 and Figure 2—figure supplement 3 ) . Since the upstream nucleotides were not present in the binary complex , the change in the structure of the C-terminal region of Pum may be induced by the presence of the upstream RNA and/or Nos . With these conformational changes , Pum and Nos interact with one another and together recognize RNA immediately 5´ of the core PRE motif . Using an electrophoretic mobility shift assay ( EMSA ) , we demonstrated that Nos ZC binds stably and tightly to a Pum-hb NRE2 RNA complex and cooperatively strengthens the binding affinity of Pum for hb RNA . The RNA-binding domain of Pum ( PUM-HD ) bound hb NRE2 RNA , and addition of equimolar Nos ZC to the binding reaction further retarded the hb NRE2 RNA mobility , indicating formation of a Nos-Pum-RNA ternary complex ( Figure 3A , B , and Figure 3―figure supplement 1 ) . Disruption of either ZF of Nos ( C319Y or C354Y ) eliminated ternary complex formation ( Figure 3B ) . In addition , the Pum-hb NRE2 RNA interaction is essential for Nos binding , as RNA-binding deficient Pum mutR7 did not support a ternary complex ( Figure 3B ) . Nos did not shift hb NRE2 RNA on its own ( Figure 3B ) , even at protein concentrations of one micromolar ( Figure 3C ) . We applied the EMSA quantitatively and found that Nos binds with high affinity to the Pum-hb NRE2 RNA complex ( Figure 3C and D ) and increases Pum binding affinity for hb NRE2 RNA by 3-fold ( Figure 3E and F ) . These data establish that the requirements for cooperative assembly of the Nos-Pum-hb NRE2 complex in vitro mirror those for Nos-enhanced , Pum-mediated repression in cells and embryos , and therefore complex formation reflects repression activity . 10 . 7554/eLife . 17096 . 012Figure 3 . Nanos increases the binding affinity of Pumilio for hunchback RNA . ( A ) Diagram of recombinant proteins and RNA ligand used for EMSAs . The amino acid residue boundaries of the Pum RNA-binding domain ( PUM-HD , yellow ) are represented relative to full-length Pum . For simplicity , we refer to the PUM-HD as Pum . The Z and C regions are shown in the context of full-length Nos . Dashed lines outline regions excluded from the recombinant proteins used for EMSAs . The RNA sequence of the Cy5-labeled hb NRE2 is shown with the PRE sequence highlighted ( yellow ) . ( B ) A representative EMSA with hb NRE2 RNA is shown . Nos and Pum test protein concentrations are indicated above the gel . ( C ) A representative EMSA measuring binding of Nos to the Pum – hb NRE2 complex . Nos was titrated into binding reactions with a constant concentration of Pum ( 100 nM ) . The mean observed dissociation constant ( Kd ) with standard deviation ( SD ) from triplicate experiments is shown below the gel . ( D ) Graph of fraction bound for Nos-Pum-hb NRE2 complex in response to Nos concentration . Mean and standard error of the mean ( SEM ) values from triplicate EMSAs are plotted . ( E ) Representative EMSAs measuring binding to hb NRE2 RNA , performed at the same time under identical conditions , titrating Pum in the presence or absence of Nos . The mean Kd with SD from triplicate experiments is shown below the gel . ( F ) Graph of fraction bound of Nos-Pum-hb NRE2 and Pum – hb NRE2 complexes in response to Pum concentration . Mean and SEM values from triplicate EMSAs are plotted . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 01210 . 7554/eLife . 17096 . 013Figure 3—figure supplement 1 . Recombinant purified Pum and Nos test proteins . Coomassie blue-stained SDS-polyacrylamide gel loaded with equivalent amounts of purified recombinant Pum and Nos WT and mutant test proteins used for the EMSAs . Molecular weights of protein markers are shown on the left . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 013 The protein-protein interactions observed between Nos and Pum are focused between the C-terminal end of Nos ( I376Nos to E385Nos ) and the non-RNA-binding convex surface of Pum in repeats R7 and R8 ( Figure 4A and B ) . For example , the side chain of Q1337Pum is within hydrogen bonding distance of main chain N and O atoms of I376Nos . In addition , F1367Pum forms part of a hydrophobic pocket that interacts with M378Nos in the Nos C-terminus ( Figure 4B ) . 10 . 7554/eLife . 17096 . 014Figure 4 . Interactions between Nanos and Pumilio are necessary for repression . ( A ) Diagrams of Nos and Pum proteins highlighting residues involved in protein-protein interaction . The amino acid sequence of the C-terminal region of Nos is shown . Residues 376–382 that are deleted in the nosL7 fly mutant , a strong allele for defective abdominal segmentation , are colored red , and residues 383–393 are colored purple . Residues that form the Nos C-terminal α helix are in boldface . ( B ) View of the interface between Nos ( blue with red and purple C-terminal region ) and Pum ( yellow ) . Interacting residues in Nos and Pum are shown in stick representation , and the hb NRE2 RNA is shown as a cartoon representation . ( C and D ) Percent repression values are graphed for the RnLuc WT hb 3´UTR reporter with negative control Halo-tag alone ( Halo ) and variants of Halo-Nos test proteins . Nos test proteins included full-length Nos ( NZC ) , a truncation of the C-terminal region ( NZ ) , deletions ( Δ ) of the indicated amino acids and specific amino acid substitutions in the context of full-length Nos . Labels are colored as in panels A and B . Mean and SEM values from quadruplicate samples are shown . Expression of test proteins was visualized by TMR fluorescent labeling or anti-V5 western blotting of the Halo-tag fusion proteins . Statistical analysis of the data is reported in Figure 4―source data 1 . ( E and F ) Representative EMSAs comparing ternary complex formation by WT Pum or the mutant F1367S Pum ( panel D ) or the mutant Q1337A Pum ( panel E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 01410 . 7554/eLife . 17096 . 015Figure 4—source data 1 . Values and statistical analysis of luciferase reporter assays . The average value of the relative response ratio of Renilla to Firefly luciferase activities ( Rn/FF ) , average percent repression values ( %Repress ) , and Standard Errors of the Mean ( SEM ) are listed for each condition in the experiments ( from four technical replicates , n = 4 ) . The p-values ( p-val ) resulting from two-tailed t-tests between each measurement and the indicated control are represented in bold ( significant ) or italics ( not significant ) . Experimental values derived from the same experiment are outlined in boxes . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 015 To examine the roles of the interaction between Nos and Pum for repression activity and complex formation , we measured the effects of targeted deletions of Nos on repression activity and found that interactions between the Nos C-terminal α helix and Pum are critical for repression . Since deletion of the C-terminal region severely limited Nos repression of the hb 3´UTR reporter in cells ( Figure 1D ) , we probed this interaction more precisely . We tested a Δ376–382 deletion , which eliminated much of the Pum-binding interface and corresponds to the lesion in the nosL7 allele that disrupts Nos function in vivo ( Curtis et al . , 1997 ) . This deletion impaired Nos repression activity to a similar extent as deletion of the entire C-terminal region: 27% repression compared to 14% repression for Nos NZ ( Figure 4C ) . Since our crystal structure indicates that the deleted region includes part of the C-terminal helix that interacts with Pum , it is possible that the protein , although expressed , could be incorrectly folded . We further examined this region by introducing single amino acid substitutions , including the I376Nos and M378Nos residues that contact Pum in the structure ( Figure 4B ) . Repression activity of Nos I376A was diminished to 46% , and more so for Nos M378A ( 29% ) , whereas Nos I382A caused a modest decrease to 60% , relative to 70% repression by wild type Nos ( Figure 4D ) . A Δ383–393 deletion , which removed the final three ordered residues in the structure and eight subsequent residues , also diminished repression activity to 43% ( Figure 4C ) . In contrast , a Δ394–401 deletion retained full repression activity ( Figure 4C ) . These C-terminal eight residues of Nos were disordered in the ternary complex structure , and therefore did not contact Pum . These results confirm the functional importance of the observed protein contacts between Pum and Nos for regulation in cells . We also found that single amino acid substitutions in Pum disrupt formation of the repression complex . No ternary complex was formed with Pum F1367S ( Figure 4E ) , an R7-R8 loop mutant that binds RNA , but does not interact with Nos in a yeast 3-hybrid assay ( Edwards et al . , 2001 ) or respond to Nos in cells ( Weidmann and Goldstrohm , 2012 ) . Similarly , a Q1337APum mutation eliminated ternary complex formation ( Figure 4F ) . Importantly , both Pum mutants retained the ability to bind to hb NRE2 RNA ( Figure 4E and F ) . Thus , Nos must interact with both repeat R7 and the R7-R8 loop of Pum to form a stable ternary complex . The crystal structure of the Nos-Pum-hb RNA complex reveals that Nos binds to three nucleotides upstream of the core PRE when it joins the ternary complex , and we find that repression activity is highly sensitive to mutation of the interface . The first base upstream of the PRE , -1U , is inserted into a hydrophobic binding pocket formed by F321Nos , T366Nos , and Y369Nos ( Figure 5A and Figure 2―figure supplement 1 ) . The O4 atom of −1U is near the main chain N atom of T366Nos . Nos also contacts the bases and backbone atoms of −2A and −3A ( Figure 5B ) . Three residues within the rearranged C-terminal region of Pum , T1415Pum , K1377Pum , and K1413Pum , appear to approach the phosphate groups of −2A , −3A , and −4A , respectively ( Figure 2―figure supplement 1 and Figure 2―figure supplement 3 ) . Interaction of Nos and Pum with the RNA nucleotides upstream of the PRE explains how Nos strengthens the overall ternary complex . 10 . 7554/eLife . 17096 . 016Figure 5 . Nanos Zn finger interaction with RNA extends the RNA-binding site and is critical for repression . ( A ) Interaction of Nos ZFs with the -1U of hb NRE2 . In addition to the base contacts noted in the text , the OH group of Y369Nos and the NH2 group of K368Nos interact with the phosphate group of −1U . ( B ) Interaction of Nos ZFs with the −2A and −3A nucleotides . In panels A and B , important interactions between nucleotide and amino acid residues are shown . Zn atoms are shown as grey spheres with coordination by CCHC residues indicated by yellow dashed lines . ( C ) Percent repression values are graphed for the RnLuc WT hb 3´UTR reporter expression with negative control Halo-tag alone ( Halo ) and Halo-tag fusions of WT Nos or mutant Nos . Mutated residues are shown in panels A and B . Protein expression was confirmed by western blotting for the V5 epitope tag on each test protein . Statistical analysis of the data is reported in Figure 5―source data 1 . ( D ) Sequences of hb NRE2 derivatives tested in EMSA ( panel E ) and reporter expression assay ( panel F ) . The PRE core and the Nos binding site ( NBS ) , derived from the crystal structure , are colored yellow and blue , respectively . Nucleotide changes in each RNA relative to the WT hb NRE2 sequence are marked by red lowercase letters . ( E ) Representative EMSA measuring ternary complex formation using indicated combinations of Nos and Pum with the RNA ligands shown in panel D . ( F ) Percent repression values for RnLuc reporters bearing a minimal 3´UTR containing a single hb NRE2 element or its mutant variants ( panel D ) with Halo or Halo-Nos are shown . Expression of test proteins was visualized by TMR fluorescent labeling of the Halo-tag fusion proteins . For panels C and F , mean and SEM values from quadruplicate samples are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 01610 . 7554/eLife . 17096 . 017Figure 5—source data 1 . Values and statistical analysis of luciferase reporter assays . The average value of the relative response ratio of Renilla to Firefly luciferase activities ( Rn/FF ) , average percent repression values ( %Repress ) , and Standard Errors of the Mean ( SEM ) are listed for each condition in the experiments ( from four technical replicates , n = 4 ) . The p-values ( p-val ) resulting from two-tailed t-tests between each measurement and the indicated control are represented in bold ( significant ) or italics ( not significant ) . Experimental values derived from the same experiment are outlined in boxes . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 017 To determine the importance of Nos-RNA interactions , we measured the effect of single amino acid substitutions on cellular repression activity ( Figure 5C ) . Individual mutations that disrupted the hydrophobic binding pocket for the −1U base ( F321A , T366A , and Y369A ) abrogated repression activity . Another mutant , K368Q , designed to eliminate a salt bridge interaction with the phosphate group of −1U , reduced repression activity to 32% vs 75% for WT protein . Mutations targeting interactions with nucleotides−2A and −3A ( N325A and Y352A ) had smaller , but measurable effects on repression activity . The effects of these single residue substitutions indicate the importance of Nos recognition of the −1 nucleotide and interactions with other upstream NRE nucleotides . We next investigated whether the sequence of the upstream nucleotides ( defined structurally as the Nos binding site , NBS ) is important for ternary complex formation and repression activity . Substitution of both the −1 and −2 positions of hb NRE2 with cytosine had been shown previously to disrupt abdominal segmentation , but did not affect Pum RNA association ( Murata and Wharton , 1995 ) , so we designed hb NRE2 RNAs that substituted cytosine bases at either the −1 or −2 position ( Figure 5D , −1C and −2C ) . Neither mutation hindered RNA binding by Pum , but both mutations blocked ternary complex formation ( Figure 5E ) . We then probed whether the cytosine substitutions in the NBS affect mRNA regulation in cells using reporters bearing a 3´UTR with a single hb NRE ( RnLuc 1x hb NRE2 ) . Expression of full-length Nos resulted in 50% repression of WT reporter activity compared to the negative control Halo-tag protein alone ( Figure 5F ) , similar to a mutant hb 3´UTR reporter with only a functional NRE2 sequence ( Figure 1C , mt1 ) . In contrast , Nos did not repress mutant −1C or −2C reporters ( Figure 5F ) , consistent with disruption of ternary complex formation by these substitutions . Thus , the identities of the nucleotides in the NBS are critical for repression activity . Although we did not observe sequence-specific contacts to the −2A base in our crystal structure of the ternary complex , this likely reflects the modest 3 . 7 Å resolution that was not sufficient to resolve all direct contacts or identify water molecules that may mediate protein-RNA interaction . Given the importance of the NBS sequence for Nos-enhanced Pum regulation , we examined whether differences in natural Nos-Pum mRNA target sequences affect regulatory activity . Using EMSAs , we found that Nos induced formation of a Nos-Pum-RNA ternary complex , even when Pum alone did not bind stably to RNA . The Cyclin B ( CycB ) NRE contains a PRE that diverges from the consensus with uracils in place of adenines at positions +6 and +8 , and it has a different NBS sequence ( Figure 6A ) . Pum alone did not form a stable complex with CycB NRE RNA ( Figure 6A , left , and 6B ) , but remarkably , addition of Nos ZC resulted in ternary complex formation with an apparent Kd of 12 nM ( Figure 6A , right , and 6B ) , similar to the 8 . 7 nM Kd for hb NRE2 . We next tested complex formation with sequences from the hb NRE1 ( Figure 6―figure supplement 1 ) and bicoid NRE ( Figure 6—figure supplement 2 ) , which match the PRE consensus sequence , but have a uracil or guanine , respectively , at the +5 position . For each of these sequences , Pum unexpectedly did not form a stable binary complex , but as with the CycB sequence , Nos induced ternary complex formation . 10 . 7554/eLife . 17096 . 018Figure 6 . Nanos alters Pumilio RNA-binding specificity . ( A ) A representative EMSA with increasing concentrations of Pum in the presence or absence of Nos , performed with radiolabeled CycB NRE RNA shown at the top . The PRE and the NBS are highlighted in yellow and blue , respectively . The mean observed Kd values with SD from triplicate experiments are shown below the gel . ( B ) Graph of fraction bound for complexes in panel A . Mean and SEM values from triplicate EMSAs are plotted . ( C ) Superposition of crystal structures of ternary complexes of Pum and Nos with hb NRE2 RNA ( yellow protein with orange RNA ) and CycB NRE RNA ( light blue protein and RNA ) . Nos is represented as a blue ribbon diagram . Zn atoms are shown as grey spheres . ( D ) Interaction of Pum with nucleotides 4–8 of hb NRE2 RNA ternary complex . ( E ) Interaction of Pum with nucleotides 4–7 of CycB NRE RNA within the Nos-Pum-CycB NRE RNA ternary complex . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 01810 . 7554/eLife . 17096 . 019Figure 6—figure supplement 1 . Nos promotes ternary complex formation with Pum and the hb NRE1 RNA . Representative EMSA with increasing concentrations of Pum in the presence or absence of Nos . Radiolabeled RNA sequence used is shown at the top . The PRE and NBS sequences are highlighted in yellow and blue , respectively . The mean observed dissociation constant ( Kd ) with SD from triplicate experiments is shown below the gel . Graphs of the fraction bound of Nos-Pum-NRE and Pum–NRE complexes in response to titration of Pum . Mean and SEM values from triplicate EMSA experiments are plotted . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 01910 . 7554/eLife . 17096 . 020Figure 6—figure supplement 2 . Nos promotes ternary complex formation with Pum and the bcd NRE RNA . Representative EMSA with increasing concentrations of Pum in the presence or absence of Nos . Radiolabeled RNA sequence used is shown at the top . The PRE and NBS sequences are highlighted in yellow and blue , respectively . The mean observed dissociation constant ( Kd ) with SD from triplicate experiments is shown below the gel . Graphs of the fraction bound of Nos-Pum-NRE and Pum–NRE complexes in response to titration of Pum . Mean and SEM values from triplicate EMSA experiments are plotted . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 02010 . 7554/eLife . 17096 . 021Figure 6—figure supplement 3 . Nos promotes ternary complex formation with Pum and the hb NRE2 +3G RNA . Representative EMSA with increasing concentrations of Pum in the presence or absence of Nos . Radiolabeled RNA sequence used is shown at the top . The PRE and NBS sequences are highlighted in yellow and blue , respectively , whereas mutated positions are in lowercase red letters . The mean observed dissociation constant ( Kd ) with SD from triplicate experiments is shown below the gel . Graphs of the fraction bound of Nos-Pum-NRE and Pum–NRE complexes in response to titration of Pum . Mean and SEM values from triplicate EMSA experiments are plotted . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 02110 . 7554/eLife . 17096 . 022Figure 6—figure supplement 4 . Nos and Pum do not bind the mutant hb NRE2 RNA . Representative EMSAs with increasing concentrations of Pum in the presence or absence of Nos . Radiolabeled RNA sequence used is shown at the top . The PRE and NBS sequences are highlighted in yellow and blue , respectively , whereas mutated positions are in lowercase red letters . N/D indicates that a Kd was not determined . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 02210 . 7554/eLife . 17096 . 023Figure 6—figure supplement 5 . hb and CycB RNAs form different conformations in complex with Pum and Nos . Crystal structures of hb and CycB RNAs from ternary complexes with Pum and Nos are shown superimposed with 2Fo-Fc electron density maps contoured at 1 . 2 σ . The models fit well with their respective maps ( hb RNA:hb map or CycB RNA:CycB map ) , but fit poorly when superimposed with the alternate map ( hb RNA:CycB map or CycB RNA:hb map ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 023 We then investigated the effect of specific RNA mutations on formation of the Nos-Pum-RNA ternary complex , focusing on the conserved UGU motif that is the hallmark of PUF protein binding sites . As expected , mutating +3U to G ( hb PRE2 +3G ) prevented stable binding of Pum alone ( Figure 6―figure supplement 3 ) . Yet , similar to the Cyclin B and bicoid NREs , addition of Nos ZC conferred formation of a ternary complex with a Kd of 59 . 9 nM . In contrast , mutating UGU to ACA ( hb PRE2 ACA ) abolished both Pum binding and ternary complex formation ( Figure 6―figure supplement 4 ) . Taken together , our results demonstrate that Nos can stabilize binding of Pum to RNAs containing a wider range of consensus or divergent NREs , but cannot overcome complete disruption of the UGU trinucleotide sequence . To gain molecular insight into how Nos enhances Pum recognition of different NRE sequences , we determined a 4 . 0 Å crystal structure of a ternary complex with a CycB NRE RNA ( 5´-UAUUUGUAAUUUAU , core PRE underlined ) ( Table 1 ) . The protein structures are essentially unchanged compared to the complex with hb NRE2 RNA; however , differences in Pum binding to bases +5 to +8 result in distinct conformations of hb and CycB RNAs in this region ( Figure 6C and Figure 6―figure supplement 5 ) . Pum binds to the hb PRE2 using the base-omission mode observed for human Pumilio1 ( Lu and Hall , 2011 ) , where bases +4 and +5 stack directly ( Figure 6D ) . In contrast , Pum appears to bind to the CycB PRE using the 1 repeat:1 RNA base PUF recognition mode with R1271Pum sandwiched between bases +4 and +5 ( Figure 6E ) . Pum binds specifically to the 3´ AUA sequence at positions +6 to +8 of hb PRE2 RNA ( Figure 6D ) . However , for the CycB RNA , recognition of +6U is suboptimal and the +8U nucleotide is disordered in the ternary complex ( Figure 6E ) . Thus , Nos stabilizes binding of Pum to RNAs that do not match the PRE consensus sequence in the 3´ half , reducing Pum specificity to allow regulation of a broader range of mRNA targets than Pum alone . To define changes in Pum specificity induced by Nos , we examined sequence preferences using SEQRS ( in vitro selection , high-throughput sequencing of RNA , and sequence specificity landscapes ) ( Campbell et al . , 2012a ) ( Figure 7A ) . Pum alone reproducibly enriched a motif matching the PRE consensus ( Figure 7B , Figure 7―figure supplement 1 ) . Strikingly , addition of Nos to immobilized Pum enriched A/U-rich sequences upstream of the 5´-UGUA of the PRE ( Figure 7C , Figure 7―figure supplement 2 , 3 ) , consistent with Nos recognition of the NBS . Sequence selection at the 3´ end of the Pum motif weakened . Comparison of control SEQRS analyses demonstrated specificity of the interactions . The Pum motif was highly enriched by wild-type Pum alone and , to a lesser degree , in the presence of Nos ( Figure 7D ) , the Nos-Pum motif was highly enriched by the wild-type ternary complex , whereas Nos alone or the RNA-binding defective Pum did not enrich either motif . 10 . 7554/eLife . 17096 . 024Figure 7 . SEQRS analysis of Nos and Pum reveals specificity of RNA-binding activities . ( A ) Diagram of the SEQRS procedure . ( B ) Motif from SEQRS analysis of Pum . ( C ) Motif from SEQRS analysis of Nos-Pum complex . ( D ) The Nos-Pum and Pum motifs are preferentially enriched by their corresponding samples relative to three negative control conditions . These controls are Nos alone , an RNA-binding defective Pum mutR7 , or Nos combined with Pum mutR7 . Sequences are reported in Figure 7—source data 2 . ( E ) Enrichment of Pum and Nos-Pum binding sites in 3´UTRs bound by Pum in vivo ( Gerber et al . , 2006 ) , relative to 3´UTRs of all annotated Drosophila mRNAs . P values were determined using a chi-squared test . Test values are provided in Figure 7—source data 2 . ( F ) Gene ontology enrichment of target mRNAs from Drosophila adults or embryos with Nos-Pum or Pum motifs . The ten most significantly enriched terms are shown for each category of target mRNA , ranked according to P-values . Complete tables of the gene ontology enrichment analysis is provide in Figure 7—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 02410 . 7554/eLife . 17096 . 025Figure 7—source data 1 . Related to Figure 7F . Gene ontology enrichment results for Pum targets from Drosophila adults and embryos that contain SEQRS-derived Pum motif , Nos-Pum motif , or both motifs . ( A ) Target gene lists used in gene ontology analysis . ( B ) Gene ontology enrichment analysis of target mRNAs from adults that contain the Nos-Pum motif . ( C ) Gene ontology enrichment analysis of target mRNAs from adults that contain the Pum motif . ( D ) Gene ontology enrichment analysis of target mRNAs from adults that contain the Nos-Pum and Pum motif . No categories were significantly enriched . ( E ) Gene ontology enrichment analysis of target mRNAs from embryos that contain the Nos-Pum motif . ( F ) Gene ontology enrichment analysis of target mRNAs from embryos that contain the Pum motif . No categories were significantly enriched . ( G ) Gene ontology enrichment analysis of target mRNAs from embryos that contain both the Nos-Pum and Pum motif . No categories were significantly enriched . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 02510 . 7554/eLife . 17096 . 026Figure 7—source data 2 . Related to Figure 7A–E . SEQRS sequences for Pum and Nos-Pum ternary complex and statistical analysis of motif enrichment in target mRNA 3´UTRs . ( A ) 20-mer sequences corresponding to the random region of the SEQRS library are reported for two replicates designated as A or B . ( B ) Top 100 enriched sequences for Pum and Nos-Pum complex . ( C ) Motif enrichment data and statistics for Pumilio target mRNAs from Drosophila embryos and adults based on Gerber et al , 2006 . Sequences were obtained following five rounds of selection . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 02610 . 7554/eLife . 17096 . 027Figure 7—figure supplement 1 . Comparison of the reproducibility of two replicates of Pum SEQRS . Scatter plots of all possible 8-mer sequences following five rounds of selection . R2 values are based on linear regression analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 02710 . 7554/eLife . 17096 . 028Figure 7—figure supplement 2 . Comparison of sequences selected in SEQRS for Pum and the Nos-Pum complex . Scatter plots of all possible 8-mer sequences following five rounds of selection . Orange data points indicate sequences corresponding to ( a ) hb NRE1 , ( b ) bcd NRE , ( c ) CycB NRE , and ( d ) the hb NRE2 . R2 values are based on linear regression analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 02810 . 7554/eLife . 17096 . 029Figure 7—figure supplement 3 . Analysis of upstream nucleotides enriched in SEQRS by Nos in the ternary complex relative to Pum alone . Relative enrichment values for sequences matching the pattern 5´-NNNNUGUA for the Nos-Pum complex were calculated after subtraction of the sequences bound by Pum alone . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 02910 . 7554/eLife . 17096 . 030Figure 7—figure supplement 4 . Venn diagrams reveal differences in the extent of motif overlap in Pum bound transcripts in embryo ( above ) versus adult ( below ) ( Gerber et al . , 2006 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 030 We examined whether the Nos-Pum motif was enriched in Pum target mRNAs identified from Drosophila embryos and adults ( Gerber et al . , 2006 ) and observed enrichment of the Nos-Pum motif and , to a lesser degree , the Pum motif , in targets from both embryos and adults ( Figure 7E , Figure 7―figure supplement 4 ) . More Pum target mRNAs bear the Nos-Pum motif than the Pum motif , indicating that Nos expands the range of mRNAs regulated by Pum . Notably , few mRNAs have consensus PREs with upstream NBS motifs ( i . e . both Nos-Pum and Pum motifs ) , such as hb mRNA . Most mRNAs with Pum motifs lack the upstream NBS , suggesting that they may be targeted by Pum alone or with other partners . Finally , we performed gene ontology analyses of target mRNAs bearing Pum or Nos-Pum motifs and observed that significantly enriched terms match the known functions of Nos and Pum , including body pattern formation and germline development , and also suggest new collaborative functions including regulation of cell division , receptor protein signaling , and cell fate determination ( Figure 7F ) . To evaluate the ability of Nos to enhance repression of mRNA targets with different NREs , we measured repression of reporter mRNAs containing the NREs for which we had examined ternary complex formation ( Figure 8A ) . Because hb mRNA responds to the Nos concentration gradient in the Drosophila embryo , we varied the amounts of transfected Nos expression vector to produce different levels of Nos protein ( Figure 8B ) . We found that Nos-enhanced repression of the 1x hb NRE2 reporter was dose dependent , increasing from 12% with 1 ng of transfected Nos expression vector to 68% with 200 ng ( Figure 8C ) . Nos also elicited dose-dependent repression of reporters bearing the 1x hb NRE1 or bcd NRE , which were not stably bound by Pum alone , but supported ternary complex formation . In contrast , the Nos gradient did not enhance repression of reporters bearing the 1x CycB NRE or 1x hb NRE2 +3G mutant , relative to the background repression of a control reporter lacking a PRE ( MCS ) or the 1x hb NRE2 ACA mutant that abrogated Pum binding and ternary complex formation . Background repression of the hb NRE2 ACA reporter originating from binding to degenerate motifs in the RnLuc transcript may have limited the sensitivity of the 1x NRE reporters . 10 . 7554/eLife . 17096 . 031Figure 8 . Nanos expands the Pumilio mRNA target repertoire in cells . ( A ) Diagram of the RnLuc 1x NRE reporters with minimal 3´UTRs containing WT or mutant ( +3G and ACA ) hb NRE2 sequences , the hb NRE1 sequence , or NRE sequences from the Cyclin B ( CycB ) and bicoid ( bcd ) mRNAs . The PRE sequence is yellow , whereas mutated positions are in lowercase red letters . ( B ) Dose-dependent expression of Halo-tag ( Halo ) and Halo-Nos in cells from transfected Nos plasmid was detected by fluorescent labeling with TMR ( top panel ) and western blotting with anti-V5 antibody ( middle panel ) . As a loading control , actin was detected by western blotting ( bottom panel ) . ( C ) Percent repression values are graphed for the variant RnLuc 1x NRE reporters with increasing amounts of transfected Nos or negative control Halo . ( D ) Percent repression values are graphed for the indicated RnLuc 3x NRE reporters with increasing amounts of transfected Nos expression plasmid . For Panels C and D , mean and SEM values from quadruplicate samples are shown . Statistical analysis of the data is reported in Figure 8―source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 03110 . 7554/eLife . 17096 . 032Figure 8—source data 1 . Values and statistical analysis of luciferase reporter assays . The average value of the relative response ratio of Renilla to Firefly luciferase activities ( Rn/FF ) , average percent repression values ( %Repress ) , and Standard Errors of the Mean ( SEM ) are listed for each condition in the experiments ( from 4 technical replicates , n = 4 ) . The p-values ( p-val ) resulting from two-tailed t-tests between each measurement and the indicated control are represented in bold ( significant ) or italics ( not significant ) . Experimental values derived from the same experiment are outlined in boxes . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 032 To improve the signal-to-noise ratio , we increased the number of NREs in each reporter to three adjacent sites , which dramatically increased cellular repression with lower amounts of Nos expression: 10 ng of transfected Nos expression vector resulted in 79% repression for the 3x hb NRE2 reporter ( Figure 8D ) versus 45% repression for the 1x hb NRE2 reporter ( Figure 8C ) . Background repression of the 3x hb NRE2 ACA mutant remained low . Nos-enhanced repression of the 3x CycB NRE and mutant hb NRE2 +3G reporters was dose-dependent and significantly above background ( Figure 8D ) . Thus , Nos enhances cellular repression of mRNAs with NREs that diverge from the PRE consensus and do not form stable complexes with Pum alone . Moreover , multiple weak NREs , such as 3x CycB NRE or 3x hb NRE2 +3G , confer substantial repression that approaches the level of a single strong NRE ( e . g . 1x hb NRE2 ) . As a consequence , Nos appears to expand the Pum target mRNA repertoire beyond those with perfect 5´-UGUAHAUA core PRE consensus sequences by stabilizing Nos-Pum-RNA complexes . Hundreds of proteins that bind mRNAs in cells have now been identified , and large-scale efforts to find their specificities and mRNA targets are underway ( Baltz et al . , 2012; Castello et al . , 2012; Gerstberger et al . , 2014; Sundararaman et al . , 2016 ) . To understand their regulatory functions , the emerging challenge is to classify these regulatory RBPs by their protein folds , specificities , in vivo target RNAs , and expression patterns . Furthermore , as the regulatory targets of individual RBPs are established , it will be crucial to examine overlapping networks to identify cooperative regulation by multiple RBPs . However , our current knowledge of how multiple RBPs collaboratively regulate an mRNA is limited . In this context , the mechanisms of combinatorial regulation by Nos and Pum revealed in this study , which visually capture decades of genetic and biochemical data , provide a foundation to derive general principles for how RBPs co-regulate target mRNAs . Nos uses its three functional regions: N , Z , and C ( Figure 1 ) to elicit repression of target mRNAs , and each region illustrates principles of combinatorial control . The Nos tandem ZFs form a module that adds protein-RNA contacts that strengthen RNA binding , using both side chain and main chain atoms to interact with the NBS . Nanos orthologs are found throughout Bilateria , and the tandem CCHC domains define the ZF-Nanos superfamily ( PF05741 ) based on unique sequence , spacing and length compared to other ZFs . Interestingly , these ZFs were reported to be unique to Nos orthologs , and no structural homology to other RNA-binding ZFs was detected using a DALI structural search ( Hashimoto et al . , 2010; Holm and Rosenström , 2010 ) . However , we manually compared the Nos ZFs with CCHC Zn knuckles ( ZKs ) from HIV nucleocapsid protein ( HIVnc ) ( De Guzman et al . , 1998 ) and found that Nos ZF2 is strikingly similar to the HIVnc ZKs ( Figure 9 ) . 10 . 7554/eLife . 17096 . 033Figure 9 . Nanos Zn finger 2 is structurally homologous to HIV nucleocapsid protein Zn knuckles . ( A ) Ribbon diagrams of Nos ZFs 1 and 2 . The Nos ZFs follow the same structural topology , but ZF1 has longer loops than ZF2 that are N- and C-terminal to the Zn-coordinating histidine residue . ( B ) Ribbon diagram of HIV nucleocapsid ( HIVnc ) protein Zn knuckle ( ZK ) 2 . The solution structures of HIV ZK1 and ZK2 are similar ( rmsd 0 . 72 Å , 136 atoms ) . ( C ) Superposition of Nos ZF2 and HIVnc ZK2 ( rmsd 1 . 2 Å , 88 atoms ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17096 . 033 Protein interactions between Nos and Pum further promote activity of the repression complex and , as revealed by the crystal structures , are mediated by the C-terminal region of Nos and repeats R7 and R8 of Pum . Incorporation of Nos into the Pum-RNA complex induced conformational changes in Pum that added additional RNA contacts . Protein interaction modules may also strengthen regulation by enhancing recruitment of cofactors . The N-terminal region of Nos , not included in our crystal structures , increases repression by the Nos-Pum complex by recruiting the CCR4-NOT deadenylase complex ( Kadyrova et al . , 2007; Raisch et al . , 2016 ) , which strengthens the independent repression activity of Pum ( Weidmann et al . , 2012 , 2014 ) . The functional importance of the interactions visualized by the Nos-Pum-RNA structures is corroborated by our findings and previous genetic and biochemical data . One of the first nos mutant alleles identified , nosL7 , causes a lethal loss of abdominal segments during embryogenesis and encodes a Nos protein with a deletion of residues 376–382 near the C terminus ( Arrizabalaga and Lehmann , 1999; Lehmann , 1988 ) . This deletion prevents formation of a Nos-Pum-RNA complex as assessed by yeast three-hybrid and in vitro pull-down assays ( Sonoda and Wharton , 1999 ) . Our crystal structures illustrate how contacts between the C-terminal region of Nos and Pum residues F1367Pum and Q1337Pum are lost with the nosL7 deletion . A Pum mutation , F1367S , which blocks Pum and Nos association in a yeast 3-hybrid assay ( Edwards et al . , 2001 ) and impairs Nos enhancement of repression ( Weidmann and Goldstrohm , 2012 ) , also abrogated ternary complex formation in vitro ( Figure 4E ) . Similarly , mutation of Q1337Pum prevented incorporation of Nos into the ternary complex ( Figure 4F ) . In addition , F1367Pum is within the loop between Pum repeats R7 and R8 ( Figure 2—figure supplement 2 ) , explaining why insertion of four residues in the R7-R8 loop in the PumMlu mutant disrupts complex formation , hb repression and abdominal segmentation ( Sonoda and Wharton , 1999; Wharton et al . , 1998 ) . These results show the crucial function of the interaction between the Nos C-terminal region and the convex surface presented by Pum repeats R7 and R8 . Based on our current understanding of the Pum-Nos interface and phylogenetic comparisons , the Drosophila Nos-Pum interaction is conserved among Dipteran orthologs , but it is not possible to predict conservation of the interaction in vertebrates . The Nos C-terminal region that contacts Pum is conserved within Diptera , but substantially diverges beyond that order ( Curtis et al . , 1995 ) . Vertebrates have three Nos orthologs which share the tandem Nos ZFs but diverge substantially in their C termini . For example , the C termini of human Nos proteins are not homologous to each other or Drosophila Nos . The amino acid sequence of the 42-residue C-terminus of human NOS1 is enriched in prolines and is conserved throughout vertebrate NOS1 orthologs . The sequence of the 23-residue C-terminus of human NOS2 is enriched in arginines and is conserved among vertebrate NOS2 orthologs . Likewise , the sequence of the 63-residue C-terminus of human NOS3 is only shared by vertebrate NOS3 proteins . We also analyzed the conservation of the Nos-binding interface of Pum proteins and find that the contacts points observed in the Nos-Pum structure are conserved in Dipteran orthologs , but diverge throughout the tree of life . For example , residues equivalent to Pum Q1337 and F1367 change to proline and methionine , respectively , in vertebrate Pum homologs . In addition , vertebrate PUM proteins possess a three residue ( GPH ) insertion in the R7-R8 loop region that could augment protein contacts with vertebrate Nos proteins . Interestingly , deletion of these residues allows human Pum1 to interact with Drosophila Nos ( Sonoda and Wharton , 1999 ) . Given these differences in Nos and Pum proteins , several possible evolutionary scenarios are worth consideration . First , the direct interaction between Drosophila Nos and Pum might be unique to Dipterans; however , Nos and Pum homologs from C . elegans , Xenopus and humans have been reported to interact and a simple loop deletion allows human Pum1 to interact with Drosophila Nos ( Jaruzelska et al . , 2003; Kraemer et al . , 1999; Lolicato et al . , 2008; Nakahata et al . , 2001; Sonoda and Wharton 1999 ) . Instead , the Nos and Pum contacts likely coevolved as the number of Nos and Pum homologs increased , perhaps restricting interactions between particular Nos and Pum proteins . Nos and Pum homologs may also interact indirectly , mediated by a bridging partner ( s ) , as was suggested for C . elegans ( Kraemer et al . , 1999 ) and mouse ( Suzuki et al . , 2016 ) . We found that Nos stabilizes the ternary complex and adds 5´ sequence specificity . Nos specifically recognizes the 5´ sequence only in the context of the ternary complex and also induces a localized conformational change of Pum that adds contacts with the phosphate backbone of nucleotides -1 to -4 . Previous reports indicated that the Nos tandem ZFs in isolation exhibit non-specific RNA binding ( Curtis et al . , 1997; Hashimoto et al . , 2010 ) . With our structure-based definition of the NBS , we find that the mutations in the RNAs tested by Curtis et al . were outside the NBS and thus would not have detected sequence-specific differences in Nos binding . With our findings , we can now attribute the negative effect of NBS mutations on abdominal segmentation in vivo ( Wharton et al . , 1998 ) and ternary complex formation in yeast three-hybrid assays ( Sonoda and Wharton , 1999 ) to the loss of RNA recognition by Nos . Intriguingly , these newly identified Nos and Pum contacts with the 5´ NBS relax the sequence recognition requirements at the 3´ end of the PRE ( Figure 7C ) , thereby allowing Nos and Pum to regulate mRNA targets bearing imperfect PREs , including CycB and bcd , that are not bound stably by Pum alone . As a result , the cooperative activity of a second RBP ( Nos ) adds upstream RNA sequence specificity , which alters mRNA target selection of the primary RBP ( Pum ) and diversifies the range of target sequences . Cooperative RNA recognition by Nos and Pum is reminiscent of cooperative binding of msl2 RNA by Sex lethal ( Sxl ) and Upstream of N-Ras ( Unr ) ( Hennig et al . , 2014 ) , yet the mechanisms and effects of cooperative recognition by Nos and Pum display novel distinguishing characteristics . Similar to Nos and Pum , Sxl and Unr proteins interact with each other only in the presence of target RNA , cooperatively interacting with a regulatory element that contains binding sites for each protein . Sxl and Unr recognize the RNA nucleotides between their individual binding sites , forming a sandwich or a “triple zipper” arrangement that creates a longer recognition element than either protein alone . Nos and Pum recognize their respective NBS and PRE elements; however , Nos recognition of the 5´ NBS and clamp-like binding to Pum and RNA relaxes the sequence requirements of Pum interaction with the 3´ end of the PRE , as with CycB mRNA , which effectively shifts the recognition sequence , rather than extending it . As a consequence , Nos and Pum cooperatively bind to RNAs that neither protein stably binds on its own . Nos appears to dramatically alter the repertoire of transcripts bound by Pum . Our analysis of Pum target mRNAs using the SEQRS-derived Nos-Pum motif indicates that the majority of Pum target mRNAs contain the Nos-Pum motif but lack the full canonical PRE Pum recognition sequence . This suggests that , in addition to strengthening Pum repression activity , Nos can alter the identity of mRNAs regulated by Pum . Moreover , enrichment of specific gene ontology terms in the mRNAs bearing the Nos-Pum motif suggest that Nos can alter the biological functions controlled by Pum ( Figure 7 ) . These analyses allude to the utility of the combinatorial specificity in vivo and explain broadened regulation by Nos and Pum in Drosophila embryos and adults . In addition to altering sequence specificity , our studies highlight other advantages of combinatorial RNA target regulation by Nos and Pum in vivo . First , repression is responsive to the Nos protein concentration . In the Drosophila embryo , the Nos protein gradient is highest at the posterior end where hb expression must be repressed for abdomen formation ( Barker et al . , 1992 ) . We showed that higher Nos levels induced greater repression of reporters bearing hb NREs , mirroring the effect in the embryonic posterior . Interestingly , the expression of Nos is dynamic over the course of development and likely modulates Pum activity temporally as well as spatially ( Wang and Lehmann , 1991 ) . Second , repression is affected by the sequence of the NREs; reporters bearing imperfect PREs were also repressed , but required higher Nos protein levels for equivalent effect . Third , the number of Nos-Pum binding sites in the mRNA modulates regulatory activity . For example , hb mRNA bears two NREs , each with consensus NBS and PRE sites . We found that each site was highly responsive to Nos-enhanced , Pum-mediated repression , and multiple sites conferred greater responsiveness . Repression of the CycB PRE reporter required higher concentrations of transfected Nos , mirroring the requirement for concentrated Nos protein in pole cells where CycB translation is repressed ( Asaoka-Taguchi et al . , 1999; Kadyrova et al . , 2007 ) . Inclusion of three CycB PREs supported repression at lower Nos concentrations , consistent with previous studies showing how the number and quality of PREs in the hb 3´UTR , and the amount of Nos protein expressed , confer the precise level of regulation upon the hb transcript in embryonic development ( Wharton and Struhl , 1991 ) . Indeed , although the CycB mRNA 3´ UTR contains no perfect PRE , it does contain 7 sequence elements with an NBS upstream of a partial PRE with the 5´ UGU sequence . Together , these results define multiple parameters that contribute to biologically relevant levels of regulatory activity . Other RBPs collaborate with Pum to regulate mRNAs . For example , the Brain Tumor ( Brat ) protein contributes to repression of hb expression in the embryo by recognizing an element located upstream of the NBS and PRE in each NRE ( i . e . the so-called Box A motif ) ( Laver et al . , 2015; Loedige et al . , 2014; Sonoda and Wharton , 2001 ) . Moreover , additional partner proteins can modulate RNA binding of PUF proteins ( Campbell et al . , 2012a , 2012b; Menichelli et al . , 2013 ) and the Nos-Pum-RNA ternary complex suggests common mechanisms of co-regulation . For instance , RNA-binding affinity and specificity of the C . elegans PUF protein , FBF-2 , is modulated by interactions with CPB-1 ( Cytoplasmic Polyadenylation element Binding protein 1 ) ( Menichelli et al . , 2013 ) . Intriguingly , CPB-1 and other partners bind to a region of FBF-2 that corresponds to the R7-R8 loop of Pum that binds Nos ( Campbell et al . , 2012b; Menichelli et al . , 2013; Wu et al . , 2013 ) . Thus , the Nos-Pum-RNA complex exemplifies a protein-protein interaction hotspot ( Campbell et al . , 2012b ) . Our results indicate that formation of PUF protein-partner complexes can alter the PUF protein conformation to create additional protein-RNA contacts that strengthen RNA binding while modulating target specificity and regulation . In conclusion , the partnership of Nos and Pum illustrates the profound influence combinatorial control can have on gene expression , demonstrating how location-specific regulation is achieved through the action of RBPs with different spatial distributions , how the addition of a second RBP shifts the RNA recognition motif of the first RBP to modulate target selection , and how regulatory sensitivity can be adjusted by the number and quality of binding sites within a target mRNA and by RBP expression levels . This paradigm emphasizes the importance of understanding how the multitude of RNA-binding factors collaborate to control mRNA stability , translation , processing , and localization . Reporter plasmids pAc5 . 1 FFluc , pAc5 . 1 RnLuc hb 3´UTR were previously described ( Weidmann and Goldstrohm , 2012 , Weidmann et al . , 2014 ) , as was the control plasmid pIZ Halo-tag . The pIZ Halo-Nos expression vector ( NZC ) was created by inserting the Drosophila Nanos cDNA ( NP_001262723 . 1 ) into the XbaI site of pIZ Halo-tag , which contains an N-terminal Halo-tag with a TEV protease cleavage site and a C-terminal V5 epitope . The Nos sequence was amplified from whole fly cDNA and corresponds to isoform Nos-PB , which lacks an alternatively spliced exon encoding a 19 amino acid sequence aa14-VGVANPPSLAQSGKIFQLQ-32 present in the N-terminus of Nos-PA ( NP_476658 . 1 ) . For consistency with the originally identified domain boundaries and mutants , the reported amino acid positions correspond to Nos-PA ( e . g . C319Y and C354Y correspond to C300Y and C335Y of Nos-BP ) . Using the Nos plasmid as a template , the C319Y , C354Y , I376A , M378A , and I382A mutations were generated via QuikChange site-directed mutagenesis ( QC-SDM , Agilent ) . The following mutations and truncations were created using inverse PCR from the Nos plasmid template: F321A , N325A , Y352A , T366A , K368Q , Y369A , NZ ( aa1-373 ) , ZC ( aa289-401 ) , N ( aa1-294 ) , Z ( aa289-373 ) , C ( aa374-401 ) , Δ376–382 ( aa1-375 + 383–401 ) , Δ383–393 ( aa1-382 + 394–401 ) , and Δ394–401 ( aa1-393 ) . Bacterial expression and purification of recombinant Pum or Nos for EMSAs was achieved using pFN18K ( Promega ) with an N-terminal Halo-tag and a TEV cleavage site . The Pum RNA-binding domain sequence , encoding aa1091-1426 of NP_001262403 . 1 , including an N-terminal triple FLAG tag was inserted into pFN18K to create pFN18K Pum plasmid . QC-SDM was used to generate the F1367S Pum mutant and the RNA-binding defective mutR7 ( wherein RNA recognition amino acids are mutated: S1342A N1343A E1346A ) plasmids . Inverse PCR was used to create Pum Q1337A from the wild type template . The same strategy was applied to generate pFN18K NosZC and the C319Y and C354Y mutant vectors with appended C-terminal V5 epitopes . For crystallographic studies , Drosophila Pum RNA-binding domain ( amino acids 1091–1426 ) and Nanos ZF domain ( amino acids 289–401 ) were subcloned into the pSMT3 vector with an N-terminal His6-SUMO tag ( kindly provided by Christopher Lima , Memorial Sloan Kettering Cancer Center , New York ) . Reporters used for Nos enhancement of Pum repression in Figure 8 were made in the pAc5 . 4 vector , wherein a cryptic cleavage/polyadenylation element intrinsic to pAc5 . 1 vector was removed and a degenerate PRE motif in the RnLuc ORF was inactivated by introducing synonymous codons . Complementary DNA oligos ( IDT ) bearing wild type and mutant NREs , listed below , were inserted into XhoI and Not1 restriction sites within the 3´UTR . The 3x hb NRE2 , 3x CycB NRE , 3x hb NRE2 +3G , and 3x hb NRE2 ACA reporters were generated in an identical fashion using oligos with three repeated NRE elements . RNA was purified from 2 million D . mel-2 cells transfected with the plasmids indicated in Figure 1―figure supplement 1 . Cells were harvested at 1000 × g for 3 min , washed twice in PBS , and RNA was purified from cell pellets using TRIzol reagent ( Life Technologies ) . Total RNA preparations were then analyzed by Northern blotting as previously described ( Blewett and Goldstrohm , 2012 ) . RNA was separated in a denaturing 0 . 85% agarose gel containg 1x MOPS and formaldehyde . RNA was transferred by blotting to an Immobilon NY+ membrane ( Millipore ) . Membranes were then UV-crosslinked and probed for the RNAs indicated in the figure . For RnLuc reporter , a 32P body-labeled , antisense RNA probe was created by in vitro transcription . The following primers were used to amplify templates for creation of RnLuc RNA probes . The T7 promoter sequence is underlined and gene specific regions are bolded . RnLuc 3´ forward primer: 5´-GGGCGAGGTTAGACGGCCTACCCT RnLuc 3´ reverse primer: 5´-GGATCCTAATACGACTCACTATAGGGCGGCCAGCGGCCTTGG The 7SL RNA was detected on northern blots using a 32P 5’ end-labeled DNA oligo with the following sequence . 7SL Probe: 5´-CACCCCTGGCCCGGTTCATCCCTCCTTAGCCAACCTGAATGCCACGG . The radioactive blots were exposed to a storage phosphor screen . The signal on the screen was captured with a Typhoon Trio imager ( GE Healthcare ) and subsequently quantified using ImageQuant TL Software ( GE Healthcare ) . Each RnLuc signal was normalized to the 7SL signal . Statistical analysis of Northern blot data from three replicate cell cultures is reported in Figure 1―figure supplement 1—source data 1 . The oligos used to create RnLuc 1xNRE reporter plasmids used for Nos enhancement of Pum repression are as follows ( Restriction site overhangs are indicated in bold , PRE sequences underlined , mutations lowercase ) : hb NRE2 Forward: 5´-TCGACGAAAATTGTACATAAGCC hb NRE2 Reverse: 5´-GGCCGGCTTATGTACAATTTTCG hb NRE2 +3G Forward: 5´-TCGACGAAAATTGgACATAAGCC hb NRE2 +3G Reverse: 5´-GGCCGGCTTATGTcCAATTTTCG hb NRE2 ACA Forward: 5´-TCGACGAAAATacaACATAAGCC hb NRE2 ACA Reverse: 5´-GGCCGGCTTATGTtgtATTTTCG hb NRE1 Forward: 5´-TCGACCAGAATTGTATATATTCG hb NRE1 Reverse: 5´-GGCCCGAATATATACAATTCTG bcd NRE Forward: 5´-TCGAAAGTGATTGTAGATATCTA bcd NRE Reverse: 5´-GGCCTAGATATCTACAATCACTT CycB NRE Forward: 5´-TCGAGACTATTTGTAATTTATATC CycB NRE Reverse: 5´-GGCCGATATAAATTACAAATAGTC hb NRE2 -1C Forward: 5´-TCGACGAAAAcTGTACATAAGCC hb NRE2 -1C Reverse: 5´-GGCCGGCTTATGTACAgTTTTCG hb NRE2 -2C Forward: 5´-TCGACGAAAcTTGTACATAAGCC hb NRE2 -2C Reverse: 5´-GGCCGGCTTATGTACAAgTTTCG Synthetic RNAs ( IDT ) used in EMSA experiments include the following ( with PRE elements underlined and mutations in lowercase bold ) : Cy5- hb NRE2 RNA: 5´-Cy5-rUUGUUGUCGAAAAUUGUACAUAAGCC . hb NRE2 RNA: 5´-rAAAUUGUACAUAAGCC hb NRE2 +3G RNA: 5´-rAAAUUGgACAUAAGCC hb NRE2 ACA RNA: 5´-rAAAUacaACAUAAGCC hb NRE1 RNA: 5´-rGAAUUGUAUAUAUUCG bcd NRE RNA: 5´-rUGAUUGUAGAUAUCUA CycB NRE RNA: 5´-rUAUUUGUAAUUUAUAUC hb NRE2 -1C RNA: 5´-rAAAcUGUACAUAAGCC hb NRE2 -2C RNA: 5´-rAAcUUGUACAUAAGCC D . mel-2 cells were cultured at 28°C in Sf-900 III SFM ( Life Technologies ) with 50 Units/mL penicillin and 50 µg/mL streptomycin . Transfections were performed as described ( Weidmann and Goldstrohm , 2012; Weidmann et al . , 2014 ) using Effectene ( Qiagen ) . Each transfected well of a 6-well plate contained 5 ng Firefly Luciferase internal control plasmid ( pAc5 . 1 FFLuc ) , 10 ng of the indicated RnLuc reporter plasmid , and 200 ng total of protein expression vector , 43–44 µl of EC buffer , 1 . 6 µl of enhancer , 2 µl of Effectene , 300 µl of new Sf-900 III SFM , and 1 . 6 mL of D . mel-2 cells ( 1 × 106 cells/mL ) . For Nos experiments , 10 ng of pIZ Nos expression vector ( unless otherwise noted ) was balanced with empty pIZ vector for a total mass of 200 ng . For each Nos transfection gradient , pIZ was also used to balance the total mass of transfected expression vector to 200 ng . Transfection conditions for the experiment in Figure 4D differed in the following manner: each well of a 6-well plate contained 5 ng Firefly Luciferase internal control plasmid ( pAc5 . 4 FFLuc ) , 10 ng of the reporter gene pAc5 . 4 RnLuc hb 3'UTR , and 100 ng of the indicated protein expression plasmid . Total transfected DNA was set at 3 µg per well , balanced by pIZ plasmid . Fugene HD ( Promega ) transfection reagent was used at a 4:1 ratio ( µl Fugene HD: µg of DNA ) , prepared in 150 µL Sf-900 III SFM media and incubated for 15 min at room temperature prior to application to 2x106 cells in 2 ml of media . Luciferase reporter assays were performed using the Dual-Glo assay ( Promega ) and a GloMax Discover luminometer . A relative response ratio ( RRR ) was calculated from RnLuc/FFLuc signals for each sample . Percent repression values were calculated as previously described ( Van Etten et al . , 2013; Weidmann and Goldstrohm , 2012 ) . The pIZ-Halo-tag vector served as the negative control for Halo-NZC Nos constructs . For the Nanos gradient experiments in Figure 8 , the 0 ng condition , corresponding to 200 ng of transfected pIZ plasmid , served as the negative control . Data and statistical analyses of all reporter assays are reported in Figure 1―source data 1 , Figure 4―source data 1 , Figure 5―source data 1 , and Figure 8―source data 1 . Four replicate cell cultures were analyzed in each experiment as indicated in the figure legends . Results were validated in multiple independent experiments performed on different days . Recombinant Pum and Nos for EMSAs were expressed in KRX E . coli cells ( Promega ) in 2xYT media with 25 µg/mL Kanamycin and 2 mM MgSO4 at 37°C to OD600 of 0 . 7–0 . 9 , at which point protein expression was induced with 0 . 1% ( w/v ) rhamnose for 3 hr . Cell pellets were washed with 50 mM Tris-HCl , pH 8 . 0 , 10% [w/v] sucrose and pelleted again . Pellets were suspended in 25 mL of 50 mM Tris-HCl pH 8 . 0 , 0 . 5 mM EDTA , 2 mM MgCl2 , 150 mM NaCl , 1 mM DTT , 0 . 05% ( v/v ) Igepal CA-630 , 1 mM PMSF , 10 µg/ml aprotinin , 10 µg/ml pepstatin , and 10 µg/ml leupeptin . To lyse cells , lysozyme was added to a final concentration of 0 . 5 mg/mL and cells were incubated at 4°C for 30 min with gentle rocking . MgCl2 was increased to 7 mM and DNase I ( Roche ) was added to 10 µg/mL followed by incubation for 20 min . Lysates were cleared at 50 , 000xg for 30 min at 4°C . Halo-tag containing proteins were purified using HaloLink Resin ( Promega ) at 4°C . Beads were washed 3 times with 50 mM Tris-HCl pH 8 . 0 , 0 . 5 mM EDTA , 2 mM MgCl2 , 1 M NaCl , 1 mM DTT , 0 . 5% [v/v] Igepal CA-630 ) and 3 times with Elution Buffer ( 50 mM Tris-HCl , pH 7 . 6 , 150 mM NaCl , 1 mM DTT , 20% [v/v] glycerol ) . After washing , beads were resuspended in Elution Buffer with 30 U of AcTEV protease ( Invitrogen ) , cleavage proceeded for 24 hr at 4°C , and beads were removed by centrifugation through a micro-spin column ( Bio-Rad ) . All RNA-binding reactions were performed in RNA-binding Buffer ( 50 mM Tris-HCl pH 7 . 6 , 150 mM NaCl , 2 mM DTT , 2 µg/mL BSA , 0 . 01% [v/v] Igepal CA-630 , 0 . 02% bromophenol blue , 20% [v/v] glycerol ) . Reactions were equilibrated for 3 hr at 4°C . 5% native polyacrylamide TBE mini-PROTEAN gels ( Bio-Rad ) were pre-run for 3 hr at 50V before loading 5 µl of each sample and then run at 50V for 2–2 . 25 hr at 4°C . For EMSAs with fluorescently-labeled hb NRE2 RNA , reactions contained 1 nM target RNA and the concentrations of PUM-HD and NosZC are as noted in Figure 3B . For Kd measurements shown in Figure 3C and D , Pum was held constant at 100 nM , and NosZC concentrations included 0 , 0 . 5 , 1 , 2 , 5 , 10 , 20 , 50 , 100 , 200 , 500 , and 1000 nM . In the RNA + NosZC lane , RNA was at 1 nM and NosZC was at 1000 nM . For Kd measurements with radioactive RNA oligos , reactions contained 0 . 3 nM RNA that were labeled at their 5´ ends using T4 Kinase ( NEB ) with ATP [γ-32P] ( Perkin-Elmer ) . Where indicated for hb NRE2 RNA , the concentration of NosZC was held constant at 400 nM while Pum was titration included concentrations of 0 , 0 . 2 , 0 . 5 , 1 , 2 , 5 , 10 , 20 , 50 , 75 , 100 , 150 , and 200 nM . For all other RNAs tested in the presence of Nos , the reactions contained 1 µM NosZC while Pum concentration was titrated from 0 , 2 , 5 , 10 , 20 , 50 , 100 , 200 , 300 , 400 , 500 , 750 , and 1000 nM . Gels containing radioactive RNAs were dried onto Whatmann filter paper . The radioactive gels were then exposed to a storage phosphor screen ( GE ) for 16 hr . EMSAs were imaged with a Typhoon Trio imager ( GE Healthcare ) and subsequently quantified using ImageQuant TL software . Fraction bound values from three replicate EMSAs were plotted against titrated protein concentration , and Kd was calculated via nonlinear regression analysis for one site interaction with GraphPad Prism software ( GraphPad Software , Inc . ) . Pum and Nos proteins ( pSMT3 vectors ) were overexpressed in E . coli strain BL21-CodonPlus ( DE3 ) -RIL ( Agilent ) at 16°C for 20 hr after induction with 0 . 2 mM IPTG . The cell pellet was resuspended in lysis buffer containing 20 mM Tris ( pH 8 . 0 ) , 0 . 5 M NaCl , 20 mM imidazole , 5% ( v/v ) glycerol , and 0 . 1% ( v/v ) β-mercaptoethanol . Both proteins were purified by a similar procedure involving four sequential steps: a Ni-NTA chelating column , Ulp1 overnight incubation to remove the His6-SUMO tag , a heparin column and lastly a Superdex 75 column ( GE Healthcare ) . The Ni elution buffer contained 20 mM Tris ( pH 8 . 0 ) , 50 mM NaCl , 0 . 2 M imidazole , and 1 mM DTT . A gradient elution was run on the heparin column with buffer A containing 20 mM Tris ( pH 8 . 0 ) and 1 mM DTT and buffer B containing additional 1 M NaCl . The Superdex 75 column buffer contains 20 mM Hepes ( pH 7 . 4 ) , 0 . 15 M NaCl , and 2 mM DTT . For the ternary complex formation , a 16-nt hb NRE2 RNA ( 5´-AAAUUGUACAUAAGCC ) and a 14-nt CycB NRE RNA ( 5´-UAUUUGUAAUUUAU ) were used , separately . Purified Pum and Nos were concentrated and mixed together with RNA in a molar ratio of 1:1:1 . 1 . After overnight incubation at 4°C , the mixture was loaded onto a Superdex 200 10/300 GL column . Peak fractions containing ternary complexes were concentrated to OD280 ~ 7 in a buffer of 20 mM HEPES ( pH 7 . 4 ) , 0 . 15 M NaCl , and 2 mM DTT for crystallization . For the Pum-RNA binary complex , concentrated Pum ( OD280~4 . 0 ) was incubated with 8-nt hb PRE2 RNA ( 5´-UGUACAUA ) in a molar ratio of 1:1 . 2 on ice for 2 hr . The mixture was directly put into crystallization trays . Binary complexes were also formed with 12-nt hb NRE2 RNA ( 5´-AAAUUGUACAUA ) , but no crystals were obtained . Crystals of Nos-Pum-hb NRE2 RNA complex were obtained by hanging drop vapor diffusion , mixing 2 μL of sample and 2 μL of reservoir solution [1 . 1 M ( NH4 ) 2SO4 , 0 . 1 M MES , pH 5 . 6] at 20°C . Crystals of Pum-Nos-CycB NRE RNA complex were obtained in the condition of 0 . 9 M ( NH4 ) 2SO4 , 0 . 1 M MES , pH 5 . 6 by microseeding using the hb complex crystals . Crystals of Pum-hb PRE2 RNA binary complex were obtained in the condition of 15% ( w/v ) PEG 3350 , 0 . 1 M bis-tris , pH 6 . 5 and 0 . 2 M NH4OAc . All crystals were transferred to a cryo-solution containing the reservoir solution with additional 15–20% ( v/v ) glycerol , and flash frozen in liquid nitrogen . X-ray diffraction data were collected at the SER-CAT Beamline 22-ID at the Advanced Photon Source , Argonne National Laboratories . Crystals of ternary complexes belong to P6522 space group with one complex in an asymmetric unit . The structure of Pum-Nos-hb NRE2 RNA complex was determined by molecular replacement using the structures of Drosophila Pum ( PDB: 3H3D ) and zebrafish Nanos ZF domain ( PDB: 3ALR ) as the search model with Phaser . Iterative model building was done with COOT and Phenix ( Adams et al . , 2002; Emsley and Cowtan , 2004 ) . The structure of Pum-Nos-CycB NRE RNA complex was determined by molecular replacement using the hb complex as the search model . Residues 1092–1419 in Pum , residues 316–385 in Nos and RNA bases from −4 to +8 are modeled in the hb NRE2 structure . Residues 1092–1102 and 1121–1418 in Pum , residues 316–384 in Nos , and RNA bases from −4 to +7 are modeled in the CycB NRE structure . Crystals of Pum-hb PRE2 RNA binary complex belong to C2 space group with one complex in an asymmetric unit . The structure was determined by molecular replacement using the structure of Drosophila Pum ( PDB: 3H3D ) as the search model . Residues 1090–1426 in Pum and RNA bases +1 to +8 of PRE2 are modeled in the binary structure . Data collection and refinement statistics are presented in Table 1 . Western blotting from luciferase assay samples was performed as previously described ( Weidmann and Goldstrohm , 2012 , Weidmann et al . , 2014 ) . Where indicated , proteins were detected using a V5 monoclonal antibody ( Invitrogen ) and horseradish peroxidase conjugated goat anti-mouse IgG ( Thermo Scientific ) . Signals were detected using either Pierce ECL Western blotting substrate ( Thermo ) or Immobilon western chemiluminescent substrate ( Millipore ) and autoradiography film . Protein extracts of D . mel-2 cells expressing Halo-tag fusions were prepared as previously described ( Weidmann and Goldstrohm , 2012 ) . Extracts were then incubated with 100 nM Halo-tag TMR Ligand ( Promega ) for 30 min on ice , protected from light . After labeling , extracts were separated via SDS polyacrylamide gel electrophoresis and labeled proteins were imaged with a Typhoon Trio imager ( GE Healthcare ) . SEQRS was conducted as described with minor modifications ( Campbell et al . , 2014 ) on the following samples: 1 ) wild type Pum alone , 2 ) wild type Nos alone , 3 ) wild type Pum with Nos , 4 ) Pum mut R7 alone and 5 ) Pum mutR7 with Nos . The proteins were purified as described above for EMSA experiments except that Magnetic Halolink beads ( Promega ) were used and the Pum test proteins remained covalently bound via N-terminal Halotag to the beads . For Nos alone ( sample 2 ) , Nos remained linked to the magnetic beads . For the other samples that contained Nos ( samples 3 and 5 ) , Nos protein was cleaved from the beads using TEV protease and equimolar amount was added to the Pum-linked beads . The initial RNA library was transcribed from 1 μg of input dsDNA using the AmpliScribe T7-Flash Transcription Kit ( Epicentre ) . 200 ng of DNase treated RNA library was added to 100 nM of Halo-tagged proteins immobilized onto magnetic resin ( Promega ) . The volume of each binding reaction was 100 μl in SEQRS buffer containing 200 ng yeast tRNA competitor and 0 . 1 units of RNase inhibitor ( Promega ) . The samples were incubated for 30 min at 22°C prior to magnetic capture of the protein-RNA complex . The binding reaction was aspirated and the beads were washed four times with 200 μl of ice cold SEQRS buffer . After the final wash step , resin was suspended in elution buffer ( 1 mM Tris pH 8 . 0 ) containing 10 pmol of the reverse transcription primer . Samples were heated to 65°C for 10 min and then cooled on ice . A 5 μl aliquot of the sample was added to a 10 μl ImProm-II reverse transcription reaction ( Promega ) . The ssDNA product was used as a template for 25 cycles of PCR using a 50 μl GoTaq reaction ( Promega ) . Sequencing data were processed as described ( Campbell et al . , 2012a ) . Sequence logos corresponding to consensus binding motifs were generated by MEME analysis of the 100 most-enriched sequences ( reported in Figure 7—source data 2 ) ( Bailey et al . , 2006 ) . Enrichment of Pum and Nos-Pum binding sites in 3´UTRs was analyzed for all mRNAs and for mRNAs bound by Pum in vivo using the dataset from Gerber et al . , 2006 . Pattern matching to the Pum and Nos-Pum motifs reported in Figure 7 were preformed using the grep Perl function from command line . Significance values compared to all 3'UTR sequences was determined using chi-squares test using GraphPad Prism ( reported in Figure 7—source data 2 ) . Gene ontology enrichment analysis was performed using DAVID ( Huang et al . , 2009 , 2008 ) and Venn diagrams were generated using Venn Diagram Plotter ( Pacific Northwest National Laboratory ) .	Molecules of DNA contain the instructions needed to make proteins inside cells . Proteins perform many different roles and each needs to be produced at the right time and in the right amounts to enable the cells to survive . The DNA is first copied to make molecules of ribonucleic acid ( RNA ) , which are then used as templates to make the proteins . One way to control protein production is to regulate the RNA molecules . A family of proteins called RNA-binding proteins can recognise and bind to specific RNA molecules and determine whether a RNA molecule is destroyed , used to produce proteins , or stored for later use . In effect , these RNA-binding proteins act as switches that turn protein production on or off . Nanos and Pumilio are two RNA-binding proteins that are found in many organisms , including humans and other animals . Genetic studies in fruit flies show that these two proteins influence development , the nervous system and the behaviour of stem cells by switching off the production of certain proteins . To investigate how Nanos and Pumilio work together to regulate protein production , Weidmann , Qiu et al . used a variety of techniques to study the activity of these proteins in cells taken from fruit fly embryos . The experiments reveal that Nanos acts like a clamp to hold Pumilio close to specific RNAs , which allows Pumilio to switch off the production of the corresponding proteins . The presence of Nanos allows Pumilio to regulate RNAs that it cannot bind to alone . Therefore , the experiments show that by working together with Nanos , Pumilio is able to regulate a wider variety of RNAs than it would otherwise be able to . These findings provide a molecular understanding for why fruit fly mutants that lack Nanos or Pumilio have severe body defects and reduced fertility . The next challenge is to identify the specific RNAs targeted by Nanos and Pumilio in stem cells , the nervous system and during development .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2016	Drosophila Nanos acts as a molecular clamp that modulates the RNA-binding and repression activities of Pumilio
Mutations in CHD7 are the major cause of CHARGE syndrome , an autosomal dominant disorder with an estimated prevalence of 1/15 , 000 . We have little understanding of the disruptions in the developmental programme that underpin brain defects associated with this syndrome . Using mouse models , we show that Chd7 haploinsufficiency results in reduced Fgf8 expression in the isthmus organiser ( IsO ) , an embryonic signalling centre that directs early cerebellar development . Consistent with this observation , Chd7 and Fgf8 loss-of-function alleles interact during cerebellar development . CHD7 associates with Otx2 and Gbx2 regulatory elements and altered expression of these homeobox genes implicates CHD7 in the maintenance of cerebellar identity during embryogenesis . Finally , we report cerebellar vermis hypoplasia in 35% of CHARGE syndrome patients with a proven CHD7 mutation . These observations provide key insights into the molecular aetiology of cerebellar defects in CHARGE syndrome and link reduced FGF signalling to cerebellar vermis hypoplasia in a human syndrome . The segmental organisation of the embryonic neural tube is imparted by the action of homeobox genes that show defined expression patterns along its anterior-posterior axis , in combination with growth factors secreted from distinct organising centres ( Kiecker and Lumsden , 2012 ) . The cerebellum is derived from dorsal rhombomere 1 ( r1 ) , the anterior-most segment of the embryonic hindbrain . The survival and patterning of r1 is controlled by Fibroblast Growth Factor 8 ( FGF8 ) , secreted from the isthmus organiser ( IsO ) , an organising centre located at the boundary between the embryonic midbrain ( mesencephalon , mes ) and r1 ( reviewed by Nakamura et al . , 2005; Martinez et al . , 2013 ) . The IsO forms at the expression boundary of two homeobox genes: Otx2 ( Orthodenticle Homeobox 2 ) in the anterior neural tube and , Gbx2 ( Gastrulation Brain Homeobox 2 ) , in the posterior neural tube . Fgf8 expression in the IsO is initiated at early ( 3–5 ) somite stages in the mouse embryo , resulting in a stable gene-regulatory network at the IsO , where ( 1 ) cross-repressive interactions between Otx2 and Gbx2 maintain the IsO , ( 2 ) Otx2 represses Fgf8 expression , thus restricting it to r1 , and ( 3 ) Fgf8 and Gbx2 contribute to the maintenance of each other’s expression ( reviewed by Joyner et al . , 2000; Martinez et al . , 2013 ) . Studies in the mouse embryo have shown that the level of FGF gene expression and signalling from the IsO has to be tightly controlled to ensure normal cerebellar development . Altered FGF signalling in the mes/r1 region preferentially affects the development of the medial cerebellum , the vermis ( Xu et al . , 2000; Trokovic et al . , 2003; Basson et al . , 2008; Yu et al . , 2011 ) , which is derived from precursors in the most anterior part of r1 , closest to the source of FGF8 expression ( Sgaier et al . , 2005 ) . The observation that reduced FGF signalling results in hypoplasia of the cerebellar vermis in mice raises the possibility that reduced FGF signalling might underlie vermis hypoplasia in certain human conditions . However , studies in mice have also found that FGF signalling has many essential roles during development and even small reductions in Fgf8 expression during embryonic development are incompatible with postnatal survival ( Meyers et al . , 1998 ) . These findings suggest that mutations causing sufficiently reduced Fgf8 expression or signalling throughout the whole embryo to result in cerebellar defects are unlikely to yield viable offspring . Rather , it seems more likely that a disruption of the mechanisms that regulate local Fgf8 expression at the IsO will be responsible for cerebellar vermis hypoplasia in humans . CHARGE syndrome ( MIM#214800 ) is an autosomal dominant disorder with an estimated prevalence of 1/15 , 000 . Central nervous system defects have been reported in CHARGE ( Coloboma of the eye , Heart defects , Atresia of the choanae , Retarded growth and development , Genital anomalies and Ear malformations or deafness ) syndrome ( Lin et al . , 1990; Tellier et al . , 1998; Becker et al . , 2001; Issekutz et al . , 2005; Sanlaville et al . , 2006; Sanlaville and Verloes , 2007; Bergman et al . , 2011 ) , including reports of cerebellar defects in pre-term CHARGE fetuses ( Becker et al . , 2001; Sanlaville et al . , 2006; Legendre et al . , 2012 ) . Depending on the clinical selection , 60–90% of the individuals suspected for CHARGE syndrome have de novo , heterozygous mutations in the CHD7 ( Chromodomain helicase DNA-binding protein 7 , MIM#608892 ) gene ( Vissers et al . , 2004; Bilan et al . , 2012; Janssen et al . , 2012 ) . CHD7 is a member of the SNF2H-like chromatin-remodelling family and has been shown to function as a ‘transcriptional rheostat’ by maintaining appropriate levels of developmental gene expression ( Schnetz et al . , 2010 ) . A number of clinical findings led us to hypothesise that some developmental defects in CHARGE syndrome might be caused by insufficient FGF signalling levels . For example , CHARGE syndrome shows significant clinical overlap with 22q11 . 2 deletion and Kallmann syndromes , both of which have been linked to reduced FGF signalling ( Scambler , 2010; Miraoui et al . , 2013; Corsten-Janssen et al . , 2013; Randall et al . , 2009 ) . We therefore set out to test the hypothesis that CHD7 is required for normal levels of Fgf8 expression during development by focusing on the embryonic IsO and cerebellar development . We previously reported that mice heterozygous for the Chd7XK403 gene-trap allele ( henceforth referred to as Chd7+/− mice ) phenocopy several aspects of CHARGE syndrome ( Randall et al . , 2009 ) . To determine whether FGF signalling at the IsO was affected by Chd7 deletion , we first visualised the expression of Fgf8 in E9 . 5 embryos by in situ hybridisation . Fgf8 expression in the IsO appeared reduced in Chd7+/− embryos and was substantially downregulated in Chd7−/− embryos ( Figure 1A–C ) . Quantitative RT-PCR analysis confirmed that Fgf8 transcripts were reduced by 20% in Chd7+/− embryos , and by 40% in Chd7−/− embryos ( Figure 1D ) . Furthermore , Fgf8 expression was reduced by 80% in Chd7+/−;Fgf8+/− embryos , compared to 40% reduction in Chd7+/+;Fgf8+/− embryos ( Figure 1D ) . To ask whether this synergistic genetic interaction between Chd7 and Fgf8 loss-of-function alleles translated to defects in FGF signalling , the expression of the FGF target gene Etv5 was analysed ( Roehl and Nusslein-Volhard , 2001; Yu et al . , 2011 ) . Whereas Etv5 expression was clearly diminished in Chd7−/− embryos compared to wildtype controls ( Figure 1E , G ) , it did not appear substantially reduced in Chd7+/− embryos ( Figure 1F ) , an observation confirmed by quantitative RT-PCR ( Figure 1H ) . However , quantitative analyses showed that Etv5 expression was reduced by 50% in Chd7+/−;Fgf8+/− embryos , compared to wildtype levels in Chd7+/− and Fgf8+/− embryos ( Figure 1H ) . These data identified CHD7 as an upstream regulator of Fgf8 in the IsO and revealed a synergistic relationship between the Chd7 and Fgf8 genes . 10 . 7554/eLife . 01305 . 003Figure 1 . Reduced Fgf8 expression and FGF signalling during early cerebellar development in Chd7-deficient embryos . ( A–C ) In situ hybridisation for Fgf8 at E9 . 5 shows a Chd7 gene dosage-dependent reduction in Fgf8 expression in the mid-hindbrain isthmus organiser ( IsO , arrows ) . Scale bar = 0 . 5 mm . ( D ) Quantification of Fgf8 transcript levels in the mes/r1 region of E9 . 5 embryos . ( E–G ) Expression of the FGF-regulated gene Etv5 in E9 . 5 mouse embryos visualised by in situ hybridisation . ( H ) Quantification of Etv5 gene expression in mes/r1 tissue confirms the in situ hybridisation data and indicates a significant reduction in FGF signalling in Chd7+/−;Fgf8+/− and Chd7−/− embryos . Data represents mean ± standard error of the mean ( SEM ) from three individual samples for each genotype . p<0 . 05 , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 01305 . 003 Previous studies have shown that the medial cerebellum , the cerebellar vermis , is most sensitive to perturbations in FGF signalling during development ( Broccoli et al . , 1999; Xu et al . , 2000; Trokovic et al . , 2003; Basson et al . , 2008; Yu et al . , 2011 ) , hence we predicted that Chd7 deficiency will predispose embryos to cerebellar vermis defects . Cerebellar size was normal in Chd7+/− and Fgf8+/− animals compared to wildtype littermates ( Figure 2A–C ) , consistent with the observation that FGF signalling was not substantially reduced in these mutants ( Figure 1H ) . To accurately compare the sizes of the cerebellar regions between mice , the volumes of cerebellar hemispheres , paravermis and vermis were calculated from surface area measurements taken from serial sections through postnatal day ( P ) 21 cerebella . This analysis confirmed that cerebellar size was not significantly altered in Chd7+/− or Fgf8+/− mice ( Figure 2E ) . Furthermore , cerebellar foliation in the vermis and hemispheres appeared normal in the mutants ( Figure 2A’–C’ , A”–C” ) . As Chd7−/− embryos die by E11 . 5 , cerebellar development could not be analysed in these mutants . However , Chd7+/−;Fgf8+/− animals survive and an analysis of cerebellar size revealed a significant reduction in size owing to vermis aplasia ( Figure 2D , D’ , E ) . The cerebellar hemispheres were of normal size ( Figure 2E ) and had normal foliation compared to the controls ( Figure 2D” ) . Cerebellar vermis aplasia in Chd7+/−;Fgf8+/− animals was already present at birth , confirming that defects arose during embryonic development ( Figure 2F–I , F’–I’ , red asterisk ) . We also noted that the posterior midbrain ( inferior colliculus ) , another region that is particularly sensitive to FGF signalling levels , was abnormal in Chd7+/−;Fgf8+/− mutants ( Figure 2I’ , black asterisk ) . These observations provided functional evidence for a synergistic Chd7-Fgf8 interaction and indicated that the potential phenotypic consequences of diminished Fgf8 expression in Chd7+/− embryos could be revealed by Fgf8 heterozygosity . 10 . 7554/eLife . 01305 . 004Figure 2 . Chd7 and Fgf8 loss-of-function alleles interact to cause cerebellar vermis aplasia in the mouse . ( A–D ) Wholemount views of the mouse cerebellum at P21 . The cerebellar vermis is indicated by a double-headed arrow . Chd7+/− animals have normal cerebella , indistinguishable from wildtype and Fgf8+/− control littermates . Chd7+/−;Fgf8+/− animals exhibit vermis aplasia ( asterisk in D ) . Scale bar = 5 mm . ( A’-D’ ) Cresyl violet-stained sagittal sections through the cerebellar vermis . Note the absence of cerebellar vermis tissue in Chd7+/−;Fgf8+/− embryos ( D’ ) . ( A”–D” ) Sagittal sections through cerebellar hemispheres . ( E ) Measurements of cerebellar vermis , paravermis and hemisphere sizes in brains from the indicated genotypes . The data represents the mean of three samples with error bars indicating SEM . p<0 . 001 . ( F–I ) Wholemount views of cerebella at birth ( P0 ) , with vermis indicated by arrows . ( F’–I’ ) Sagittal sections through P0 brains with inferior colliculus ( IC ) and cerebellum ( Cb ) indicated . Note the loss of cerebellar vermis ( red asterisk ) and abnormal IC ( black asterisk ) in Chd7+/−;Fgf8+/− animals ( I’ ) . Scale bar = 1 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 01305 . 004 The Chd7 gene encodes a SNF2H-like chromatin remodelling factor that is characterised by the presence of tandem chromodomains in its N-terminal region . Genome-wide chromatin immunoprecipitation studies in cell lines have shown that CHD7 is recruited to distal gene regulatory elements , presumably through interactions between CHD7 chromodomains and methylated lysine 4 residues in histone 3 ( H3K4me ) , present at regulatory elements ( Schnetz et al . , 2009 , 2010; Engelen et al . , 2011 ) . The Drosophila homologue of the CHD7 subfamily , kismet , was identified as a Trithorax gene and kismet mutants have reduced expression of homeotic genes and consequent transformations of body segments to more anterior structures ( Daubresse et al . , 1999 ) . We therefore asked whether CHD7 has a role in maintaining the expression of homeobox genes that impart regional identity in the developing neural tube . The homeobox genes Otx2 and Gbx2 influence anterior and posterior identity in the developing embryo , respectively , position the IsO and regulate the levels of Fgf8 expression ( Broccoli et al . , 1999; Hidalgo-Sanchez et al . , 1999; Millet et al . , 1999; Joyner et al . , 2000; Heimbucher et al . , 2007 ) . The analysis of Chd7−/− mouse embryos at E8 . 25 ( 4ss ) , shortly after the initiation of Fgf8 expression in the mes/r1 region revealed Otx2 upregulation and posterior expansion of its expression ( Figure 3A , B , arrow indicating expanded expression ) . To investigate how the altered Otx2 expression domain related spatially to other hindbrain regions , we combined Otx2 in situ hybridisation with markers to visualise r3+r5 ( Krox20 ) and r2 ( Hoxa2 ) in the same embryo . Although Krox20 expression is reduced in Chd7 mutant embryos ( Alavizadeh et al . , 2001 ) , r3 was still clearly marked , confirming the expansion of Otx2 expression towards r3 ( Figure 3C , D ) . Combined Otx2/Hoxa2 in situ hybridisation experiments suggested that the expansion of Otx2 expression included most of r1 , as indicated by the absence of the Otx2/Hoxa2-negative r1 in Chd7−/− embryos ( Figure 3E , F ) . 10 . 7554/eLife . 01305 . 005Figure 3 . Chd7 loss results in Otx2 de-repression , loss of rhombomere 1 identity and reduced Fgf8 expression . ( A and B ) In situ hybridisation for Otx2 in 4 somite stage ( ss ) embryos . Note the posterior expansion of Otx2 expression in the mutant embryo ( arrow in B ) . ( C and D ) In situ hybridisation for Otx2 and Krox20 to mark the forebrain/mesencephalon and rhombomeres 3 and 5 ( r3 and r5 ) , respectively in 6 ss embryos . Note the posterior expansion of Otx2 ( arrow ) towards r3 . ( E and F ) In situ hybridisation for Otx2 and Hoxa2 , to mark the forebrain/mesencephalon and r2 , respectively in 6 ss embryos . Note the posterior expansion of Otx2 ( arrow ) and apparent loss of the Otx2-Hoxa2-negative r1 in the Chd7−/− embryo . ( G and H ) Fgf8 in situ hybridisation on 6 ss embryos . Note the initiation of Fgf8 expression at the correct position in the mutant ( H ) , despite posteriorised Otx2 expression . ( I and J ) Side-by-side comparison of Fgf8 and Otx2/Hoxa2 expression in 6 ss Chd7+/+ and Chd7−/− embryos . Note the posterior expansion of Otx2 expression ( white arrow ) and downregulated Fgf8 expression in the Chd7−/− embryos , compared to wildtype controls . Also note that Fgf8 expression is initiated at the correct position in the Chd7−/− embryo , with no evidence of a repositioning of the IsO in response to posterior expansion of Otx2 at this stage of development . ( K and L ) In situ hybridisation for Gbx2 suggesting the loss of r1 identity by E9 . ( M ) Summary of regulatory interactions at the IsO in Chd7+/+ vs Chd7−/− embryos . The loss of Otx2 repression and Gbx2 maintenance by CHD7 are predicted to result in reduced Fgf8 expression in Chd7-deficient embryos . mes = mesencephalon , r1 = rhombomere 1 , r2 = rhombomere 2 , IsO = isthmus organiser . DOI: http://dx . doi . org/10 . 7554/eLife . 01305 . 005 We further confirmed that the posterior expansion of Otx2 expression in these early Chd7−/− embryos was associated with reduced Fgf8 expression ( Figure 3G , H ) , in agreement with previously reported repression of Fgf8 expression by OTX2 ( Acampora et al . , 2001; Heimbucher et al . , 2007 ) . A side-by-side comparison of stage-matched embryos indicated that Fgf8 expression was initiated at the correct position in Chd7−/− embryos ( Figure 3I , J ) . This observation showed that the posterior expansion of Otx2 expression was not associated with a re-positioning of the IsO and indicated that Otx2 was mis-expressed in the anterior hindbrain of Chd7−/− embryos . Therefore , we asked whether the abnormal expansion of Otx2 expression into the hindbrain just posterior to the IsO affected the identity of r1 . Indeed , expression of the homeobox gene Gbx2 , a marker of r1 , was downregulated in Chd7−/− embryos ( Figure 3K , L ) . These findings are consistent with the known regulatory interactions between Otx2 , Gbx2 and Fgf8 ( Figure 3M ) ( Broccoli et al . , 1999; Millet et al . , 1999 ) . We conclude that CHD7 functions as a key regulator of homeobox gene expression in the early neural tube and that the loss of Chd7 results in the altered expression of Otx2 and Gbx2 , and the concomitant transformation of r1 into a more anterior identity . Interestingly , the effect of Chd7 mutation on Otx2 expression appears to be highly context-dependent as Otx2 is reported to be downregulated in the otic and olfactory regions of Chd7-deficient embryos ( Hurd et al . , 2010; Layman et al . , 2011 ) . The data presented thus far indicated that CHD7 is required for normal Otx2 and Gbx2 gene expression . We therefore sought evidence for CHD7 recruitment to Otx2 and Gbx2 regulatory regions . CHD7-associated chromatin was isolated from the mes/r1 region of E9 . 5 embryos by chromatin immunoprecipitation ( ChIP ) , and genomic DNA fragments ( indicated as #1–#10 in Figure 4A ) quantified by qPCR . Specific CHD7 binding was observed at three Otx2 enhancer elements identified by Kurokawa et al . ( Kurokawa et al . , 2004a , 2004b ) . The FM1 enhancer , located ∼71–73 kb upstream ( #8 in Figure 4A ) , and the FM2 enhancer ( #1 in Figure 4A ) , located ∼118 kb downstream of Otx2 are by themselves sufficient to direct gene expression to the forebrain and midbrain after E9 , and deletion of both enhancers together results in a smaller rostral brain and expanded r1 ( Sakurai et al . , 2010 ) . DNA fragments within these Otx2 enhancers ( #1 , #8 ) were specifically enriched by CHD7 ChIP ( Figure 4A ) . In addition , CHD7 was also present at the AN enhancer ( #9 and #10 in Figure 4A ) that can direct gene expression to the epiblast and anterior neuroectoderm prior to E9 . 0 ( Kurokawa et al . , 2004b ) , consistent with the observation that Otx2 expression was altered in E7 . 5 embryos ( data not shown ) . We also detected CHD7 association with regions downstream of Otx2 ( #3 ) , and the promotor regions for Otx2 . 1 ( #6 ) and Otx2 . 2 ( #7 ) transcripts . No specific enrichment was detected at two negative control regions ( #2 and #4 ) . These data suggested that CHD7 is recruited to several key Otx2 regulatory elements in the embryonic mes/r1 region . 10 . 7554/eLife . 01305 . 006Figure 4 . Association of CHD7 with Otx2 and Gbx2 regulatory regions in the mes/r1 region . ( A ) Genomic map of the mouse Otx2 locus . The transcriptional start sites of Otx2 . 1 and Otx2 . 2 transcripts are indicated by arrows and exons by tan-coloured boxes . Positions on chromosome 14 indicated above the diagram are according to the mm9 genome assembly and numbers below the horizontal lines indicate approximate positions relative to the Otx2 . 2 transcriptional start site . Known Otx2 enhancer regions FM1 , FM2 and AN are indicated by blue boxes ( Kurokawa et al . , 2004a , 2004b ) . The location of DNA fragments amplified by qPCR after ChIP are indicated by rectangular boxes numbered #1–#10 . Open boxes indicate negative control regions . ChIP-qPCR data are presented in a graph , with % of input DNA on the Y-axis and amplified region on the X-axis . Results from ChIP reactions using a CHD7-specific antiserum are in magenta and control Ig in turquoise . Error bars indicate standard deviation from reactions performed in triplicate . ( B ) Genomic map of the mouse Gbx2 locus with the transcriptional start site ( TSS ) indicated by an arrow and exons by tan-coloured boxes . Positions on chromosome 1 indicated above the diagram are according to the mm9 genome assembly and numbers below the horizontal lines indicate approximate positions relative to the TSS . The location of DNA fragments amplified by qPCR after ChIP are indicated by rectangular boxes numbered #1–#6 . Open boxes indicate negative control regions . ChIP-qPCR data are presented in a graph , with % of input DNA on the Y-axis and amplified region on the X-axis . Results from ChIP reactions using a CHD7-specific antiserum are in magenta and control Ig in turquoise . Error bars indicate standard deviation from reactions performed in triplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 01305 . 006 A regulatory region 6 kb upstream of zebrafish Gbx2 capable of driving gene expression in r1 , has been described by Islam et al . ( 2006 ) . ChIP-qPCR experiments identified substantial CHD7 recruitment to a region 5–6 . 25 kb upstream ( #3 , #4 and #5 in Figure 4B ) as well as 3 . 7 kb upstream of Gbx2 ( #1 ) in mes/r1 tissue . These observations suggested that CHD7 might regulate Gbx2 expression in r1 by interacting with Gbx2 regulatory elements . However , further experiments will be required to test whether these regions do indeed control Gbx2 expression in mouse r1 . Taken together , our observations support the supposition that homeobox genes represent key CHD7 targets . The mechanisms controlling CHD7 recruitment to regulatory regions and the action whereby CHD7 might affect gene expression in the embryo remain to be elucidated . Cerebellar defects have been reported in pre-term CHARGE fetuses ( Becker et al . , 2001; Sanlaville et al . , 2006; Legendre et al . , 2012 ) . To determine whether cerebellar defects are a common post-natal feature of CHARGE syndrome , we systematically examined cerebellar structure in a cohort of 20 patients with CHARGE syndrome and mutations in the CHD7 gene . MRI scans revealed cerebellar defects in 55% ( 11/20 ) of these patients ( Figure 5; Table 1 ) . Patients exhibited cerebellar vermis hypoplasia , varying from slight to pronounced hypoplasia ( 35% , 7/20 , Figure 5B , C ) and an anticlockwise rotated vermis ( 35% , 7/20 , Figure 5B , C’ ) . As a consequence of these abnormalities , fluid-filled spaces surrounding the cerebellum appeared larger . Examples of large foramen of Magendi and fourth ventricle ( 50% , 10/20 ) and large subcerebellar cistern ( 25% , 4/20 ) are indicated in Figure 5B–D . Thus , cerebellar defects in CHARGE syndrome have some clinical similarities to Dandy-Walker malformations ( vermis hypoplasia and anticlockwise rotated vermis ) , without the overt posterior fossa enlargement typical of Dandy-Walker malformation ( Doherty et al . , 2013 ) . Two patients with vermis hypoplasia exhibited broad gait or ataxia , consistent with defects that disrupt cerebellar function ( Table 1 ) . Furthermore , 25% ( 5/20 ) of the patients had foliation abnormalities ( Figure 5D , D’ , Table 1 ) , implying additional roles for CHD7 during the process of foliation . We conclude that a substantial proportion of patients with CHARGE syndrome present with cerebellar vermis hypoplasia . The incomplete penetrance of cerebellar vermis defects in patients with CHD7 mutations is consistent with our studies in the mouse , which suggests that mutations in FGF pathway genes are likely to substantially modify the severity of cerebellar defects in CHARGE syndrome . 10 . 7554/eLife . 01305 . 007Figure 5 . Representative sagittal MRI scans of CHARGE syndrome patients . ( A ) Sagittal T1 scan of patient #18 showing a normal vermis with a normal position , foramen of Magendi ( asterisk ) and subcerebellar cistern ( SC ) . The orientation of the cerebellum relative to the brainstem is indicated by two parallel white lines . ( B ) Sagittal T1 scan of patient #5 showing pronounced vermis hypoplasia with an anticlockwise rotated axis relative to the axis of the brainstem ( arrow ) , and ensuing large foramen of Magendi ( asterisk ) and subcerebellar cistern ( SC ) . Cerebellar hemispheres are normal ( not shown ) . ( C and C’ ) Illustrative sagittal T1 MRI images of patient #3 showing a slightly hypoplastic vermis . The white lines and arrow in C’ indicate the anticlockwise-rotated axis of the vermis compared to the axis of the brainstem , with ensuing large foramen of Magendi ( asterisk ) and subcerebellar cistern ( SC ) indicated in C . ( D ) Transverse Inversion Recovery MRI image of patient #10 showing abnormal foliation in the caudal cerebellar hemispheres extending into the cerebellar tonsils ( arrow ) . Also note a wide foramen of Magendi ( asterisk ) . ( D’ ) Transverse T2 MRI image of patient #11 , with abnormal foliation in the anterior vermis indicated by an arrow . ( E ) Transverse Inversion Recovery image and ( E’ ) T2 MRI image of a control patient with normal cerebellum . DOI: http://dx . doi . org/10 . 7554/eLife . 01305 . 00710 . 7554/eLife . 01305 . 008Table 1 . Cerebellar findings on MRI scansDOI: http://dx . doi . org/10 . 7554/eLife . 01305 . 008PatientSex; age at MRI ( y;m ) CerebellumSuggestive neurological features ( age at last examination , y;m ) CHD7 mutation1M ( 1;1 ) Pronounced vermis hypoplasia with anticlockwise rotated axis , large foramen of Magendi and large subcerebellar cistern , fissure vermisNone ( 1;1 ) nonsense934C>T2M ( 0;1 ) Slight caudal vermis hypoplasia with slightly anticlockwise rotated axis , abnormal foliation , large foramen of Magendi , normal subcerebellar cisternAtaxic gait ( 4;4 ) nonsense7160C>A3M ( 1;0 ) Slight caudal vermis hypoplasia with anticlockwise rotated axis , large foramen of Magendi , large subcerebellar cistern ( Figure 5C , C’ ) None ( 12;4 ) deletion3202- ? 8994 ? del4F ( 0;3 ) Slight caudal vermis hypoplasia , with anticlockwise rotated axis , large foramen of Magendi , normal subcerebellar cisternNone ( 2;2 ) frameshift7106delT5M ( 5;7 ) Pronounced vermis hypoplasia , with anticlockwise rotated axis , large foramen of Magendi and large subcerebellar cistern ( Figure 5B ) None ( 7;10 ) frameshift4779delT6M ( 0;1 ) Slight caudal vermis hypoplasia , with anticlockwise rotated axis , large foramen of Magendi and large subcerebellar cisternNone ( 5;2 ) frameshift5680_5681delAG7F ( 2;9 ) Slight caudal vermis hypoplasia , with slightly anticlockwise rotated axis , large foramen of Magendi and large subcerebellar cisternBroad gait ( 11;6 ) missense3973T>G8M ( 1;8 ) Large foramen of Magendi , large fourth ventricle ( only on sagittal scans ) , normal subcerebellar cisternNone ( 12;2 ) splice site5535-7G>A9M ( 2;2 ) Large foramen of Magendi , large fourth ventricle ( only on sagittal scans ) , normal subcerebellar cistern . Abnormal foliation in anterior vermisNone ( 6;2 ) nonsense3173T>A10F ( 1;1 ) Abnormal foliation caudal cerebellar hemispheres and tonsils , large foramen of Magendi ( Figure 5D ) None ( 13;0 ) splice site UV3340A>T11F ( 15;10 ) Abnormal foliation in anterior vermis ( Figure 5D’ ) None ( 18;0 ) splice site3990-1G>C12M ( 10;3 ) Abnormal foliation in anterior vermisMotor dyspraxia ( 16;10 ) frameshift5564dupC13M ( 0;1 ) Normal ( indented cranial pons ) None ( 0;11 ) frameshift1820_1821insTTGT14F ( 15;10 ) Normal ( large fourth ventricle ) None ( 20;6 ) nonsense4015C>T15F ( 0;1 ) Normal , ( split caudal vermis ) None ( 5;9 ) nonsense7879C>T16M ( 0;6 ) NormalBroad gait ( 10;6 ) splice site2238+1 G>A17M ( 1;10 ) NormalNone ( 6;4 ) nonsense1480C>T18F ( 2;10 ) Normal ( Figure 5A ) None ( 17;3 ) frameshift7769delA19M ( 1;0 ) NormalNone ( 16;9 ) nonsense1714C>T20M ( 6;3 ) NormalNone ( 12;10 ) splice site2443+5 G>C*all children show motor delay due to vestibular defects . In summary , this study identifies the chromatin-remodelling factor CHD7 as a key upstream regulator of homeobox gene expression and positional identity in the early neural tube and demonstrates a connection between CHD7 haplo-insufficiency , reduced FGF signalling and cerebellar defects in a human syndrome . We propose that CHD7 remodels chromatin at multiple Otx2 and Gbx2 regulatory elements , thereby modifying higher order chromatin architecture and interactions with tissue-specific transcription factors at these loci . Although we cannot completely rule out the possibility that CHD7 also directly fine-tunes Fgf8 expression in addition to affecting Otx2 and Gbx2 expression , the finding that Fgf8 expression is not substantially changed in the pharyngeal region of Chd7−/− embryos ( compare e . g . , Figure 3G with H ) , suggest that such effects will have to be mediated by CHD7 recruitment to tissue-specific Fgf8 regulatory elements . Our findings predict that mutations and epigenetic alterations of OTX2 and GBX2 regulatory regions are likely to contribute to cerebellar hypoplasia in humans and that OTX2 , GBX2 and FGF8 deregulation might underlie other developmental defects associated with CHARGE syndrome . The Chd7XK403 and Fgf8lacZ/+ loss-of-function alleles were maintained on C57BL/6J and C57BL/6J × DBA/2J F1 backgrounds for these studies ( Ilagan et al . , 2006; Randall et al . , 2009 ) . Tail DNA preparations were genotyped by PCR as described in the original publications . All animal procedures were approved by the UK Home Office . Brains were dissected in ice-cold PBS , fixed in 4% paraformaldehyde ( PFA ) overnight at 4°C , before dehydration and embedding in paraffin wax . Volumetric measurements were carried out on P0 and P21 cerebella . Serial , sagittal 10 μm sections of the cerebellum were dried overnight at 42°C , rehydrated and stained with 0 . 1% cresyl violet . Images of stained sections were taken and the cerebellar surface area on each section traced and measured by ImageJ . The total volume of each cerebellar region was calculated by multiplying the total surface area of all sections from the same region by the thickness of the sections . Vermis sections were selected as the most medial sections with clearly visible 10 lobules; paravermis sections were adjacent to the vermis sections , with diminishing lobules VIII , IX and X; sections lateral to paravermis were hemisphere sections . Noon on the day a vaginal plug was observed was defined as embryonic day ( E ) 0 . 5 . Somite-stage embryos were staged more accurately by counting the number of somite pairs . After dissection in ice-cold PBS , embryos were fixed overnight in 4% PFA at 4°C , gradually dehydrated in a methanol series and in situ hybridisation carried out using standard procedures ( Wilkinson et al . , 1989b ) . Digoxigenin-labelled antisense probes for Etv5 ( Hippenmeyer et al . , 2002 ) , Fgf8 ( Crossley and Martin , 1995 ) , Gbx2 ( Wassarman et al . , 1997 ) , Hoxa2 ( Wilkinson et al . , 1989a ) and Otx2 ( Simeone et al . , 1993 ) were prepared using previously published constructs . Total RNA was extracted from the mes/r1 region of at least three E9 . 5 embryos of each genotype using Trizol ( Invitrogen , UK ) with the addition of 20 µg Ultrapure Glycogen ( Life Technologies , UK ) . A total of 200 ng of RNA was used for first-strand DNA synthesis with the nanoScript Precision RT kit ( PrimerDesign Ltd . , UK ) using random hexamer primers . cDNA synthesis reactions without reverse transcriptase enzyme ( no RT ) were used as controls for quantitative RT-PCR . Quantitative RT-PCR was performed on a Rotor-Gene Q ( Qiagen ) using Precision qPCR MasterMix kit with SYBR green ( Primerdesign Ltd . , UK ) . All reactions were performed in triplicate . Cq threshold values were determined manually and all were at least 5 Cq values below no RT controls . The Cq values for each sample was normalised to the internal control gene Ywhaz ( primers provided by Primerdesign ) to give the ΔCq value . ΔΔCq values were calculated relative to wildtype samples . The primer sequences used were: Fgf8: forward 5′-AGGTCTCTACATCTGCATGAAC-3′ , reverse 5′-TGTTCTCCAGCACGATCTCT-3′; Etv5: forward 5′-GCAGTTTGTCCCAGATTTTCA-3′ , reverse 5′-GCAGCTCCCGTTTGATCTT-3′ . The embryonic mes/r1 region was dissected from E9 . 5 CD1 embryos , disrupted by trituration with a 23 G and 25 G needle , fixed for 10 min with 4% PFA , snap-frozen and stored at −80°C until use . After cell lysis and isolation of nuclei , samples were sonicated in a Bioruptor UCD-300 in 10 mM TRIS pH8 , 1 mM EDTA , 0 . 5 mM EGTA , 0 . 5% N-lauroylsarcosine to 200–500 bp fragment size . Chromatin was immunoprecipiated with antiserum to CHD7 ( ab31824 , Abcam , UK ) and control Ig ( Abcam , UK ) . Complexes were captured with Protein G Dynabeads , washed with modified RIPA buffer ( 50 mM HEPES pH7 . 5 , 1 mM EDTA , 0 . 3% Sodium deoxycholate , 1% NP40 , 250 mM LiCl ) , eluted in 50 mM TRIS pH8 , 10 mM EDTA , 1% SDS , cross-links reversed by overnight incubation at 65°C and DNA precipitated after phenol-chloroform extraction . Unique DNA fragments were amplified and quantified by qPCR using the primers in Tables 2 and 3 . Data were quantified relative to input DNA ( % of input ) . 10 . 7554/eLife . 01305 . 009Table 2 . Otx2 qPCR primersDOI: http://dx . doi . org/10 . 7554/eLife . 01305 . 009RegionForwardReverse#1AAACTCACCATAATCCTCCTGCCTCCTCCCCTTCTCCTCTAAACAGC#2CTGCTCTCCTCAACCTTCAGACTCTTGCGTGCCTTACCTTACCG#3CAACCACTCAAGTCAAGCCTATCTGTCTTCCTCTGCCTCCCAAGTTC#4CTGGCTGGTGGCTTCTGATTTTAGGTATCGCCAGGTTGCC#5ACACCAACTTGCTGAACAACATCCAGACTACTAATTAGGTGAAAATGA#6GAAAACCAAAACCCAAACCACGGAATGGAATCCTTAGCAAGCGG#7AACAGGCTTGTGTCCGTCTACGCGCTTTCTCAGCAAATCTCCC#8CATTTTCTTGCCGTCCTGCCAAAGTGTGCCTCCTGTGGTTCC#9AAAAACACTGGGGAAGAAAGGGAAATAAGAGTCAGAAGAGCGGTGC#10GCTGAATCAAACATGAATGAGCCCTGGGAGTAGACAACTGAGACA10 . 7554/eLife . 01305 . 010Table 3 . Gbx2 qPCR primersDOI: http://dx . doi . org/10 . 7554/eLife . 01305 . 010RegionForward ( 5’–3’ ) Reverse ( 5’–3’ ) #1CCCTTGGCTGGCTTTGAAATTCTGCCTTTTGTCCTGGAGA#2TGAATCCATAGCTTACCCGCAGGAACAAAGGGGGAAAGAA#3CCAGGCTTTCATCTCTCGCAATAGGCCAAGCTAAGCACCC#4GGGAATGGTGGAATGAATGGCTGAGGAGTGTGCTGAAGGGACAAC#5GTTGGCTGCCCTTTTCTTCAACCTCCATCTCCTCAGGCTA#6TGTAAACACTCCCTTCCCCGTATCCCACCCTAAACCGAAATGCG All patients included in this study are known at the Dutch expert clinic for CHARGE syndrome located at the University Medical Center Groningen ( coordinated by CvRA ) . All patients had a pathogenic mutation in CHD7 ( Table 1 , see also www . CHD7 . org ) . Patients were all evaluated in person by CvRA . Patients and/or parents gave written consent for the collection and analysis of detailed phenotypic information according to national ethical guidelines . Phenotypic information collected also included radiological images . All information is stored in a secure database under a unique patient identification number . The MRI scans were made at different hospitals . A total of 23 MRI scans could be collected . Only MRI scans that allowed a reliable interpretation of the cerebellum , that is the presence of sagittal and axial images of the cerebellum , were included in this study ( n = 20 ) . All cerebellar images were evaluated on visual basis by an experienced neuroradiologist ( LCM ) . MRI images of CHARGE patients were compared with images of age-matched controls . All observations were recorded by MTYW .	CHARGE syndrome is a rare genetic condition that causes various developmental abnormalities , including heart defects , deafness and neurological defects . In most cases , it is caused by mutations in a human gene called CHD7 . CHD7 is known to control the expression of other genes during embryonic development , but the molecular mechanisms by which mutations in CHD7 lead to the neural defects found in CHARGE syndrome are unclear . During embryonic development , the neural tube—the precursor to the nervous system—is divided into segments , which give rise to different neural structures . The r1 segment , for example , forms the cerebellum , and the secretion of a protein called FGF8 ( short for fibroblast growth factor 8 ) by a nearby structure called the isthmus organiser has an important role in this process . Since a reduction in FGF8 causes defects similar to those found in CHARGE syndrome , Yu et al . decided to investigate if the FGF signalling pathway was involved in this syndrome . Mice should have two working copies of the Chd7 gene , and mice that lack one of these suffer from symptoms similar to those of humans with CHARGE syndrome . Yu et al . examined the embryos of these mice and found that the isthmus organiser produced less FGF8 . Embryos with no working copies of the gene completely lost the r1 segment . The loss of this segment appeared to be caused by changes in the expression of homeobox genes ( the genes that determine the identity of brain segments ) . Embryos that did not have any working copies of the Chd7 gene died early in development , which made further studies impossible . However , embryos that had one working copy of the Chd7 gene survived , and Yu et al . took advantage of this to study the effects of reduced FGF8 expression on these mice . These experiments showed that mice with just one working copy of the Fgf8 gene and one working copy of the Chd7 gene had a small cerebellar vermis . This part of the cerebellum is known to be very sensitive to changes in FGF8 signalling . Yu et al . then used an MRI scanner to look at the cerebellar vermis in patients with CHARGE syndrome , and found that more than half of the patients had abnormal cerebella . In addition to confirming that studies on mouse embryos can provide insights into human disease , the work of Yu et al . add defects in the cerebellar vermis to the list of developmental abnormalities associated with CHARGE syndrome . The next step will be to test if any mutations in the human FGF8 gene can contribute to cerebellar defects in CHARGE syndrome , and to investigate if any other developmental defects in CHARGE syndrome are associated with abnormal FGF8 levels .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2013	Deregulated FGF and homeotic gene expression underlies cerebellar vermis hypoplasia in CHARGE syndrome
The maintenance of excitatory and inhibitory balance in the brain is essential for its function . Here we find that the developmental axon guidance receptor Roundabout 2 ( Robo2 ) is critical for the maintenance of inhibitory synapses in the adult ventral tegmental area ( VTA ) , a brain region important for the production of the neurotransmitter dopamine . Following selective genetic inactivation of Robo2 in the adult VTA of mice , reduced inhibitory control results in altered neural activity patterns , enhanced phasic dopamine release , behavioral hyperactivity , associative learning deficits , and a paradoxical inversion of psychostimulant responses . These behavioral phenotypes could be phenocopied by selective inactivation of synaptic transmission from local GABAergic neurons of the VTA , demonstrating an important function for Robo2 in regulating the excitatory and inhibitory balance of the adult brain . The midbrain dopamine system , consisting of the VTA and substantia nigra pars compacta ( SNc ) , is essential for motor function , motivation , reward , learning , and memory . Alterations in the activity patterns of dopamine neurons have been proposed as a key contributor to mental illness ( Grace , 1991 ) . In addition to dopamine , accumulating evidence points to an essential balance of excitatory and inhibitory neurotransmitter systems in the brain to allow for proper function . Shifts in this balance are increasingly linked to a variety of mental disorders including schizophrenia ( Eichler and Meier , 2008 ) and autism ( Nelson and Valakh , 2015 ) . The VTA dopamine system is also broadly implicated in addiction where changes in inhibitory and excitatory strength are proposed to underlie drug-seeking behavior ( Chen et al . , 2010 ) . The cellular mechanisms that maintain the excitatory and inhibitory synaptic control of the adult VTA are not well known , but resolving these processes has important implications for resolving the molecular regulation of circuit connectivity . To identify additional genes that may regulate synaptic connectivity in the midbrain dopamine system , we surveyed the Allen Institute mouse brain expression atlas ( Lein et al . , 2007 ) for genes with enriched or partially enriched expression in the VTA/SNc that had previously defined roles in axonal pathfinding , synaptogenesis , or plasticity , and have linkage to mental illness . One of the genes that we identified encodes for the axon guidance receptor Robo2 . Robo2 was first discovered as one of the four mammalian homologs to the Drosophila Roundabout gene ( Robo ) , which was named because of the unique axonal recrossing pattern generated at the CNS midline ( Seeger et al . , 1993; Kidd et al . , 1998; Brose et al . , 1999; Simpson et al . , 2000 ) . Further studies found that Robo family members perform additional axon guidance functions to regulate brain wiring broadly throughout the nervous system ( Ypsilanti and Chedotal , 2014 ) , including midbrain dopamine neurons ( Dugan et al . , 2011 ) . Additional studies expanded the essential functions of Robo2 to include cell migration , synaptogenesis , synaptic plasticity , neuronal survival , and dendritic patterning during prenatal and early postnatal development ( Shen and Cowan , 2010; Gibson et al . , 2014; Koropouli and Kolodkin , 2014; Ypsilanti and Chedotal , 2014 ) . Previous studies demonstrate that the neuronal axon guidance receptor deleted in colorectal cancer ( DCC ) is also highly expressed in the adult dopamine neurons , is dynamically regulated by amphetamine ( Yetnikoff et al . , 2007 ) , and mice with haploinsufficiency for DCC demonstrate blunted responses to the psychomotor activating effects of amphetamine ( Flores et al . , 2005 ) . Consistent with the continued function of DCC in the adult nervous system , it has recently been demonstrated that DCC regulates excitatory synaptic transmission in the adult hippocampus ( Horn et al . , 2013 ) . These findings are consistent with a role for adult expression of neuronal pathfinding genes in the regulation of the adult dopamine system . During early development of the spinal cord , Robo and DCC signaling have opposing roles in commissural axon guidance , with DCC activation promoting axon growth toward the midline and Robo serving as a chemorepellant to prevent recrossing ( Tessier-Lavigne and Goodman , 1996 ) . Robo receptors are activated upon binding chemorepellants of the Slit family , Slit1 , Slit2 , and Slit3 ( Tessier-Lavigne and Goodman , 1996 ) , which cause growth cone collapse through cytoskeletal remodeling via Rho-family GTPase signaling ( Wong et al . , 2001 ) . All three genes for the Slit ligands and the receptors Robo1 and Robo2 ( Marillat et al . , 2002 ) are expressed in the adult midbrain , including the VTA and SNc; however , the function of these ligands and receptors in this context and their cell-specific requirements has yet to be established . Of further interest , genes involved in Robo signaling have been linked to schizophrenia , autism , conduct disorder , intellectual disability , language impairment , and bipolar depression ( Anitha et al . , 2008; Potkin et al . , 2009 , 2010; Viding et al . , 2010; Suda et al . , 2011; Dadds et al . , 2013; St Pourcain et al . , 2014; Wang et al . , 2014; Okbay et al . , 2016 ) . Based on the expression of Robo2 in the adult limbic system of the brain , its role in establishing the early wiring of the midbrain dopamine system ( Dugan et al . , 2011 ) , and its opposing function related to netrin/DCC signaling , we hypothesized that this gene may play an important role in maintaining or modulating synaptic connectivity of the adult dopamine system . To test this hypothesis , we conditionally inactivated the Robo2 gene exclusively in the adult VTA of mice . We find that Robo2 regulates the inhibitory synaptic connectivity of the VTA , and that disruption of this connectivity has profound behavioral effects including altered psychomotor activity and impaired learning . Most intriguing was a paradoxical inversion of behavioral responses to the psychostimulants cocaine and amphetamine , whereby these drugs calmed hyperactivity in mice lacking Robo2 in the adult midbrain . These findings strongly implicate Robo2 signaling in maintaining balanced control of the VTA that is highly relevant for behavioral disorders linked to altered psychomotor control and cognition such as autism , ADHD , addiction , and schizophrenia . Analysis of the Allen Institute mouse brain expression atlas ( Lein et al . , 2007 ) reveals that Robo2 is broadly expressed in the adult brain with the highest expression in the ventral midbrain , hippocampus and cerebellum , similar to previous reports ( Marillat et al . , 2002 ) ( Figure 1—figure supplement 1A ) . To confirm the expression of Robo2 in the VTA , we performed immunostaining for Robo2 and tyrosine hydroxylase ( TH , a dopamine neuron marker ) on brain sections from adult mice ( >8 weeks old ) . Robo2 expression was observed throughout the VTA , largely localized to TH-positive neurons , though several TH-negative neurons were observed to also express Robo2 ( TH+Robo2+ = 75 . 28 ± 0 . 56% , TH+Robo2− = 7 . 81 ± 1 . 46% , and TH-Robo2+ = 16 . 90 ± 1 . 77% , n = 4 ) ( Figure 1A ) . 10 . 7554/eLife . 23858 . 003Figure 1 . Robo2 VTA mutants have a reduction in the frequency of sIPSCs . ( A ) Robo2 is expressed in the VTA , predominantly in dopamine neurons ( TH+ cells , arrows ) , but also non-dopamine neurons ( TH- cells , arrowheads ) . Scale bar = 100 µm ( Robo2 ) , 10 µm ( merge/zoom ) . ( B ) The genomic structure of the Robo2lox/lox allele before and after Cre recombination , adapted from Lu et al . ( 2007 ) . ( C ) Schematic of strategy to virally inactivate Robo2 in the VTA ( left ) . Representative image ( right ) showing expression of Cre-GFP and TH in the VTA . Scale bar = 100 µm . ( D ) sIPSC representative traces for control and mutant Ih positive cell . ( E ) sIPSC frequency ( left ) and cumulative distribution plot ( right ) for control ( n = 13 ) and mutant ( n = 16 ) Ih positive cells . Unpaired t-test , t ( 27 ) = 3 . 944 , *p<0 . 001 . Two-way repeated measures ANOVA , genotype x time interaction , F ( 27 , 100 ) = 17 . 30 , p<0 . 0001 . ( F ) sIPSC amplitude ( left ) and cumulative distribution plot ( right ) for control and mutant Ih positive cells . ( G ) sIPSC representative trace for control and mutant Ih negative cell . ( H ) sIPSC frequency ( left ) and cumulative distribution plot ( right ) for control ( n = 10 ) and mutant ( n = 14 ) Ih negative cells . Unpaired t-test , t ( 22 ) = 3 . 459 , p<0 . 01 . Two-way repeated measures ANOVA , genotype x time interaction , F ( 22 , 100 ) =11 . 40 , ***p<0 . 0001 . ( I ) sIPSC amplitude ( left ) and cumulative distribution plot ( right ) for control and mutant Ih negative cells . Bars represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 23858 . 00310 . 7554/eLife . 23858 . 004Figure 1—figure supplement 1 . Robo2 VTA mutants . ( A ) In situ hybridization for Robo2 in the rostral , medial , and caudal VTA ( image credit: Allen Institute ) . Scale bar = 500 µm . ( B ) Traces of the viral transduction of AAV1-Cre-GFP in Robo2lox/lox animals ( n = 10 ) in the rostral , medial , and caudal VTA . Each color represents one animal and corresponds across the three sections and to the cell count quantification in ( C ) . Scale bar = 500 µm . ( C ) Cell count of AAV1-Cre-GFP positive cells in the rostral , medial , and caudal VTA . ( D ) Slc6a3 ( DAT ) mRNA is enriched in midbrain dopamine ( n = 3 animals ) but not midbrain GABA neurons ( n = 3 animals ) . ( E ) Slc32a1 ( Vgat ) mRNA is enriched in midbrain GABA but not midbrain dopamine neurons . ( F ) Robo2 mRNA is enriched in both midbrain dopamine and midbrain GABA neurons . ( H ) The distribution of cell types in Robo2lox/lox animals injected with either AAV1-∆Cre-GFP ( n = 4 ) or AAV1-Cre-GFP ( n = 4 ) . Bars represent mean ± SEM . ( G ) Expression of the control virus , AAV1-∆Cre-GFP does not affect expression of Robo2 ( arrows in top ) . AAV1-Cre-GFP causes loss of Robo2 in GFP+ cells of the VTA ( arrowheads in bottom ) , but Robo2 expression is still expressed in GFP- cells . Scale bar = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 23858 . 00410 . 7554/eLife . 23858 . 005Figure 1—figure supplement 2 . Robo2 VTA mutants have normal sEPSCs . ( A ) Representative trace from a recorded Ih positive cell . ( B ) sEPSC representative traces for control and mutant Ih positive cell . ( C ) sEPSC frequency ( left ) and cumulative distribution plot ( right ) for control ( n = 10 ) and mutant ( n = 11 ) Ih positive cells . ( D ) sEPSC amplitude ( left ) and cumulative distribution plot ( right ) of control and mutant Ih positive cells . ( E ) Representative trace from a recorded Ih negative cell . ( F ) sEPSC representative trace for control and mutant Ih negative cell . ( G ) sEPSC frequency ( left ) and cumulative distribution plot ( right ) for control and mutant Ih negative cells . ( H ) sEPSC amplitude ( left ) and cumulative distribution plot ( right ) for control and mutant Ih negative cells . Bars represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 23858 . 00510 . 7554/eLife . 23858 . 006Figure 1—figure supplement 3 . Robo2 VTA mutants . ( A ) Representative examples of asymmetrical and symmetrical synapses in control and mutant mice . Blue shading indicates presynaptic terminal . Pink shading indicates postsynaptic terminal . Scale bar = 500 nm . ( B ) Average number of synapses of specified type observed from control ( n = 16 images from 2 mice ) and mutant ( n = 14 images from 2 mice ) mice . Synapses were identified by the presence of synaptic vesicles juxtaposed to a clearly identifiable postsynaptic site in the presence or absence of a postsynaptic density . Symmetrical synapses , control versus mutant: Unpaired t-test , t ( 28 ) = 0 . 08934 , p=0 . 9295 . Asymmetrical synapses , control versus mutant: Unpaired t-test , t ( 28 ) = 1 . 119 , p=0 . 2725 . ( C ) Average waveform of Ih positive cells in control ( n = 13 ) and mutant ( n = 16 ) mice . ( D ) Quantification of tau value in Ih positive cells in controls ( n = 13 ) and mutants ( n = 16 ) . Unpaired t-test , t ( 27 ) = 1 . 619 , p=0 . 1171 . ( E ) Average waveform of Ih negative cell in control ( n = 10 ) and mutant ( n = 14 ) mice . ( F ) Quantification of tau value in Ih negative cell in controls ( n = 10 ) and mutants ( n = 14 ) . Unpaired t-test , t ( 22 ) = 0 . 6904 , p=0 . 4972 . Bars represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 23858 . 006 To confirm Robo2 expression in the midbrain , we performed Ribotag , a method which allows for cell-type specific enrichment of actively translating mRNAs ( Sanz et al . , 2009 ) . Mice containing the floxed HA-tagged Ribosomal Protein L22 ( Rpl22 ) gene were crossed to mice containing Cre recombinase expression under the control of the endogenous dopamine transporter locus ( Slc6a3Cre/+ ) , or the vesicular GABA transporter ( Slc32a1Cre/+ ) , allowing for isolation of dopamine and GABA neuron translatomes , respectively ( Figure 1—figure supplement 1D–E ) . Tissue punches of the adult midbrain from double heterozygous animals ( Slc6a3Cre/+::Rpl22lox-HA/+ , n = 3; Slc32a1Cre/+::Rpl22lox-HA/+ , n = 3 ) were immunoprecipitated ( IP ) to isolate Rpl22-HA tagged polyribosomes . TaqMan real time PCR was used to test for the relative enrichment of Robo2 in the IP fraction relative to total mRNA ( input ) . We found Robo2 to have enriched expression in the IP fraction relative to the input in both mouse lines ( Figure 1—figure supplement 1F ) confirming Robo2 expression in both dopamine and GABA neurons of the midbrain neurons . Based on previous findings that DCC regulates excitatory synaptic strength in the adult hippocampus ( Horn et al . , 2013 ) , we hypothesized that Robo2 may also play an important role in this process . To establish whether Robo2 in the adult VTA plays a functional role in synaptic function , we measured spontaneous excitatory and inhibitory postsynaptic currents ( sEPSCs and sIPSCs , respectively ) in putative dopamine and non-dopamine neurons following Robo2 inactivation . To inactivate Robo2 , an adeno-associated viral vector ( AAV ) containing an expression cassette for a GFP-tagged Cre recombinase ( AAV1-Cre-GFP ) was injected into the VTA of mice with a Cre-conditional allele for Robo2 ( ‘mutant mice’ ) ( Figure 1B–C and Figure 1—figure supplement 1B–C ) . The generation of Robo2lox/lox mice were described previously ( Lu et al . , 2007 ) . In this line , loxP recognition sites are inserted to flank exon 5 of Robo2 , and following Cre-mediated recombination , a frameshift occurs resulting in a premature stop codon in exon 6 ( Figure 1B ) . We confirmed a specific reduction in Robo2 protein in cells expressing Cre-GFP but not in cells expressing the enzymatically dead version of Cre , ∆Cre-GFP ( Figure 1—figure supplement 1G–H ) . Putative dopamine neurons in an acute slice preparation from AAV1-Cre-GFP::Robo2lox/lox ( mutant ) and AAV1-∆Cre-GFP::Robo2lox/lox ( control ) were identified by the presence of an afterhyperpolarization induced current ( Ih ) as described ( Johnson and North , 1992 ) ( Figure 1—figure supplement 2A ) , though this is not a definitive feature of all dopamine producing neurons ( Margolis et al . , 2006 ) . Surprisingly , we did not observe significant changes in either sEPSC frequency or amplitude ( Figure 1—figure supplement 2B-DA-C ) . In contrast , we did observe a significant reduction in sIPSC frequency in mutant mice relative to controls ( Figure 1D–E ) , with no change in sIPSC amplitude ( Figure 1D and F ) . Due to expression in putative non-dopamine neurons , as well as putative dopamine neurons , Robo2 signaling might also alter the connectivity and/or physiology of these cells . To test this , we recorded sIPSCs and sEPSCs on putative non-dopamine neurons in the VTA ( Ih-negative; Figure 1—figure supplement 2E ) . Similar to putative dopamine neurons , sIPSC frequency was reduced in Ih-negative neurons and sIPSC amplitude was not altered ( Figure 1G–I ) . We also did not observe differences in sEPSCs in these neurons ( Figure 1—figure supplement 2F-HD-F ) . A reduction in sIPSC frequency following Robo2 inactivation is consistent with a reduction in inhibitory synaptic input to midbrain neurons . Ultrastructural analysis of synaptic contacts using transmission electron microscopy did not reveal gross morphological differences in either symmetrical ( inhibitory ) or asymmetrical ( excitatory ) synapses ( Figure 1—figure supplement 3A ) , nor the number of observable synapses ( Figure 1—figure supplement 3B ) ; indicating that the reduction in sIPSC frequency is not due to a gross loss of inhibitory synaptic connectivity . Additionally , we did not observe differences in the inactivation kinetics of sIPSCs suggesting that the subunit composition of the postsynaptic GABAA receptors is not altered ( Figure 1—figure supplement 3C–F ) . GABAergic synaptic transmission potently regulates dopamine neuron activity patterns ( Paladini and Tepper , 1999 ) . To determine whether altered inhibitory synaptic inputs to dopamine neurons alters their activity in vivo , we recorded single-unit activity from the VTA of control and mutant mice ( Figure 2—figure supplement 1A ) . Putative dopamine neurons were identified by end of session sensitivity to the D2 agonist quinpirole that could be reversed with the D2 antagonist eticlopride ( Figure 2A–B ) . No differences in quinpirole sensitivity were observed between the two groups ( Figure 2—figure supplement 1D–F ) . We observed no difference between controls and Robo2 VTA mutants in the basal firing rate ( Figure 2C ) , nor did we detect significant differences in the number of burst firing , or phasic events ( Figure 2D ) . In contrast , we did observe a significant increase in the duration of burst events ( Figure 2E ) and the number of spikes per burst ( Figure 2F ) , consistent with a reduced inhibitory termination of burst epochs . 10 . 7554/eLife . 23858 . 007Figure 2 . Robo2 VTA mutants show increased burst activity . ( A , B ) Firing rate histogram of a representative cell displaying inhibition to quinpirole that is reversed with eticlopride in control ( A ) and mutant animal ( B ) . ( Left inset ) Waveform of represented cell on each wire of the tetrode . ( Right inset ) Cumulative spikes of represented cell across time ( min ) . Scale bar = 0 . 5 ms . ( C ) Firing rate of dopamine cells in controls ( n = 42 ) and mutants ( n = 50 ) is unchanged . ( D ) Burst rate of dopamine cells . ( E ) Burst duration of dopamine cells is longer in mutant cells . Unpaired t-test , t ( 90 ) = 2 . 17 , p<0 . 05 . ( F ) Spikes per burst in dopamine cells are increased in mutants . Unpaired t-test , t ( 90 ) = 2 . 33 , p<0 . 05 . ( G ) Firing rate of non-dopamine cells during in vivo electrophysiological recordings in controls ( n = 61 ) and mutants ( n = 71 ) . ( H ) Total normalized distribution of the inter-spike interval ( ISI ) in controls and mutants . ( I ) Skew of the ISI in non-dopamine cells is reduced in Robo2 VTA mutants . Unpaired t-test , t ( 130 ) = 2 . 42 , p<0 . 05 . ( J ) Coefficient of variation of the ISI in non-dopamine cells is reduced in mutants . Unpaired t-test , t ( 130 ) = 2 . 66 , p<0 . 01 . Bar represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 23858 . 00710 . 7554/eLife . 23858 . 008Figure 2—figure supplement 1 . in vivo electrophysiology . ( A ) Schematic showing tetrode placement in controls and mutants . ( B ) Representative image showing placement of tetrode in the VTA for in vivo electrophysiological recordings . ( C ) The distribution of dopamine and non-dopamine cells recorded showed no difference between controls ( n = 103 ) and mutants ( n = 121 ) . ( D ) Cumulative cell count of dopamine and non-dopamine cells in response to quinpirole ( right ) or eticlopride ( top ) or both ( left ) in control animals . ( E ) Cumulative cell count of dopamine and non-dopamine cells in response to quinpirole ( right ) or eticlopride ( top ) or both ( left ) in mutant animals . ( F ) The average change in dopamine cells in response to quinpirole ( top ) or eticlopride ( bottom ) showed no difference between controls and mutants . Bars represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 23858 . 008 To determine whether activity patterns of putative non-dopamine neurons are altered in vivo , we analyzed firing of quinpirole insensitive neurons recorded on the same tetrodes as quinpirole sensitive neurons . Similar to putative dopamine neurons , quinpirole insensitive neurons did not show changes in overall firing rate ( Figure 2G ) . However , these cells did show a significant change in the skew of their interspike-interval distribution ( ISI , Figure 2H–I ) . Consistent with this change in ISI distribution , quinpirole insensitive neurons showed a reduction in the coefficient of variation of the ISI ( CV-ISI , Figure 2J ) , indicative of a reduced number of inhibitory pauses . We next sought to determine whether changes in dopamine activity patterns result in altered dopamine release . To test this , we performed fast scan cyclic voltammetry ( FSCV ) to monitor dopamine release in the nucleus accumbens ( NAc ) in Robo2 mutant mice . Dopamine release was measured in response to increasing stimulus intensity delivered to an excitatory afferent input , the pedunculopontine tegmental area ( PPTg ) in urethane-anesthetized animals ( Figure 3A–B; Figure 3—figure supplement 1A ) as previously described ( Soden et al . , 2013 ) . At lower stimulus intensities , dopamine release was significantly elevated in Robo2 mutant mice compared to controls ( Figure 3C–H ) . We did not observe significant differences in dopamine release evoked by direct stimulation of the medial forebrain bundle ( MFB; Figure 3—figure supplement 1E–J ) , suggesting that dopamine transporter function is not altered in Robo2 mutant mice . Consistent with a lack of observed differences in the overall firing rate of dopamine neurons ( Figure 2C ) , we did not detect differences in tonic dopamine release as measured by microdialysis ( Figure 3—figure supplement 1K–L ) . 10 . 7554/eLife . 23858 . 009Figure 3 . Robo2 VTA mutants have increased phasic dopamine release . ( A ) Schematic of experimental design showing stimulation of the PPTg and FSCV recording in the NAc . ( B ) Representative voltammogram of dopamine recorded in the NAc in control ( left ) and mutant ( right ) . ( C–G ) Average dopamine oxidation current recorded in NAc after PPTg stimulation is increased in mutants ( n = 6 ) compared to controls ( n = 5 ) . ( C ) 50 μA: Two-way repeated measures ANOVA , effect of genotype , F ( 1 , 9 ) = 12 . 75 , p<0 . 01 . ( D ) 100 μA: Two-way repeated measures ANOVA , effect of genotype , F ( 1 , 9 ) = 5 . 84 , p<0 . 05 . ( E ) 150 μA: Two-way repeated measures ANOVA , effect of genotype , F ( 1 , 9 ) = 8 . 65 , p<0 . 05 . ( F ) 200 μA . ( G ) 250 μA . ( H ) Area under the curve of the dopamine current across all stimulation intensities . Two-way repeated measures ANOVA , effect of genotype , F ( 1 , 9 ) = 5 . 99 , p<0 . 05 . Bar represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 23858 . 00910 . 7554/eLife . 23858 . 010Figure 3—figure supplement 1 . Similar dopamine levels were recorded after MFB stimulation or microdialysis . ( A ) Schematic showing the recording electrode sites in the controls and mutants . ( B ) Representative image showing recording electrode placement in the NAc . ( C ) Representative image showing stimulating electrode placement in the PPTg . ( D ) Representative image showing stimulating electrode placement in the MFB . ( E–I ) Average dopamine oxidation current recorded in NAc after MFB stimulation shows no difference between controls ( n = 4 ) and mutants ( n = 4 ) . ( E ) 50 μA: Two-way repeated measures ANOVA , effect of genotype , F ( 1 , 6 ) = 2 . 54 , p=0 . 1620 . ( F ) 100 μA: Two-way repeated measures ANOVA , effect of genotype , F ( 1 , 6 ) = 3 . 49 , p=0 . 1110 . ( G ) 150 μA: Two-way repeated measures ANOVA , effect of genotype , F ( 1 , 6 ) = 2 . 08 , p=0 . 1997 . ( H ) 200 μA: Two-way repeated measures ANOVA , effect of genotype , F ( 1 , 6 ) = 0 . 53 , p=0 . 4958 . ( I ) 250 μA: Two-way repeated measures ANOVA , effect of genotype , F ( 1 , 6 ) = 0 . 04 , p=0 . 8467 . ( J ) Area under the curve of the dopamine current across all stimulation intensities after MFB stimulation shows no difference between controls and mutants . Two-way repeated measures ANOVA , Genotype X stimulation intensity interaction , F ( 4 , 24 ) = 0 . 06 , p=0 . 9930 . ( K , L ) Tonic dopamine levels calculated as percent area ratio of dopamine to D4-dopamine across four time points ( K ) and averaged ( L ) for controls ( n = 5 ) and mutants ( n = 5 ) . Unpaired t-test , t ( 8 ) = 0 . 1853 , p=0 . 8576 . Bars represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 23858 . 010 To test whether inactivation of Robo2 in the adult VTA has a functional impact on dopamine-dependent behavioral regulation , we monitored Robo2 mutant and control mice in a series of behavioral tests . Both AAV1-∆Cre-GFP injected Robo2lox/lox mice and AAV1-Cre-GFP injected Robo2+/+ mice were used as controls . We first tested locomotor activity across the day-night cycle . Robo2 mutants were significantly hyperactive compared to controls ( Figure 4A ) , particularly during the night ( Figure 4B ) . Consistent with altered locomotion in mutant mice , we observed reduced forelimb separation and greater print separation ( Figure 4—figure supplement 1A–F ) . 10 . 7554/eLife . 23858 . 011Figure 4 . Behavioral characterization of Robo2 VTA mutants show phenotypes in locomotion and cocaine sensitization . ( A ) Locomotion measured across three consecutive days and nights showing mutants ( n = 31 ) are hyperactive relative to controls ( n = 36 ) . Two-way repeated measures ANOVA , genotype x time interaction , F ( 252 , 16380 ) = 2 . 53 , *p<0 . 0001 . ( B ) Total distance traveled across three consecutive nights . Two-way repeated ANOVA , effect of genotype , F ( 1 , 65 ) = 12 . 49 , p<0 . 001 , Bonferroni’s multiple comparison test , p<0 . 001 , p<0 . 01 . ( C–D ) Locomotor response to cocaine ( 20 mg/kg ) in controls ( n = 20 ) and mutants ( n = 11 ) on day 1 ( C ) , Two-way repeated measures ANOVA , genotype x time interaction , F ( 35 , 1015 ) = 7 . 10 , **p<0 . 0001 and day 5 ( D ) , Two-way repeated measures ANOVA , genotype x time interaction , F ( 35 , 1015 ) = 8 . 23 , p<0 . 0001 . ( E ) Normalized locomotor response to cocaine by subtracting 60 min pre-cocaine from 60 min post-cocaine . Two-way repeated measures ANOVA , effect of genotype , F ( 1 , 29 ) = 18 . 26 , p<0 . 001 , Bonferroni’s multiple comparison test , *p<0 . 0001 , p<0 . 01 . ( F ) Locomotor response to amphetamine ( 2 . 5 mg/kg ) in controls ( n = 19 ) and mutants ( n = 17 ) . Two-way repeated measures ANOVA , genotype x time interaction , F ( 35 , 1190 ) = 6 . 43 , **p<0 . 0001 . ( G ) Locomotor response to the DAT blocker , GBR-12909 ( 10 mg/kg ) in controls ( n = 19 ) and mutants ( n = 17 ) . Two-way repeated measures ANOVA , genotype x time interaction , F ( 35 , 1190 ) = 2 . 94 , p<0 . 0001 . ( H ) Normalized locomotor response to amphetamine and GBR-12909 by subtracting 60 min pre-drug from 60 min post-drug . AMP: Unpaired t-test , t ( 34 ) = 2 . 968 p<0 . 01 , . GBR: Unpaired t-test , t ( 34 ) = 2 . 061 , p<0 . 05 . ( I ) Discrimination between a CSp and CSm is impaired in AAV1-Cre-GFP::Robo2lox/lox animals ( n = 16 ) compared to controls ( n = 22 ) . Two-way repeated measures ANOVA , genotype x time interaction , F ( 6 , 216 ) = 2 . 40 , p<0 . 05 , Bonferroni’s multiple comparison test , *p<0 . 01 and p<0 . 05 . Bars represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 23858 . 01110 . 7554/eLife . 23858 . 012Figure 4—figure supplement 1 . Robo2 VTA mutants have altered gait . ( A ) Hindleg stride length in controls ( n = 8 animals ) and mutants ( n = 8 animals ) . ( B ) Foreleg stride length . ( C ) Hindleg base . ( D ) Foreleg base , unpaired t-test , t ( 14 ) = 2 . 469 , p<0 . 05 . ( E ) Print separation , unpaired t-test , t ( 14 ) = 2 . 992 , p<0 . 01 . ( F ) Representative footprints from control and mutant showing forelimbs ( green ) and hindlimbs ( red ) . Scale bar = 1 cm . Bars represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 23858 . 01210 . 7554/eLife . 23858 . 013Figure 4—figure supplement 2 . Additional pharmacological studies on Robo2 VTA mutant mice . ( A ) Conditioned place preference ( CPP ) to cocaine ( 15 mg/kg ) in controls ( n = 11 ) and mutants ( n = 10 ) showed no difference between genotypes . ( B ) Locomotor response to saline in controls ( n = 28 ) and mutants ( n = 28 ) , Two-way repeated ANOVA , effect of genotype , F ( 1 , 54 ) = 5 . 19 , p<0 . 05 . ( C ) Locomotor response to fluoxetine ( 20 mg/kg ) in controls ( n = 7 ) and mutants ( n = 8 ) . Two-way repeated ANOVA , genotype x time interaction , F ( 35 , 455 ) = 2 . 17 , **p<0 . 001 . ( D ) Locomotor response to nisoxetine ( 10 mg/kg ) in controls ( n = 13 ) and mutants ( n = 11 ) . ( E ) Normalized locomotor response to saline , fluoxetine , and nisoxetine by subtracting 90 min pre-injection from 90 min post-injection . ( Nisoxetine ) unpaired t-test , t ( 13 ) = 2 . 780 , p<0 . 05 . ( F ) Robo2 VTA mutants have no significant difference in head entries during Pavlovian conditioning between control and mutant mice . ( G ) Acoustic pre-pulse inhibition at different pre-pulse intensities is not different between control ( n = 5 ) and mutant mice ( n = 5 ) . ( H ) Response to NMDA receptor antagonist MK-801 ( 0 . 2 mg/kg ) shows a difference in mutant mice ( n = 8 ) compared to control mice ( n = 7 ) , Two-way repeated measures ANOVA , genotype x time interaction , F ( 35 , 455 ) = 1 . 47 , p<0 . 05 . Bars represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 23858 . 013 Increased locomotor activity in Robo2 mutant mice suggests these animals may have enhanced psychomotor activation . To test this , control and Robo2 mutant mice were subjected to daily injections of cocaine ( 20 mg/kg ) for 5 days . In control mice , we observed a characteristic locomotor sensitization to cocaine ( Figure 4C–E ) . In contrast , Robo2 mutant mice showed an inversion in the normal locomotor response to cocaine ( Figure 4C–E ) , decreasing rather than increasing their activity; this pattern persisted across days . To determine whether the inverted psychomotor response to cocaine in Robo2 mutant mice leads to an anhedonic as opposed to a hedonic state , we measured conditioned place preference ( CPP ) to cocaine . Although cocaine CPP was not as robust in mutant mice compared to controls ( Figure 4—figure supplement 2A ) , it did not lead to an avoidance of the cocaine-paired side , suggesting that although their psychomotor response is altered , the hedonic aspect of cocaine is not significantly changed . An inverted psychostimulant response to cocaine is remarkably similar to that reported following genetic inactivation of the dopamine transporter ( Slc6a3-KO mice ) ( Gainetdinov et al . , 1999 ) . Analysis of Slc6a3-KO mice demonstrated that these animals not only have inverted responses to psychostimulants , but also a variety of monoamine transport blockers ( Gainetdinov et al . , 1999 ) . Similar to their response to cocaine , Robo2 mutant mice had inverted locomotor responses to amphetamine ( AMP ) and the DAT blocker GBR12909 ( GBR; Figure 4F–H ) . Robo2 mutant mice also had a differential response to the serotonin transport blocker fluoxetine with the mutants showing greater inhibition , but the norepinephrine transport blocker nisoxetine reduced activity equally in control and mutant mice ( Figure 4—figure supplement 2B–E ) . In addition to regulating psychomotor activation , VTA dopamine neurons play an essential role in regulating reward-related learning . To establish whether alterations in the dopamine system of Robo2 mutant mice affect this process , we monitored their performance in a Pavlovian associative learning paradigm . Robo2 mutant mice showed significant deficits in this task compared to control mice , with mutant mice failing to discriminate between predictive ( CSplus ) and non-predictive ( CSminus ) cues ( Figure 4I ) . This impairment was not associated with a reduction in reward seeking as total head entries into the reward receptacle did not differ ( Figure 4—figure supplement 2F ) . Hyperactivity and cognitive or learning impairments associated with enhanced phasic dopamine release may be related to a psychosis-like state in mutant mice ( Grace , 1991 ) . To address this , we monitored pre-pulse inhibition of the acoustic startle reflex and enhanced sensitivity to the psychomimetic NMDA receptor antagonist MK-801 , two metrics used to assess psychosis-related phenotypes in animal models ( Geyer , 2008 ) . Robo2 mutant mice did not differ from control mice in PPI , and although they are hyperactive at baseline , they do not show increased sensitivity to MK-801 . ( Figure 4—figure supplement 2G–H ) . Alterations in inhibitory control of dopamine neurons , enhanced phasic dopamine release , behavioral responses to psychostimulant drugs , and impaired reward learning all point to dopaminergic phenotypes following Robo2 inactivation in the VTA . To establish whether the observed behavioral phenotypes are the result of inactivation in dopamine producing neurons of adult mice , we crossed Robo2lox/lox mice to mice containing an inducible Cre coupled to the estrogen receptor ( iCreER ) under control of the dopamine transporter promotor ( Slc6A3-iCre/ERT2 ) . Treatment of adult mice with the estrogen receptor agonist tamoxifen ( 75 mg/kg x 5 days ) resulted in Cre-mediated recombination in dopamine neurons of the VTA and SNc ( Figure 5A ) . Behavioral analysis of tamoxifen treated DAT-iCreER::Robo2lox/lox mice did not result in hyperactivity , altered psychostimulant response to cocaine , or learning deficits relative to sham injected Slc6A3-iCre/ERT2::Robo2lox/lox mice ( Figure 5B–C; Figure 5—figure supplement 1A–D ) . 10 . 7554/eLife . 23858 . 014Figure 5 . Adult or embryonic inactivation of Robo2 in dopamine neurons does not recapitulate the Robo2 VTA mutant phenotypes . ( A ) Schematic showing expression of tamoxifen-inducible Cre specifically in adult dopamine neurons . Representative images from Slc6a3-iCreERT2 crossed to a Cre-dependent reporter line , Gt ( ROSA ) 26Sortm ( CAG-tdTomato ) Hze , shows induction of the tdTomato reporter is specific to animals receiving tamoxifen injections . Scale bar = 100 µm . ( B ) Locomotion measured across three consecutive days and nights showing mutants ( n = 12 ) are normal compared to control animals ( n = 12 ) . ( C ) Discrimination between CSp and CSm ( left ) is normal in Slc6a3-iCreERT2::Robo2lox/lox animals ( n = 12 ) compared to controls ( n = 12 ) . ( D ) Schematic showing expression of Cre in dopamine neurons starting in embryonic development and continuing into adulthood . ( E ) Locomotion measured across three consecutive days and nights showing mutants ( n = 12 ) are normal compared to control animals ( n = 11 ) . ( F ) Discrimination between CSp and CSm ( left ) is normal in Slc6a3-Cre::Robo2∆/lox animals ( n = 11 ) compared to controls ( n = 12 ) . ( G ) Schematic of strategy for injecting AAV1-Cre-GFP into the midbrain of Slc6a3-Cre; Robo2∆/lox animals . ( H ) Locomotion measured across three consecutive days and nights showing mutants ( n = 13 ) are hyperactive relative to controls ( n = 15 ) . Two-way repeated measures ANOVA , genotype x time interaction , F ( 252 , 6552 ) = 3 . 17 , **p<0 . 0001 . ( I ) Discrimination between a CSp and CSm is impaired in AAV1-CreGFP::Slc6a3-Cre::Robo2∆/lox animals ( n = 6 ) compared to controls ( n = 8 ) . Two-way repeated measures ANOVA , genotype x time interaction , F ( 6 , 72 ) = 2 . 68 , p<0 . 05 , Bonferroni’s multiple comparison test , p<0 . 01 . ( J , K ) Locomotor response to cocaine ( 20 mg/kg ) in controls ( n = 15 ) and mutants ( n = 13 ) on day 1 ( d ) , Two-way repeated measures ANOVA , genotype x time interaction , F ( 35 , 910 ) =6 . 57 , p<0 . 0001 and day 5 ( e ) , Two-way repeated measures ANOVA , genotype x time interaction , F ( 35 , 910 ) = 17 . 03 , p<0 . 0001 . ( L ) Normalized locomotor response to cocaine by subtracting 60 min pre-cocaine from 60 min post-cocaine on day 1 and day 5 , Two-way repeated measures ANOVA , genotype x time interaction , F ( 1 , 26 ) = 10 . 97 , p<0 . 01 , Bonferroni’s multiple comparison test , **p<0 . 0001 , p<0 . 01 . Bars represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 23858 . 01410 . 7554/eLife . 23858 . 015Figure 5—figure supplement 1 . Embryonic or adult inactivation of Robo2 does not recapitulate the Robo2 VTA mutant phenotypes . ( A , B ) Locomotor response to cocaine ( 20 mg/kg ) in controls ( n = 12 ) and mutants ( n = 12 ) on day 1 ( A ) and day 5 ( B ) . ( C ) Normalized locomotor response to cocaine by subtracting 60 min pre-cocaine from 60 min post-cocaine . ( D ) There was no difference in head entries during Pavlovian conditioning between controls ( n = 12 ) and mutants ( n = 12 ) . ( E ) Schematic of AAV1-FLEX-Rpl22HA injection into the VTA of Slc6a3-Cre::Robo2△/lox and control mice ( right ) . Inactivation of Robo2 in dopamine neurons resulted in a significant de-enrichment of ribosomal-associated Robo2 mRNA , p<0 . 01 . ( F , G ) Locomotor response to cocaine ( 20 mg/kg ) in controls ( n = 11 ) and mutants ( n = 12 ) on day 1 ( F ) and day 5 ( G ) . ( H ) Normalized locomotor response to cocaine by subtracting 60 min pre-cocaine from 60 min post-cocaine . ( I ) There was no difference in head entries during Pavlovian conditioning between controls ( n = 11 ) and mutants ( n = 12 ) . ( J ) Total distance traveled across three consecutive nights . Two-way repeated measures ANOVA , effect of genotype , F ( 1 , 26 ) = 50 . 42 , p<0 . 0001 , Bonferroni’s multiple comparison test , p<0 . 0001 . ( K ) There was no difference in head entries during Pavlovian conditioning between controls ( n = 6 ) and mutants ( n = 8 ) . Bars represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 23858 . 015 The observed lack of behavioral effects in Slc6A3-iCre/ERT2::Robo2lox/lox mice was unexpected , but could reflect an inefficiency in Robo2 inactivation . The conventional Cre-driver mouse line ( Slc6A3-Cre ) has been previously shown to effectively inactivate genes in virtually all dopamine producing neurons ( Engblom et al . , 2008; Zweifel et al . , 2008 ) . One caveat to this mouse line is that the early expression of Cre may result in developmental compensations ( Engblom et al . , 2008 ) . Nonetheless , we generated conditional , non-inducible knockouts of Robo2 in dopamine neurons ( Slc6A3-Cre::Robo2∆/lox mice , Figure 5D ) . Consistent with inactivation of Robo2 , Ribotag analysis following injection of an AAV containing Cre-dependent Rpl22HA ( AAV1-FLEX-Rpl22HA ) demonstrated a significant reduction in Robo2 mRNA in dopamine neurons ( Figure 5—figure supplement 1E ) ; consistent with nonsense-mediated mRNA decay ( Gibson et al . , 2014 ) . Similar to Slc6A3-iCre/ERT2::Robo2lox/lox mice , Slc6A3-Cre::Robo2∆/lox mice did not show hyperactivity , altered psychostimulant response to cocaine , or learning deficits relative to control mice ( DAT-iCreER::Robo2∆/+ ) ( Figure 5E–F; Figure 5—figure supplement 1F–I ) . The lack of observed effects in Slc6A3::Robo2∆/lox mice could be the result of developmental compensation as discussed above , and if this were true then injection of AAV1-Cre-GFP into the VTA of Slc6A3::Robo2∆/lox mice should not result in behavioral alterations due to an occlusion-like effect . To test this , adult Slc6A3::Robo2∆/lox mice were injected with AAV1-Cre-GFP into the VTA ( AAV1-Cre-GFP::Slc6A3-Cre::Robo2∆/lox mice , Figure 5G ) . Similar to injection of AAV1-Cre-GFP into the VTA of Robo2lox/lox mice , AAV1-Cre-GFP::Slc6A3-Cre::Robo2∆/lox mice were hyperactive , had inverted cocaine responses , and deficits in Pavlovian associative learning ( Figure 5H–L; Figure 5—figure supplement 1J–K ) . These results indicate that Robo2 inactivation is required in non-dopamine producing neurons to induce the observed behavioral phenotypes . Changes in the inhibitory control of both dopamine and non-dopamine neurons in the VTA following Robo2 inactivation suggests that local GABAergic control of the VTA is responsible for the observed behavioral alterations . To test this , we selectively silenced GABAergic neurons in the VTA through conditional expression of the light-chain of tetanus toxin fused to GFP ( GFP-TetX ) to block synaptic transmission ( Kim et al . , 2009 ) . An AAV1 vector containing a Cre-dependent expression cassette for GFP-TetX ( AAV1-FLEX-GFP-TetX , [Han et al . , 2015] ) was injected into the VTA of mice expressing Cre under the control of the endogenous VGAT locus , Slc6a32A1-Cre mice ( Figure 6A ) . Unexpectedly , inactivation of GABAergic transmission in the VTA of Slc6a32A1-Cre::AAV1-FLEX-GFP-TetX mice resulted in rapid weight loss and compromised health in a number of animals ( 11 out of 20 fell below 80% of their pre-surgery bodyweight and were removed from the study ) that could not be further assessed for behavioral changes . Of the remaining Slc6a32A1-Cre::AAV1-FLEX-GFP-TetX mice ( 9 out of 20 ) that did not show dramatic weight loss , possibly due to a reduced efficiency of GABA inactivation , we observed hyperactivity and inverted locomotor responses to cocaine ( Figure 6B–E ) . Post-hoc histological analysis showed robust viral expression within the VTA , so it is unclear why the surviving animals were less affected; however , behaviorally testing the survivors possibly caused a selection bias that underestimated the severity of the phenotype . These findings suggest that reduced GABAergic control of the VTA following Robo2 inactivation is likely the cause of hyperactivity and altered behavioral responses to psychostimulants . 10 . 7554/eLife . 23858 . 016Figure 6 . Midbrain GABA neuron inactivation recapitulates Robo2 VTA mutant phenotypes . ( A ) Schematic of strategy for injecting AAV1-GFP-TetX into the midbrain of Vgat-Cre animals ( left ) . Schematic of Cre-dependent AAV1-GFP-TetX ( top ) . Representative image ( bottom ) showing viral transduction is specific to non-dopamine neurons ( TH- neurons ) . Scale bar = 100 µm ( 25 µm in zoom ) . ( B ) Locomotion measured across three consecutive days and nights showing mutants ( n = 9 ) are hyperactive relative to controls ( n = 6 ) . Two-way repeated measures ANOVA , genotype x time interaction , F ( 252 , 3276 ) = 1 . 35 , p<0 . 001 . ( C , D ) Locomotor response to cocaine ( 20 mg/kg ) in controls ( n = 6 ) and mutants ( n = 9 ) on day 1 ( C ) , Two-way repeated measures ANOVA , genotype x time interaction , F ( 35 , 455 ) = 1 . 34 , p=0 . 0987 , and on day 5 ( D ) , Two-way repeated measures ANOVA , genotype x time interaction , F ( 35 , 455 ) = 2 . 86 , p<0 . 0001 . ( E ) Normalized locomotor response to cocaine by subtracting 60 min pre-cocaine from 60 min post-cocaine on day 1 and day 5 , Two-way repeated measures ANOVA , genotype x time interaction , F ( 1 , 13 ) = 5 . 35 , p<0 . 05 . Bars represent mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 23858 . 016 Our results provide the first demonstration that Robo2 has a critical function beyond early development of the nervous system by potently regulating GABAergic synapses . The observed behavioral phenotype of inverted psychostimulant responses is a rare genetic finding . Numerous other genetic models have been described that cause hyperactivity , but few have inverted psychostimulant responses ( Leo and Gainetdinov , 2013 ) . Given the link between hyperactivity and paradoxical calming by psychostimulants in ADHD , it is attractive to think that Robo2 regulation of inhibitory control of the midbrain may play an important role in this disorder . Although selective inactivation of Robo2 exclusively in the adult VTA is not a true genetic model of ADHD , our results strongly implicate balanced excitatory and inhibitory control of the midbrain as a likely source of ADHD-like phenotypes . Similar to inactivation of the dopamine transporter ( Gainetdinov et al . , 1999 ) , we observed increased sensitivity to the serotonin transport blocker fluoxetine , suggesting the hyperactivity associated with loss of Robo2 is due to a potential enhancement of the serotonin system . We find that Robo2 is highly expressed in the VTA in both dopamine and non-dopamine producing neurons . Given the dopamine-related nature of the phenotypes we observed , we were surprised to find that inactivation of Robo2 exclusively in dopamine neurons was not sufficient to invoke these phenotypes , particularly given the partial enrichment of Robo2 mRNA in these cells . Instead our data suggests that Robo2 is required in multiple cell types for the observed behavioral changes . There are numerous cell types within the VTA , including dopamine neurons , GABAergic interneurons , GABAergic projection neurons , and glutamatergic neurons ( Barker et al . , 2016 ) . Because the phenotype emerged only when dopamine and non-dopamine neurons were genetically targeted , it is possible that Robo2 is required globally within the VTA . Unfortunately , Cre driver lines that allow for the inducible expression of Cre recombinase only in GABA or glutamate neurons of the VTA do not currently exist that would allow us to test this in greater detail . However , given that many of the observed behavioral deficits are dopamine-dependent and the genetic manipulation alters synaptic physiology in both dopamine and non-dopamine neurons , it is highly likely that the observed deficits are the results of a global change in GABAergic connectivity in the VTA . Consistent with the function of GABA in this process , reducing synaptic release from midbrain GABA neurons recapitulated the adult VTA Robo2 mutant phenotype , though the effects were much more severe overall . These findings point to a key role of GABAergic transmission in the VTA for behavioral hyperactivity and learning deficits . Given the high degree of specificity for partial GABA agonists in targeting GABA receptors with specific subunit compositions ( Hoestgaard-Jensen et al . , 2014 ) , drugs targeting specific GABAA receptors in the VTA may be a promising therapeutic avenue for multiple disorders . Additional studies are required to understand the molecular mechanisms whereby Robo2 signaling controls inhibitory synapses in the VTA . The reduced frequency of sIPSCs following Robo2 inactivation is consistent with a requirement for Robo2 in presynaptic GABA release . It was recently shown that Robo receptors interact with A-kinase anchoring proteins ( AKAPs ) ( Samelson et al . , 2015 ) , placing Robo receptors in key scaffolding complexes . It has also been shown that AKAP 79/150 is expressed in dopamine and non-dopamine neurons in the VTA where it regulates GABAergic synaptic transmission . Intriguingly , disruption of AKAP function induces a long-term depression like effect on GABAergic synapse ( Dacher et al . , 2013 ) . Thus , an intriguing possibility is that Robo2 receptors interact with AKAPs to stabilize inhibitory synaptic connectivity . Another , non-mutually exclusive possibility is that Robo2 interacts with Robo1 either in cis or in trans to stabilize synapses . Both Robo1 and Robo2 are expressed in the VTA , and it has been demonstrated in Drosophila that during commissural axon pathfinding , Robo2 interacts with Robo1 in trans to suppress the repulsive properties of Robo1 ( Evans et al . , 2015 ) . Future exploration into these potential mechanisms will provide important insight into this process . In addition to expression in the ventral midbrain , Robo2 is highly expressed in other brain regions , most notably the limbic system including the prefrontal cortex , striatum , and hippocampus . Whether adult Robo2 signaling functions in a similar fashion in these regions will be important to elucidate , particularly given the linkage between Slit/Robo signaling and mental illness . Our results also suggest that Robo2 receptors may play an important role in regulating synaptic changes associated with drugs of abuse . Interestingly , during cocaine sensitization , changes in spine density in the striatum results from inhibition of the small Rho-family GTPase , Rac1 , which is a known effector molecule for Robo signaling , and is a potent regulator of spine turnover during development ( Dietz et al . , 2012 ) . In addition , cocaine sensitization reduces inhibitory synaptic transmission in the VTA ( Liu et al . , 2005 ) that may be mediated in part by Robo2 signaling . Determining how Slit/Robo signaling regulates synaptic processes in other cell types will be an important next step towards further resolving its function in the adult nervous system . Animals were group housed ( max 5 animals/cage ) , maintained on a 12:12 light:dark cycle , and given ad libitum access to food and water except during food restriction when they were restricted to 85% of their ad libitum bodyweight . All behavioral experiments were performed during the light cycle . Equal numbers of C57Bl/6 male and female mice ( 8–12 weeks old ) were used for all experiments . Animals were assigned to experimental groups to randomize for sex , age , and genotype . The experimenter was blinded to the genotype during data collection , and all data was generated from at least two independent experiments . The sample sizes were chosen based upon the predicted variability of the assay . Two to six independent experiments were performed for each behavioral test . In at least one of these experiments , the animals underwent a single behavioral test . When multiple tests were performed with the same group of animals , an additional cohort of animals had the behavioral tests reversed to counterbalance any interaction between tests; however , all pharmacology studies occurred subsequent to non-invasive behavioral tests ( ie Pavlovian conditioning ) . The Robo2lox/lox mouse line was generated as described ( Lu et al . , 2007 ) . The delta allele of Robo2∆/lox was generated by crossing the Robo+/lox to the Meox2-Cre mouse line to recombine the Robo2 lox allele . This approach was taken with the Slc6a3-Cre line to prevent mosaicism caused by potential transient Cre expression in the gametes of the Slc6a3-Cre line . The Slc6a3-iCreERT2 , Slc6a32A1-Cre , and Gt ( ROSA ) 26Sortm ( CAG-tdTomato ) Hze mouse lines were obtained from Jackson Laboratory . To induce expression of Slc6a3-iCreERT2 , daily injections of tamoxifen at 75 mg/kg were performed for five days ( Sigma-Aldrich ) . All experiments were done in accordance with protocols approved by the University of Washington Animal Care and Use Committee . AAV1 virus was prepared as described ( Gore et al . , 2013 ) . For stereotaxic injections into the brain , either AAV1-Cre-GFP , AAV1-∆Cre-GFP , or AAV1-FLEX-GFP-TetX was injected in the VTA ( Relative to bregma: M-L = 0 . 5 , A-P = −3 . 25F , D-V = −4 . 5 , F= ( Bregma-Lambda ) /4 . 21 ) . For the generation of control mice , either a virus containing a partially deleted , nonfunctional version of Cre ( AAV1-∆Cre-GFP ) was injected into Robo2lox/lox animals ( control for Robo2lox/lox genotype ) or AAV-Cre-GFP was injected into wild type mice ( control for viral Cre expression ) . Equal numbers of these two control groups were generated and showed no statistical difference , so they are presented as a single combined ‘control’ group . Animals were allowed to recover for 2 weeks prior to testing to allow for expression of the virus . Exclusion criteria was based upon histological examination that verified the expression of the virus in the VTA . For Slc6a3Cre;Rpl22HA+/lox and Slc32a1Cre;Rpl22HA+/lox experiments , tissue was collected from mice 8–10 weeks of age . For AAV1-FLEX-Rpl22HA injected Slc6a3Cre::Robo2∆/lox injected mice , tissue was collected 3 weeks following injection to allow for sufficient HA-tagged Rpl22 incorporation . Tissue punches of the ventral midbrain were homogenized as described previously ( Sanz et al . , 2009 ) . Homogenized tissue from individual mice were incubated with 5 µl of anti-HA antibody ( Covance ) coupled to 200 µl of magnetic beads ( Pierce ) overnight at 4°C . Following elution from magnetic beads , RNA from both immunoprecipitated ( IP ) samples and non-HA tagged ( input ) samples was obtained using the RNeasy micro kit ( Qiagen ) according to manufacturer's directions . Total RNA was quantified using a Ribogreen RNA assay kit ( Invitrogen ) . cDNA was generated with Superscript IV ( Invitrogen ) using oligo dT primers from equal amounts of starting RNA , according to the manufacturer protocol . IP purified mRNA was converted to cDNA and analyzed via qRT-PCR . TaqMan ( Applied Biosystems ) primers against Robo2 and Actb were used to detect expression levels . Relative expression values were obtained using the comparative CT method and normalized to Actb levels . Fold enrichment was calculated as the IP versus input ratio and represented the amount of the transcript in the targeted cell type ( IP ) when compared to equal amounts of RNA from the input . Whole-cell recordings were made using an Axopatch 700B amplifier ( Molecular Devices ) with filtering at 1 KHz using 4–6 MΩ electrodes . Horizontal brain slices ( 200 μm ) were prepared from 6 week old mice in an ice slush solution containing ( in mM ) : 92 NMDG , 2 . 5 KCl , 1 . 25 NaH2PO4 , 30 NaHCO3 , 20 HEPES , 25 glucose , 2 thiourea , 5 Na-ascorbate , 3 Na-pyruvate , 0 . 5 CaCl2 , 10 MgSO4 , pH 7 . 3–7 . 4 ( Ting et al . , 2014 ) . Slices recovered for ~12 min in the same solution at 32°C and then were transferred to a room temperature solution including ( in mM ) : 92 NaCl , 2 . 5 KCl , 1 . 25 NaH2PO4 , 30 NaHCO3 , 20 HEPES , 25 glucose , 2 thiouria , 5 Na-ascorbate , 3 Na-pyruvate , 2 CaCl2 , 2 MgSO4 . Slices recovered for an additional 60 min . All solutions were continually bubbled with O2/CO2 , and all recordings were made in ACSF containing ( in mM ) : 126 NaCl , 2 . 5 KCl , 1 . 2 NaH2PO4 , 1 . 2 MgCl2 11 D-glucose , 18 NaHCO3 , 2 . 4 CaCl2 at 32°C continually perfused over slices at a rate of ~2 ml/min . Ih currents were induced by 2 s hyperpolarizing voltage steps from −70 mV to −120 mV . All recorded cells were medial and proximal to the MT , where the presence of an Ih current correlates highly with dopamine neuron identity ( Margolis et al . , 2006 ) . For recording spontaneous EPSCs , electrodes were filled with an internal solution containing ( in mM ) : 130 K-gluconate , 10 HEPES , 5 NaCl , 1 EGTA , 5 Mg-ATP , 0 . 5 Na-GTP , pH 7 . 3 , 280 mOsm . 200 mM picrotoxin was included in the ACSF to inhibit GABAA receptor-mediated events . For recording spontaneous IPSCs , electrodes were filled with an internal solution containing ( in mM ) : 135 KCl , 12 NaCl , 0 . 05 EGTA , 100 HEPES , 0 . 2 Mg-ATP , 0 . 02 Na-GTP , pH 7 . 3 , 280 mOsm . 2 mM kynurenic acid was included in the ACSF to inhibit AMPA and NMDA receptor-mediated events . Cells were held at −60 mV , and access resistance was monitored throughout all experiments . To calculate the Tau ( τw ) for the sIPSCs , we calculated the weighted time constant , τw , for 10–90% of the rise time by generating an average waveform for each cell using the equation , τw = ( A1 τ1 ) + ( A2 τ2 ) , where τ1 and τ2 are the time constants of the first and second exponential functions , respectively , and A1 and A2 are the proportion of the amplitude of the IPSC of the first and second exponential functions , respectively as described ( Eyre et al . , 2012 ) . Four recording tetrodes were made from tungsten wires ( 25 µm in diameter; California Fine Wire Company ) and mounted on a custom-built Microdrive ( Jo et al . , 2013 , Jo and Mizumori , 2016 ) . Each wire was gold-plated to adjust its final impedance to 200–400 KΩ ( tested at 1 kHz ) . The microdrive was surgically implanted in one hemisphere of the VTA after bilateral virus infusion of AAV-Cre-GFP or AAV- ∆Cre-GFP into the VTA . Two weeks after surgery , each mouse was placed in a rectangular box ( 21 . 6 cm × 17 . 8 cm ) and single-unit activity was daily monitored as follows: the microdrive was connected to a preamplifier and its outputs were transferred to a RZ5 BioAmp processor ( Tucker-Davis Technologies ) . Unit signals were filtered between 0 . 3 and 5 kHz , and digitized at 24 kHz . Tetrodes were advanced in 40 µm increments , up to 160 µm per day . Once stable and isolated units were found , their sensitivity to quinpirole and eticlopride was tested to identify putative dopamine cells ( Zweifel et al . , 2009 ) . A VTA cell was classified as putative dopaminergic if it exhibited a low basal firing rate ( less than 12 Hz ) and showed severe inhibition by quinpirole ( 0 . 2 mg/ml , i . p . ; ≥70% inhibition of the basal frequency ) and subsequent recovery by eticlopride ( 0 . 5 mg/ml , i . p . ; ≥70% the basal frequency ) . The other neurons that did not meet the criteria were considered as putative non-dopaminergic cells . VTA neurons were recorded for 20 min to determine baseline firing properties . Spikes from these single neurons were isolated by cluster analysis using Offline Sorter ( Plexon ) . Subsequent data analyses , such as comparison of average firing rates and burst rates were performed with Matlab software ( Mathworks , Zweifel , 2017 https://github . com/zweifellab/ephys with a copy archived at https://github . com/elifesciences-publications/ephy ) . The onset of a burst was identified as two consecutive spikes with an interspike interval of <80 ms and its termination was defined as the an interspike interval of >160 ms ( Grace and Bunney , 1984 ) . FSCV was performed using carbon fiber microelectrodes encased in fused-silica capillary tubing ( Polymicro Technologies ) ( Clark et al . , 2010 ) . The carbon fiber microelectrode was positioned in the dorsomedial striatum at stereotaxic coordinates: A-P = 1 . 2 mm , M-L = 1 . 2 mm , D-V = 3 . 5 below the dura and a waveform was applied at a frequency of 60 Hz . After 40 min , the waveform frequency was reduced to 10 Hz for 10 min or until baseline recordings stabilized . Next , the carbon fiber microelectrode was lowered into the nucleus accumbens in 0 . 05 mm increments until maximum dopamine release was observed ( D-V coordinates ranged from 3 . 60–3 . 98 below bregma , average 3 . 80 ) . An Ag/AgCl reference electrode was placed in the contralateral hemisphere . The stimulation electrode ( Plastics One ) was placed above the PPTg at stereotaxic coordinates: A-P: −4 . 6 mm , M-L: 0 . 8 mm and lowered in 0 . 1 mm increments from 2 . 5 mm below the dura until optimal dopamine release was observed after 1 s stimulations of 60 Hz ( average D-V coordinate: 2 . 7 mm below the dura ) . For the MFB stimulations , the stereotaxic coordinates were A-P: 2 . 4 mm , M-L: 1 . 1 mm , DV: 0 . 1 increments from 4 mm . Microdialysis-UPLC experiments were modified for mouse tissue from a protocol previously described ( Schindler et al . , 2016 ) . In brief , a 1 mm long microdialysis probe ( BASi instruments ) was equilibrated in artificial cerebrospinal fluid ( ACSF ) containing 154 . 7 mM Na+ , 0 . 82 mM Mg+2 , 2 . 9 mM K+ , 132 . 49 mM Cl− , 1 . 1 mM Ca+2 , and 5 . 9 mM D-glucose , and pH to 7 . 4 with CO2 . The probe was inserted unilaterally into the nucleus accumbens ( M-L = +1 . 0 , A-P = +1 . 3F , D-V = 4 . 0 , F= ( distance between lambda and bregma ) /4 . 21 ) . The probe was flushed with ACSF at a rate of 2 µl/min for 90 min , followed by 1 µl/min for 30 min . Four 20 min fractions were collected . An equal volume of dialysate was mixed with an internal standard solution containing 100 pg/µl of deuterated D4 ( 2- ( 3 , 4-dihydroxyphenyl ) ethyl-1 , 1 , 2 , 2-d4-amine HCL ) 0 . 01 N perchloric acid and frozen on dry ice until run on the UPLC . For the detection of dopamine on the UPLC , serial dilutions of purified D4 dopamine samples of known concentrations were first analyzed to detect the retention time for dopamine on a Xevo TQ-S mass spectrometer . Samples for controls and Robo2 VTA mutants were then ran in an interleaved manner to control for any column artifact . The dopamine and D4 peaks were identified and the area under the curve was calculated for each sample . Animals were deeply anesthetized with beuthanasia and transcardially perfused with 4% paraformaldehyde ( PFA ) . The brain was removed and fixed overnight in 4% PFA before being cryoprotected in 30% sucrose . 30 μm frozen sections were made on a Leica cryostat and fluorescence imaging was performed on a Nikon upright microscope . For immunohistochemistry , the following antibodies were used: Tyrosine hydroxylase ( rabbit polyclonal , 1:2000 , Millipore , or mouse monoclonal , 1:2000 , Millipore ) , GFP ( mouse monoclonal , 1:1000 , Invitrogen or chicken polyclonal , 1:1000 , Abcam ) , Robo2 ( rabbit polyclonal , 1:1000 , Abcam ( discontinued ) , and dsRed ( rabbit , 1:1000 , Clontech ) . For transmission electron microscopy , tissue was fixed overnight in 4% glutaraldehyde in 0 . 1 M sodium cacodylate buffer , pH 7 . 2 , and then postfixed in 1% osmium tetroxide , washed , en bloc stained in 1% uranyl acetate , embedded in epoxy resin , and then sectioned to 80 nm . Statistical analyses were performed in Prism ( GraphPad ) , Excel ( Microsoft ) , and Matlab ( Mathworks ) . For data that was normally distributed with equal variance between groups , an unpaired t-test or two-way repeated measures ANOVA was performed . For data that was not normally distributed ( Figure 1—figure supplement 2E ) , a non-parametric Mann-Whitney test was performed . The sample sizes were chosen based upon previously established variability of the assay . The images were processed using either Photoshop ( Adobe ) .	Although no two people are alike , we all share the same basic brain structure . This similarity arises because the same developmental program takes place in every human embryo . Specific genes are activated in a designated sequence to generate the structure of a typical human brain . But what happens to these genes when development is complete – do they remain active in the adult brain ? A gene known as Robo2 encodes a protein that helps neurons find their way through the developing brain . Many of these neurons will ultimately form part of the brain’s reward system . This is a network of brain regions that communicate with one another using a chemical called dopamine . The reward system contributes to motivation , learning and memory , and also underlies drug addiction . In certain mental illnesses such as Parkinson’s disease and schizophrenia , the dopamine-producing neurons in the reward system work incorrectly or die . To find out whether Robo2 is active in the mature nervous system , Gore et al . used genetic techniques to selectively remove the gene from the reward system of adult mice . Doing so reduced the ability of the dopamine neurons within the reward system to inhibit one another , which in turn increased their activity . This changed the behavior of the mice , making them hyperactive and less able to learn and remember . Cocaine makes normal mice more active; however , mice that lacked the Robo2 gene became less active when given cocaine . Overall , the work of Gore et al . suggests that developmental axon guidance genes remain important in the adult brain . Studying developmental genes such as Robo2 may therefore open up new treatment possibilities for a number of mental illnesses and brain disorders .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Roundabout receptor 2 maintains inhibitory control of the adult midbrain
East Asians ( EAs ) experience worse metabolic health outcomes compared to other ethnic groups at lower body mass indices; however , the potential role of the gut microbiota in contributing to these health disparities remains unknown . We conducted a multi-omic study of 46 lean and obese East Asian and White participants living in the San Francisco Bay Area , revealing marked differences between ethnic groups in bacterial richness and community structure . White individuals were enriched for the mucin-degrading Akkermansia muciniphila . East Asian subjects had increased levels of multiple bacterial phyla , fermentative pathways detected by metagenomics , and the short-chain fatty acid end-products acetate , propionate , and isobutyrate . Differences in the gut microbiota between the East Asian and White subjects could not be explained by dietary intake , were more pronounced in lean individuals , and were associated with current geographical location . Microbiome transplantations into germ-free mice demonstrated stable diet- and host genotype-independent differences between the gut microbiotas of East Asian and White individuals that differentially impact host body composition . Taken together , our findings add to the growing body of literature describing microbiome variations between ethnicities and provide a starting point for defining the mechanisms through which the microbiome may shape disparate health outcomes in East Asians . Culture-independent surveys have emphasized differences in gut microbial community structure between countries ( Hehemann et al . , 2010; Vangay et al . , 2018; Yatsunenko et al . , 2012 ) , but the factors that contribute to these differences are poorly understood . Diet is a common hypothesis for geographical variations in the gut microbiota ( De Filippo et al . , 2010; Devoto et al . , 2019 ) , based upon extensive data from intervention experiments in humans and mouse models ( Bisanz et al . , 2019; Carmody et al . , 2015; David et al . , 2014; Gehrig et al . , 2019 ) . However , diet is just one of the many factors that distinguishes human populations at the global scale , motivating the desire for a more holistic approach . Self-identified race/ethnicity ( SIRE ) provides a useful alternative , as it integrates the broader national or cultural tradition of a given social group and is closely tied to both dietary intake and genetic ancestry . Multiple studies have reported associations between the gut microbiota and ethnicity in China ( Khine et al . , 2019 ) , the Netherlands ( Deschasaux et al . , 2018 ) , Singapore ( Xu et al . , 2020 ) , and the United States ( Brooks et al . , 2018; Sordillo et al . , 2017 ) . In contrast , a recent study of Asian immigrants suggested that once an individual relocates to a new country , the microbiota rapidly assumes the structure of the country of residence ( Vangay et al . , 2018 ) . Thus , the degree to which microbiome signatures of ethnicity persist following immigration and their consequences for host pathophysiology remain an open question . The links between ethnicity and metabolic disease are well established . For example , East Asian ( EA ) subjects are more likely to develop health-related metabolic complications at lower body mass index ( BMI ) compared to their White ( W ) counterparts ( Gu et al . , 2006; Zheng et al . , 2011 ) . Moreover , Asian Americans have persistent ethnic differences in metabolic phenotypes following immigration ( Jih et al . , 2014 ) , including a decoupling of BMI from total body fat percentage ( Alba et al . , 2018 ) . The mechanisms contributing to these ethnic differences in fat accrual remain unknown . Human genetic polymorphisms may play a role ( Wen et al . , 2010; Xiang et al . , 2004 ) ; however , putative alleles are often shared between members of different ethnic groups ( Gravel et al . , 2011 ) . The gut microbiome might offer a possible explanation for differences in metabolic disease rates across ethnic groups ( He et al . , 2018 ) , but there has been a relative scarcity of microbiome studies in this area ( Gaulke and Sharpton , 2018 ) . These observations led us to hypothesize that ethnicity-associated differences in host metabolic phenotypes may be determined by corresponding differences in the gut microbiome . First , we sought to better understand the extent to which ethnicity is linked to the human gut microbiome in states of health and disease . We conducted a cross-sectional multi-omic analysis of the gut microbiome using paired 16S rRNA gene sequencing ( 16S-seq ) , metagenomics , and metabolomics from the Inflammation , Diabetes , Ethnicity , and Obesity ( IDEO ) cohort at the University of California , San Francisco . IDEO includes rich metabolic , dietary , and socioeconomic metadata ( Alba et al . , 2018 ) , a restricted geographical distribution within the San Francisco Bay Area , and a balanced distribution of EA and W individuals that are both lean and obese ( Supplementary file 1A ) . We report marked differences in gut microbial richness , community structure , and metabolic end-products between EA and W individuals in the IDEO cohort . We then used microbiome transplantations to assess the stability of ethnicity-associated differences in the gut microbiota in the context of genetically identical mice fed the same diet . We also explored the functional consequences of these differences for host metabolic phenotypes . Our results emphasize the importance of considering ethnicity in microbiome research and further complicate prior links between metabolic disease and the gut microbiome ( Ley et al . , 2006; Turnbaugh et al . , 2008; Wu et al . , 2020 ) , which may be markedly different across diverse ethnic groups . Ethnicity was associated with inter-individual variations in the human gut microbiota . Principal coordinates analysis of PhILR Euclidean distances from 16S-seq data ( Supplementary file 1B , n=22 EA , 24 W subjects ) revealed a subtle but significant separation between the gut microbiotas of EA and W subjects ( p=0 . 006 , R2=0 . 046 , ADONIS; Figure 1A ) . Statistical significance was robust to the distance metric used ( Supplementary file 1C ) . Bacterial diversity was significantly higher in W individuals across three distinct metrics: Faith’s phylogenetic diversity , amplicon sequence variant ( ASV ) richness , and Shannon diversity ( Figure 1B ) . Six bacterial phyla were significantly different between ethnicities ( Figure 1C ) , of which only one phylum , Verrucomicrobiota , was significantly enriched in W subjects . Phylogenetic analyses of all ASVs revealed marked variations in the direction of change across different phyla between EA and W subjects ( Figure 1—figure supplement 1A ) , indicating that the phylum level trends ( Figure 1C ) resulted from the integration of subtle shifts across multiple component members ( Figure 1D–F ) . Several significant differences were detectable at the genus level ( Figure 1D–E ) , including Blautia , Bacteroides , and Streptococcus which were significantly enriched in EA subjects . We also identified two ASVs that were significantly different between ethnicities: Blautia obeum and a Streptococcus species , both enriched in EA subjects ( Figure 1F ) . There were no significant differences between ethnicities in 16S rRNA copy number ( Figure 1—figure supplement 1F ) . Next , we used a random forest classifier to define biomarkers in the gut microbiota that distinguish EA and W subjects ( Figure 1—figure supplement 1B-D ) . Classifiers employing ASV data and PhILR transformed phylogenetic nodes were trained using leave-one-out cross-validation . B . obeum ( ASV1 ) was the top contributor to the resulting classifier , followed by Anaerostipes hadrus ( ASV45 ) and then Streptococcus parasanguinis ( ASV110 ) ( Figure 1—figure supplement 1B ) . Both classifiers demonstrated the ability to distinguish between ethnic groups , with PhILR transformed phylogenetic nodes achieving a higher area under the curve compared to ASVs ( Figure 1—figure supplement 1C , D ) . The majority ( 18/23 ) of the top ASVs identified by our classifier were also significantly different between ethnicities ( Figure 1—figure supplement 1E ) . Metagenomic sequencing provided independent confirmation of differences in the gut microbiome between ethnicities ( Supplementary file 1B , n=21 EA , 24 W subjects ) . Consistent with our 16S-seq analysis , we detected a difference in the gut microbiomes between ethnicities based upon metagenomic species abundances ( p=0 . 003 , R2=0 . 047 , ADONIS , Figure 2A ) and gene families ( p=0 . 029 , R2=0 . 036 , ADONIS ) . Ethnicity explained more variation in species abundances than a selection of demographic , laboratory , lifestyle , and metabolic metadata ( Figure 2B ) . Visualization of diversity and species assignments within each phylum revealed marked variation in the magnitude and direction of change between individuals of a given ethnicity ( Figure 2C ) . Genera that were found to be significantly different between ethnicities in our metagenomic data included Akkermansia and an unspecified Erysipelotrichaceae genera ( Figure 2D ) elevated in W individuals . Four bacterial species were significantly different between ethnicities in our metagenomic data: W individuals had higher levels of A . muciniphila , Bacteroidales bacterium ph8 , and Roseburia hominis , and lower levels of Ruminococcus gnavus , compared to EA individuals ( Figure 2E ) . Next , we used nuclear magnetic resonance ( NMR ) -based stool metabolomics to gain insight into the potential functional consequences of ethnicity-associated differences in the human gut microbiome ( Supplementary file 1B , n=10 subjects/ethnicity ) . Metabolite profiles were more strongly associated with ethnicity ( p=0 . 008 , R2=0 . 128 , ADONIS; Figure 3A ) than community structure ( R2=0 . 029–0 . 055 , ADONIS; Supplementary file 1C ) or gene abundance ( p=0 . 029 , R2=0 . 036 , ADONIS ) . Feature annotations revealed elevated levels of the branched-chain amino acid ( BCAA ) valine and the short-chain fatty acids ( SCFAs ) acetate and propionate in EA subjects ( Figure 3B and Supplementary file 1D ) . In contrast , proline , formate , alanine , xanthine , and hypoxanthine were found at higher levels in W subjects ( Figure 3B ) . To assess the statistical significance and reproducibility of these trends , we used targeted gas chromatography mass spectrometry ( GC-MS ) and UPLC-MS/MS to quantify a panel of BCAAs , SCFAs , and bile acids ( Supplementary file 1E ) . Confirming our NMR data , EA subjects had significantly higher levels of stool acetate ( Figure 3C ) and propionate ( Figure 3D ) ; however , we did not detect any significant differences in BCAAs or bile acids ( Figure 3—figure supplement 1 ) . Isobutyrate ( which was not detected by NMR ) was also significantly higher in EA subjects ( Figure 3E ) . In agreement with these metabolite levels , a targeted re-analysis of our metagenomic data revealed a significant enrichment in two SCFA-related pathways: ‘pyruvate fermentation to butanoate’ ( p=0 . 023 , fold-difference=2 . 216 ) and ‘superpathway of Clostridium acetobutylicum acidogenic fermentation’ ( p=0 . 023 , fold-difference=2 . 182 ) . Consistent with prior work ( Le Chatelier et al . , 2013; Turnbaugh et al . , 2008 ) , we found that gut bacterial richness in W individuals was significantly associated with both BMI ( Figure 4A ) and body fat percentage ( Figure 4B ) . Remarkably , these associations were undetectable in EA subjects ( Figure 4A and B ) even when other metrics of bacterial diversity were used ( Figure 4—figure supplement 1 ) , with the single exception of a negative correlation between Shannon diversity and BMI in EA subjects ( Figure 4—figure supplement 1C ) . Re-analysis of our data separating lean and obese individuals revealed that the previously observed differences between ethnic groups were driven by lean individuals . Compared to lean EA individuals , lean W subjects had significantly higher bacterial diversity ( Figure 4C ) and more marked differences in gut microbial community structure ( p=0 . 0003 , R2=0 . 122 , ADONIS; Figure 4D ) and metabolite profiles ( p=0 . 010 , R2=0 . 293 , ADONIS; Figure 4E ) . By contrast , obese W versus EA individuals were not different across any of these metrics ( Figure 4C–E ) , except for lower Shannon diversity in obese EA compared to W individuals ( Figure 4C ) . We also detected differences in the gut microbiotas of lean EA and W individuals at the phylum ( Figure 5A ) and genus ( Figure 5B ) levels that were largely consistent with our original analysis of the full data set ( Figure 1C and E ) . More modest differences in the gut microbiota between ethnicities were observed in obese subjects ( Figure 5A and C ) . Next , we sought to understand the potential drivers of differences in the gut microbiome between ethnic groups in lean individuals within the IDEO cohort . Consistent with prior studies ( Falony et al . , 2016 ) , PERMANOVA analysis of our full 16S-seq data set revealed that diabetes ( Forslund et al . , 2015 ) , age ( Ghosh et al . , 2020 ) , metformin use ( Wu et al . , 2017 ) , and statin intake ( Vieira-Silva et al . , 2020 ) were significantly associated with variance in the PhILR Euclidean distances ( Figure 6—figure supplement 1 ) . Metagenomic sequencing of the IDEO cohort with subsequent PERMANOVA analysis confirmed significant associations with ethnicity and statin use , while also highlighting significant associations with HOMA-IR and BMI ( Figure 2B ) , consistent with prior reports ( Liu et al . , 2017; Zouiouich et al . , 2021 ) . While several factors linked to body composition were different between obese EA and W subjects using a nominal p-value , only triglyceride levels were significantly different between lean EA and W subjects and this trend did not survive multiple testing correction ( Supplementary file 1A ) . Although everyone in the cohort was recruited from the San Francisco Bay Area , birth location varied widely ( Figure 6—figure supplement 2 ) . There was no significant difference in the proportion of subjects born in the United States between ethnicities ( 75% W , 54 . 5% EA; p=0 . 15 , Pearson’s χ2 test ) . There was also no significant difference in the geographical distance between birth location and San Francisco [W median 2 , 318 ( 2 . 2–6 , 906 ) miles; EA median 1 , 986 ( 2 . 2–6 , 906 ) miles; p=0 . 69 , Wilcoxon rank-sum test] or the amount of time spent in the San Francisco Bay Area at the time of sampling [W median 270 ( 8 . 00–741 ) months; EA median 282 . 5 ( 8 . 50–777 ) months; p=0 . 42 , Wilcoxon rank-sum test] . Surprisingly , we did not detect any significant differences in either short- ( Supplementary file 1F ) or long-term ( Supplementary file 1G ) dietary intake between ethnicities . Consistent with this , procrustes analysis did not reveal any significant associations between dietary intake and gut microbial community structure: procrustes p=0 . 280 ( DHQIII ) and p=0 . 080 ( ASA24 ) relative to PhILR transformed 16S-seq ASV data . The Spearman Mantel statistic was also non-significant [r=0 . 0524 , p=0 . 243 ( DHQIII ) and r=−0 . 0173 , p=0 . 590 ( ASA24 ) ] , relative to PhILR transformed 16S-seq ASV data . Despite the lack of an overall association between reported dietary intake and the gut microbiota , we were able to identify 12 ASVs and 7 metagenomic species associated with dietary intake in lean W individuals ( Figure 6—figure supplement 3A ) . We also detected 20 significant species-level associations in lean EA subjects ( Figure 6—figure supplement 3B ) . There were no overlapping associations between ethnicities . Given the marked variation in the gut microbiome at the continental scale ( Hehemann et al . , 2010; Vangay et al . , 2018; Yatsunenko et al . , 2012 ) , we hypothesized that the observed differences in lean EA and W individuals may be influenced by a participant’s current address at the time of sampling . Consistent with this hypothesis , we found clear trends in ethnic group composition across ZIP codes in the IDEO cohort ( Figure 6A and B ) that were mirrored by the 2018 US census data ( Pearson r=0 . 52 , p=0 . 026 for neighborhoods with greater than 50% W subjects; Figure 6D ) . Obese individuals from both ethnicities and lean W subjects tended to live closer to the center of San Francisco relative to lean EA subjects ( Figure 6C ) . Distance between the current ZIP code and the center of San Francisco and duration of residency within San Francisco were both associated with gut microbial community structure ( Figure 6E and F ) . The association between the current address and the gut microbiota was robust to the central point used , as evidenced by using the Bay Bridge as the central reference point ( p=0 . 008 , rho=0 . 394 , Spearman correlation ) . Taken together , our results support the hypothesis that there are stable ethnicity-associated signatures within the gut microbiota of lean EA versus W individuals that are independent of diet . To experimentally test this hypothesis , we transplanted the gut microbiotas of two representative lean W and lean EA individuals into germ-free male C57BL/6J mice fed a low-fat , high-plant-polysaccharide ( LFPP ) diet ( two independent experiments; per group n = 12 mice , 2 donors; per donor n=6 mice , 1 isolator; Figure 7—figure supplement 1A , B ) . The donors for this and the subsequent experiment were matched for their metabolic and other phenotypes to minimize potential confounding factors ( Supplementary file 1H and I ) . Despite maintaining the genetically identical recipient mice on the same autoclaved LFPP diet , we detected significant differences in gut microbial community structure ( Figure 7A ) , bacterial richness ( Figure 7C ) , and taxonomic abundance ( Figure 7D and E and Supplementary file 1J ) between the two ethnicity-specific recipient groups . These differences recapitulated key aspects of the gut microbiota observed in the IDEO cohort , including significantly lower bacterial richness ( Figure 7C ) and higher abundance of Bacteroides ( Figure 7D and E ) in recipient mice transplanted with microbiota from EA compared to W donors . Next , we sought to assess the reproducibility of these findings across multiple donors and in the context of a distinctive dietary pressure . We fed 20 germ-free male mice a high-fat , high-sugar ( HFHS ) diet for 4 weeks prior to colonization with a gut microbiota from 1 of 5 W and 5 EA donors . Mice were maintained on the HFHS diet following colonization ( per group n=10 mice , 5 donors; per donor n=2 mice , 1 cage; Figure 7—figure supplement 1C ) . This experiment replicated our original findings on the LFPP diet , including significantly altered gut microbial community structure between ethnicities ( Figure 7F ) , significantly increased richness in mice receiving W donor microbiota ( Figure 7H ) , and a trend toward higher levels of Bacteroides in mice receiving the gut microbiotas of EA donors ( Figure 7I and J ) . Of note , the variance explained by ethnicity was lower in mice fed the HFHS diet ( R2=0 . 126 ) than the LFPP diet ( R2=0 . 384 ) , potentially suggesting that in the context of human obesity , excessive fat and sugar consumption may serve to diminish the signal otherwise associated with ethnicity . As expected ( Nayak et al . , 2021; Turnbaugh et al . , 2009; Walter et al . , 2020 ) , the input donor microbiota was distinct from that of the recipient mice ( Figure 7B and G ) ; however , there was no difference between ethnic groups in the efficiency of engraftment ( Figure 7—figure supplement 2 ) . In a pooled analysis of all gnotobiotic experiments accounting for one donor for multiple recipient mice , ethnicity and diet were both significantly associated with variations in the gut microbiota ( Figure 7—figure supplement 3 ) , consistent with the extensive published data demonstrating the rapid and reproducible impact of an HFHS diet on the mouse and human gut microbiota ( Bisanz et al . , 2019 ) . Finally , mice transplanted with gut microbiomes of EA and W individuals displayed differences in body composition . LFPP fed mice that received W donor microbiota had significantly increased adiposity in conjunction with decreased lean mass , relative to LFPP fed mice that received the EA donor microbiota ( Figure 8A–C ) . Although these trends were mirrored in recipient mice that fed the HFHS diet ( Figure 8E–G ) , they did not reach statistical significance . There were no significant differences in glucose tolerance in either experiment ( Figure 8D and H ) . Taken together , these results suggest that dietary input may mask the metabolic consequences of ethnicity-associated differences in the gut microbiota . Our results support the utility of considering ethnicity as a covariate in microbiome studies , due to the ability to detect signals that are difficult to capture by more specific metadata such as individual dietary intake values . On the other hand , these findings raise the importance of dissecting the sociological and biological components of ethnicity with the goal of identifying factors that shape the gut microbiota , either alone or in combination . This emerging area of microbiome research is just one component in the broader efforts to explore the boundaries and mechanistic underpinning of ethnicity with respect to multiple ethnic groups . The IDEO cohort provides a valuable research tool to conduct prospective longitudinal and intervention studies examining diabetes in diverse participants . More broadly , IDEO provides a framework to approach other disease states where self-identified race or ethnicity are thought to contribute to health outcomes related to the microbiome , including the use of gnotobiotic mouse models to examine the specific role of microbial communities in contributing to phenotypes linked to ethnicity . By understanding the biological features that drive differences between ethnic groups , we may be able to achieve similar health outcomes and to support more precise therapies informed by a broader appreciation of both microbial and human diversity . The IDEO cohort was established to explore the pathogenesis of obesity and metabolic diseases in highly vulnerable segments of the population . It includes men and women of multiple ethnicities recruited from the general medicine , endocrinology , diabetes , general surgery , and bariatric surgery clinics at the University of California San Francisco ( UCSF ) and Zuckerberg San Francisco General Hospital and by public advertisements throughout the local San Francisco Bay Area . All study participants were part of the IDEO cohort , which has been previously described ( Alba et al . , 2018; Oguri et al . , 2020 ) . Briefly , IDEO consists of 25–65-year-old men and women of multiple ethnicities and across a wide BMI range ( 18 . 5–52 kg/m2 ) living in the San Francisco Bay Area . Using IDEO , we recruited both lean and obese W and EA individuals into this study based on World Health Organization cutoffs: W/EA BMI≤24 . 9 kg/m2 ( lean ) ; W BMI≥30 kg/m2 ( obese ) ; and EA BMI≥27 . 5 kg/m2 ( obese ) ( Hsu et al . , 2015; Jih et al . , 2014; Expert Consultation , 2004 ) . To avoid bias toward non-English speaking participants , all documents including flyers , screening questionnaires , and consents were available in Cantonese and Mandarin . Potential participants completed screening questionnaires and exclusion criteria were assessed in more detail . These included acute or chronic infections , current medications with a recognized impact on the immune system , recent antibiotic use , current smoking , recent changes in weight , active liver disease or liver failure , chronic kidney disease ( eGFR <30 ml/min/1 . 73 m2 ) , history of cancer and chemotherapy therapy within the past 5 years , psychiatric and neurological disorders , prior bariatric surgery , and weight >159 kg ( the DXA scanner weight limit ) . Whereas exclusion criteria inherently lend bias toward healthy individuals , this is done to limit the confounding effects of a wide variety of chronic diseases and environmental exposures on the comparisons being made . IDEO also limited bias by standardizing how individuals are asked to self-identify race/ethnicity . Individuals are asked to respond to two separate questions about ethnicity ( e . g . , ‘are you of Hispanic , Latino , or Spanish origin ? ’ ) and race ( ‘What is your race ? ’ ) . Hispanic/LatinX individuals were enrolled as part of a separate IDEO sub-study from the topic of this manuscript . Participants are also asked questions about their parents’ race and ethnic background . Each participant consented to take part in the study , which was approved by the UCSF Committee on Human Research . We utilized demographic , medical , dietary , and lifestyle metadata on each participant that were part of their initial recruitment into IDEO , as previously reported ( Alba et al . , 2018; Oguri et al . , 2020 ) . Participants with Type 2 Diabetes ( T2D ) were classified in accordance with American Diabetes Association Standards of Medical Care guidelines ( American Diabetes Association , 2019 ) , defined by having glycated hemoglobin ( HbA1c ) ≥6 . 5% or the combination of a prior diagnosis of T2D and the active use of an antidiabetic medication . For stool sample collection , participants took home or were mailed a stool sample collection kit and detailed instructions on how to collect the specimen . All samples were collected at home , stored at room temperature , and brought to the UCSF Clinical Research Center by the participants within 24 hr of defecation . Samples were aliquoted and stored at –80°C . We leveraged host phenotypic and demographic data from IDEO , which was the focus of two previous studies ( Alba et al . , 2018; Oguri et al . , 2020 ) . For the convenience of the reader , we restate our methods here . Height and weight were measured using a standard stadiometer and scale , and BMI ( kg/m2 ) was calculated from two averaged measurements . Waist and hip circumferences ( to the nearest 0 . 5 cm ) were measured using a plastic tape meter at the level of the umbilicus and of the greater trochanters , respectively , and waist-to-hip ratio ( WHR ) was calculated . Blood pressure was measured with a standard mercury sphygmomanometer on the left arm after at least 10 min of rest . Mean values were determined from two independent measurements . Blood samples were collected after an overnight fast and analyzed for plasma glucose , insulin , serum total cholesterol , high-density lipoprotein ( HDL ) cholesterol , and triglycerides . Low-density lipoprotein ( LDL ) cholesterol was estimated according to the Friedewald formula ( Friedewald et al . , 1972 ) . Insulin resistance was estimated by the homeostatic model assessment of insulin resistance ( HOMA-IR ) index calculated from fasting glucose and insulin values ( Matthews et al . , 1985 ) . Two obese subjects on insulin were included in the HOMA-IR analysis ( 1 EA , 1 W ) . Body composition of the subjects was estimated by Dual-Energy X-ray Absorptiometry ( DEXA ) using a Hologic Horizon/A scanner ( 3 min whole-body scan<0 . 1 mGy ) per manufacturer protocol . A single technologist analyzed all DEXA measurements using Hologic Apex software ( 13 . 6 . 0 . 4:3 ) following the International Society for Clinical Densitometry guidelines . Visceral adipose tissue ( VAT ) was estimated from a 5-cm-wide region across the abdomen just above the iliac crest , coincident with the fourth lumbar vertebrae , to avoid interference from iliac crest bone pixels and matching the region commonly used to analyze VAT mass by CT scan ( Bredella et al . , 2013; Kaul et al . , 2012; Neeland et al . , 2016 ) . The short version of the International Physical Activity Questionnaire ( IPAQ ) was used to assess the habitual physical activity levels of the participants . The IPAQ total score is expressed in metabolic equivalent ( MET ) -min/week ( Craig et al . , 2003 ) . IDEO participants completed two dietary questionnaires , as previously described ( Alba et al . , 2018; Oguri et al . , 2020 ) , allowing for the assessment of usual total fiber intake and fiber from specific sources , as well as macronutrient , phytochemical , vitamin , and mineral uptake . The first instrument was an Automated Self-Administered 24 hr Dietary Assessment ( ASA24 ) ( McClung et al . , 2018; Park et al . , 2018; Timon et al . , 2016 ) , which queries intake over a 24-hr period . The 24 hr recalls and supplement data were manually entered in the ASA24 Dietary Assessment Tool ( v . 2016 ) , an electronic data collection and dietary analysis program . ASA24 employs research-based strategies to enhance dietary recall using a respondent-driven approach allowing initial recall to be self-defined . The second instrument was the National Cancer Institute’s Diet History Questionnaire III ( DHQIII ) ( National Cancer Institute , 2020; Millen et al . , 2006 ) . The DHQIII queries one’s usual diet over the past month . Completing the DHQIII was associated with participant survey fatigue and completion rates were accordingly only 42% after one phone-based administration of the instrument , although they improved to 79% by the 2nd session and reached 100% within four sessions over a 5-month period . Due to the effort needed to achieve DHQIII completion , we modified our protocol to request completion of the simpler ASA24 at three separate times , at appointments where there were computers and personnel assistance for online completion , in addition to completion of the DHQIII questionnaire . By combining both instruments , we were able to reliably obtain complete dietary information on all participants . Human stool samples were homogenized with bead beating for 5 min ( Mini-Beadbeater-96 , BioSpec ) using beads of mixed size and material ( Lysing Matrix E 2 ml Tube , MP Biomedicals ) in the digestion solution and lysis buffer of a Wizard SV 96 Genomic DNA Kit ( Promega ) . The samples were centrifuged for 10 min at 16 , 000×g and the supernatant was transferred to the binding plate . The DNA was then purified according to the manufacturer’s instructions . Mouse fecal pellets were homogenized with bead beating for 5 min ( Mini-Beadbeater-96 , BioSpec ) using the ZR BashingBead lysis matrix containing 0 . 1 and 0 . 5 mm beads ( ZR-96 BashingBead Lysis Rack , Zymo Research ) and the lysis solution provided in the ZymoBIOMICS 96 MagBead DNA Kit ( Zymo Research ) . The samples were centrifuged for 5 min at 3000×g and the supernatant was transferred to 1 ml deep-well plates . The DNA was then purified using the ZymoBIOMICS 96 MagBead DNA Kit ( Zymo Research ) according to the manufacturer’s instructions . For human samples , 16S rRNA gene amplification was carried out using GoLay-barcoded 515F/806R primers ( Caporaso et al . , 2012 ) targeting the V4 region of the 16S rRNA gene according to the methods of the Earth Microbiome Project ( earthmicrobiome . org ) ( Supplementary file 1B ) . Briefly , 2 µl of DNA was combined with 25 µl of AmpliTaq Gold 360 Master Mix ( Thermo Fisher Scientific ) , 5 µl of primers ( 2 µM each GoLay-barcoded 515/806R ) , and 18 µl H2O . Amplification was as follows: 10 min 95°C , 30× ( 30 s 95°C , 30 s 50°C , 30 s 72°C ) , and 7 min 72°C . Amplicons were quantified with PicoGreen ( Quant-It dsDNA; Life Technologies ) and pooled at equimolar concentrations . Aliquots of the pool were then column ( MinElute PCR Purification Kit; Qiagen ) and gel purified ( QIAquick Gel Extraction Kit; Qiagen ) . Libraries were then quantified ( KAPA Library Quantification Kit; Illumina ) and sequenced with a 600 cycle MiSeq Reagent Kit ( 250×150; Illumina ) with ~15% PhiX spike-in . For mouse samples , 16S rRNA gene amplification was carried out as per reference protocol and primers ( Gohl et al . , 2016 ) . In brief , the V4 region of the 16S rRNA gene was amplified with 515F/806R primers containing common adaptor sequences , and then the Illumina flow cell adaptors and dual indices were added in a secondary amplification step ( see Supplementary file 1I for index sequences ) . Amplicons were pooled and normalized using the SequalPrep Normalization Plate Kit ( Invitrogen ) . Aliquots of the pool were then column ( MinElute PCR Purification Kit , Qiagen ) and gel purified ( QIAquick Gel Extraction Kit , Qiagen ) . Libraries were then quantified and sequenced with a 600 cycle MiSeq Reagent Kit ( 270×270; Illumina ) with ~15% PhiX spike-in . Demultiplexed sequencing reads were processed using QIIME2 v2020 . 2 ( Bolyen et al . , 2019 ) with denoising by DADA2 ( Callahan et al . , 2016 ) . Taxonomy was assigned using the DADA2 implementation of the RDP classifier ( Wang et al . , 2007 ) using the DADA2 formatted training sets for SILVA version 138 ( benjjneb . github . io/dada2/assign . html ) . For ASV analyses , we utilized quality scores to set truncation and trim parameters . The reverse read of human 16S data suffered from low sequence quality and reduced the overall ASV counts , so we therefore analyzed only the forward reads , although a separate analysis using merged forward and reverse reads complemented the findings we report in this manuscript . For the manuscript , forward reads were truncated to 220 base pairs and underwent an additional 5 base pairs of trimming for 16S analysis of human stool . For gnotobiotic mice , forward and reverse reads were truncated to 200 and 150 base pairs , respectively . ASVs were filtered such that they were present in more than one sample with at least a total of 10 reads across all samples . Alpha diversity metrics were calculated on subsampled reads using Vegan ( Dixon , 2003 ) and Picante ( Kembel et al . , 2010 ) R packages . The PhILR Euclidean distance was calculated by first carrying out the phylogenetic isometric log ratio transformation ( philr , PhILR [Silverman et al . , 2017] ) followed by calculating the Euclidean distance ( vegdist , Vegan [Dixon , 2003] ) . Principal coordinates analysis was carried out using the pcoa function of APE ( Paradis et al . , 2004 ) . ADONIS calculations were carried out ( adonis , Vegan ) with 999 replications on each distance metric . The permutational space for the adonis calculation for the three combined gnotobiotic experiments was restricted by donor identifier to account for multiple recipient mice for a given donor and applied to Figure 7—figure supplement 3 using setblocks to define permutations and specifying these blocks in the command . Centered log2-ratio ( CLR ) normalized abundances were calculated using the Make . CLR function in MicrobeR package ( Bisanz , 2017 ) with count zero multiplicative replacement ( zCompositions; Martín-Fernández et al . , 2014 ) . ALDEx2 ( Fernandes et al . , 2013 ) was used to analyze differential abundances of count data , using features that represented at least 0 . 05% of total sequencing reads . Corrections for multiple hypotheses using the Benjamini-Hochberg method ( Benjamini and Hochberg , 1995 ) were performed where applicable . Where described , a false discovery rate ( FDR ) indicates the Benjamini-Hochberg adjusted p-value for an FDR ( 0 . 1 unless otherwise specified ) . Analysis of distance matrices and alpha diversity mirror prior analyses developed in the Turnbaugh lab and were adapted to the current manuscript ( Bisanz et al . , 2019 ) . Calculations of associations between ASVs and ASA24 questionnaire data were completed by calculating a Spearman rank correlation and then adjusting the p-value for a Benjamini-Hochberg FDR using the cor_pmat function in the R package ggcorrplot ( Kassambara and Kassambara , 2019 ) for all CLR transformed ASVs detected between ethnic groups . Shotgun data for each ethnicity was processed using Metaphlan2 and the species associations were calculated for relative abundance by ASA24 questionnaire data separate from the ASV data . The randomForest package ( Liaw and Wiener M , 2002 ) was employed to generate random forest classifiers . Given the total number of samples ( n=46 ) , we generated 46 classifiers trained on a subset of 45 samples and used each classifier to predict the sample left out . AUCs are visualized utilizing the pROC ( Robin et al . , 2011 ) and ROCR ( Sing et al . , 2005 ) packages . Whole-genome shotgun libraries were prepared using the Nextera XT DNA Library Prep Kit ( Illumina ) . Paired ends of all libraries were sequenced on the NovaSeq 6000 platform in a single sequencing run ( n=45 subjects; see Supplementary file 1B for relevant metadata and statistics ) . Illumina reads underwent quality trimming and adaptor removal using fastp ( Chen et al . , 2018 ) and host read removal using BMTagger v1 . 1 . 0 ( ftp . ncbi . nlm . nih . gov/pub/agarwala/bmtagger/ ) in the metaWRAP pipeline ( github . com/bxlab/metaWRAP ) ( Uritskiy et al . , 2018 ) . Metagenomic samples were taxonomically profiled using MetaPhlan2 v2 . 7 . 7 ( Truong et al . , 2015 ) and functionally profiled using HUMAnN2 v0 . 11 . 2 ( Franzosa et al . , 2018 ) , both with default parameters . Principal coordinates analysis on MetaPhlan2 species-level abundances was carried out using Bray Curtis distances and the pcoa function of APE ( Paradis et al . , 2004 ) . Metaphlan2 abundance outputs were converted to counts and subsampled to even sample depth . Differences between groups were determined utilizing the Aldex2 package as described above . Tables of gene family abundances from HUMAnN2 were regrouped to KEGG orthologous groups using humann2_regroup_table . Functional pathways relating to SCFA production were manually curated from the pathway outputs from HUMANn2 and normalized by the estimated genome equivalents in each microbial community obtained from MicrobeCensus ( Nayfach and Pollard , 2015 ) . Absolute 16S rRNA gene copy number was derived by adjustments for dilutions during DNA extraction and template normalization dividing by the total fecal mass used for DNA extraction in grams . Quantification of bacterial load was conducted using quantitative PCR ( qPCR ) given stool samples were frozen for the IDEO cohort as described above and bacterial lysis was achieved with a preparation including both bead beating and a detergent . Differences in 16S rRNA gene copy number between bacterial strains may have masked more subtle differences in colonization level . qPCR was performed on DNA extracted from the human stool samples . DNA templates were diluted 1:10 into a 96-well plate . Samples were aliquoted in a 384-well plate , and PCR primers and iTaq Universal Probes Supermix were added utilizing an Opentrons OT-2 instrument then analyzed on a Bio-Rad CFX384 thermocycler with an annealing temperature of 60°C . The following primers including a FAM labeled PCR probe was used for quantification: 891F , TGGAGCATGTGGTTTAATTCGA; 1003R , TGCGGGACTTAACCCAACA; 1002P , [6FAM]CACGAGCTGACGACARCCATGCA[BHQ1] . Absolute quantifications were determined against a standard curve of purified 8F/1542R amplified Vibrio casei DNA . Reactions identified as inappropriately amplified by the instrument were rejected , and the mean values were used for downstream analysis . Absolute 16S rRNA gene copy number was derived by adjustments for dilutions during DNA extraction and template normalization dividing by the total fecal mass used for DNA extraction in grams . Quantification of bacterial load was conducted using qPCR given stool samples were frozen for the IDEO cohort as described above and bacterial lysis was achieved with a preparation including both bead beating and a detergent . NMR spectroscopy was performed at 298K on a Bruker Avance III 600 MHz spectrometer configured with a 5 mm inverse cryogenic probe ( Bruker Biospin , Germany ) as previously described ( Cai et al . , 2017 ) . Lean and obese EA and W individuals ( n=20 total individuals , five in each group ) were selected and matched based on body composition and metabolic parameters . Stool samples from these subjects were subjected to NMR-based metabolomics . 50 mg of human feces were extracted with 1 ml of phosphate buffer ( K2HPO4/NaH2PO4 , 0 . 1 M , pH 7 . 4 , 50% v/v D2O ) containing 0 . 005% sodium 3- ( trimethylsilyl ) [2 , 2 , 3 , 3–2 H4] propionate ( TSP-d4 ) as a chemical shift reference ( δ 0 . 00 ) . Samples were freeze-thawed three times with liquid nitrogen and water bath for thorough extraction , then homogenized ( 6500 rpm , 1 cycle , 60 s ) and centrifuged ( 11 , 180×g , 4°C , 10 min ) . The supernatants were transferred to a new 2 ml tube . An additional 600 μl of phosphate-buffered saline was added to the pellets , followed by the same extraction procedure described above . Combined fecal extracts were centrifuged ( 11 , 180×g , 4°C , 10 min ) , 600 μl of the supernatant was transferred to a 5 mm NMR tube ( Norell , Morganton , NC ) for NMR spectroscopy analysis . A standard one-dimensional NOESY pulse sequence noesypr1d ( recycle delay-90°-t1-90°-tm-90°-acquisition ) was used with a 90 pulse length of approximately 10 s ( –9 . 6 dbW ) and 64 transients are recorded into 32k data points with a spectral width of 9 . 6 kHz . NMR spectra were processed as previously described ( Cai et al . , 2017 ) . First , spectra quality was improved with Topspin 3 . 0 ( Bruker Biospin , Germany ) for phase and baseline correction and chemical shift calibration . AMIX software ( version: 3 . 9 . 14 , Bruker Biospin , Germany ) was used for bucketing ( bucket width 0 . 004 ppm ) , removal of interfering signal , and scaling ( total intensity ) . Relative concentrations of identified metabolites were obtained by normalized peak area . Targeted analysis of SCFAs and BCAAs was performed with an Agilent 7890A gas chromatograph coupled with an Agilent 5975 mass spectrometer ( Agilent Technologies , Santa Clara , CA ) using a propyl esterification method as previously described ( Cai et al . , 2017 ) . 50 mg of human fecal samples were pre-weighed , mixed with 1 ml of 0 . 005 M NaOH containing 10 μg/ml caproic acid-6 , 6 , 6-d3 ( internal standard ) and 1 . 0 mm diameter zirconia/silica beads ( BioSpec , Bartlesville , OK ) . The mixture was thoroughly homogenized and centrifuged ( 13 , 200×g , 4°C , 20 min ) . 500 μl of supernatant was transferred to a 20 ml glass scintillation vial . 500 μl of 1-propanol/pyridine ( v/v=3/2 ) solvent was added into the vial , followed by a slow adding of an aliquot of 100 μl of esterification reagent propyl chloroformate . After a brief vortex of the mixture for 1 min , samples were derivatized at 60°C for 1 hr . After derivatization , samples were extracted with hexane in a two-step procedure ( 300 μl + 200 μl ) as described ( Zheng et al . , 2013 ) . First , 300 μl of hexane was added to the sample , briefly vortexed and centrifuged ( 2000×g , 4°C , 5 min ) , and 300 μl of the upper layer was transferred to a glass autosampler vial . Second , an additional 200 μl of hexane was added to the sample , vortexed , centrifuged , and the 200 μl upper layer was transferred to the glass autosampler vial . A combination of 500 μl of extracts were obtained for GC-MS analysis . A calibration curve of each SCFA and BCAA was generated with series dilution of the standard for absolute quantitation of the biological concentration of SCFAs and BCAAs in human fecal samples . Bile acid quantitation was performed with an ACQUITY ultra high pressure liquid chromatography ( UHPLC ) system using a Ethylene Bridged Hybrid C8 column ( 1 . 7 µm , 100 mm×2 . 1 mm ) coupled with a Xevo TQ-S mass spectrometer equipped with an electrospray ionization source operating in negative mode ( All Waters , Milford , MA ) as previously described ( Sarafian et al . , 2015 ) . Selected ion monitoring for non-conjugated bile acids and multiple reaction monitoring for conjugated bile acids was used . 50 mg of human fecal sample was pre-weighed , mixed with 1 ml of pre-cooled methanol containing 0 . 5 μM of stable-isotope-labeled bile acids ( internal standards ) , and 1 . 0 mm diameter zirconia/silica beads ( BioSpec , Bartlesville , OK ) , followed by thorough homogenization and centrifugation . Supernatant was transferred to an autosampler vial for analysis . 100 µl of serum was extracted by adding 200 µl pre-cooled methanol containing 0 . 5 μM deuterated bile acids as internal standards . Following centrifugation , the supernatant of the extract was transferred to an autosampler vial for quantitation . Calibration curves of individual bile acids were drafted with bile acid standards for quantitation of the biological abundance of bile acids . All mouse experiments were approved by the UCSF Institutional Animal Care and Use Committee and performed accordingly . Germ-free mice were maintained within the UCSF Gnotobiotic Core Facility and fed ad libitum autoclaved standard chow diet ( Lab Diet 5021 ) . Germ-free adult male C57BL/6J mice between 6 and 10 weeks of age were used for all the experiments described in this paper . 10 lean subjects in our IDEO cohort were selected as donors for the microbiota transplantation experiments , including 5 EA and 5 W donors . The selected donors for gnotobiotic experiments were matched for phenotypic data to the degree possible ( Supplementary file 1H ) . Stool samples to be used for transplantation were resuspended in 10 volumes ( by weight ) of brain heart infusion media in an anaerobic Coy chamber . Each diluted sample was vortexed for 1 min and left to settle for 5 min , and a single 200 µl aliquot of the clarified supernatant was administered by oral gavage into each germ-free mouse recipient . In experiments LFPP1 and LFPP2 , microbiome transplantations were performed for two donors per experiment ( 1 W , 1 EA ) with gnotobiotic mice housed in sterile isolators ( CBC flexible , softwall isolator ) and maintained on ad libitum standard chow also known as low-fat , high-plant-polysaccharide ( LFPP ) diet . In LFPP1 , six germ-free mice per colonization group received an aliquot of stool from a donor of either ethnicity and body composition ( measured using EchoMRI ) were recorded on the day of colonization and at 6 weeks post-transplantation ( per group n=6 recipient mice , 1 isolator , 2 cages ) . In LFPP2 , we shortened the colonization time to 3 weeks and used two new donor samples . For the third experiment ( HFHS experiment ) , mice were weaned onto an irradiated HFHS diet ( TD . 88137 , Envigo ) for 4 weeks prior to colonization and housed in pairs in Tecniplast IsoCages . The same four donors from LFPP1 and LFPP2 were included in the HFHS experiment , in addition to six new donors ( per donor n=2 recipient mice , 1 IsoCage ) . Body weight and body composition were recorded on the day of colonization and again at 3 weeks post-transplantation . Mice were maintained on the HFHS diet throughout the experiment . All samples were sequenced in a single pool ( Supplementary file 1I ) . For comparisons between donors and recipient mice , donors and recipient mice were subsampled to even sequencing depth and paired between donor and recipient mice ( range: 18 , 544–78 , 361 sequencing reads/sample ) . Food was removed from mice 10 hr ( LFPP1 experiment ) or 4 hr ( HFHS experiment ) prior to assessment of glucose tolerance . Mice received i . p . injections of D-glucose ( 2 mg/kg ) , followed by the repeated collection of blood by tail nick and determination of glucose levels by handheld glucometer ( Abbott Diabetes Care ) over a 2-hr period . Map tiles and distance data were obtained using GGMap ( Kahle and Wickham , 2013 ) , OpenStreet Maps ( Fellows and Stotz , 2016 ) , and the Imap R ( Wallace , 2012 ) packages . GGMap was employed using a Google Cloud API key and the final map tiles were obtained in July 2020 ( Kahle and Wickham , 2013 ) . Spearman ranked correlation coefficients ( rho ) were calculated as embedded in the ggpubr ( Kassambara , 2018 ) R package . 2018 US Census data for EA and W subjects was obtained ( B02001 table for race , data . census . gov ) for the ZIP codes available in our study and using the leaflet ( Cheng et al . , 2018 ) package . The census data used is included as part of Supplementary file 1B to aid in reproduction . Each census region is plotted as a percentage of W individuals over a denominator of W and EA subjects . The leaflet package utilized ZIP Code Tabulation Areas ( ZCTAs ) from the 2010 census . We extracted all ZCTAs starting with 9 , and the resulting 29 ZIP codes that overlap with IDEO subjects were analyzed ( Supplementary file 1B ) . Two ZCTAs ( 95687 and 95401 ) were primarily W when comparing W and EA subjects . There were two W subjects recruited from these ZTCAs . These ZIP codes are cutoff based on the zoom magnification for that figure and as a result ZTCAs for 27 individuals are plotted . Distance to a central point in SF was calculated . The point of reference was latitude=37 . 7585102 , longitude=−122 . 4539916 . DHQIII and ASA24 data were analyzed using a Euclidean distance matrix . These transformations were completed using the cluster package ( Maechler et al . , 2021 ) . Subsequent analysis was completed using the vegan package ( Dixon , 2003; Oksanen et al . , 2013 ) . Procrustes transformations were performed using 16S-seq data from human subjects , which was then subjected to a PhILR transformation . The resulting matrix was rotated against the distance matrix for ASA24 or DHQIII questionnaire data using the procrustes command in the vegan R package using 999 permutations . Mantel statistics were calculated utilizing the mantel command of the vegan package . Picante ( Kembel et al . , 2010 ) , PhILR ( Silverman et al . , 2017 ) , MicrobeR ( Bisanz , 2017 ) , ALDEx2 ( Fernandes et al . , 2013 ) , ggcorrplot ( Kassambara and Kassambara , 2019 ) , randomForest ( Liaw and Wiener M , 2002 ) , GGMap ( Kahle and Wickham , 2013 ) , OpenStreetMap ( Fellows and Stotz , 2016 ) , IMap ( Wallace , 2012 ) , ggpubr ( Kassambara , 2018 ) , leaflet ( Cheng et al . , 2018 ) , cluster ( Maechler et al . , 2021 ) , readxl ( Wickham and Bryan , 2017 ) , Rtsne ( Krijthe , 2015 ) , vegan ( Dixon , 2003; Oksanen et al . , 2013 ) , ape ( Paradis and Schliep , 2019 ) , tigris ( Walker , 2018 ) , lmerTest ( Kuznetsova et al . , 2017 ) , qiime2R ( Bisanz , 2018 ) , gghighlight ( Yutani , 2018 ) , Phyloseq ( McMurdie et al . , 2013 ) , Janitor ( Firke , 2018 ) , table1 ( Rich , 2020 ) , and ggplot2 ( Wickham , 2016 ) . Statistical analysis of the human data was performed using the table1 package in R ( STATCorp LLC , College Station , TX ) . Human data were presented as mean ± SD . Unpaired independent Student’s t-tests were used to compare differences between the two groups in the case of continuous data and in the case of categorical data the χ2 test was utilized for Supplementary file 1A . For a given lean or obese categories between ethnicity tests were adjusted for a Benjamini-Hochberg FDR utilizing the command p . adjust in R , which is indicated as an adjusted p-value in the tables and none were significant as described in the table legend . In Supplementary file 1G , H , no values met an adjusted p-value cutoff of <0 . 1 . In Supplementary file 1A and p-values indicated by numbers were pooled together for adjustments and those represented by symbols were separately pooled together for adjustment . All microbiome-related analyses were carried out in R version 3 . 5 . 3 or 4 . 0 . 2 . Where indicated , Wilcoxon rank-sum tests were calculated . A Benjamini-Hochberg adjusted p-value ( FDR ) of 0 . 1 was used as the cutoff for statistical significance unless stated otherwise . Statistical analysis of glucose tolerance tests was carried out using linear mixed-effects models with the lmerTest ( Kuznetsova et al . , 2017 ) R package and mouse as random effect . Graphical representation was carried out using ggplot2 . Boxplots indicate the interquartile range ( 25th to 75th percentiles ) , with the center line indicating the median and whiskers representing 1 . 5× the interquartile range . All 16S-seq and metagenomic sequencing data generated in the preparation of this manuscript have been deposited in NCBI’s Sequence Read Archive under accession number PRJNA665061 . Metabolomics results and metadata are available within this manuscript ( Supplementary file 1 ) . Code for our manuscript and a more comprehensive metadata table is available on GitHub ( https://github . com/turnbaughlab/2021_IDEO , Upadhyay and Turnbaugh , 2021; copy archived at swh:1:rev:07f9ee797d57620e10734bef4d893bf51662559c ) .	The community of microbes living in the human gut varies based on where a person lives , in part because of differences in diets but also due to factors still incompletely understood . In turn , this ‘microbiome’ may have wide-ranging effects on health and diseases such as obesity and diabetes . Many scientists want to understand how differences in the microbiome emerge between people , and whether this may explain why certain diseases are more common in specific populations . Self-identified race or ethnicity can be a useful tool in that effort , as it can serve as a proxy for cultural habits ( such as diets ) or genetic information . In the United States , self-identified East Asian Americans often have worse ‘metabolic health’ ( e . g . levels of sugar or certain fat molecules in the blood ) at a lower weight than those identifying as White . Ang , Alba , Upadhyay et al . investigated whether this health disparity was linked to variation in the gut microbiome . Samples were collected from 46 lean and obese individuals living in the San Francisco Bay Area who identified as White or East Asian . The analyses showed that while the gut microbiome of White participants changed in association with obesity , the microbiomes of East Asian participants were distinct from their White counterparts even at normal weight , with features mirroring what was seen in White individuals in the context of obesity . Although these differences were connected to people’s current address , they were not attributable to dietary differences . Ang , Alba , Upadhyay et al . then transplanted the microbiome of the participants into genetically identical mice with microbe-free guts . The differences between the gut microbiomes of White and East Asian participants persisted in recipient animals . When fed the same diet , the mice also gained different amounts of weight depending on the ethnic identity of the microbial donor . These results show that self-identified ethnicity may be an important variable to consider in microbiome studies , alongside other factors such as geography . Ultimately , this research may help to design better , more personalized treatments for an array of conditions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2021	The East Asian gut microbiome is distinct from colocalized White subjects and connected to metabolic health
Extinction of fear responses is critical for adaptive behavior and deficits in this form of safety learning are hallmark of anxiety disorders . However , the neuronal mechanisms that initiate extinction learning are largely unknown . Here we show , using single-unit electrophysiology and cell-type specific fiber photometry , that dopamine neurons in the ventral tegmental area ( VTA ) are activated by the omission of the aversive unconditioned stimulus ( US ) during fear extinction . This dopamine signal occurred specifically during the beginning of extinction when the US omission is unexpected , and correlated strongly with extinction learning . Furthermore , temporally-specific optogenetic inhibition or excitation of dopamine neurons at the time of the US omission revealed that this dopamine signal is both necessary for , and sufficient to accelerate , normal fear extinction learning . These results identify a prediction error-like neuronal signal that is necessary to initiate fear extinction and reveal a crucial role of DA neurons in this form of safety learning . The ability to learn which stimuli predict danger is crucial for survival but it is equally important to adapt behavior when those stimuli no longer represent a threat . One classic example of this is fear extinction learning , during which the repeated presentation of a stimulus ( conditioned stimulus , CS ) that no longer predicts an aversive outcome ( unconditioned stimulus , US ) leads to a gradual decrease in learned fear responses . Many anxiety disorders , such as post-traumatic stress disorder , are characterized by impaired extinction learning ( Craske et al . , 2017; Graham and Milad , 2011; Mahan and Ressler , 2012; Milad and Quirk , 2012; Pitman et al . , 2012 ) and thus understanding the neural basis of fear extinction has clinical significance . A large body of evidence indicates that fear extinction represents new learning rather than forgetting or the erasure of the original fear memory ( Bouton et al . , 2006; Myers and Davis , 2007 ) . In order to initiate extinction learning , the absence of the expected aversive outcome must be detected and signaled to the brain regions mediating fear extinction . Decades of research on fear extinction has revealed that a distributed network of brain structures including the amygdala , medial prefrontal cortex and hippocampus mediates the acquisition , consolidation and retrieval of fear extinction memories ( Duvarci and Pare , 2014; Maren et al . , 2013; Pape and Pare , 2010; Sotres-Bayon and Quirk , 2010; Tovote et al . , 2015 ) . However , none of these structures have been shown to signal the absence of the expected aversive outcome during fear extinction . The neural substrates of such a signal that could initiate extinction learning have therefore remained elusive . New learning is initiated when outcomes violate expectations ( Rescorla and Wagner , 1972 ) . Such violations are thought to cause ‘prediction error’ signals that initiate the neural processes which ultimately lead to changes in behavior ( Friston , 2012; den Ouden et al . , 2012 ) . During fear extinction , the absence of the US is an unexpected event and likely generates a prediction error signal that initiates extinction learning . More specifically , the omission of the aversive US can be conceptualized as a better-than-expected outcome . It is well-established that the activity of midbrain dopamine ( DA ) neurons represents the degree to which outcomes are better or worse than expected ( Bayer and Glimcher , 2005; Eshel et al . , 2015; Eshel et al . , 2016; Schultz et al . , 1997; Schultz and Dickinson , 2000 ) . For example , many DA neurons increase their firing to rewards that are either unexpected or better than expected and this DA signal is sufficient to drive reinforcement learning ( Steinberg et al . , 2013 ) . Based on our data from human studies , we have previously proposed that DA neurons could provide a prediction error-like signal during the aversive US omission to initiate fear extinction ( Raczka et al . , 2011 ) . Consistent with this , an increase in DA release has been observed in the nucleus accumbens ( NAc ) during fear extinction ( Badrinarayan et al . , 2012 ) and pharmacological blockade of DA receptors in the NAc impairs fear extinction ( Holtzman-Assif et al . , 2010 ) . However , the electrical activity of DA neurons during fear extinction − particularly at the time of the aversive US omission − and its relationship to extinction learning , is incompletely understood . In this study , we hypothesized that the unexpected omission of the aversive US activates DA neurons in the ventral tegmental area ( VTA ) and that this signal is necessary to initiate normal fear extinction learning . To test this hypothesis , we used in vivo single-unit recordings , DA neuron-specific calcium recordings and bi-directional optogenetic manipulations in behaving mice to examine and causally test the role of VTA DA neurons in fear extinction . We first examined whether DA neurons in the VTA are activated by the unexpected omission of the aversive US during fear extinction . We recorded the single-unit spiking activity of VTA neurons ( Figure 1A–C;Figure 1—figure supplement 1 ) in mice ( n = 11 ) that were trained in a fear conditioning paradigm ( Figure 1D–F;Figure 1—figure supplement 2A ) where a tone ( CS ) was paired with an aversive foot shock ( US ) . Twenty-four hours after fear conditioning , mice received an extinction session consisting of CS presentations in the absence of the aversive US . A total of 43 ( out of 90 ) and 40 ( out of 75 ) VTA neurons classified as ‘putative’ DA neurons ( see Materials and methods; Figure 1—figure supplement 3 ) were recorded during day 1 and day 2 , respectively . Analysis of neuronal firing rates during the time of the US omission revealed that 25% of putative DA neurons ( 10 of 40 ) exhibited a significant increase in firing rate to the omission of the aversive US during the early extinction trials ( E-Ext: average of first 10 CSs ) when the US omission was unexpected ( Figure 1G-I;Figure 1—figure supplement 3B; see Materials and methods for details ) . This was not simply a response to the CS offset since only 2 . 3% of neurons ( 1 of 43; Figure 1H;Figure 1—figure supplement 3A ) showed increased firing at the end of the CS during tone habituation ( Hab ) . On the other hand , during late extinction trials ( L-Ext: average of last 10 CSs ) , when the US omission was no longer unexpected and animals showed significant extinction of fear responses ( Figure 1F ) , only 7 . 5% of putative DA neurons ( 3 of 40; Figure 1G-I ) showed an increase in firing to the absence of the US . Importantly , there was no difference in freezing levels during the CS and the post-CS period in E-Ext ( Figure 1—figure supplement 2B ) suggesting that the observed increase in DA neuron firing was not due to an increase in movement when the CS terminates . Furthermore , analysis of the distribution of z-scores during the US omission ( Figure 2A ) revealed that none of the putative DA neurons ( 0 of 40 ) showed a selective decrease in firing to the US omission ( see Materials and methods for details ) during either E-Ext or L-Ext . This suggests that the dominant response of putative DA neurons to US omission was excitatory . Consistent with this , we observed a significant increase in firing to the US omission at the population level when we examined the average response of all putative DA neurons during E-Ext ( paired t-test , t ( 39 ) = 2 . 22 , p = 0 . 03 ) , but not L-Ext ( paired t-test , t ( 39 ) = 0 . 20 , p = 0 . 83 ) or Hab ( paired t-test , t ( 42 ) = 0 . 71 , p = 0 . 47; Figure 1J ) . Local circuit interactions between DA and GABA ( γ-aminobutyric acid ) neurons in the VTA underlie reinforcement learning ( Cohen et al . , 2012; Eshel et al . , 2015 ) . It is therefore possible that the DA and local GABA neurons also interact to drive fear extinction learning . For instance , the increased firing of putative DA neurons could be mediated by disinhibition resulting from inhibition of local GABA neurons at the time of the US omission . To test this possibility , we analyzed the activity of putative non-DA neurons – the subset of VTA neurons which likely dominantly includes GABA cells – during the time of the US omission and found that 8 . 5% of neurons ( 3 of 35 ) showed a selective decrease in firing to the US omission during E-Ext ( Figure 1—figure supplement 3B , Figure 2B ) . However , this was not significantly different from the proportion of non-DA neurons ( 4 . 2% , 2 of 47 neurons ) showing decreased firing at the CS offset during Hab ( Figure 1—figure supplement 3B , Figure 2B; Fisher’s exact test , p = 0 . 64 ) . These results therefore suggest that the excitation observed in the putative DA neurons during US omission is unlikely mediated by the activity of the local GABA neurons . Moreover , only a small proportion of putative non-DA neurons showed increased firing at the time of the US omission during E-Ext ( 2 . 8% , 1 of 35 neurons; Figure 1—figure supplement 3B , Figure 2B ) . This proportion was again not different from the 2 . 1% of neurons ( 1 of 47 ) that showed increased firing during Hab ( Figure 1—figure supplement 3B , Figure 2B; Fisher’s exact test , p = 1 ) . Together , these results suggest that the putative non-DA , likely local GABA , neurons in VTA do not change their firing at the time of the US omission and that the US omission is signaled specifically by the putative DA neurons in VTA . In contrast to the uniform response observed during US omission , the responses of putative DA neurons during the CS were diverse . Figure 1—figure supplement 4 shows example neurons that displayed excitation , inhibition and also biphasic response during the CS . Whereas some neurons showed sustained responses to the CS ( Figure 1—figure supplement 4B ) , others showed a transient response at the onset of the CS ( Figure 1—figure supplement 4C ) . In order to quantify these CS-evoked responses , we calculated the mean response to the CS for each neuron during Hab , E-Ext and L-Ext ( Figure 1—figure supplement 4A ) . We found that 5% ( 2 of 40 ) of putative DA neurons exhibited CS-evoked excitation and 7 . 5% ( 3 of 40 ) CS-evoked inhibition during E-Ext . However these percentages were not significantly different from the proportion of cells that showed excitation ( 6 . 9% , 3 of 43 neurons; Fisher’s exact test , p = 1 ) or inhibition ( 0% , 0 of 43; Fisher’s exact test , p = 0 . 1 ) during Hab . Furthermore , when we examined the population activity by averaging the response of all putative DA neurons , there was no significant change in the average response during the CS ( paired t-tests , Hab: t ( 42 ) = 0 . 58 , p = 0 . 56 , E-Ext: t ( 39 ) = 1 . 89 , p = 0 . 065 , L-Ext: t ( 39 ) = 1 . 23 , p = 0 . 22; Figure 1J;Figure 1—figure supplement 4A ) . These results suggest that the average activity of putative DA neurons in the VTA did not change during the CS even though different subsets of cells showed CS-evoked excitation and inhibition . A lack of strong responses to the CS might be due to the cued fear conditioning paradigm that we used in our study . It has recently been shown that DA neurons respond strongly to the CS when a discriminative fear conditioning task is used and the strength of their response increases with increasing discrimination between the aversive and the safe CS ( Jo et al . , 2018 ) . The above results show that putative DA , but not non-DA , neurons in the VTA signal the omission of the aversive US during fear extinction , specifically during the beginning of extinction learning when the US omission is unexpected . To further confirm that DA neurons signal the unexpected omission of the US , we next measured activity-dependent calcium signals selectively in DA neurons using fiber photometry . To this end , a Cre-dependent adeno-associated virus ( AAV ) expressing the genetically encoded calcium ( Ca+2 ) indicator gCaMP6 was injected , and an optical fiber implanted , in the VTA of transgenic mice expressing Cre recombinase under the control of the dopamine transporter ( Dat ) promoter ( DAT-Cre mice; Figure 3A–B and Figure 3—figure supplement 1A ) . In these mice , Cre expression is highly selective for dopamine neurons ( Lammel et al . , 2015 ) . Accordingly , we observed a high degree of overlap between Cre-dependent gCaMP6 expression and immunohistochemical staining against tyrosine hydroxylase ( TH; Figure 3—figure supplement 1B–C ) . In control mice , we injected a Cre-dependent AAV expressing GFP to examine whether changes in fluorescence could be independent of neuronal activity . Recordings from gCaMP6-expressing animals revealed transient increases in fluorescence whereas such increases were absent in mice expressing GFP ( Figure 3C ) . Furthermore , consistent with electrophysiological studies in VTA DA neurons ( Eshel et al . , 2015; Eshel et al . , 2016; Roesch et al . , 2007 ) , we also confirmed that reward delivery caused large increases in fluorescence in the gCaMP6-expressing mice ( Figure 3—figure supplement 2 , also see Materials and methods ) , indicating that this Ca+2 signal is indeed generated by DA neuron activity . Both gCaMP6- and GFP-expressing animals underwent the same fear conditioning protocol as in the electrophysiology experiment ( Figure 3D–E ) and showed comparable levels of freezing to the CS across sessions ( two-way repeated measures ANOVA; main effect of group: F1 , 24 = 0 . 04 , p = 0 . 83; group × trial interaction: F2 , 24 = 0 . 04 , p = 0 . 96; Figure 3F and Figure 3—figure supplement 3A ) . During E-Ext , we observed a significant increase in the Ca+2 signal of gCaMP6 animals at the time of the US omission compared to the pre-CS baseline ( p<0 . 01 , sign-rank test ) and the GFP control group ( p<0 . 01 , rank-sum test; Figure 3G ) . On the other hand , during L-Ext , when the US omission was no longer unexpected , the Ca+2 signal did not change during the post-CS period ( p = 0 . 85 , sign-rank test; Figure 3G ) . No changes in fluorescence were observed in GFP-expressing animals during either E-Ext or L-Ext ( Figure 3G ) . Furthermore , we did not observe any change in the Ca+2 signal during Hab ( gCaMP6 group , p = 0 . 32 , sign-rank test; Figure 3G ) suggesting that the increase during US omission is unlikely a response to the CS offset . Furthermore , freezing levels during the CS and the post-CS period in E-Ext were comparable ( Figure 3—figure supplement 3B ) ruling out the possibility that an increase in movement when CS terminates might have resulted in the observed increase in Ca+2 signal at the time of the US omission during E-Ext . Notably , these results obtained by measuring the population Ca+2 signal from DA neurons are consistent with the results of the electrophysiology recordings where we found a significant increase in the average population activity of all putative DA neurons at the time of the US omission during E-Ext ( Figure 1J ) . If activation of DA neurons at the time of the unexpected US omission drives extinction learning then this Ca+2 signal during E-Ext should be larger in animals that exhibit better extinction learning . To test this , we took advantage of the variability in the extinction learning rates of individual mice and asked whether they were correlated with the Ca+2 signal at the time of the US omission during E-Ext . This revealed a significant correlation between the Ca+2 signal during E-Ext and the change in freezing from E-Ext to L-Ext ( Spearman’s correlation = = 0 . 83 , p = 0 . 0037; Figure 3H ) . A significant correlation was also observed between the Ca+2 signal during E-Ext and the freezing levels during L-Ext ( Spearman’s correlation = 0 . 92 , p = 0 . 0003 ) . However , it is possible that the variation in the Ca+2 signal during E-Ext might be due to the variation in recording locations in the VTA across animals rather than reflecting the relationship with extinction learning . We reasoned that differences in recording locations would likely result in variation in reward responses . We therefore examined the correlation between the change in freezing from E-Ext to L-Ext and reward responses and did not find a significant relationship between these two variables ( Spearman’s correlation = 0 . 32 , p = 0 . 36; Figure 3I ) . These results therefore suggest that the magnitude of the Ca+2 signal during E-Ext correlated with the level of extinction learning and the variations in the magnitude of this signal were not due to differences in the recording location . Contrasting with the uniform increase in DA neuron activity during US omission ( Figure 3G ) , the responses to the CS varied across animals ( Figure 4A ) . Some animals showed increased and some decreased fluorescence during the CS in E-Ext ( Figure 4A–B ) . Accordingly , consistent with single unit results which showed no change in CS-evoked population activity of putative DA neurons ( Figure 1J; Figure 1—figure supplement 4A ) , the Ca+2 signal evoked by the CS was not significantly different from the baseline during E-Ext ( p = 1 . 0 , sign-rank test ) or Hab ( p = 0 . 43 , sign-rank test ) when we averaged the CS responses of all animals ( Figure 4A ) . Overall , these findings together with our single-unit results demonstrate that DA neurons signal the unexpected omission of the aversive US during fear extinction and that the magnitude of this DA signal predicts the strength of extinction learning . We next asked whether the observed increase in DA neuron firing at the time of the unexpected US omission is necessary for fear extinction learning . To address this question , we optogenetically inhibited DA neurons in the VTA at the time of the US omission during fear extinction . DAT-cre mice received bilateral injections of a Cre-dependent AAV expressing either the light-activated inhibitory opsin halorhodopsin fused with enhanced yellow fluorescent protein ( NpHR-eYFP ) or eYFP only ( eYFP control ) into the VTA , as well as bilateral implantation of optical fibers above the VTA to allow for selective inhibition of VTA DA neurons ( Figure 5A–C and Figure 5—figure supplement 1A ) . We observed a high degree of overlap between Cre-dependent NpHR-eYFP expression and immunohistochemical staining against TH ( Figure 5—figure supplement 1B–C ) suggesting DA neuron-selective expression . Furthermore , we confirmed that optical stimulation of NpHR inhibits DA neuron firing in awake DAT-cre mice ( Figure 5—figure supplement 2 ) . Mice were trained in a fear conditioning protocol ( Figure 5D ) consisting of 4 CS-US pairings on day 1 . Twenty-four hours after fear conditioning , mice received an extinction session . In the experimental group expressing NpHR-eYFP light was delivered bilaterally to the VTA to inhibit DA neurons specifically at the end of each CS presentation , that is during the time of the US omission ( Paired-NpHR , n = 7; Figure 5E ) . The behavior of the experimental group was compared to two control groups: one group consisted of mice expressing eYFP only which received the identical light delivery ( Paired-eYFP , n = 7 ) and a second group consisted of mice expressing NpHR-eYFP that received light delivery to inhibit DA neurons during the intertrial intervals ( ITIs; Unpaired-NpHR , n = 8; Figure 5F ) . Compared to the two control groups , the Paired-NpHR group exhibited high freezing levels to the CS throughout the extinction session , suggesting impaired extinction learning ( Figure 5G ) . A two-way repeated measures ANOVA revealed a significant main effect of group ( F2 , 456 = 7 . 03 , p = 0 . 0052 ) and a significant interaction between group and trial ( F48 , 456 = 2 . 19 , p<0 . 0001 ) . Consistent with this , there was a significant difference between the Paired-NpHR group and the Paired-eYFP ( p<0 . 001 ) or Unpaired-NpHR ( p<0 . 001 ) controls during L-Ext ( Figure 5I ) . Furthermore , impaired extinction learning resulted in a weaker memory for extinction when tested the next day ( extinction recall test; two-way repeated measures ANOVA; main effect of group: F2 , 456 = 7 . 1 , p = 0 . 005; group × trial interaction: F48 , 456 = 1 . 49 , p = 0 . 02; Figure 5G ) . Consistently , during the early extinction recall trials ( E-Ext Rec: first 10 CSs ) the Paired-NpHR group froze significantly more compared to the Paired-eYFP ( p<0 . 001 ) or the Unpaired-NpHR ( p<0 . 001 ) controls ( Figure 5J ) . In contrast to the Paired-NpHR group , the Unpaired-NpHR group behaved comparable to the Paired-eYFP control group during both extinction and extinction recall tests ( Figure 5G–J ) suggesting that optical inhibition of DA neurons per se did not result in a nonspecific increase in freezing levels , and that the behavioral effect was dependent on the temporally specific inhibition of DA neurons during the time of the US omission . Furthermore , there was no difference between the groups in their fear acquisition on day 1 ( two-way repeated measures ANOVA , no main effect of group , F2 , 57 = 0 . 3 , p = 0 . 74 or no group × trial interaction , F6 , 57 = 0 . 94 , p = 0 . 47; Figure 5G ) and all groups showed comparable levels of freezing at the start of extinction ( first CS; one-way ANOVA , F2 , 19 = 2 . 06 , p = 0 . 15; Figure 5H ) before any experimental manipulation took place , ruling out the possibility that differences in the strength of fear memory on day 2 between groups might have caused the observed effect . Importantly , freezing levels of the NpHR group at the beginning of extinction recall were comparable to freezing levels at the beginning of extinction ( paired t-test comparing first CS of extinction and first CS of extinction recall , t ( 6 ) = 0 . 56 , p = 0 . 59 ) , suggesting that no significant extinction learning happened in these animals . Furthermore , we found that the extinction rate of the Paired-NpHR group during extinction recall , in the absence of optogenetic inhibition , was comparable to the extinction rate of the Paired-eYFP and Unpaired-NpHR groups during extinction ( two-way repeated measures ANOVA; no main effect of group F2 , 456 = 1 . 02 , p = 0 . 37 and group × trial interaction: F48 , 456 = 0 . 86 , p = 0 . 73 ) . This suggests that our manipulation did not have a nonspecific long-term effect on the ability of the Paired-NpHR group to exhibit extinction learning . These results also suggest that our optogenetic manipulation likely did not affect the strength of the fear memory . Taken together , these findings demonstrate that DA neuron activation by the unexpected omission of the US is necessary for fear extinction learning . If DA neuron firing at the time of the unexpected US omission drives fear extinction , then enhancing this DA signal should accelerate fear extinction learning . To test this , we optogenetically excited DA neurons precisely at the time of the US omission during fear extinction learning . DAT-cre mice were bilaterally injected with a Cre-dependent AAV expressing either channelrhodopsin-2 ( ChR2 ) fused with eYFP ( ChR2-eYFP ) or eYFP only ( eYFP control ) into the VTA , and implanted bilaterally with optical fibers above VTA ( Figure 6A–C; Figure 6—figure supplement 1A ) . There was again a high level of overlap between Cre-dependent ChR2-eYFP expression and immunohistochemical staining against TH ( Figure 6—figure supplement 1B–C ) suggesting DA neuron-selective expression of ChR2 . Furthermore , we confirmed that optical stimulation of ChR2 induces firing of DA neurons in awake DAT-cre mice ( Figure 6—figure supplement 2 ) . Mice were trained in a fear conditioning protocol ( Figure 6D ) similar to the optogenetic inhibition experiment . The experimental group consisted of ChR2-eYFP expressing mice which received light stimulation of DA neurons specifically at the time of the US omission ( Paired-ChR2 , n = 7; Figure 6E ) . Two control groups , one expressing eYFP only that received the identical light delivery ( Paired-eYFP , n = 7 ) and the other expressing ChR2-eYFP that received light excitation during the ITIs ( Unpaired-ChR2 , n = 7; Figure 6F ) , were used to control for nonspecific effects of light and DA neuron stimulation , respectively . As expected , all groups showed a gradual decrease in freezing to the CS during the extinction session . However , in the Paired-ChR2 group , freezing decreased faster than in the control groups suggesting accelerated extinction learning ( Figure 6G ) . A two-way repeated measures ANOVA comparing freezing levels confirmed this observation by revealing a significant main effect of group ( F2 , 432 = 4 . 1 , p = 0 . 03 ) . Comparison of freezing levels in the three groups , particularly during E-Ext , revealed a significant difference between the Paired-ChR2 group and the Paired-eYFP ( p<0 . 05 ) or Unpaired-ChR2 ( p<0 . 05 ) controls ( Figure 6I ) suggesting accelerated extinction learning . On the other hand , the two control groups behaved comparably ( p>0 . 05 ) . These results suggest that optical excitation of DA neurons per se did not result in a nonspecific decrease in freezing levels and that the temporally precise excitation during the US omission is necessary for the observed behavioral effect . Furthermore , there was no difference between the groups in their fear acquisition on day 1 ( two-way repeated measures ANOVA , no main effect of group , F2 , 54 = 0 . 17 , p = 0 . 84 and no group × trial interaction , F6 , 54 = 0 . 65 , p = 0 . 68 , Figure 6G ) and all groups showed comparable levels of freezing during the first CS of extinction ( one-way ANOVA , F2 , 18 = 0 . 81 , p = 0 . 66 , Figure 6H ) before light stimulation began . Thus , excitation of DA neurons precisely at the time of the unexpected US omission is sufficient to accelerate fear extinction learning . Finally , Paired-ChR2 mice spent less time freezing to the CS compared to the control groups during the extinction recall test ( two-way repeated measures ANOVA; main effect of group: F2 , 162 = 5 . 5 , p = 0 . 013 and group × trial interaction: F18 , 162 = 2 . 52 , p = 0 . 0011 , Figure 6G ) suggesting that the accelerated extinction learning resulted in a stronger extinction memory . An alternative possibility is that the low level of freezing in the Paired-ChR2 group during Ext Rec was due to an effect of optical stimulation of DA neurons on the fear memory . For instance , the optogenetic manipulation could result in the erasure of the fear memory by impairing the memory reconsolidation process ( Nader , 2015 ) rather than accelerating extinction learning and strengthening extinction memory . To rule this possibility out , we tested the animals on a fear renewal test on day 4 by presenting 5 CSs in the conditioning context . It is well established that extinction learning is context-dependent such that if the animals are tested in a different context than the one they are extinguished in , fear responses return , a phenomenon called fear renewal ( Bouton , 2004 ) . Therefore , if low level of freezing during Ext Rec was due to impaired reconsolidation then we would expect to see impaired fear renewal in Paired-ChR2 group ( Duvarci and Nader , 2004 ) . However , we found that all groups showed high freezing to the CS during the renewal test and that there was no difference in the freezing levels between the groups ( Figure 6—figure supplement 3; two-way repeated measures ANOVA , no significant effect of group , F2 , 72 = 0 . 33 , p = 0 . 72 and group × trial interaction , F8 , 72 = 0 . 86 , p = 0 . 55 ) suggesting that all three groups showed comparable levels of fear renewal . This suggests that the low level of freezing in Paired-ChR2 group during extinction recall was not due to an affect of our optogenetic manipulation on the fear memory but rather was due to enhanced extinction learning and memory formation . Taken together , these findings demonstrate that increasing DA neuron activity at the time of US omission — and thus enhancing an endogenous extinction mechanism — is sufficient to accelerate extinction learning and strengthen extinction memory . Here we demonstrated that DA neurons were activated by the omission of the aversive US during fear extinction , specifically during the beginning of extinction when the US omission is most unexpected . Importantly , the magnitude of this DA signal correlated with the strength of extinction learning . Furthermore , temporally specific optogenetic inhibition of DA neurons at the time of the US omission prevented extinction , demonstrating that this signal is necessary for normal fear extinction . Conversely , enhancing this DA signal using temporally-specific optogenetic excitation was sufficient to accelerate extinction learning . Together , these results identify a crucial role of DA neurons in signaling the unexpected omission of aversive outcomes and thereby driving fear extinction learning . Previous studies have shown that DA neurons encode a reward prediction error , or the discrepancy between expected and actual rewards , which acts as a teaching signal for reinforcement learning ( Bayer and Glimcher , 2005; Eshel et al . , 2015; Eshel et al . , 2016; Schultz et al . , 1997; Steinberg et al . , 2013 ) . Specifically , presentation of unexpected or better than expected rewards induces increased firing in DA neurons ( Bayer and Glimcher , 2005; Eshel et al . , 2016; Roesch et al . , 2007; Schultz et al . , 1997 ) . Our results suggest that DA neurons might also signal a better than expected outcome during fear extinction . We found that a subpopulation of putative DA neurons in the VTA increased their firing selectively at the time of the US omission during fear extinction . This US omission-responsive firing was observed specifically during the early trials of extinction when the absence of the US was unexpected , and was significantly reduced during the late stages of extinction when the US omission was no longer unexpected and animals showed significant extinction of fear responses . Importantly , these responses were not observed in putative non-DA neurons . This suggests that DA neurons encode a prediction error-like signal during fear extinction learning . Ca+2 recordings selectively in DA neurons further confirmed these results and revealed that this DA signal correlated with the strength of extinction learning . Recent studies have shown that DA neurons not only encode reward prediction errors but also signal prediction errors to gate fear learning ( Groessl et al . , 2018 ) and to drive threat avoidance ( Menegas et al . , 2018 ) . Interestingly , omission of aversive stimuli during fear extinction in fruit flies is encoded by the DA system that mediates reward , but not aversive , learning ( Felsenberg et al . , 2018 ) . Whether the DA signal during fear extinction in mammals is also similar to prediction error signals for reward and mediated by the brain’s reward circuitry ( Wise , 2002 ) will be important questions for future studies . Consistent with our results , previous studies using partial reinforcement paradigms have shown that DA neurons exhibit increased firing to the unexpected omission of aversive stimuli ( Matsumoto and Hikosaka , 2009; Matsumoto et al . , 2016; but see Tian and Uchida , 2015 ) . However , responses to aversive US omission in these studies were much smaller compared to what we observed . There are several differences between these studies and ours that might account for this . The aversive US used in our study is a painful footshock whereas an air-puff was used in previous studies ( Matsumoto and Hikosaka , 2009; Matsumoto et al . , 2016 ) . Furthermore , it has been shown that the valence of the testing context influences DA neuron responses to the omission of the aversive US . Matsumoto and colleagues ( 2016 ) found a significant increase in DA neuron responses to the omission of aversive airpuff in a low reward , but not high reward , context . In our study , the animals were only fear conditioned and no reward learning happened prior to or during fear conditioning . Therefore , the context was exclusively aversive . In addition , the CS-US contingency during fear conditioning was higher in our study than in partial reinforcement tasks used in previous studies . The CS predicted the shock with 100% probability at the end of fear conditioning and therefore , the omission of the US was fully unexpected at the beginning of fear extinction in our study . On the other hand , in the studies using partial reinforcement tasks the CS-US contingency was 25–90% and the omission of the US was performed intermittently ( Matsumoto and Hikosaka , 2009; Matsumoto et al . , 2016; Tian and Uchida , 2015 ) . The US omission was therefore arguably less unexpected . Together , these factors might therefore account for the differences in DA neuron responses between these studies and ours . Overall , our findings suggest that DA neurons not only signal better than expected rewards ( Bayer and Glimcher , 2005; Eshel et al . , 2016; Roesch et al . , 2007; Schultz et al . , 1997 ) but also better than expected outcomes more generally , such as the omission of an aversive event . Detecting the discrepancy between expected and actual outcomes is critical for new learning ( Rescorla and Wagner , 1972 ) . During fear extinction , the omission of the aversive US is an unexpected outcome which initiates new learning about the CS , specifically that it no longer predicts danger . However , how this learning is initiated at the neuronal level has remained unknown . Here , by using bidirectional optogenetic manipulations , we demonstrated that the DA signal during the omission of the aversive US drives normal fear extinction learning: inhibiting this DA signal prevented , while enhancing it accelerated , normal extinction learning . Our results are consistent with previous findings establishing the causal role of DA neurons in reinforcement learning ( Chang et al . , 2016; Steinberg et al . , 2013; Tsai et al . , 2009 ) and further extend their role to safety learning . Consistent with our findings , a recent study inhibited VTA DA neurons during US omission in rats and also found reduced fear extinction learning ( Luo et al . , 2018 ) . Our study replicates this finding in mice and further extends it by showing that enhancing DA neuron activity at the time of the US omission is also sufficient to accelerate normal extinction learning . Consistent with our results , it has been shown that enhancement of DA signaling by L-DOPA administration during extinction was sufficient to initiate fear extinction learning in a mouse model of impaired extinction learning ( Whittle et al . , 2016 ) . Our results therefore suggest that enhancement of DA signaling during extinction could be a potential strategy for the treatment of anxiety disorders . Fear extinction is mediated by a network of brain structures consisting mainly of the amygdala and the infra-limbic ( IL ) subregion of the medial prefrontal cortex ( Duvarci and Pare , 2014; Maren et al . , 2013; Pape and Pare , 2010; Sotres-Bayon and Quirk , 2010; Tovote et al . , 2015 ) . Although neuronal activation has been observed in these structures during different stages of fear extinction , they occur during the CS and later than the DA signal we observed in our study . In the basolateral amygdala , a subpopulation of neurons termed 'extinction neurons' increases their firing to the CS during extinction learning ( Amano et al . , 2011; Herry et al . , 2008 ) . However , these neurons become CS responsive late in the extinction session right before the animals show a decrease in fear responses ( Herry et al . , 2008 ) suggesting that these neurons likely mediate inhibition of fear responses . In the IL , a structure necessary for consolidation of extinction memories ( Sotres-Bayon and Quirk , 2010 ) , increased firing to the CS is observed during extinction recall ( Milad and Quirk , 2002 ) . In contrast , the DA signal that we demonstrated in our study occurs at the early trials of the extinction session , supporting our conclusion that this signal initiates extinction learning . Plasticity in the amygdala and IL underlie acquisition and consolidation of fear extinction memories ( Duvarci and Pare , 2014; Maren et al . , 2013; Pape and Pare , 2010; Sotres-Bayon and Quirk , 2010; Tovote et al . , 2015 ) . Furthermore , DA signaling in the amygdala and IL has been shown to play an important role in fear extinction ( Abraham et al . , 2014; Haaker et al . , 2013; Hikind and Maroun , 2008; Mueller et al . , 2010; Shi et al . , 2017; Whittle et al . , 2016 ) . Interestingly , optogenetic inhibition of DA neurons during US omission has been found to prevent extinction-related plasticity in the amygdala and IL ( Luo et al . , 2018 ) . However , it is unclear how this DA signal induces plasticity in the amygdala and IL to underlie extinction memory . This can be mediated through direct DA projections to these structures or indirectly through a multi-synaptic circuit mechanism . The first step in addressing this issue will be to identify the projection target of these DA neurons that signal the omission of the US during fear extinction . Recent studies have shown that midbrain DA neurons form functionally distinct and mostly non-overlapping subpopulations based on their projection targets ( Beier et al . , 2015; Lammel et al . , 2008; Lammel et al . , 2011; Lerner et al . , 2015; Lynd-Balta and Haber , 1994; Menegas et al . , 2015; Menegas et al . , 2017; Parker et al . , 2016; Roeper , 2013 ) . Therefore , an important question is which subpopulation of VTA DA neurons generates the response to the US omission . One possible candidate is the subpopulation projecting to the NAc . Supporting this possibility , an increase in DA release around the time of the CS offset during fear extinction has been observed in the NAc ( Badrinarayan et al . , 2012 ) . In particular , this was observed only during the early trials of extinction , consistent with our results . DA signaling in the NAc has also been shown to be important for relief learning ( Mayer et al . , 2018 ) and avoidance behavior ( Gentry et al . , 2016; Oleson et al . , 2012 ) . Whether fear extinction learning and relief and avoidance learning share related mechanisms and involve overlapping subpopulations of DA neurons projecting to NAc is not known and will be an important question for future research . Notably , pharmacological blockade of DA receptors in the NAc have been found to impair fear extinction learning ( Holtzman-Assif et al . , 2010 ) . Furthermore , we have observed that fear extinction learning in humans is accompanied by a prediction error-like activation in the ventral striatum ( Raczka et al . , 2011 ) . However , at odds with these findings , inhibition of DA terminals in NAc or DA neurons projecting to the medial shell of NAc during US omission did not affect fear extinction learning , although it did impair consolidation of extinction memory , in a recent study ( Luo et al . , 2018 ) . Our single unit results demonstrate that a small subpopulation of DA neurons mediate this DA signal to drive extinction learning . Therefore , it is possible that this subpopulation of DA neurons projects to a specific subregion of NAc that was not targeted by Luo et al . ( 2018 ) . In addition to NAc , other possible candidates include the DA neurons that project to the amygdala and/or IL . Identifying which projection-defined subpopulation of DA neurons signals the omission of the US to initiate fear extinction learning will be an important question for future research . In conclusion , our study identifies a prediction error-like signal encoded by DA neurons that is necessary to initiate fear extinction learning . Furthermore , we found that enhancing this DA signal is sufficient to accelerate extinction learning and strengthen extinction memory consolidation . Deficits in fear extinction learning are thought to underlie anxiety disorders ( Craske et al . , 2017; Graham and Milad , 2011; Mahan and Ressler , 2012; Milad and Quirk , 2012; Pitman et al . , 2012 ) . Our study therefore has therapeutic implications for anxiety disorders by identifying DA neuron activity as a potential target for novel treatments . All procedures were conducted in accordance with the guidelines of the German Animal Protection Act and were approved by the local authorities ( Regierungsprasidium Darmstadt; protocol number 1038 ) . Male C57BL/6N mice ( Charles River ) , aged 3 months at the start of experiments , were used in the in vivo electrophysiology experiment . Male heterozygous DAT-Cre mice ( Zhuang et al . , 2005; backcrossed with C57BL/6N ) aged 3–6 months at the start of experiments were used in the photometry and optogenetics experiments . All experimental groups were matched for age . All mice were individually housed on a 12 hr light/dark cycle . All experiments were performed during the light cycle . AAV5-EF1a-DIO-hChR2 ( H134R ) -eYFP , AAV5-EF1a-DIO-eNpHR3 . 0-eYFP , AAV5-EF1a-DIO- eYFP and AAV5-CAG-Flex-GFP were produced and packaged by the University of North Carolina Vector Core . AAV5-CAG-Flex-GCaMP6f-WPRE-SV40 and AAV5-CAG-Flex- GCaMP6s-WPRE-SV40 were produced and packaged by the University of Pennsylvania Vector Core . Animals were anesthetized using isoflurane ( 1–2% ) and placed in a stereotaxic frame . At the onset of anesthesia , all animals received subcutaneous injections of carprofen ( 4 mg/kg ) and dexamethasone ( 2 mg/kg ) . The animal’s temperature was maintained for the duration of the surgical procedure using a heating blanket . Anesthesia levels were monitored throughout the surgery and the concentration of isoflurane adjusted so that the breathing rate never fell below 1 Hz . After exposing the skull surface , craniotomies were made overlying the VTA ( 3 . 2 mm posterior to bregma and 0 . 5 mm lateral to the midline ) . For in vivo single-unit recordings , we used a moveable bundle of 5–8 stereotrodes made by twisting together two 0 . 0005 inch tungsten wires ( M219350 , California Fine Wire ) . The stereotrode bundle was attached to a custom-made microdrive that made it possible to advance the electrodes along the dorsoventral axis . On the day of implantation , the stereotrodes were gold-plated to reduce the impedance to 0 . 2–0 . 3 MΩ at 1 kHz . The stereotrode bundle was inserted through the craniotomy above the VTA to a depth of 3 . 9–4 . 0 mm below bregma . All electrode wires were connected to an electrode interface board ( EIB-16; Neuralynx ) for relaying electrophysiological signals to the data acquisition system . The microdrive was anchored to the skull using skull screws and dental cement ( Paladur ) . For the photometry experiments , DAT-cre mice were injected unilaterally with 1 μl of AAV5-CAG-Flex-GCaMP6f-WPRE-SV40 ( final titer 2 . 7 × 1012 pp per ml ) or AAV5-CAG-Flex- GCaMP6s-WPRE-SV40 ( final titer 6 . 4 × 1012 pp per ml ) or AAV5-CAG-Flex-GFP ( final titer 4 . 5 × 1012 pp per ml ) . Viruses were injected in the VTA ( 3 . 2 mm posterior to bregma , 0 . 5 mm lateral to the midline and 4 . 5 mm ventral to bregma ) at 50 nl/min using a 10 μl syringe with a 33-gauge needle controlled by an injection pump . The needle was left in place for an additional 10–15 min before slowly being withdrawn . Following infusion of the virus , an optical fiber ( 400 μm core diameter , 0 . 48 NA , Doric Lenses ) was slowly inserted through the same craniotomy into the VTA at a depth of 4 . 1–4 . 3 mm below bregma . The optical fiber was then anchored to the skull using skull screws and dental cement ( Paladur ) . For optogenetic experiments , DAT-cre mice were injected bilaterally in the VTA with 1 μl of AAV5-EF1a-DIO-hChR2 ( H134R ) -eYFP ( final titer 4 . 3 × 1012 pp per ml ) , AAV5-EF1a-DIO- eNpHR3 . 0-eYFP ( final titer 4 × 1012 pp per ml ) or AAV5-EF1a-DIO-eYFP ( final titer 4 . 4 × 1012 pp per ml ) per hemisphere using the coordinates described above . Optical fibers ( 200 μm core diameter , 0 . 22 NA , Thorlabs ) were implanted bilaterally above the VTA to a depth of 3 . 9–4 . 0 mm below bregma as described above . Fear conditioning and extinction took place in two different contexts ( A and B ) . Context A consisted of a square chamber with an electrical grid floor ( Med Associates ) used to deliver footshock . Context B consisted of a white teflon cylindrical chamber with bedding material on the floor . The chambers were located inside a sound attenuating box and were cleaned with 50% ethanol and 1% acetic acid before and after each session . All mice were habituated to contexts A and B for 10–15 min each in a counterbalanced fashion . For electrophysiology and photometry experiments , mice received a tone habituation session followed by fear conditioning on day 1 . Tone habituation started following a 2 min baseline period in context A and consisted of 10 presentations of the CS ( 4 kHz tone , 75 dB , 10 s ) with a random intertrial interval ( ITI ) of 40–120 s . Fear conditioning consisted of five pairings of the CS with a US ( 1 s footshock , 0 . 55 mA , ITI: 40–120 s ) . The onset of the US coincided with the offset of the CS . On day 2 , mice received an extinction session consisting of 25–30 and 30 presentations of the CS alone in context B in the electrophysiology and photometry experiments , respectively . For optogenetic experiments , on day 1 mice underwent fear conditioning consisting of four pairings of a CS ( 4 kHz tone , 75 dB , 30 s ) with a US ( 1 s footshock ) with a random ITI of 40–120 s in context A . The US intensity was 0 . 45 mA and 0 . 5 mA in the optogenetic inhibition and excitation experiments , respectively . On day 2 , mice received an extinction session consisting of 25 presentations of the CS alone in context B . Twenty-four hours later , mice received an extinction recall test in context B . The extinction recall test consisted of 25 and 10 presentations of the CS alone in the optogenetic inhibition and excitation experiments , respectively . The behavior of mice was recorded to video and scored by an experienced observer blind to the experimental condition . Behavioral freezing , defined as the absence of all bodily movements except breathing-related movement ( Blanchard and Blanchard , 1972 ) , was used as the measure of fear . Animals that showed low conditioned fear ( <50% freezing ) at the start of the extinction session ( First CS of extinction ) were excluded from the study . This criterion led to the exclusion of 1 mouse from photometry and one mouse from optogenetics experiments . Following one week of recovery from surgery , animals were habituated to handling and moving with the wire tether connecting the microdrive to the recording system . Before fear conditioning began , stereotrodes were slowly advanced until single-units with low baseline firing rate ( <10 Hz ) were observed , to increase the probability of recording putative DA neurons during the task . The spike waveforms tended to change from day 1 ( tone habituation and conditioning ) to day 2 ( extinction ) even when electrodes were not moved or in case cells were lost overnight , we advanced ( 40–80 μm ) the microdrive to find new cells on day 2 . We therefore treated the cell populations recorded on these two days as independent . On each day of the fear conditioning task , single-units were first recorded for 5–10 min while animals were in their homecage to assess their baseline firing rates . Neural data were acquired using a 16-channel headstage ( HS-18 , Neuralynx ) that was connected to the electrode interface board and relayed the signals to a Digital Lynx SX ( Neuralynx ) data acquisition system . To extract putative spikes , neural signals were bandpass filtered between 600 and 6000 Hz , and waveforms that passed the threshold ( 50–60 μV ) were digitized at 30 kHz . One stereotrode channel that did not have any apparent units was used as reference . We also recorded the neural signals bandpass filtered between 1 and 6000 Hz to analyze the spike waveforms . In order to verify recording locations , current ( 50 mA , 10 s ) was passed through one of the stereotrode channels to produce a lesion in the recording site at the end of the experiment . Spike waveforms were sorted offline into single-unit clusters using SpikeSort3D ( Neuralynx ) . Only well-isolated single-units that displayed a clear refractory period ( >1 ms ) were included in subsequent analysis , performed using scripts custom-written in MATLAB ( MathWorks ) . The VTA contains different cell types , including DA neurons and gamma-aminobutyric acid ( GABA ) neurons ( Morales and Margolis , 2017 ) . Midbrain DA neurons typically exhibit baseline firing rates below 10 Hz in awake animals ( Jin and Costa , 2010; Li et al . , 2012 ) and long duration action potentials ( Grace and Bunney , 1980; Jin and Costa , 2010; Li et al . , 2012 ) . We therefore classified neurons whose baseline firing rate ( measured in the home cage ) were below 10 Hz and exhibited long duration action potentials ( peak-to-peak duration >450 µs ) as putative DA neurons ( Figure 1—figure supplement 3 ) . Spike waveforms were analyzed using the neural signals bandpass filtered between 1 and 6000 Hz . To analyze firing rates during the task , we constructed peri-stimulus time histograms ( PSTHs ) aligned to the onset of the CS ( −5 s to +15 s ) using 1 s bins . PSTHs were calculated separately for early extinction trials ( E-Ext: average of first 10 CSs ) , late extinction trials ( L-Ext: average of last 10 CSs ) and tone habituation ( Hab: average of 10 CSs ) . These PSTHs were then normalized with a z-score transformation by subtracting the baseline firing rate ( 5 s pre-CS period ) from each individual 1 s bin and then dividing this difference by the standard deviation of the baseline . A neuron was classified as US omission excited if it met two criteria: 1 ) z-score was greater than 2 at the time of the US omission ( during the 1 s bin following the offset of the CS ) ; 2 ) the z-score had to increase by at least two from the last bin of the CS to the US omission bin to ensure that the increase in firing was specific to the omission of the US and not a result of sustained increase in firing to the CS . Conversely , a neuron was classified as US omission inhibited if ( 1 ) z-score was smaller than −2 at the time of the US omission ( during the 1 s bin following the offset of the CS ) ; ( 2 ) the z-score had to decrease by at least two from the last bin of the CS to the US omission bin to ensure that the decrease in firing was specific to the omission of the US and not a result of sustained decrease in firing to the CS . To quantify responses to the CS , we averaged the z-scores during the entire CS period ( 10 s ) for Hab , E-Ext and L-Ext trials and obtained an average z-score for each neuron . The neurons were classified as excited or inhibited by the CS if they showed average z-scores greater than two or smaller than −2 , respectively . Animals were injected with viral vectors and implanted with an optical fiber in the VTA , as described above . After a waiting period of 3–4 weeks to allow for surgical recovery and virus expression , animals were connected to a 400 μm patch cord ( Doric Lenses ) . Fluorescence was measured by delivering 465 nm excitation light through the patch cord and separating the emission light at 525 nm with a beamsplitter ( Fluorescence MiniCube FMC5 , Doric Lenses ) . The emission light was collected using a Femtowatt Silicon Photoreceiver ( Model # 2151 , Newport ) . The voltage output of the photoreceiver was then digitized at 2 kHz ( Digital Lynx SX , Neuralynx ) . Using the same approach we also measured the isosbestic ( activity-independent ) fluorescence of gCaMP6 at 430 nm using 405 nm excitation light . After animals were habituated to handling and being connected to the patch cord , they underwent the fear conditioning protocol . At the end of fear conditioning experiments , animals were tested on an operant conditioning task ( Figure 3—figure supplement 2 ) . To this end , animals were placed in an operant chamber containing a water delivery port . Each nosepoke into the delivery port that followed the previous nosepoke by a variable inter-trial interval ( 3–5 s ) triggered the delivery of liquid reward ( 10% sucrose solution ) with a 50% probability . This task was used to verify that reward delivery caused large increases in fluorescence in gCaMP6-expressing mice , as expected based on previous electrophysiology ( Bayer and Glimcher , 2005; Eshel et al . , 2015; Eshel et al . , 2016; Roesch et al . , 2007; Schultz et al . , 1997 ) and fiber photometry ( Menegas et al . , 2017; Parker et al . , 2016; Soares et al . , 2016 ) studies , suggesting that this Ca+2 signal is indeed generated by DA neuron activity . The voltage output of the photoreceiver , representing fluctuations in fluorescence , was first low-pass filtered at 4 Hz and then downsampled to 10 Hz . The change in fluorescence evoked by the CS ( dF/F ) was then calculated by subtracting from each trace the baseline fluorescence ( average during the 5 s before CS onset ) and dividing it by the baseline fluorescence . dF/F traces were then averaged separately for each animal for early extinction trials ( E-Ext: average of first 10 CSs ) , late extinction trials ( L-Ext: average of last 10 CSs ) and tone habituation ( Hab: average of 10 CSs ) . To examine responses to US omission we further averaged dF/F values in the 5 s following CS offset . Similar results were obtained from animals expressing gCaMP6f ( n = 5 ) and gCaMP6s ( n = 5 ) , therefore the data was pooled . To quantify responses to the CS , we averaged the dF/F values during the CS for each animal for Hab , E-Ext and L-Ext . To quantify responses to reward ( Figure 3—figure supplement 2 ) , average dF/F was calculated separately for rewarded and unrewarded nosepokes using the baseline fluorescence 3 s before noseport entry . For bilateral optogenetic manipulations during behavior , the implanted optical fibers ( 200 μm core diameter , 0 . 22 NA , Thorlabs ) were connected to 200 μm patch cords ( Thorlabs ) with zirconia sleeves and the patch cords were connected to a light splitting rotary joint ( FRJ 1 × 2 i , Doric Lenses ) that was connected to a laser with a 200 μm patch cord ( Thorlabs ) . For mice expressing the light-activated inhibitory opsin halorhodopsin ( NpHR-eYFP ) and their eYFP controls , yellow light was delivered from a DPSS 594 nm laser ( Omicron ) . Laser power at the tip of the optic fiber was 10–15 mW . To inhibit DA neurons around the time of the US omission during the extinction session on day 2 , the laser was turned on from 1 s before to 2 s after the CS offset . The laser was then turned off gradually using a 1 s ramp to avoid rebound excitation ( Mahn et al . , 2016; Figure 5—figure supplement 2 ) . For mice in the Unpaired-NpHR control group , the light was delivered the same way during the ITIs . For mice expressing the light activated excitatory opsin channelrhodopsin-2 ( ChR2 ) and their eYFP controls , blue light pulses were delivered from a 473 nm laser ( LuXx473 , Omicron ) . The laser power at the tip of the optic fiber was 5–10 mW . To phasically activate DA neurons at the time of the US omission , 5 ms light pulses were delivered at 20 Hz for 1 s beginning at the offset of the CS . The mice in the Unpaired-ChR2 group received delivery of the same light pulses during the ITIs . Using the surgical procedures described above DAT-cre mice were injected in the VTA with 1 μl of AAV5-EF1a-DIO-hChR2 ( H134R ) -eYFP ( final titer 4 . 3 × 1012 pp per ml ) or AAV5-EF1a-DIO- eNpHR3 . 0-eYFP ( final titer 4 × 1012 pp per ml ) per hemisphere using the coordinates described above . Two to three weeks after virus injection , animals were anesthetized and placed in a stereotaxic frame with the skull exposed . A stainless-steel head post ( Luigs and Neumann ) was then cemented to the exposed skull . The area of the skull overlying VTA was left free of cement but covered with a silicon elastomer ( Kwik-Sil , World Precision Instruments ) . Skull screws were inserted over the frontal cortex and cerebellum to serve as reference and ground , respectively , and to provide anchoring support for the cement . Following recovery from surgery , animals were handled and habituated to being head-fixed , by inserting the head post into a matching head post holder ( Luigs and Neumann ) . Following 2–3 days of habituation to being head-fixed , animals underwent another surgery to prepare a craniotomy over the VTA . Animals were anesthetized with isoflurane and placed in a stereotaxic frame . The Kwik-Sil was removed from the skull and a small craniotomy was then made in the skull over the VTA and then sealed with Kwik-Sil . The following day , the animals were head-fixed , the Kwik-Sil removed , and a 32-channel optrode ( silicon probe with an optical fiber; A1 × 32-Edge-5mm-20–177-OA32 , NeuroNexus ) was lowered into the VTA using a micromanipulator ( SM-8 , Luigs and Neumann ) . The optrode was then advanced to a depth of ~4700 μm below the brain surface to span the dorsal-ventral extend of the VTA . Following final placement of the optrode and a brief waiting period ( ∼15 min ) , neural activity was recorded while laser pulses ( yellow light: 10–15 mW , blue light: 5–10 mW ) were delivered through a patch cord ( Neuronexus ) connected to the optic fiber . Electrophysiological signals were filtered between 1 and 6000 Hz , digitized at 30 kHz using a digitizing headstage ( RHD2132 Amplifier Board , Intan Technologies ) , and acquired using a USB interface board ( RHD2000 , Intan Technologies ) . Once the recording was over , optrode was removed from the brain and then was penetrated to another location in the VTA . Multiple penetrations and recordings were performed during a session . Before the final penetration , the silicone probe was coated with a fluorescent dye ( DiI , Life Technologies ) to assist with the identification of the recording location . At the end of the experiments , mice were deeply anesthetized with sodium pentobarbital and were transcardially perfused with 4% paraformaldehyde and 15% picric acid in phosphate-buffered saline ( PBS ) . Brains were removed , post-fixed overnight and coronal brain slices ( 60 µm ) were sectioned using a vibratome ( VT1000S , Leica ) . Standard immunohistochemical procedures were performed on free-floating brain slices . Briefly , sections were rinsed with PBS and then incubated in a blocking solution ( 10% horse serum , 0 . 5% Triton X-100% and 0 . 2% BSA in PBS ) for 1 hr at room temperature . Slices were then incubated in a carrier solution ( 1% horse serum , 0 . 5% Triton X-100% and 0 . 2% BSA in PBS ) containing the primary antibody overnight at room temperature . The next day , the sections were washed in PBS and then incubated in the same carrier solution containing the secondary antibody overnight at room temperature . The following primary antibodies were used: polyclonal rabbit anti-tyrosine hydroxylase ( TH , catalog # 657012 , 1:1000 , Calbiochem ) , monoclonal mouse anti-TH ( catalog # MAB318 , 1:1000 , Millipore ) , polyclonal rabbit anti-GFP ( catalog # A11122 , 1:1000 , Life Technology ) . The following secondary antibodies were used: Alexa Fluor 568 goat anti-rabbit ( catalog # A11011 , 1:1000 , Thermo Fisher Scientific , Invitrogen ) , Alexa Fluor 568 goat anti-mouse ( catalog # A11004 , Thermo Fisher Scientific , Invitrogen ) , and Alexa Fluor 488 goat anti-rabbit ( catalog # A11008 , 1:1000 , Thermo Fisher Scientific , Invitrogen ) . Finally , sections were washed with PBS , mounted on slides and coverslipped with a 4’ , 6-diamidin-2-phenylindol ( DAPI ) containing medium ( VECTASHIELD , Vector Laboratories ) or incubated for 10 min in 0 . 1 M PBS containing 0 . 02% DAPI ( catalog # D1306 , Molecular Probes , Invitrogen ) , washed for 10 min in PBS , mounted on slides and coverslipped . Animals with incorrect electrode/optical fiber placements were excluded from data analysis . A total of four mice were excluded from the study due to incorrect placement of electrodes/optical fibers . We excluded one animal from the electrophysiology experiment because the electrodes were placed outside ( posterior to ) the VTA . Three animals were excluded from the fiber photometry experiment . In one animal the optical fiber was placed too dorsal and in the other two animals too ventral in the VTA . In these animals , the neuronal activity dependent fluorescence signal was undetectable or too low . No animals were excluded from optogenetic experiments . Data were statistically analyzed using GraphPad Prism ( GraphPad Software ) and MATLAB ( Mathworks ) . All statistical tests were two-tailed and had an α level of 0 . 05 . All error bars show s . e . m . All ANOVAs were followed by Bonferroni post hoc tests if significant main or interaction effects were detected . No statistical methods were used to predetermine sample size , but our sample sizes were similar to those generally used in the fear conditioning field . Animals were randomly assigned to experimental groups before the start of each experiment after ensuring that all experimental groups were matched for age . All results were obtained using groups of mice that were run in several cohorts .	To survive , animals must identify and react to stimuli in their environment that signal danger . But they must also adapt their behavior when those stimuli no longer signal danger – hiding whenever you hear a loud noise might keep you safe , but it also prevents you from searching for food . In the laboratory , we can study this form of learning using procedures called fear conditioning and extinction . During fear conditioning , animals learn that a stimulus , such as a tone , signals that an unpleasant event is about to occur . That event might involve receiving a mild shock to the foot , for example . After experiencing the tone and shock paired together multiple times , animals will initially show signs of fear – such as freezing – when they hear the tone . But if later the tone occurs without being followed by the shock , these fear responses fade . This fading process is called extinction . Extinction does not involve erasing the old memory about the tone-shock relationship . That is , it is not a form of forgetting . Instead , the animals learn that the tone no longer signals an impending shock . By monitoring brain activity in mice trained to associate a shock with a tone , Salinas-Hernández et al . reveal how the brain begins to learn that the shock no longer follows the tone . When the mice do not receive the anticipated shock to the foot , a group of brain cells that produce the chemical dopamine increase their activity . These neurons also fire whenever animals receive a reward , particularly one that exceeds their expectations . The more the dopamine neurons fire , the faster the mice reduce their fear responses to the tone . Preventing the neurons from increasing their activity prevents the mice from extinguishing their fear memory . By contrast , activating the neurons speeds up the extinction process . Understanding how the brain extinguishes learned fear responses has therapeutic implications . Many anxiety disorders , such as post-traumatic stress disorder , involve impaired fear extinction learning . Indeed , exposure therapy – used to treat anxiety disorders such as phobias – is a form of fear extinction . Manipulating the activity of dopamine neurons during extinction could therefore help to treat anxiety disorders .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Dopamine neurons drive fear extinction learning by signaling the omission of expected aversive outcomes
Although numerous long noncoding RNAs ( lncRNAs ) have been identified , our understanding of their roles in mammalian physiology remains limited . Here , we investigated the physiologic function of the conserved lncRNA Norad in vivo . Deletion of Norad in mice results in genomic instability and mitochondrial dysfunction , leading to a dramatic multi-system degenerative phenotype resembling premature aging . Loss of tissue homeostasis in Norad-deficient animals is attributable to augmented activity of PUMILIO proteins , which act as post-transcriptional repressors of target mRNAs to which they bind . Norad is the preferred RNA target of PUMILIO2 ( PUM2 ) in mouse tissues and , upon loss of Norad , PUM2 hyperactively represses key genes required for mitosis and mitochondrial function . Accordingly , enforced Pum2 expression fully phenocopies Norad deletion , resulting in rapid-onset aging-associated phenotypes . These findings provide new insights and open new lines of investigation into the roles of noncoding RNAs and RNA binding proteins in normal physiology and aging . Long noncoding RNAs ( lncRNAs ) comprise a heterogeneous class of transcripts that are defined by a sequence length greater than 200 nucleotides and the lack of a translated open-reading frame ( ORF ) . lncRNAs have been proposed to perform a variety of cellular functions including regulation of gene expression in cis and trans , modulation of functions of RNAs and proteins to which they bind , and organization of nuclear architecture ( Batista and Chang , 2013 ) . Although they have been estimated to number in the tens of thousands ( Harrow et al . , 2012 ) , the biological significance of the vast majority of lncRNAs remains to be established . This is due , in part , to the generally low abundance and poor evolutionary conservation of most lncRNAs , which has limited our ability to interrogate their biochemical functions as well as their biologic roles in vivo using model organisms . Moreover , while genetic studies in mice have uncovered important functions for selected mammalian lncRNA loci in development and disease states ( Arun et al . , 2016; Sauvageau et al . , 2013 ) , it has often been challenging to connect specific lncRNA-driven phenotypes to defined RNA-mediated functions . As a result , our broad understanding of how the molecular pathways controlled by lncRNAs impact development and physiology remains limited . Noncoding RNA activated by DNA damage ( NORAD ) is a recently described lncRNA that is distinguished from the majority of transcripts in this class due to its high abundance in mammalian cells and strong evolutionary conservation across mammalian species ( Lee et al . , 2016; Tichon et al . , 2016 ) . Studies in human cells have established that this RNA functions as a strong negative regulator of PUMILIO1 ( PUM1 ) and PUMILIO2 ( PUM2 ) , RNA binding proteins ( RBPs ) that belong to the deeply conserved family of Pumilio and Fem3 binding factor ( PUF ) proteins . PUM1/2 bind specifically to the eight nucleotide ( nt ) PUMILIO response element ( PRE ) ( UGUANAUA ) , which is often located in the 3’ UTR of mRNAs . Binding of PUM1/2 to these sites triggers accelerated deadenylation , reduced translation , and turnover of mRNA targets ( Miller and Olivas , 2011; Quenault et al . , 2011 ) . With the capacity to bind a large fraction of PUM1/2 within the cell , NORAD limits the availability of these proteins to repress target mRNAs ( Lee et al . , 2016; Tichon et al . , 2016 ) . Consequently , inactivation of NORAD results in PUMILIO hyperactivity with augmented repression of a program of target mRNAs that includes key regulators of mitosis , DNA repair , and DNA replication . Dysregulation of these genes results in dramatic genomic instability in NORAD-deficient cells ( Lee et al . , 2016 ) . In accordance with this model , PUM2 overexpression is sufficient to phenocopy , while PUM1/2 loss-of-function is sufficient to suppress , the NORAD knockout phenotype in human cells . Recent work has identified additional RNA-binding proteins that interact with NORAD including SAM68 , which facilitates PUMILIO antagonism by this lncRNA ( Tichon et al . , 2018 ) , and RBMX , an RNA binding protein that contributes to the DNA damage response ( Munschauer et al . , 2018 ) . While it has not yet been demonstrated that NORAD:RBMX interaction is essential to maintain genomic stability , this intriguing interaction raises the possibility of additional functions for NORAD beyond regulation of PUMILIO activity . Although PUF proteins are deeply conserved across eukaryotic species , the emergence of NORAD specifically within mammals suggests the existence of strong selective pressure to maintain tight control of PUMILIO activity within this lineage . In mice , PUM1 and PUM2 loss-of-function has been linked to behavioral abnormalities , elevated neuronal excitability , and impaired neurogenesis , while inactivation of PUM1 reduces fertility in males and females ( Goldstrohm et al . , 2018 ) . Interestingly , mammalian neurons are exquisitely sensitive to reduced dosage of PUMILIO , with only a 25% to 50% reduction in PUM1 expression resulting in neurodegeneration in both human and mouse brain ( Gennarino et al . , 2018; Gennarino et al . , 2015 ) . The existence of NORAD suggests that hyperactivity of PUMILIO , expected to occur in the absence of this lncRNA , may also result in deleterious effects . While studies in cell lines have demonstrated that NORAD loss or PUMILIO overexpression results in genomic instability , the consequences of mammalian PUMILIO hyperactivity in vivo have yet to be examined . Here , we report an investigation of the physiologic role of the Norad-PUMILIO axis through the generation and characterization of Norad-deficient and Pum2 transgenic mouse lines . While deletion of Norad does not overtly impact development , mice lacking this lncRNA develop a dramatic multi-system degenerative phenotype that resembles premature aging . Loss of Norad results in PUMILIO hyperactivity and repression of genes that are essential for normal mitosis , leading to the accumulation of aneuploid cells in Norad-deficient tissues . Unexpectedly , Norad loss also causes striking mitochondrial dysfunction , associated with repression of PUMILIO targets that regulate mitochondrial homeostasis . Importantly , transgenic expression of Pum2 is sufficient to fully phenocopy Norad loss of function , triggering genomic instability , mitochondrial dysfunction , and rapidly-advancing aging-associated phenotypes . These findings demonstrate the importance of Norad in maintaining tight control of PUMILIO activity in vivo and establish a critical role for this lncRNA-RBP regulatory interaction in mammalian physiology . A mouse ortholog of NORAD ( 2900097C17Rik or Norad ) , exhibiting 61% nucleotide identity with its human counterpart , is clearly identifiable on mouse chromosome 2 ( Figure 1A ) . Like the human transcript , mouse Norad shows minimal protein-coding potential as assessed by PhyloCSF , a metric that discriminates between coding and noncoding sequences based on their evolutionary signatures ( Lin et al . , 2011 ) ( Figure 1—figure supplement 1A ) . Both human and mouse NORAD/Norad are ubiquitously expressed throughout the body , with highest abundance in brain ( Figure 1B and Figure 1—figure supplement 1B ) . RNA fluorescent in situ hybridization ( RNA FISH ) in mouse embryonic fibroblasts ( MEFs ) , as well as fractionation experiments in a panel of mouse cell lines , revealed a predominately cytoplasmic localization of Norad ( Figure 1C and Figure 1—figure supplement 1C ) , mirroring the localization of the human transcript ( Lee et al . , 2016; Tichon et al . , 2016 ) . To investigate the function of the mouse Norad ortholog , clustered regularly interspaced short palindromic repeats ( CRISPR ) /Cas9-mediated genome editing was employed to delete the lncRNA-encoding sequence from the mouse genome , yielding three independent knockout lines ( Figure 1D–E and Figure 1—figure supplement 2A ) . Importantly , expression of the neighboring genes Epb41l1 and Cnbd2 was unaffected in Norad–/– brain and spleen ( Figure 1—figure supplement 2B ) , demonstrating that deletion of the Norad locus does not perturb neighboring gene expression . Norad–/– mice were viable , fertile , and born at the expected Mendelian frequency ( Figure 2—figure supplement 1A ) . Early in life , Norad-deficient mice were indistinguishable from wild-type littermates , but by 6 months of age , the onset of a multi-system degenerative phenotype resembling premature aging became apparent , with approximately 50% of the mice developing severe manifestations by 1 year of age ( Figure 2A–B ) . This phenotype was characterized by accelerated alopecia and graying of fur in male Norad–/– mice ( Figure 2A and Figure 2—figure supplement 1B ) , while both male and female Norad–/– mice displayed pronounced kyphosis ( Figure 2B and Figure 2—figure supplement 1C ) . Increased kyphosis was also evident in Norad+/– animals , indicating a dose-dependent effect of Norad loss of function . Although body weight was comparable between cohorts of Norad+/+ , Norad+/– , and Norad–/– mice at 1 year of age ( Figure 2—figure supplement 1D ) , mice that developed outward features of aging such as kyphosis also exhibited significant weight loss accompanied by loss of total body fat and subcutaneous fat ( Figure 2C–E ) . Abnormalities were also apparent in Norad-deficient skeletal muscle , with marked switching of fast twitch glycolytic ( FTG ) fibers to slow twitch oxidative ( STO ) fibers ( Figure 2F ) , a phenomenon associated with normal muscle aging ( Ciciliot et al . , 2013 ) . In addition , Norad–/– mice showed accelerated onset of aging-associated pathologies within the central nervous system ( CNS ) ( Gray and Woulfe , 2005; Samorajski et al . , 1968; Son et al . , 2012 ) , including condensed neuronal cell bodies and an accumulation of lipofuscin and vacuoles within spinal motor neurons ( Figure 2G and Figure 2—figure supplement 1E ) . Similar pathologies were evident in neurons in the dorsal root ganglia , brain stem , and cerebellum ( data not shown ) . These changes were accompanied by an overall reduction in neuronal density in the spinal cord ( Figure 2—figure supplement 1F ) . Finally , Norad–/– mice showed decreased overall survival over a 2-year interval ( Figure 2H ) . These findings demonstrate that Norad is essential to suppress widespread aging-associated degenerative phenotypes in mice . Previous work established that NORAD is the preferred binding partner of PUM2 in human cells ( Lee et al . , 2016 ) . NORAD knockout or knockdown triggers PUMILIO hyperactivity and a consequent loss of genomic stability due to excessive PUMILIO-mediated repression of a set of target mRNAs that are critical for normal mitosis ( Lee et al . , 2016; Tichon et al . , 2016 ) . Given the prior demonstration that genomic instability in mice causes aging-associated phenotypes ( Baker et al . , 2004; Baker et al . , 2006 ) , the regulation of PUMILIO activity by this lncRNA , and the resulting effects of PUMILIO hyperactivity , could potentially underlie the phenotype of Norad-deficient animals . To assess whether Norad regulates PUMILIO activity in mice , we performed enhanced UV crosslinking immunoprecipitation coupled with high-throughput sequencing ( eCLIP ) ( Van Nostrand et al . , 2016 ) to assess the transcriptome-wide interactions of PUM2 with target RNAs in Norad+/+ and Norad–/– mice ( Supplementary file 1 ) . Brain was chosen for these experiments because Norad shows the highest expression in this tissue ( Figure 1B and Figure 1—figure supplement 1B ) , pathologic changes are present in the CNS of Norad–/– mice ( Figure 2G and Figure 2—figure supplement 1E–F ) , and mammalian PUMILIO proteins have been implicated in various neuronal functions ( Driscoll et al . , 2013; Gennarino et al . , 2018; Gennarino et al . , 2015; Siemen et al . , 2011; Vessey et al . , 2010; Vessey et al . , 2006; Zhang et al . , 2017 ) . Like human NORAD , the mouse transcript is highly enriched for PREs , harboring 11 perfect matches to the canonical PRE consensus and an additional 3 PREs conforming to a slightly relaxed consensus sequence ( UGUANAUN ) ( Figure 3A ) . Accordingly , robust binding of PUM2 to Norad was detectable by eCLIP , with the majority of binding occurring in the vicinity of PRE sequences . Strikingly , Norad was by far the most highly bound PUM2 target in the transcriptome ( Figure 3B ) , exhibiting at least 1000 times greater CLIP signal than 95% of all PUM2-bound mRNAs . Thus , as observed in human cell lines , but to a much greater extent in vivo , Norad is the preferred RNA target of PUM2 in mouse brain . We next examined the transcriptome-wide interactions of PUM2 with target mRNAs . Notably , the relaxed PRE consensus was the most enriched sequence motif detected in PUM2-bound mRNA 3’ UTRs , supporting the reliability of this eCLIP dataset ( Figure 3B ) . To estimate the apparent PUM2 CLIP signal for each target in Norad+/+ and Norad–/– brain , the average reads in CLIP clusters in a given target 3’ UTR normalized to the expression level of the target was determined for each condition , an approach used previously to estimate relative binding in CLIP data ( Bosson et al . , 2014 ) . This analysis suggested that PUM2 target occupancy was significantly increased in Norad–/– brain , consistent with PUMILIO hyperactivity ( Figure 3C ) . Further supporting this conclusion , RNA-seq revealed that expression of PUM2 CLIP targets was significantly decreased in Norad–/– brain ( Figure 3D ) . Augmented repression of PUM2 CLIP targets was even more apparent when specifically examining targets whose PUM2 binding was measurably increased in the Norad–/– condition ( Figure 3E ) . Together , these data strongly support a conserved function for Norad as a negative regulator of PUMILIO activity in vivo . To assess whether Norad loss of function results in genomic instability , DNA FISH was used to quantify the number of marker chromosomes in primary hematopoietic cells , a representative mitotic tissue . This analysis revealed a significant increase in aneuploid lymphocytes and splenocytes in 3-month and 12-month-old Norad–/– mice ( Figure 4A ) . To directly determine whether loss of Norad results in mitotic abnormalities , MEFs were examined using DNA FISH and live cell imaging . In contrast to lymphocytes or splenocytes , we observed a high rate of polyploidization in all MEF lines tested , as previously reported ( Todaro and Green , 1963 ) ( Figure 4—figure supplement 1A–B ) . We therefore excluded tetraploid and octaploid cells from those scored as aneuploid ( higher ploidy was rarely observed ) . Despite applying these stringent criteria , we detected a significant increase in aneuploidy in Norad–/– MEFs ( Figure 4—figure supplement 1A ) . Moreover , time lapse microscopy revealed an increased occurrence of anaphase bridges and lagging chromosomes as Norad–/– MEFs underwent mitosis ( Figure 4B ) . Thus , Norad-deficient mouse tissues and cells exhibit genomic instability . In human cell lines with NORAD deficiency , genome instability has been linked to repression of a program of genes that are required for mitosis , DNA repair , and DNA replication due to PUMILIO hyperactivity ( Lee et al . , 2016 ) . Consistent with these findings , RNA-seq revealed significant repression of a similar set of genes in Norad–/– spleens ( Figure 4C–D ) . Moreover , brain PUM2 CLIP targets with two or more PREs in their 3’ UTRs were downregulated in Norad–/– spleens , providing evidence of PUMILIO hyperactivity in this tissue ( Figure 4E ) . Taken together , these data support a conserved , essential role for the Norad-PUMILIO axis in the maintenance of genomic stability in mammals . Phenotypic analyses of Norad-deficient mice unexpectedly revealed that in addition to genomic instability , widespread mitochondrial dysfunction was evident in knockout tissues . Overt mitochondrial abnormalities were observed in skeletal muscle of 12-month-old Norad–/– mice , including large accumulations of subsarcolemmal mitochondria ( Figure 5A–B ) accompanied by a significant increase in mitochondrial DNA ( mtDNA ) content ( Figure 5—figure supplement 1A ) . Ultrastructurally , these mitochondria appeared irregular in shape and enlarged , with loss of cristae ( Figure 5B ) . Similarly irregular and enlarged mitochondria were observed in spinal neurons of Norad–/– mice ( Figure 5C ) . These structural abnormalities were accompanied by evidence of reduced mitochondrial function , such as decreased cytochrome c oxidase ( COX; also known as Complex IV of the electron transport chain ) activity in spinal neurons ( Figure 5D ) . Additionally , rare COX-negative fibers were observed in Norad-deficient but not wild-type skeletal muscle ( Figure 5—figure supplement 1B ) . These findings were noteworthy given the extensive evidence linking a decrease in mitochondrial function to aging-associated phenotypes ( Sun et al . , 2016 ) and the previous demonstration that mice lacking proofreading activity of the mtDNA polymerase , which consequently accumulate mtDNA mutations and deletions , exhibit a premature aging phenotype with many similarities to that seen in Norad–/– mice ( Trifunovic et al . , 2004 ) . A major consequence of mitochondrial dysfunction that is believed to play a role in cellular damage and aging is the accumulation of reactive oxygen species ( ROS ) ( Raha and Robinson , 2000; Zorov et al . , 2014 ) . Indeed , brain and spinal cord of Norad–/– mice show evidence of oxidative damage , including elevated levels of 3-nitrotyrosine ( 3-NT ) , 4-hydroxynonenal ( 4-HNE ) , and 8-hydroxy-2'-deoxyguanosine/8-hydroxyguanosine ( 8-OHdG/8-OHG ) , markers of protein , lipid , and nucleic acid oxidation , respectively ( Figure 5—figure supplement 1C–D ) . To directly assess mitochondrial function in Norad-deficient cells , respiration rates were analyzed in pairs of littermate-matched Norad+/+ and Norad–/– MEFs . Basal and maximal respiration was significantly reduced in Norad–/– cells ( Figure 5E ) , accompanied by a decrease in mitochondrial membrane potential ( MMP ) and an increase in ROS production ( Figure 5F ) . Respiration was also examined in human NORAD–/– HCT116 cells ( Lee et al . , 2016 ) . Unlike MEFs , HCT116 cells lacking NORAD exhibited a significant increase in mitochondrial content ( Figure 5—figure supplement 2A–C ) . Nevertheless , when normalized to mtDNA copy number , a similar reduction in respiration and increase in ROS was detectable in these cells ( Figure 5—figure supplement 2D–E ) . These results document a previously unrecognized requirement for Norad in the maintenance of mitochondrial homeostasis in mammalian cells and tissues . To investigate the mechanism through which Norad loss-of-function leads to mitochondrial dysfunction , we examined RNA-seq data from Norad+/+ and Norad–/– brain and spleen using Gene Set Enrichment Analysis ( GSEA ) ( Subramanian et al . , 2005 ) . Genes associated with mitochondria-related gene ontology ( GO ) terms , such as mitochondrial protein complex , electron transport chain , and oxidative phosphorylation , were significantly repressed in Norad-deficient tissues ( Figure 6—figure supplement 1A–B ) . Remarkably , identical gene sets were repressed in human NORAD–/– HCT116 cells . We further identified a set of PUM2 brain CLIP targets within these gene sets that are known to perform important functions in mitochondrial biogenesis and homeostasis , mitochondrial transport , oxidative phosphorylation , metabolism , and ROS detoxification ( Figure 6A ) . Downregulation of a representative set of these genes was validated by qRT-PCR in Norad–/– brain , spleen , and multiple independent MEF lines ( Figure 6B and Figure 6—figure supplement 1C ) . These data are consistent with a model in which PUMILIO hyperactivity in Norad-deficient cells and tissues leads to coordinated downregulation of a broad set of target genes that are critical for normal mitochondrial function . While widespread genomic instability and mitochondrial dysfunction would be predicted to result in the premature aging-like phenotype displayed by Norad–/– mice , it remained to be demonstrated whether PUMILIO hyperactivity alone could account for the full spectrum of observed phenotypes . To address this question , transgenic mice with doxycycline ( dox ) -inducible expression of FLAG-tagged PUM2 were generated and crossed to mice harboring a ubiquitously expressed reverse tetracycline-controlled transactivator 3 transgene ( CAG-rtTA3 ) ( Premsrirut et al . , 2011 ) ( Figure 7A ) . Administration of dox induced broad transgene expression in Pum2; rtTA3 double transgenic mice , as documented by FLAG immunohistochemistry ( IHC ) ( Figure 7—figure supplement 1A ) . Robust transgenic PUM2 expression was also detectable by western blot in isolated MEFs ( Figure 7—figure supplement 1B ) , although an increase in total PUM2 protein levels was surprisingly not detectable in bulk tissue ( Figure 7—figure supplement 1C ) . Importantly , RNA-seq of spleens from transgenic animals revealed significant repression of PUM2 targets , demonstrating PUM2 hyperactivity in this tissue ( Figure 7—figure supplement 1D ) . Because CAG-rtTA3 does not efficiently drive transgene expression in the CNS ( Premsrirut et al . , 2011 ) , we focused our phenotypic studies of Pum2; rtTA3 double transgenic mice on peripheral tissues . Administration of dox to young ( 8–14 week-old ) Pum2; rtTA3 double transgenic mice derived from two independent founders , but not to Pum2 or rtTA3 single transgenic controls , resulted in a striking phenotype within 2 months that closely resembled the appearance of Norad–/– mice at 1 year of age . Dox-treated Pum2; rtTA3 mice developed rapidly progressing kyphosis , alopecia , graying of fur , and loss of body fat ( Figure 7B–C and Figure 7—figure supplement 2A ) . These phenotypes were accompanied by increased aneuploidy in splenocytes ( Figure 7D ) and the accumulation of subsarcolemmal , irregularly shaped mitochondria lacking normal cristae in skeletal muscle ( Figure 7E–F ) . Further demonstrating mitochondrial abnormalities , a global reduction in COX activity was observed in Pum2; rtTA3 skeletal muscle ( Figure 7G and Figure 7—figure supplement 2B ) together with scattered necrotic and regenerating fibers ( Figure 7H and Figure 7—figure supplement 2C ) . Lastly , we directly assessed whether enforced PUM2 expression impairs mitochondrial function in MEFs and human cell lines . Transient expression of FLAG-PUM2 in MEFs or stable expression of either PUM1 or PUM2 in HCT116 significantly impaired respiration ( Figure 7I , Figure 7—figure supplement 2D , Figure 7—figure supplement 3A–C ) . Overall , these data provide compelling evidence that PUMILIO hyperactivity in Norad-deficient animals results in genomic instability , mitochondrial dysfunction , and ultimately a multi-system degenerative phenotype resembling premature aging . Although important roles for a growing number of lncRNA-encoding loci have been uncovered in development and disease states ( Anderson et al . , 2016; Arun et al . , 2016; Sauvageau et al . , 2013 ) , definitive examples of noncoding RNA-mediated functions that are essential for mammalian physiology and maintenance of homeostasis across tissues are limited . Our studies of the murine Norad ortholog reported here unequivocally establish the importance of this lncRNA , and the tight regulation of its target PUMILIO proteins , in mammalian biology and implicate the Norad-PUMILIO axis as a major regulator of aging-associated phenotypes ( Figure 7J ) . These findings provide important new insights and open new lines of investigation into the roles of noncoding RNAs and RNA binding proteins in normal physiology , aging , and disease . While PUMILIO proteins belong to a deeply conserved family of post-transcriptional regulators , obvious Norad orthologs are apparent only in mammals . Other RNAs that regulate the activity of PUMILIO-related proteins in a similar manner in non-mammalian species have not been reported . Why then did this additional layer of PUMILIO regulation evolve ? A possible answer to this question may relate to recent findings that , together with those reported here , demonstrate an exquisite sensitivity to PUMILIO dosage in mammals . Zoghbi and colleagues recently showed that slightly reduced PUM1 dosage causes neurodegeneration in human and mouse brain ( Gennarino et al . , 2018; Gennarino et al . , 2015 ) . Human subjects carrying heterozygous PUM1 deletions or missense mutations develop a neurodevelopmental disorder referred to as PUM1-associated developmental disability , ataxia , and seizure ( PADDAS ) , associated with a ~ 50% reduction in PUM1 protein , or a later onset variant known as PUM1-related cerebellar ataxia ( PRCA ) , associated with only a ~ 25% lowering of PUM1 levels ( Gennarino et al . , 2018 ) . Taken together with our findings from this study , in which we examined for the first time the effects of mammalian PUMILIO hyperactivity in vivo , we can conclude that PUMILIO activity must be maintained within a very narrow range in order to prevent widespread deleterious consequences . In light of these findings , we propose that one major function of Norad is to buffer PUMILIO activity such that it stays within this critical range . This model posits the existence of Norad-bound and free PUMILIO pools which are exchangeable and in equilibrium , ensuring a consistent amount of available PUMILIO for target mRNA engagement and preventing fluctuations in PUMILIO expression from manifesting in altered target repression . Indeed , an RNA such as Norad represents an ideal molecule to serve as a buffer of this type , as it is able to efficiently regulate the activity of a pre-existing pool of PUMILIO at the level of target engagement . In addition to providing a buffering function , it is likely that Norad is also utilized for dynamic regulation of PUMILIO activity under selected conditions . In particular , this lncRNA is known to be induced by a variety of cellular stressors , including DNA damage ( Lee et al . , 2016 ) and hypoxia ( Michalik et al . , 2014 ) , which would be expected to result in de-repression of PUMILIO targets following these stimuli . Although the functional consequences of modulating PUMILIO-mediated gene regulation under these conditions is not yet understood , continued investigation of the signaling inputs that control this system and the resulting effects on PUMILIO-regulated gene networks will be important to further elucidate the roles of this newly discovered pathway in mammalian biology . A surprising observation reported in this study was the rapid onset of dramatic premature aging-like phenotypes in Pum2 transgenic mice despite a lack of overt overexpression of PUM2 protein at the bulk tissue level ( Figure 7—figure supplement 1C ) . This finding raises the question of how transgene induction is able to drive such a striking phenotype if the protein product does not accumulate to supraphysiologic levels . An appealing hypothesis to explain this observation postulates that there are key vulnerable cell populations that become dysfunctional or damaged upon Pum2 induction . These cells may be rare or may naturally express lower levels of PUMILIO , such that a large change in PUM2 expression within them may be masked by the majority of cells in the tissue . Indeed , isolated MEFs from Pum2; rtTA3 transgenic mice show a robust increase in PUM2 protein expression ( Figure 7—figure supplement 1B ) , thereby demonstrating clear transgene activity in isolated cell types . In addition , given that production of transgenic FLAG-PUM2 is robustly detectable by IHC ( Figure 7—figure supplement 1A ) , repression of endogenous PUMILIO through a previously described negative feedback mechanism ( Galgano et al . , 2008; Kedde et al . , 2010; Morris et al . , 2008 ) may also contribute to the apparent lack of increase in total PUM2 levels in bulk tissue . Most importantly , the significant repression of PUM2 targets upon transgene induction ( Figure 7—figure supplement 1D ) demonstrates detectable PUM2 hyperactivity in Pum2; rtTA3 tissues even in the absence of a global increase in protein abundance . Finally , it is worth noting that the absolute magnitude of repression of individual PUM2 targets in Norad–/– or Pum2; rtTA3 tissues is generally low ( ~10–20% ) ( for example , see Figure 6 ) . While the complexity of bulk tissue may mask more robust repression of PUM2 targets in specific cell-types , it is likely that the phenotypic consequences of PUMILIO hyperactivity cannot be attributed to repression of individual targets . Rather , the coordinated , modest repression of a broad set of PUMILIO targets in aggregate most likely produces the dramatic phenotypes observed . Cells that are sensitive to enforced Pum2 expression likely include stem cell populations whose dysfunction could lead to loss of tissue homeostasis and degenerative phenotypes . Accordingly , PUMILIO and related proteins have been implicated as critical stem cell regulators in model organisms and mammals ( Crittenden et al . , 2002; Forbes and Lehmann , 1998; Leeb et al . , 2014; Lin and Spradling , 1997; Naudin et al . , 2017; Shigunov et al . , 2012 ) . Identification of specific cell populations that drive aging-associated phenotypes under conditions of Norad-deficiency or PUMILIO hyperactivity represents an important priority for future work as this approach may reveal new cell types whose dysfunction contributes to the natural aging-associated decline of tissue homeostasis and renewal . Analyses of Norad-deficiency and enforced Pum2 expression unexpectedly revealed that PUMILIO hyperactivity triggers the coordinated repression of a large set of PUM2 target transcripts with key roles in mitochondrial function and homeostasis , associated with widespread structural and functional mitochondrial defects . These findings were corroborated by a recent study reporting that elevated PUM2 levels in aged mice impaired mitochondrial homeostasis ( D'Amico et al . , 2019 ) . A large body of evidence has linked a decline in mitochondrial function to aging-associated phenotypes ( Sun et al . , 2016 ) , including the direct demonstration that ‘mitochondrial mutator mice’ , which harbor a mutation in the mitochondrial DNA polymerase and consequently accumulate mtDNA mutations , develop a premature aging phenotype with many features in common with Norad–/– mice ( Trifunovic et al . , 2004 ) . Thus , mitochondrial dysfunction in concert with genomic instability , another abnormality associated with premature aging in mice ( Baker et al . , 2004; Baker et al . , 2006 ) , provides a compelling mechanistic basis for the phenotype of Norad-deficient animals . The regulation of mitochondrial biogenesis and function by PUMILIO-related proteins is not restricted to mammals . The budding yeast Puf family member Puf3p preferentially associates with mRNAs that encode mitochondrial proteins and facilitates their local translation in the vicinity of the mitochondrial protein import machinery ( García-Rodríguez et al . , 2007; Gerber et al . , 2004; Saint-Georges et al . , 2008 ) . In Drosophila and cultured mammalian cells , PUMILIO proteins repress translation of mRNAs that encode mitochondria-destined proteins until these transcripts are docked at the mitochondrial outer membrane ( Gehrke et al . , 2015 ) . Together , these observations suggest a deeply conserved role for PUMILIO proteins in the regulation of mitochondrial biology across eukaryotic species . Perhaps , the most intriguing question to arise from these studies is whether dysregulation of the Norad-PUMILIO axis plays a role in physiologic aging and/or human disease . Remarkably , a recent RNA-seq study of noncoding RNA expression in the subependymal zone of human brains of increasing age reported a strong age-related decrease in NORAD expression ( Barry et al . , 2015 ) . These findings take on added significance in light of our new understanding of the consequences of disruption of the Norad-PUMILIO axis and suggest that this lncRNA , and its target PUMILIO proteins , represent new candidates whose altered expression or function may influence the normal age-related decline in tissue function . These genes also represent previously unrecognized candidates that may be mutated or otherwise disrupted in rare progeroid cases that are unlinked to the genes that are presently known to cause these disorders . Thus , further study of the Norad-PUMILIO axis , and the pathways that regulate this noncoding RNA and its target proteins , promises to reveal important and unexpected new insights into mammalian physiology and disease . All animal protocols were approved by the Institutional Animal Care and Use Committee ( IACUC ) of The University of Texas Southwestern Medical Center ( UTSW ) and The Ohio State University , Nationwide Children’s Hospital . Mice were maintained in regular housing with a 12 hr light/dark cycle and normal chow and water ad libitum . Norad–/– mice were generated in the UTSW Transgenic Core by injecting Cas9 mRNA ( Sigma-Aldrich ) together with two in vitro transcribed sgRNAs flanking the Norad locus into fertilized C57BL/6J oocytes as described ( Yang et al . , 2013 ) . Founder mice harboring deletions of Norad were maintained by backcrossing to wild-type C57BL/6J mice . Of note , Norad–/– lines were produced from three independent founder mice with Norad deletions ( Figure 1—figure supplement 2A ) . All were phenotypically indistinguishable and used for subsequent studies of Norad function . For all animal experiments , cohort sizes were not pre-determined using power calculations . Sufficiently large sample sizes were used to allow detection of statistically significant differences between experimental cohorts . Doxycycline ( dox ) -inducible FLAG-Pum2 transgenic mice ( in this study referred to as Pum2 mice ) were generated in a C57BL/6J background by the UTSW Transgenic Core using standard procedures for pronuclear injection . For the generation of the transgene vector , a cDNA clone of isoform 3 of the mouse Pum2 coding sequence ( BC041773 ) was purchased from transOMIC Technologies , verified by Sanger sequencing , PCR amplified with primers adding a FLAG tag to the protein N-terminus , and cloned into the pTRE-Tight vector ( Clontech ) . Of note , isoform 3 encodes for the shorter PUM2 variant , which can be detected as the lower of two PUM2 bands in western blots . Broad endogenous expression of this isoform was detected by PCR in all tested mouse tissues ( data not shown ) , confirming its physiologic relevance . Transgene positive Pum2 mice were crossed to a ubiquitous cytomegalovirus early enhancer element chicken beta-actin ( CAG ) promoter-driven reverse tetracycline-controlled transactivator 3 ( rtTA3 ) mouse line , which was generated in the Lowe laboratory ( Premsrirut et al . , 2011 ) and obtained from The Jackson Laboratory ( stock number 016532 ) . The resulting Pum2; rtTA3 double transgenic mice were used for subsequent experiments together with Pum2 , rtTA3 , and wild-type littermates as controls . Transgene expression was induced in 8–14 week-old mice for 1 . 5–2 months by administering 2 g/L doxycycline hydrochloride ( dox ) ( Sigma-Aldrich ) supplemented with 10 g/L sucrose ( Research Products International ) in drinking water . Mice were anesthetized with isoflurane ( Henry Schein Animal Health ) and subjected to retro-orbital bleeding . Approximately 200 μL of blood were collected , immediately heparinized with 500 USP units/mL ( Fresenius Kabi ) , transferred to 1 . 3 mL of PB Max Karyotyping Medium supplemented with 50 μg/mL lipopolysaccharide ( Gibco and Sigma-Aldrich ) , and incubated at 37°C for 48 hr with shaking . After this incubation , cells were harvested and processed for DNA FISH analysis . For the isolation of splenocytes , mice were euthanized with an overdose of isoflurane , and spleens were resected , minced with a razor blade in 1X Hank’s balanced salt solution ( HBSS ) without calcium and magnesium ( Gibco ) , passed through a 70 μm cell strainer ( Corning ) , and washed with 1X phosphate buffered saline ( PBS ) ( Sigma-Aldrich ) . Immediately after isolation , splenocytes were processed for DNA FISH analysis . Norad+/– females were bred to Norad+/– males or Pum2 females were bred to rtTA3 males and euthanized at embryonic day E14 . 5 . Under sterile conditions , the uterine horns were removed and washed once with 70% ethanol and three times with 1X PBS . The embryos were then released and the heads and all visceral organs removed . The remainder of the embryo was finely minced using razor blades and treated with 0 . 25% trypsin/EDTA ( Gibco ) at 37°C for a total of 20 min . MEF growth medium consisting of DMEM with 4 . 5 g/L glucose , L-glutamine and sodium pyruvate ( Gibco ) supplemented with 1X nonessential amino acids ( NEAA ) , 1X Antibiotic-Antimycotic ( AA ) ( all Gibco ) , and 10% fetal bovine serum ( FBS ) ( Gibco , Sigma-Aldrich ) was added to inactivate the trypsin . Tissue chunks were disrupted by vigorous pipetting , centrifuged , resuspended in MEF growth medium , and plated in T25 flasks . The next day , non-adherent tissue debris was used for genotyping , while attached cells were transferred to a fresh tissue culture dish and designated as primary MEF passage 1 ( P1 ) . Aliquots of primary MEF P1 were frozen and stored until needed . Primary MEFs were used for a maximum of passages . Immortalized MEF lines were generated by transfecting primary MEFs with pSG5-SV40-Large-T-Antigen using Lipofectamine 3000 ( Invitrogen ) according to the manufacturer’s instructions . Starting at 48 hr post transfection , cells were serially passaged 1:10 to select for SV40-immortalized MEFs . After 6 passages , all SV40 large T antigen-transfected MEF lines were regarded as immortalized and this passage was designated as immortalized MEF P1 . Of note , the genders of the MEF lines used in this study are not known . All MEF lines were tested and confirmed to be mycoplasma free . All established mouse cell lines ( Neuro-2a , CT26 , and Hepa1-6 ) were obtained from ATCC and cultured in DMEM with 4 . 5 g/L glucose , L-glutamine and sodium pyruvate ( Gibco ) supplemented with 1X AA ( Gibco ) , and 10% FBS ( Sigma-Aldrich ) . The male colon cancer cell line HCT116 was obtained from ATCC ( CCL-247 ) and cultured in McCoy’s 5a medium ( Gibco ) supplemented with 1X AA ( Gibco ) and 10% FBS ( Gibco , Sigma-Aldrich ) . The cell line was authenticated by ATCC using short tandem repeat ( STR ) analysis in November 2017 . The generation of HCT116 NORAD–/– clones via transcription activator-like effector nuclease ( TALEN ) -mediated insertion of a Lox-Stop-Lox cassette as well as HCT116 PUM1 and PUM2 overexpression clones using lentiviral transduction was described previously ( Lee et al . , 2016 ) . Cell lines were tested and confirmed to be mycoplasma free . Total RNA was isolated from cells or tissues using the miRNeasy Mini Kit ( Qiagen ) following the manufacturer’s instructions including an on-column DNAse I digest to remove genomic DNA contamination . Complementary DNA ( cDNA ) was generated from 1 μg of total RNA using the SuperScript III First-Strand Synthesis SuperMix for qRT-PCR ( Invitrogen ) according to the manufacturer’s protocol . Relative Norad expression in mouse was quantified using the Applied Biosystems TaqMan assay for 2900097C17Rik ( Norad ) ( Mm04242407_s1 ) and the TaqMan Universal II Master Mix ( Applied Biosystems ) . Human NORAD was quantified using a custom TaqMan assay described previously ( Lee et al . , 2016 ) . For all other genes analyzed in this study , expression was quantified using the Power SYBR Green PCR Master Mix ( Applied Biosystems ) together with primers provided in Supplementary file 2 . RNA expression levels were normalized to 18S ribosomal RNA ( cell line studies ) or Gapdh mRNA ( in vivo studies ) using either standard curves of each gene or the comparative ΔCt method . The number of biological replicates is stated in the figure legends , each biological replicate was run with three technical replicates . Total RNA was isolated from brains and spleens of 10-week-old male Norad+/+ and Norad–/– mice ( three mice per genotype ) as well as from spleens of 16-week-old Pum2; rtTA3 and control ( Pum2 and wild-type ) mice after 4 weeks of dox treatment ( four mice per group ) using the miRNeasy Mini Kit ( Qiagen ) including a DNAse I digestion step to remove genomic DNA . RNA integrity was determined with the Agilent 2100 Bioanalyzer , and only RNA samples with an RNA integrity number ( RIN ) of greater than eight were used for subsequent analysis . Sequencing libraries were prepared with the TruSeq Stranded mRNA Library Prep Kit ( Illumina ) and sequenced using the 75 base pair ( bp ) single-read protocol on a NextSeq 500 platform ( Illumina ) . Library prep and RNA-seq were performed by the UTSW McDermott Center Next-Generation Sequencing Core . For RNA-seq in Norad+/+ and Norad–/– mice , quality assessment of the sequencing data was performed with NGS QC Toolkit ( v2 . 3 . 3 ) ( Patel and Jain , 2012 ) . Reads with more than 30% of nucleotides with a Phred quality score of less than 20 were removed from further analysis . Quality-filtered reads were then aligned to the mouse reference genome GRCm38 ( mm10 ) using Tophat2 ( v2 . 0 . 12 ) with default settings ( Kim et al . , 2013 ) . Only reads uniquely mapped to the genome were kept for future analysis . Aligned reads were counted per gene ID using featureCount ( v1 . 4 . 6 ) ( Liao et al . , 2014 ) . Differential gene expression analysis was carried out using the R package EdgeR ( v3 . 8 . 6 ) ( Robinson et al . , 2010 ) . For RNA-seq in Pum2; rtTA3 and control mice , reads were aligned to GRCm38 ( mm10 ) using HISAT2 ( v2 . 1 . 0 ) ( Pertea et al . , 2016 ) . Only reads uniquely mapped to the genome were kept for future analysis . Aligned reads were counted per gene ID using featureCount ( v1 . 6 . 0 ) . Differential gene expression analysis was carried out using EdgeR ( v3 . 24 . 0 ) . For each comparison , genes were required to have at least one read in at least one sample to be considered as expressed . Differential gene expression analysis was performed using the GLM approach following EdgeR’s official documentation . CPM ( counts per million ) and FPKM ( fragments per kilobase million ) were obtained using EdgeR and Stringtie ( v1 . 2 . 2 ) ( Pertea et al . , 2015 ) , respectively . Gene set enrichment analysis ( GSEA ) ( Subramanian et al . , 2005 ) was performed using default gene sets of gene ontology ( GO ) terms . The results obtained from the RNA-seq analyses of brain and spleen ( normalized reads as CPM ) were used as input data . The normalized enrichment scores ( NES ) as well as the false discovery rates ( FDR ) are provided in the figures . GO analysis of spleen RNA-seq data was also carried out on genes that were significantly ( p≤0 . 05 ) downregulated in Norad–/– spleens using the Database for Annotation , Visualization and Integrated Discovery ( DAVID ) ( Huang et al . , 2009a; Huang et al . , 2009b ) . PUM2 RNA interactions in the mouse brain were determined by eCLIP , following a previously published protocol ( Van Nostrand et al . , 2016 ) . In brief , brains from 3-month-old Norad+/+ and Norad–/– females ( two mice per genotype representing two biological replicates ) were resected and cut in halves . Per sample , one half of a brain was minced with razor blades in 1X ice-cold diethyl pyrocarbonate ( DEPC ) -treated PBS . The resulting tissue suspensions were UV crosslinked on ice in a Spectrolinker XL-1500 ( Spectronics ) at 254 nm three times at 400mJ/cm2 . The UV crosslinked tissues were centrifuged , snap-frozen in ethanol/dry ice , and stored at −80°C until needed . For the immunoprecipitation of PUM2 , Protein G Dynabeads ( Invitrogen ) were used together with the same polyclonal goat anti-PUM2 antibody ( sc-31535 , Santa Cruz ) used previously for PUM2 CLIP analysis in human cells ( Lee et al . , 2016 ) . For each genotype , duplicate size-matched input and immunoprecipitation samples were prepared ( four samples per genotype ) . In contrast to the original protocol ( Van Nostrand et al . , 2016 ) , in which libraries were designed for paired-end sequencing , we adapted the RNA and DNA linker sequences for single-read sequencing . For each sample , separate sequencing libraries were generated using a unique modified RiL19-new RNA linker as well as a modified rand103Tr3-new DNA linker and AR17-new reverse transcription primer ( sequences provided in Supplementary file 2 ) . PCR library amplification was performed with polyacrylamide gel electrophoresis ( PAGE ) -purified oligonucleotides containing specific indexes ( D501-D504 , D701-D703 ) . Single-read sequencing was performed on a NextSeq 500 platform using a NextSeq 500/550 High Output v2 Kit , 75 cycles ( Illumina ) in the UTSW McDermott Center Next-Generation Sequencing Core . All adapter sequences were removed using Cutadapt with an e-value set to 0 . 1 . All reads less than 18 nt after adapter trimming were discarded , and the unique molecular identifiers ( 10 nt randomers ) for PCR duplication identification were trimmed and recorded using in-house scripts . Because of the high number of NORAD pseudogenes in the human and mouse genome , we followed a similar mapping strategy to that used in our previous study ( Lee et al . , 2016 ) . Reads were first mapped to Norad ( 2900097C17Rik ) before all remaining reads , which did not align to Norad , were mapped to GRCm38 ( mm10 ) using Tophat2 ( v2 . 0 . 12 ) with default settings ( Kim et al . , 2013 ) . Only uniquely mapped genomic reads were retained . PCR duplicates were then removed based on the unique molecular identifier information using in-house scripts . All remaining reads were regarded as usable reads and subjected to cluster calling . For each IP sample , the read coverage of each nucleotide was calculated and all regions with coverages of equal or greater than three were kept as candidate bins . The read counts of each IP/input pair were obtained for every bin , requiring at least a 50% sequence overlap . The fold-changes of the normalized read counts were then calculated for each bin: normalized fold-change = ( ( reads_in_bin[IP]+1 ) /total_usable_reads[IP] ) / ( ( reads_in_bin[input]+1 ) /total_usable_reads[input] ) . Bins with fold-changes greater than or equal to four were considered as clusters . Finally , we filtered clusters for those that were detected in both biological replicates of either genotype . Clusters that overlapped with at least 30% of their length were merged . Bedgraph files of each sample were generated with BEDtools ( Quinlan and Hall , 2010 ) using reads normalized to the total usable genomic read count and visualized with the Integrative Genomics Viewer ( IGV ) . Genes with one or more CLIP clusters in their 3’ UTRs were regarded as PUM2 target genes . The number of CLIP reads per gene ( Figure 3B ) was determined by calculating the weighted sum of all reads within 3’ UTR CLIP clusters for each PUM2 target gene from both Norad+/+ CLIP replicates . To calculate normalized CLIP signal of PUM2 targets in Norad+/+ and Norad–/– brains ( Figure 3C ) , a method similar to that used by Bosson et al . was used ( Bosson et al . , 2014 ) . Genes were first filtered for those with an average expression level of at least 1 FPKM in brain RNA-seq data . The average normalized number of CLIP reads from both CLIP replicates within 3’ UTR clusters of each PUM2 target gene were then summed and divided by the gene’s expression level ( average FPKM ) in the respective genotype . For CDF plots depicting fold-changes of CLIP targets in brain ( Figure 3D–E ) , non-CLIP targets were filtered for those whose expression levels were within 25% of the log mean value for this parameter in the set of CLIP targets . Non-CLIP targets were also filtered for those whose 3’ UTR length was within 25% of the mean of 3’ UTR lengths in the set of CLIP targets . For similar CDF plots in which the expression of CLIP targets was examined in spleen ( Figure 4E and Figure 7—figure supplement 1D ) , brain CLIP targets were filtered for those with FPKM ≥ 1 in either Norad–/– or Pum2; rtTA3 spleen and the presence of at least two PREs within 100 nucleotides of CLIP clusters in the target 3’ UTR . Non-CLIP targets were filtered as described above for brain to match expression level and 3’ UTR length to the set of CLIP targets shown in each plot . Total DNA was isolated using the DNeasy Blood and Tissue Kit ( Qiagen ) according to the manufacturer’s instructions . Mitochondrial ( mtDNA ) and nuclear ( nDNA ) DNA were quantified by qPCR using either human or mouse specific primers ( Supplementary file 2 ) and the Power SYBR Green PCR Master Mix ( Applied Biosystems ) . The quantity of mtDNA was then normalized to the quantity of nDNA . Both mtDNA and nDNA concentrations were determined using standard curves . The number of replicates is provided in the respective figure legends . Mouse cell lines ( immortalized MEFs , Neuro-2a , CT26 , or Hepa1-6 ) were seeded in triplicate and harvested the next day for subcellular fractionation , which was performed as previously described ( Lee et al . , 2016 ) . Briefly , cell pellets were lysed in RLN1 buffer ( 50 mM Tris-HCl pH 8 . 0 , 140 mM NaCl , 1 . 5 mM MgCl2 , 0 . 5% NP-40 , RNAse inhibitor ) , incubated on ice for 5 min , and centrifuged . The supernatant contained the cytoplasmic fraction , while the pellet contained the nuclear fraction . Both fractions were then subjected to RNA isolation and equal cell equivalents of nuclear and cytoplasmic RNA were used in subsequent qRT-PCR reactions . All samples were tested for Norad as well as for Neat1 ( nuclear control ) and Actb ( cytoplasmic control ) . Because equal cell equivalents of nuclear and cytoplasmic RNA were used in each reaction , the sum of the expression level of each transcript in the nucleus plus cytoplasm can be set to 100% , thereby allowing determination of the percentage of each transcript localized to each compartment . Neat1 and Actb , respectively , showed the expected nuclear and cytoplasmic localization in each experiment , confirming successful subcellular fractionation . RNA FISH was performed as described previously ( Mito et al . , 2016 ) . A DIG-labeled RNA probe for mouse Norad was synthesized by in vitro transcription using a DIG-labeling mix ( Roche ) . Primers used for amplification of the DNA template are provided in Supplementary file 2 . MEFs grown on poly-L-lysine coated coverslips were fixed in 4% paraformaldehdye for 10 min followed by permeabilization in 0 . 5% Triton X-100 for 10 min . Samples were then hybridized with 10 ng/μL DIG-labeled RNA probe at 55°C for 16 hr . Following hybridization , samples were washed and treated with RNase A . DIG-labeled probes were detected using a mouse monoclonal anti-DIG primary antibody ( Roche ) and a Cy3-labeled goat anti-mouse IgG secondary antibody ( EMD Millipore ) . A Zeiss LSM700 confocal microscope was used for imaging . Cell and tissue lysates were prepared in RIPA buffer ( 50 mM Tris-HCl pH 8 . 0 , 150 mM NaCl , 1% NP-40 , 0 . 5% sodium deoxycholate , 0 . 1% sodium dodecyl sulfate ) supplemented with cOmplete Protease Inhibitor Cocktail ( Roche ) . Western blots were probed with monoclonal rabbit anti-PUM2 antibody ( ab92390 , Abcam ) , monoclonal rabbit anti-PUM1 antibody ( ab92545 , Abcam ) , polyclonal rabbit anti-FLAG ( 2368 , Cell Signaling ) , or monoclonal rabbit anti-GAPDH antibody ( 2118 , Cell Signaling ) . Bands were visualized using an IRDye 800CW donkey anti-rabbit IgG secondary antibody ( 925–32213 , Licor ) and an Odyssey CLx Imager ( Licor ) . Tissues were harvested from mice that had been treated with dox for 3 . 5–6 . 5 weeks . All tissues were fixed in 10% neutral buffered formalin ( NBF ) for 24–48 hr . Fixed samples were processed , paraffin embedded , and sectioned using standard procedures . To detect the expression of the FLAG-Pum2 transgene , immunohistochemistry ( IHC ) was performed by the UTSW Tissue Management Shared Resource using the monoclonal anti-FLAG M2 antibody ( F1804 , Sigma-Aldrich ) . Images were acquired on an AxioObserver Z1 microscope ( Zeiss ) . 12-month-old Norad+/+ and Norad–/– mice were used for semi-thin and ultrastructural analysis . For these studies , one group of mice were given xylazine/ketamine anesthesia and euthanized by cardiac perfusion with 4% paraformaldehyde followed by 5% glutaraldehyde ( both in 0 . 1 M phosphate buffer ) . Tissue samples from brain and spinal cord were removed under a dissecting microscope . A second group of mice were perfused with 4% paraformaldehyde and their muscles were removed and further fixed in situ in 5% glutaraldehyde . These tissues were dissected into small blocks and processed for plastic embedding using standard methods ( Sahenk and Mendell , 1979 ) . Thick ( 1 µm ) sections were stained with toluidine blue and selected blocks were sectioned and examined with an electron microscope ( Hitachi H7650 ) . Brain and spinal cord segments were placed in 10% NBF and processed for paraffin embedding . Brain , spinal cord , and skeletal muscle were collected from additional mice for cryostat sectioning . Muscle tissues from Pum2; rtTA3 as well as Pum2 and rtTA3 single transgenic littermates were collected and processed as for Norad–/– mice . All transgenic mice were between 4 and 6 months old and had been treated with dox for 1 . 5–2 months . To analyze neuronal cell loss in the ventral horn neuron pools , 5 µm thick , paraffin embedded and hematoxylin and eosin ( H and E ) stained lumbar spinal cord sections from Norad+/+ ( n = 5 ) and Norad–/– ( n = 5 ) mice were analyzed . Motor neuron pools in the anterior horn cell areas of both hemicords from each section were included . For each mouse , sections from 1-2 levels were photographed at 20X magnification . Only cell bodies clearly showing a nucleolus on the plane of the section were considered . Equal numbers of spinal cord levels were analyzed in each group . Mean neuronal densities of anterior horn areas per lumbar cord level were calculated . Succinate dehydrogenase ( SDH ) enzyme histochemistry was used to assess metabolic fiber type changes in the aging muscle . For this purpose , muscle fiber types were grouped into three categories: slow twitch oxidative ( STO ) , fast twitch oxidative ( FTO ) , and fast twitch glycolytic ( FTG ) . Twelve-µm-thick cross-sections from the gastrocnemius muscles of 12-month-old Norad+/+ and Norad–/– mice ( n = 4 in each group , 2 males and 2 females ) were stained for SDH activity . Three images , each representing a distinct zone of the gastrocnemius muscle ( a deep zone predominantly composed of STO , an intermediate zone showing a checkerboard appearance of STO , FTO , or FTG , and a superficial zone predominantly composed of FTG fibers ) , were taken along the midline axis at 20X magnification using an Olympus BX41 microscope . Muscle fiber types were determined and counted based on their dark ( STO ) , intermediate ( FTO ) , or light ( FTG ) SDH staining as previously reported ( Yalvac et al . , 2017 ) . Mitochondrial function was assessed in spinal cord neurons of 12-month-old Norad+/+ and Norad–/– mice using cytochrome c oxidase ( COX ) histochemistry . Twelve µm thick frozen lumbar spinal cord tissue sections were mounted onto superfrost glass slides ( Thermo Fisher ) and dried at room temperature for 1 hr before the COX enzymatic activity assay was performed . Two sections , 100 µm apart , were analyzed from each mouse . The anterior horn areas of each section were photographed at 10X magnification and the number of neurons with normal COX activity were determined . Counts from the hemicord anterior horn cell area with the higher number of neurons with normal COX activity was included in the analysis . Only cell bodies with an obviously visible nucleolus on the plane of the section and with a COX activity clearly above the background of gray matter were considered . The number of neurons with normal COX activity was counted per each section analyzed . COX activity was also qualitatively analyzed in fresh frozen gastrocnemius muscle sections from 12-month-old Norad+/+ and Norad–/– mice and 4-6 month-old Pum2; rtTA3 transgenic mice or controls that had been treated with dox for 1 . 5–2 months . DNA FISH for two representative chromosomes ( chromosomes 2 and 16 ) was performed in lymphocytes , splenocytes , and primary MEFs . Cells were first incubated in hypotonic KCl solution: lymphocytes were incubated in 75 mM KCl at 37°C for 15 min , splenocytes in 75 mM KCl at room temperature for 30 min , and MEFs in 0 . 4% KCl at room temperature for 8 min . Subsequently , cells were centrifuged and resuspended in methanol/acetic acid ( 3:1 ) , washed twice with methanol/acetic acid ( 3:1 ) , dropped onto Rite-On Micro Slides ( Gold Seal Products ) , air-dried , and either used immediately or stored at −20°C until needed . DNA FISH was performed using the Mouse IDetect Chromosome Point Probes for chromosome 2 ( red ) and 16 ( green ) ( IDMP1002-R , IDMP1016-1-G , Empire Genomics ) following the manufacturer’s protocol . Slides were mounted with ProLong Diamond Antifade Mountant with DAPI ( Invitrogen ) and analyzed on an AxioObserver Z1 microscope ( Zeiss ) using the 100X oil objective . Lymphocytes and splenocytes whose chromosome count differed from 2n for at least one of the two tested chromosomes were regarded as aneuploid or off mode . MEFs were only considered aneuploid when their chromosome count differed from 2n or a multiple of 2n in order to account for the increased polyploidy in this cell type . Primary Norad+/+ and Norad–/– MEFs ( three MEF lines per genotype , P<4 ) were grown on Lab-Tek Chambered Coverglass slides ( Thermo Fisher ) that were coated with poly-L-lysine ( Sigma-Aldrich ) . Prior to the analysis , DNA was visualized by adding 50 ng/mL Hoechst dye ( Invitrogen ) to the growth medium . Mitoses were monitored by taking fluorescence images every 5 min for ~48 hr on a Leica inverted microscope equipped with a temperature and CO2-controlled chamber , a 63X oil objective , an Evolve 512 Delta EMCCD camera , and the MetaMorph Microscopy Automation and Image Analysis Software ( Molecular Devices , LLC ) . Videos were generated from the acquired time-lapse images and analyzed for the occurrence of mitotic defects including anaphase bridges and lagging chromosomes . Norad+/+ , Norad+/– , and Norad–/– mice were continuously monitored over a period of 12 months for the onset and progression of kyphosis as well as alopecia and graying of fur . The kyphosis scoring system was adopted from a previously reported strategy ( Guyenet et al . , 2010 ) . In brief , a kyphosis score of 0 indicates no kyphosis detectable , a score of 1 indicates the presence of mild kyphosis but the mouse is still able to entirely stretch its spine , and scores of 2 and 3 indicate that there is prominent kyphosis at rest which persists in a mild ( score of 2 ) or a severe ( score of 3 ) form even when the mouse stretches its spine . Whole-body fat , as a percentage of body weight , was measured by nuclear magnetic resonance ( NMR ) in 3-month and 12-month-old Norad+/+ and Norad–/– mice , or in 4–6 month-old Pum2; rtTA3 double transgenic mice and control littermates after 1 . 5–2 months of dox treatment , using a Bruker Minispec mq10 . Subcutaneous ( s . c . ) adipose thickness of 12-month-old Norad+/+ and Norad–/– mice was determined using standard H and E-stained skin histology . For every skin sample , images were acquired at 5X magnification across the entire length of the section on an AxioObserver Z1 microscope ( Zeiss ) using the AxioVision 4 . 8 software ( Zeiss ) . In these images , the thickness of the s . c . adipose tissue was measured at 15 different points using the AxioVision 4 . 8 software . The average of the 15 measurements was then calculated to obtain the adipose thickness of each mouse . Tetramethylrhodamine ethyl ester ( TMRE ) ( ENZ-52309 , Enzo Life Sciences ) was used for measuring mitochondrial membrane potential ( MMP ) in immortalized Norad+/+ and Norad–/– MEFs . 80 × 103 cells were seeded in triplicate in 6-well plates in regular growth medium and incubated for 16–18 hr until 70–80% confluent . Cells were then trypsinized with 0 . 25% trypsin/EDTA ( Gibco ) , pelleted at 300 g , and resuspended in 500 µL of fresh growth medium containing 50 nM TMRE . Samples were then incubated at 37°C in the dark for 30 min and analyzed by flow cytometry using a BD Accuri C6 Cytometer ( BD Biosciences ) . The average and standard deviation of the mean fluorescence intensities of the three replicates was calculated for each sample , and each Norad–/– MEF line was compared to its Norad+/+ littermate control line . The Enzo Total ROS Detection Kit ( ENZ-51011 , Enzo Life Sciences ) was used for detection of ROS levels in immortalized MEFs and human HCT116 cells . 80 × 103 MEFs were plated in triplicate in -well plates in regular growth medium and incubated for 16–18 hr . 60 × 103 HCT116 cells were seeded in triplicate in 24-well plates in regular growth medium and also incubated for 16–18 hr . All cells were 70–80% confluent at the time of the assay . ROS levels were measured according to the manufacturer’s protocol . Briefly , cells were trypsinized with 0 . 25% trypsin/EDTA ( Gibco ) , pelleted at 300 g , washed once with Enzo 1X Wash Buffer , and resuspended in 300 µL of freshly prepared Enzo ROS Detection Solution ( 1 µL ROS dye in 5 mL Wash Buffer ) . Samples were incubated in the dark for 30 min and analyzed by flow cytometry using a BD Accuri C6 Cytometer ( BD Biosciences ) . The average and standard deviation of the mean fluorescence intensities of the three replicates was calculated for each sample . Each Norad–/– MEF line was compared to its Norad+/+ littermate control line . Oxidative damage was examined in the CNS of 12-month-old Norad+/+ and Norad–/– mice using IHC and immunofluorescence . Lipid peroxidation was examined in formalin-fixed paraffin embedded spinal cord sections by IHC using the polyclonal rabbit anti-4-Hydroxynonenal ( 4-HNE ) antibody ( ab46545 , Abcam ) . Nucleic acid oxidation was assessed in fresh frozen brain sections by immunofluorescence using the mouse monoclonal anti-DNA/RNA Damage antibody ( ab62623 , Abcam ) , which detects 8-OHdG/8-OHG . In addition , protein oxidation was assessed in spinal cord and brain using an enzyme-linked immunosorbent assay ( ELISA ) for 3-nitrotyrosine ( 3-NT ) ( MBS262795 , Mybiosource ) according to the manufacturer’s instructions . Oxygen consumption rates ( OCRs ) were analyzed in immortalized MEFs and human HCT116 cells using a Seahorse Bioscience XF96 Extracellular Flux Analyzer ( Seahorse Bioscience/Agilent ) . 10 × 103 cells per well were plated in Seahorse XF96 cell culture microplates ( Agilent ) in regular growth medium and incubated for 14–16 hr . For HCT116 cells , Seahorse XF96 Cell Culture Microplates were coated with poly-L-lysine ( Sigma-Aldrich ) to improve cell attachment . Prior to measurement , cells were equilibrated for 1 hr in Seahorse assay medium ( D5030 , Sigma-Aldrich ) supplemented with 10 mM glucose ( Sigma-Aldrich ) , 1 mM sodium pyruvate ( Gibco ) , and 2 mM L-glutamine ( Sigma-Aldrich ) . OCRs were monitored before and after adding the following mitochondrial inhibitors: 2 µM oligomycin ( complex V inhibitor ) , 1 µM carbonyl cyanide 3-chlorophenylhydrazone ( CCCP , uncoupler of oxidative phosphorylation ) , and 1 µM antimycin A ( complex III inhibitor ) ( all Sigma-Aldrich ) . OCRs were normalized to the amount of protein in each sample using a bicinchoninic acid ( BCA ) assay ( Thermo Fisher ) according to the manufacturer’s instructions . For HCT116 cells , OCRs were further normalized to the mtDNA content to account for differences in mitochondrial content . To overexpress PUM2 in MEFs , the same FLAG-Pum2 cDNA that was used for generating the transgenic mouse was cloned into a pBROAD3 vector ( Invivogen ) , in which the Rosa26 promoter was replaced by a strong CAG promoter ( pCAG-FLAG-Pum2 ) . A pcDNA3-EGFP vector was used for control transfections . FLAG-PUM2 and EGFP were transfected into immortalized Norad+/+ MEFs using 2 . 5 µg of plasmid DNA and Lipofectamine 3000 ( Invitrogen ) according to the manufacturer’s protocol . In brief , 100 × 103 cells were seeded into -well plates and incubated overnight . The next day , cells were transfected with either pCAG-FLAG-Pum2 or pcDNA3-EGFP and incubated again overnight . The following day , cells were collected and re-plated for a second transfection , performed identically . After overnight incubation , cells were seeded for Seahorse analysis , as described above . PUM2 overexpression was assessed after the second transfection using western blot . A comprehensive description of the RNA-seq and eCLIP analysis including the use of software is provided in the respective sections . The significance of the cumulative distribution functions was calculated using the Kolmogorov-Smirnov test and plotted in R . For all other analyses , statistical significance was analyzed using Prism 7 ( GraphPad Software ) . Student’s t tests or log-rank tests ( for survival and phenotype incidence ) were used to determine statistical significance , which is indicated as p≤0 . 05 , p≤0 . 01 , p≤0 . 001 , **p≤0 . 0001 . Data are presented as mean ± SD in all figures except Figure 5—figure supplement 2D and Figure 7—figure supplement 3C ( left graph ) where the data are presented as mean ± SEM . The numbers of replicates are stated in the figure legends . RNA-seq and eCLIP data has been deposited in the Gene Expression Omnibus ( GEO ) at NCBI ( Accession numbers GSE121684 , GSE121688 , and GSE125539 ) . Data is available for download via the following links: https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE121684 https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE121688 https://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE125539	Only a tiny portion of our genetic material contains the information required to create proteins , the workhorses of the body . The rest of our DNA , however , is not useless: some of it can be transcribed to create molecules known as non-coding RNAs , which are increasingly scrutinized by scientists . For example , a non-coding RNA called NORAD acts as a guardian of the genome by reducing the activity of a protein named PUMILIO . Without NORAD , PUMILIO becomes overactive , and this causes problems as genetic information is split between two ‘daughter cells’ when a cell divides . Defects in the amount of genetic material in cells have been linked with faster aging in animals . In addition , some studies suggest that as animals get older , the levels of NORAD in the body decrease , while the levels of PUMILIO increase . However , the precise role that NORAD may play in aging remains unclear . To address this question , Kopp et al . engineered mutant mice that lack Norad ( the mouse equivalent of human NORAD ) and carefully monitored how they grew and developed . The animals looked normal at birth , but they seemed to age faster: for instance , their fur became thin and gray , and their brains developed age-related abnormalities much sooner than normal mice . At the level of individual cells , losing Norad was also associated with problems often seen in old age . The mutant animals were more likely to have incorrect amounts of genetic information in their cells , and they had defects in the cell compartments that create the energy necessary for life . Further experiments showed that these issues were driven by PUMILIO being hyperactive . Overall , the work by Kopp et al . reveal that the non-coding RNA Norad is essential to keep PUMILIO activity in check and to prevent problems associated with aging from appearing in young animals . Further studies are now needed to take a closer look at how NORAD and other non-coding RNAs keep us healthy .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "genetics", "and", "genomics" ]	2019	PUMILIO hyperactivity drives premature aging of Norad-deficient mice
Existing in vitro models of human skeletal muscle cannot recapitulate the organization and function of native muscle , limiting their use in physiological and pharmacological studies . Here , we demonstrate engineering of electrically and chemically responsive , contractile human muscle tissues ( ‘myobundles’ ) using primary myogenic cells . These biomimetic constructs exhibit aligned architecture , multinucleated and striated myofibers , and a Pax7+ cell pool . They contract spontaneously and respond to electrical stimuli with twitch and tetanic contractions . Positive correlation between contractile force and GCaMP6-reported calcium responses enables non-invasive tracking of myobundle function and drug response . During culture , myobundles maintain functional acetylcholine receptors and structurally and functionally mature , evidenced by increased myofiber diameter and improved calcium handling and contractile strength . In response to diversely acting drugs , myobundles undergo dose-dependent hypertrophy or toxic myopathy similar to clinical outcomes . Human myobundles provide an enabling platform for predictive drug and toxicology screening and development of novel therapeutics for muscle-related disorders . Development of human in vitro systems for basic biological studies and drug discovery is motivated by the need to improve outcomes in human patients and alleviate ethical considerations demanding a reduction in the use of animals ( Dambach and Uppal , 2012; Bhatia and Ingber , 2014 ) . While significant progress has been made towards predictive in vitro models for liver , lung , and cardiac tissues ( Bhatia and Ingber , 2014 ) , a functional model of human skeletal muscle has not been described . This is of particular concern as there are a wide range of metabolic , neuromuscular , and dystrophic disorders involving skeletal muscle that are under investigation and still lacking therapies . Skeletal muscle is also central to diseases with high societal impact and those that do not have adequate animal models , including diabetes , obesity , and different dystrophies . Furthermore , through secretion of contraction-dependent myokines , skeletal muscle has been strongly implicated in organ–organ interactions including processes as diverse as cognition , inflammation , cancer , and aging ( Pedersen and Febbraio , 2012 ) . The need for an accurate preclinical model of human skeletal muscle was exemplified by the market withdrawal of cerivastatin that was well tolerated in mice but caused fatal rhabdomyolysis in humans ( von Keutz and Schluter , 1998; Thompson et al . , 2006 ) . Expansion of primary human myoblasts and formation of myotubes in two-dimensional ( 2D ) systems is well known , however , these cultures are difficult to maintain over long times , lack the architecture of native muscle , and require complex media components to initiate spontaneous contractions ( Blau and Webster , 1981; Eberli et al . , 2009; Guo et al . , 2014 ) . The contractile force of single , in vitro cultured human myofibers can be measured ( Smith et al . , 2014 ) , though such a system is limited by its inability to investigate biochemical changes or cell–matrix interactions that can be critical in different pathologies including muscle dystrophies and wasting disorders ( Ciciliot et al . , 2013 ) . While three-dimensional ( 3D ) culture models of rodent skeletal muscle have measureable contractile force ( Dennis and Kosnik , 2000; Huang et al . , 2005; Hinds et al . , 2011; Juhas et al . , 2014; Vandenburgh et al . , 2008 ) and can be applied to drug testing ( Vandenburgh et al . , 2008 ) and disease modeling ( Lee and Vandenburgh , 2013 ) , in vitro 3D systems using primary human myoblasts rely on measurements of passive force ( Powell et al . , 2002; Moon du et al . , 2008; Mudera et al . , 2010 ) which is not specific to functional skeletal muscle . Here , we describe a biomimetic human skeletal muscle culture system ( ‘myobundle’ ) amenable to studies of contractile function and biochemical changes in response to a wide range of stimuli . Conditions for primary myogenic cell expansion and 3D tissue formation were optimized to reproducibly obtain contractile myobundles consisting of aligned , cross-striated myofibers and a pool of cells expressing the satellite cell marker Pax7 . In response to electrical and pharmacological stimuli , myobundles exhibited forceful contractions and calcium transients which could be non-invasively measured to track physiological responses and functional maturation over time . Reproducible functional characteristics were obtained using cells from nine different donors and one commercial source . Similar to clinical outcomes in humans , when pharmaceutically challenged , myobundles experienced enhanced contractile performance in response to a steroid-like substance , underwent autophagic myopathy following administration of an anti-malarial agent , and exhibited statin-induced weakness and lipid accumulation . Myogenic cells were isolated from human muscle biopsies and expanded for 3–5 passages , when they contained a significant fraction of muscle precursors positive for desmin and MyoD ( Figure 1—figure supplement 1 ) . Engineered human skeletal muscle ‘myobundles’ were generated using a hydrogel molding technique ( Figure 1A , Figure 1—figure supplement 2 ) we developed for rodent cells ( Hinds et al . , 2011; Juhas et al . , 2014 ) . Following hydrogel compaction for 3–5 days , low serum media was applied to induce myofiber formation and differentiation . After an additional 3–5 days , the myobundles began to spontaneously twitch ( Video 1 ) , which was previously reported only in rodent 3D muscle constructs ( Dennis and Kosnik , 2000 ) . After 2-week culture , the myobundles contained densely packed and aligned myofibers embedded in a laminin-rich matrix ( Figure 1B ) and surrounded at the periphery by vimentin+ fibroblasts ( Figure 1—figure supplement 3A–C ) . Mature structure of the myofibers was evident by the expression of myosin heavy chain ( MYH ) , sarcomeric alpha-actinin ( SAA ) cross-striations , and multiple myogenin+ nuclei ( Figure 1C–E and Figure 1—figure supplement 2B–C ) . Of functional importance , acetylcholine receptors , which are necessary for neuromuscular junction formation , were present at the myofiber surface ( Figure 1F ) . While the majority of expanded myogenic cells fused to form myofibers , a fraction of cells continued to express the satellite cell marker Pax7 ( Figure 1G ) , suggesting regenerative capacity as described in a rat culture model ( Juhas et al . , 2014 ) . With time in culture , structural maturation of myobundles was evident from the progressive increase in myofiber diameter ( 13 . 5 ± 1 . 5 µm and 21 . 8 ± 2 . 8 µm at 1 and 4 weeks of culture , Figure 1H , Figure 1—figure supplement 3D ) and expression of the muscle-specific proteins ( MYH , SAA , and muscle creatine kinase ( MCK ) , Figure 1I ) , while myofiber length and myonuclei number ( 524 ± 70 and 7 ± 3 . 6 , respectively , at 3 weeks of differentiation ) remained relatively steady with time of culture ( Figure 1—figure supplement 4 ) . 10 . 7554/eLife . 04885 . 003Figure 1 . Structure and cellular composition of myobundles . ( A ) Human myogenic precursors were cast within a fibrin/matrigel matrix in PDMS molds and anchored to nylon frames . Once compacted , frames with myobundles were removed for free-floating culture . ( B ) F-actin+ myofibers shown within 2-week myobundles are aligned and surrounded by laminin . ( C ) Transverse myobundle cross-section showing dense , uniformly distributed myosin heavy chain ( MYH ) expressing myofibers . ( D–F ) Aligned myofibers within myobundle show striated pattern of the contractile protein sarcomeric α-actinin ( SAA ) ( D ) , myogenin ( MyoG ) positive nuclei ( E ) , and bungarotoxin-labeled acetylcholine receptors ( AChR ) ( F ) . ( G ) Pax7+ cells ( arrows ) are found abutting myofibers suggesting regenerative potential . ( H ) Myofiber diameter increases with time in culture , with significant enhancement at 3 and 4 weeks vs 1 week ( p < 0 . 05 , N = 4 donors , n > 10 myofibers per myobundle ) . ( I ) Structural maturation is also evident from increased expression of muscle markers MYH , SAA , and muscle creatine kinase ( MCK ) . Scale bars: ( B–F ) scale = 50 µm , ( G ) scale = 25 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 00310 . 7554/eLife . 04885 . 004Figure 1—figure supplement 1 . Myogenicity of donor cells during expansion . ( A ) Expanded donor cells at passage three still express the muscle precursor markers desmin and ( B ) MyoD . ( C ) After switch to low serum media in 2D , myogenic cells fuse into myotubes and express myogenin ( MyoG ) and sarcomeric alpha-actinin ( SAA ) . Scale bars: ( A ) 50 µm , ( B and C ) 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 00410 . 7554/eLife . 04885 . 005Figure 1—figure supplement 2 . Schematic of myobundle fabrication . ( A ) Machined teflon masters were used to generate multiple replicas of PDMS molds . The PDMS molds contain an outer ridge that fit laser-cut frames made of porous Cerex material . ( B ) Separate solutions containing myogenic cells and hydrogel proteins were prepared on ice and mixed immediately prior to pipetting into the PDMS mold with frame . ( C ) Images depicting the appearance of the myobundles during the course of culture . At far left , the cell/hydrogel solution is gelled at 37°C in a tissue culture incubator for 30 min . Following gelation , culture media is added to the well . Within 4 days , the myobundles compact and the edges come away from the PDMS mold . The compacted myobundles are removed from the mold and cultured free-floating . Scale bar = 5 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 00510 . 7554/eLife . 04885 . 006Figure 1—figure supplement 3 . Characterization of myobundle architecture . ( A ) Representative composite image of a myobundle consisting of aligned , F-actin+ myofibers surrounded by a layer of vimentin+ fibroblasts on the outer surface . ( B and C ) Hematoxylin and eosin stain at lower ( B ) and higher ( C ) magnification show uniform density of aligned , multinucleated myofibers at 3 weeks of culture . ( D ) F-actin+ immunostaining shows increased myofiber diameter with time in culture . Scale bars: ( A ) 500 µm , ( B ) 200 µm ( C and D ) = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 00610 . 7554/eLife . 04885 . 007Figure 1—figure supplement 4 . Characterization of myofiber length and myonuclei number . ( A ) Representative composite image of a 3-week differentiated myobundle formed using 5% GFP expressing myogenic cells to visualize individual myofibers . Scale bar = 500 µm . ( B ) The average length of GFP+ myofibers as a function of differentiation time ( 20–50 myofibers per bundle , n = 4–6 myobundles per time point ) . ( C ) Histogram of myonuclei number per GFP+ myofiber in 3-week myobundles with average and median myonuclei numbers of 7 ± 3 . 6 and 6 , respectively ( n = 127 myofibers from 6 myobundles ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 00710 . 7554/eLife . 04885 . 008Video 1 . Spontaneous contractions of human myobundles . Following 3–5 days of differentiation within the hydrogel construct , myofibers began spontaneously contracting . These contractions typically last for a few days , and are rarely seen beyond 2 weeks following differentiation . Video is shown in real time for 26 s duration and at field of view of 2 × 1 . 5 mm then 0 . 8 × 0 . 6 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 008 The amplitude of induced contractile force by electrical or chemical stimulation is a key parameter used to evaluate skeletal muscle function both in vivo and ex vivo on isolated muscle fibers ( Fuglevand et al . , 1999; Bottinelli and Reggiani , 2000 ) . To optimize contractile force output of myobundles , myogenic cells were expanded in media containing either bFGF ( Ham et al . , 1988 ) or EGF ( Cheng et al . , 2014 ) . Despite comparable myoblast purity and myofiber formation in 2D culture , myobundles made of EGF-expanded cells had superior contractile function ( Figure 2—figure supplement 1 ) . In addition to spontaneous contractions , myobundles also contracted in response to electrical stimulation ( Video 2 ) and , similar to native muscle ( Rassier et al . , 1999 ) , exhibited stronger contraction with an increase in stimulation frequency and myobundle length ( Figure 2A–B , Figure 2—figure supplement 2 ) . In concert with observed structural maturation , amplitudes of twitch and tetanus force in myobundles increased over 4 weeks in culture ( Figure 2C ) , while twitch kinetics remained unchanged ( Figure 2D ) . 10 . 7554/eLife . 04885 . 009Video 2 . Stimulated contractions of human myobundles . Myobundles respond to electrical stimulation by forceful contraction . Here , a myobundle pair is contracting in concert with 1 Hz electrical stimulation with enough force to bend the frame on which it is attached . Video is shown in real time for 17 s duration and at field of view of 25 × 25 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 00910 . 7554/eLife . 04885 . 010Figure 2 . Contractile function of myobundles . ( A ) Representative contractile force traces of a 3-week myobundle showing fusion of individual twitches into a stronger tetanic contraction induced by increased stimulation frequency . ( B ) Representative increase in both contractile ( active ) force and passive tension with increase in myobundle length for one donor at 3 weeks in culture ( n = 3 myobundles ) . ( C ) Twitch and tetanus forces increase over time in culture with significant enhancement at 4 weeks vs 1 week ( p < 0 . 05 , n = 4 myobundles ) . ( D ) Kinetics of twitch rise and relaxation do not vary over 4 weeks in culture ( n = 4 myobundles ) . ( E ) Specific twitch and tetanus force and tetanus-to-twitch ratio for different cell sources ( D1–D9 , donors 1–9; CS , commercial source , Lonza ) . ( F ) Kinetics of twitch response for different cell sources . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 01010 . 7554/eLife . 04885 . 011Figure 2—figure supplement 1 . Optimization of myogenic cell expansion using two different media . ( A ) Myogenic cells expanded in the two different media containing either bFGF or EGF had similar fractions of MyoD+ cells . ( B ) Upon differentiation in 2D culture , these cells expressed similar levels of myogenin ( n = 3 coverslips ) ( C ) Cells expanded in EGF containing media formed myobundles with significantly higher levels of contractile force than those expanded in bFGF containing media . ( N = 2 donors , n = 4 myobundles , p < 0 . 05 . ) DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 01110 . 7554/eLife . 04885 . 012Figure 2—figure supplement 2 . Force-frequency relationship of myobundles . Contractile force increases with stimulation frequency . ( n = 4 myobundles; p < 0 . 05 vs 1 Hz; #p < 0 . 05 vs 1 Hz and 5 Hz . ) DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 012 To evaluate the robustness of the developed methodology , we expanded and utilized cells from nine donor muscle samples ( obtained by needle biopsy or surgical waste ) and one commercially available myoblast source ( Lonza ) . Expanded cells from all ten sources formed functional myobundles that contracted in response to electrical stimulation with an average specific force of 2 . 1 ± 0 . 9 mN/mm2 and 7 . 0 ± 2 . 2 mN/mm2 for twitch and tetanus , respectively ( Figure 2E ) . The average tetanus force was similar to values measured in fetal human muscle ( Racca et al . , 2013 ) and an order of magnitude lower than values reported for adult muscle ( Racca et al . , 2013; Cheng et al . , 2014 ) , while average tetanus-to-twitch ratio ( 3 . 5 ± 0 . 8 , Figure 2E ) was within the normal adult range ( Cheng et al . , 2014 ) . The kinetic parameters of twitch contraction were also evaluated for each donor sample ( Figure 2F ) and were on average twofold slower than those of adult human muscle ( Fuglevand et al . , 1999 ) and comparable to those of single in vitro cultured human myotubes ( Smith et al . , 2014 ) . To expand the utility of the myobundle platform , we incorporated a capability for non-invasive real-time monitoring of calcium transients in myobundles as calcium handling is critical to normal muscle function and can be affected by pathological conditions including dystrophic disorders and malignant hyperthermia ( Berchtold et al . , 2000 ) . Expanded myogenic cells were lentivirally transduced with a calcium indicator , GCaMP6 ( Chen et al . , 2013 ) , driven by a muscle-specific promoter , MHCK7 ( Salva et al . , 2007 ) , prior to myobundle formation . As a result , robust expression of GCaMP6 in differentiated myofibers ( Figure 3A ) allowed detection of both spontaneous and electrically stimulated calcium transients in myobundles ( Figure 3B , Video 3 ) under a variety of conditions . In response to 10 Hz ( tetanic ) vs single ( twitch ) stimuli , the amplitude of calcium transients increased ( Figure 3C–D ) , as measured by normalized change in fluorescence intensity ( ΔF/F ) , similar to the increase in contractile force with tetanic stimulation . Additionally , with longer time in culture , calcium transient amplitude increased ( Figure 3D ) and correlated with the contractile force amplitude measured in the same bundles ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 04885 . 013Figure 3 . Calcium handling of myobundles . ( A ) Myofiber-specific expression of GCaMP6 in lentivirally transduced myobundles . SAA , sarcomeric α-actinin ( scale bar = 50 µm ) . ( B ) Time course of GCaMP6 fluorescence during a single electrically stimulated twitch ( scale bar = 200 µm ) . ( C ) Representative fluorescence traces from 1 Hz and 10 Hz stimulations of 2-week old myobundles . ( D ) Amplitude of electrically stimulated calcium transient increases with time of culture and myobundle maturation ( p < 0 . 05 vs 1 week , n = 4 myobundles ) . ( E ) Representative fluorescence trace of acetylcholine ( ACh , 100 mM bolus ) stimulated calcium release in a 2-week myobundle . ( F ) ACh receptor blocker tubocurarine ( 25 µM ) specifically and significantly reduces ACh induced calcium release without affecting electrically stimulated calcium transients ( p < 0 . 05 , n = 5 myobundles ) . Note that amplitude of Ach-induced calcium release is similar to that of calcium transient induced by tetanic ( 10 Hz ) electrical stimulation . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 01310 . 7554/eLife . 04885 . 014Figure 3—figure supplement 1 . Correlation of contractile force and calcium transients . Each point represents a single bundle calcium transient plotted against the corresponding force , either twitch or tetanus ( at 10 Hz ) . Data was obtained from myobundles during 4 weeks in culture prepared from the same pool of MHCK7-GCaMP6 transduced myogenic precursors . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 01410 . 7554/eLife . 04885 . 015Figure 3—figure supplement 2 . Caffeine induced calcium transients . ( A ) A bolus of caffeine of different concentrations was added to the bath during video imaging and resulted in an increase in relative fluorescence amplitude ( ΔF/F ) . ( B ) ΔF/F at 30 s following caffeine administration was calculated and positively correlated with caffeine concentration . ( Donor A , n = 3 myobundles; Donor B n = 4 myobundles; p < 0 . 05 vs 20 mM . ) DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 01510 . 7554/eLife . 04885 . 016Figure 3—figure supplement 3 . Myobundle response to acetylcholine . ( A ) Myobundles exhibited a similar amplitude of calcium release in response to acetylcholine ( Ach ) throughout 4 weeks in culture ( n = 4 ) . ( B ) Representative trace of contractile force for a 3-week myobundle in response to a bolus of acetylcholine . ( C ) The amplitude of acetylcholine induced contractile force was similar to that induced by tetanic electrical stimulation . Incubation of myobundles with the ACh receptor blocker tubocurarine reduced acetylcholine induced contractile force without affecting electrically stimulated contraction at 3 weeks in culture ( n = 5 , p < 0 . 05 ) . ( D ) Amplitudes of calcium and contractile force responses to ACh are similar to those during tetanic electrical stimulation in myobundles from different donors . ( A , n = 5 myobundles; B , n = 3 myobundles . ) DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 01610 . 7554/eLife . 04885 . 017Video 3 . GCaMP6 reported calcium release of human myobundles . Myobundles formed using myogenic precursors that were lentivirally transduced with a GCaMP6 calcium reporter contain myofibers that produce fluorescence signal in response to calcium release . Here , myobundles show calcium release in response to both twitch ( 1 Hz ) and tetanus ( 10 Hz ) electrical stimulation . Video is shown in real time for 36 s duration and at field of view of 500 × 500 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 017 We further tested the functionality of calcium-handling machinery in myobundles by biochemical stimulation with caffeine and acetylcholine ( ACh ) . By opening of ryanodine receptors , caffeine is known to generate concentration-dependent calcium release and contraction in skeletal muscle ( Moulds and Denborough , 1974; Berchtold et al . , 2000 ) , as was observed in myobundles from different donors ( Figure 3—figure supplement 2 , Video 4 ) . ACh is released at the neuromuscular junction to stimulate muscle contraction via opening of ligand-gated Na/K-permeable channels and voltage-gated Ca channels , while ACh receptors are a target of different muscle relaxants and toxins ( Kalamida et al . , 2007 ) . The degree of calcium release in response to a bolus of 10 mM ACh ( Figure 3E , Video 5 ) was comparable to that from electrically stimulated tetanus ( Figure 3F ) and unchanged throughout the entire culture period ( Figure 3—figure supplement 3A ) . Tubocurarine , a muscle relaxant , blocked ACh ( but not electrically ) induced calcium transients ( Figure 3F ) and contractions ( Figure 3—figure supplement 3B–C ) , mimicking the neuromuscular block observed in vivo ( Secher et al . , 1982 ) . 10 . 7554/eLife . 04885 . 018Video 4 . Caffeine induced calcium release in human myobundles . Similar to human skeletal muscle , application of caffeine to myobundles induces calcium release via the ryanodine receptors . This lease to an increase in fluorescence from the GCaMP calcium reporter . Video is shown in real time for 66 s duration and at field of view of 2 × 2 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 01810 . 7554/eLife . 04885 . 019Video 5 . Acetylcholine induced calcium release from human myobundles . Function of myobundle acetylcholine receptors was confirmed by GCaMP6 detected calcium release in response to a bolus of 10 mM acetylcholine . Myobundle response to acetylcoline was significantly blocked by the muscle relaxant tubocurarine similar to that observed clinically in human skeletal muscle . Video is shown in real time for 75 s duration and at field of view of 2 × 2 mm . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 019 We evaluated the potential application of myobundles as a preclinical test bed by studying their responses to three classes of pharmaceutical agents with a broad range of known effects . Statins are widely prescribed to prevent coronary artery disease , however even at normal doses some of them can induce significant myopathic weakness and rhabdomyolysis after as early as 2 weeks of use ( Dobkin , 2005; Thompson et al . , 2006 ) . We tested the effects of lovastatin and cerivastatin at their clinically-relevant dose ranges ( 100-fold higher for lovastatin due its lower bioavailability and bioactivity [Kantola et al . , 1998; Shitara and Sugiyama , 2006] ) encompassing both maximum therapeutic blood serum concentrations and higher doses known to accelerate myopathic induction ( Dobkin , 2005 ) . In our studies , 2-week application of each statin was well tolerated in the myobundles derived from two of three donors at their respective therapeutic doses , while higher doses induced significant contractile weakness in the myobundles from all donors ( Figure 4A–B ) . Unlike human myobundles that replicated clinical response , murine engineered muscle tissues in previous studies exhibited a sharp decrease in contractile function even at the lowest statin dose tested ( Vandenburgh et al . , 2008 ) . Human myobundles also recapitulated the histopathology of statin-associated myopathy characterized by dose-dependent lipid accumulation ( Thompson et al . , 2006 ) ( Figure 4C ) . 10 . 7554/eLife . 04885 . 020Figure 4 . Pharmacological validation of myobundles . ( A and B ) 2-week application of cerivastatin ( A ) and lovastatin ( B ) at increasing doses significantly reduced tetanus force , normalized to untreated or vehicle treated ( DMSO for Lovastatin ) control ( n = 4 myobundles per donor ) . ( C ) Accumulation of lipids in myobundles , evaluated by Oil Red O stain , was absent from controls , moderate at lower concentrations , and considerable at higher concentrations of both statins ( scale bar = 50 µm ) . ( D–F ) 1-week exposure of myobundles to chloroquine resulted in dose-dependent decrease of contractile force ( n = 4 myobundles per donor ) ( D ) as well as increased expression of the autophagic pathway marker LC3B-II and decreased expression of contractile protein sarcomeric α-actinin ( SAA ) ( E–F , n = 4 myobundles per donor ) . ( A–F ) ( p < 0 . 05 vs 0 µM , #p < 0 . 05 vs all other concentrations ) . ( G ) Acute ( 30-min ) and ( H ) chronic ( 2-week ) application of clenbuterol to myobundles ( shown in different donors ) results in a dose-dependent increase in contractile force with peak effects observed at 1 µM ( acute ) and 0 . 1 µM ( chronic ) and significant reduction in force generation observed at 100 µM ( acute , n = 3 myobundles; chronic , n = 4 myobundles ) . ( I ) Chronic administration of 0 . 1 µM Clenbuterol induced hypertrophy of myofibers as evident from a rightward shift in their diameter distribution and significant increase in the average myofiber diameter ( untreated , 15 . 7 ± 0 . 3 µm vs 0 . 1 µM clenbuterol , 17 . 1 ± 0 . 6 µm , p < 0 . 05 , n ≥ 55 myofibers per myobundle , pooled for 3 myobundles ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 02010 . 7554/eLife . 04885 . 021Figure 4—figure supplement 1 . Biochemical responses of human 2D myotube and 3D myobundle cultures to chloroquine . Human myogenic cells from a single donor were differentiated on 2D Matrigel coated dishes ( black ) or in 3D myobundles ( white ) then treated for 1 week with varying doses of chloroquine . The biochemical response of 2D and 3D muscle models were evaluated by western blot ( A ) and shown to be similar for ( B ) the dose-depended accumulation of LC3B-II and the reduction of contractile proteins ( C ) sarcomeric alpha actinin ( SAA ) and ( D ) myosin heavy chain ( MYH ) ( n = 4 2D wells or myobundles , #p < 0 . 05 vs control and 1 µM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 02110 . 7554/eLife . 04885 . 022Figure 4—figure supplement 2 . Improved myobundle function following clenbuterol treatment . ( A ) Clenbuterol-induced increase in force was reproduced among multiple donors following 2 week treatment ( n = 4 myobundles per donor , *p < 0 . 05 ) . Dose response for donor A is shown in Figure 4H . ( B ) Clenbuterol increased myofiber diameter , visualized by F-actin staining . Scale bar = 100 µm . Quantification of these immunostainings is shown in Figure 4I . DOI: http://dx . doi . org/10 . 7554/eLife . 04885 . 022 We also challenged myobundles with the anti-malarial agent chloroquine for 1 week to evaluate its effects on autophagy , a conserved lysosomal pathway in both physiological and pathological conditions ( Shintani and Klionsky , 2004 ) . With increasing doses of chloroquine , myobundles from all donors exhibited a decrease in contractile force generation ( Figure 4D ) , which was associated with the autophagic buildup marked by conversion of LC3B-I to LC3B-II and a decrease in the expression of the contractile protein SAA ( Figure 4E–F ) . These outcomes were consistent with autophagic-related myopathy seen in humans treated with chloroquine ( Shintani and Klionsky , 2004; Lee et al . , 2012 ) . Similar biochemical responses to chloroquine including accumulation of LC3B-II and reduction of contractile proteins was also observed in 2D cultures ( Figure 4—figure supplement 1 ) . Clenbuterol is a β2-adrenergic agonist with both short and long-term concentration-dependent effects on muscle , improving contractile force and hypertrophy at low concentrations , while inducing apoptosis and necrosis at high concentrations ( Burniston et al . , 2006 ) . Clenbuterol and other β-agonists are under investigation for prevention of muscle wasting ( Ryall and Lynch , 2008 ) , however , species-dependent differences in their anabolic effects limit the usefulness of preclinical animal studies ( Chen and Alway , 2000 ) . In our studies , both acute and chronic application of clenbuterol induced an in vivo-like biphasic dose-dependent effect on contractile force generation of myobundles ( Figure 4G–H ) with the typical anabolic response and stronger contractions at 0 . 1 µM and diminished contractile response above 1 μM . The observed positive inotropic effect of 0 . 1 µM clenbuterol was partially attributed to myofiber hypertrophy ( Figure 4I ) and was confirmed in myobundles from multiple donors , resulting in an average force increase of 43 . 2 ± 10 . 8% ( Figure 4—figure supplement 2 ) . Collectively , these results confirm the functional similarity of myobundles to human muscle tissue and validate their potential use in the future predictive studies of muscle physiology . We described the development and validation of the ‘myobundle’ , a biomimetic human skeletal muscle culture platform for clinically relevant in vitro studies of muscle physiology and drug development . Myobundles recapitulate key functional aspects of human skeletal muscle including a functioning contractile apparatus , responsive acetylcholine and β2-adrenergic receptors , and physiological calcium handling , all of which are involved in pharmacological side effects in humans ( Bowes et al . , 2012 ) . Long-term electrical and chemical responsiveness of myobundles allow for both acute and chronic physiological and pharmacological tests . Reproducibility and robustness of the system were demonstrated using biopsies from multiple donors and a commercial cell source . Correlated force generation and calcium transient responses recorded via the use of genetically encoded calcium indicators ( Chen et al . , 2013 ) enabled continuous optical monitoring of the relationship between stimuli and functional effects , thus bridging a significant gap in current testing methods ( Dambach and Uppal , 2012 ) . Existing methods to measure contractile function of human muscle in vitro rely on acute , single-time use of intact muscle fibers isolated from patient biopsies ( Bottinelli and Reggiani , 2000 ) . While 2D and 3D cultures can be used to form de novo muscle fibers from human myogenic cells , existing methods fail to reproduce a comprehensive range of myofiber physiological responses , such as twitch , tetanus , and chemically induced contractions . Compared to previous 3D culture studies ( Powell et al . , 2002; Mudera et al . , 2010 ) , a relatively high cell density , specific hydrogel and media compositions , and dynamic culture conditions ( Juhas and Bursac , 2014 ) used in our system may have all contributed to the robust formation of functional human engineered muscle . Under these conditions , the ability to generate large numbers ( >1000 ) of contractile myobundles from a single donor biopsy allowed us to perform traditional physiological and biochemical measurements in both acute and chronic settings and for multiple testing compounds and conditions . Electrically induced calcium transients and contractions ( twitch and tetanus ) as well as physiological responses to increase in muscle length and stimulation frequency ( Rassier et al . , 1999; Cheng et al . , 2014 ) were reproducibly recorded in myobundles from ten donors . The specific blockade of acetylcholine-induced but not electrically-induced calcium release by the muscle relaxant and acetylcholine receptor blocker tubocurarine mimicked responses seen in human studies ( Secher et al . , 1982 ) . Along with dose-dependent increase in calcium transient amplitude by caffeine , these experiments demonstrate that myofibers formed within the 3D myobundle culture environment exhibited intact excitation-contraction coupling and physiological responsiveness to both chemical and electrical stimuli . While repeated non-invasive interrogation of myobundle function was limited to calcium imaging , integration of smaller size myobundles with high-throughput force testing assays should be feasible as demonstrated for mouse cells ( Vandenburgh et al . , 2008 ) . The utility of myobundles as a preclinical drug testing platform was evaluated by measuring contractile and biochemical responses to statins , chloroquine , and clenbuterol . Statin myopathy is a common side effect that has been reported for all currently available statins ( Dobkin , 2005; Thompson et al . , 2006 ) . Similar to clinical reports , human myobundles showed higher sensitivity to cerivastatin than lovastatin ( Shitara and Sugiyama , 2006 ) and at excessive statin concentrations displayed progressive weakness and lipid accumulation , suggestive of equivalent mechanisms of action in vitro and in vivo . The use of myobundles allowed direct comparison of similar pharmaceuticals on the same patient or cohort , previously recommended for but unavailable for statins due to the variations among clinical trials and underreporting of symptoms ( Dobkin , 2005; Thompson et al . , 2006 ) . In response to an anti-malarial agent , chloroquine , myobundles showed induction of autophagic myopathy also observed in native muscle ( Shintani and Klionsky , 2004 ) , thus providing a potential functional screen for non-toxic modulators of autophagy . We also tested the acute and chronic responses of myobundles to β2-adrenergic agonist clenbuterol and observed myofiber hypertrophy and increased contractile strength at low clenbuterol doses followed by muscle weakness at higher doses , consistent with previous animal and human studies ( Ryall and Lynch , 2008 ) . Currently , binding affinity to β2-adrenergic receptors is one of the standard tests for drug specificity ( Bowes et al . , 2012 ) and is also a potential target for therapies in muscle wasting disorders ( Ryall and Lynch , 2008 ) . Overall , these results suggest that myobundles closely mimic the functional responses of native human muscle through multiple signaling pathways and could provide a pre-clinical assay for predictive screening of novel therapeutics for a broad range of muscle-related disorders . Our in vitro model of human skeletal muscle provides a tool for improved predictive pharmacological testing and a potential alternative to costly animal studies . Non-destructive , real-time measurement of function such as calcium handling shown here could be combined with other optically-based assays ( Kleinstreuer et al . , 2014 ) to elucidate mechanisms of drug action . The ability to measure and quantify functional endpoints in myobundles in a population- or patient-specific manner allows construction of pharmacological time- and dose–response curves previously not available for human skeletal muscle . The myobundles may be integrated with other established human micro-organ systems such as liver or heart for more predictive body-on-chip toxicology studies ( Bhatia and Ingber , 2014 ) . Functional acetylcholine receptors within myobundles are integral to studies involving the neuromuscular junctions and necessary for potential implantation of such tissues to repair muscle dysfunction or loss . Eventual applications of myobundle platform using patient-derived cells to model functional deficits observed in different muscle pathologies may allow development of more efficacious therapies and safe translation to clinics . Human skeletal muscle samples were obtained through standard needle biopsy or surgical waste under Duke University IRB approved protocols . Nine donor samples were expanded by outgrowth similar to methods previously described ( Blau and Webster , 1981 ) . Briefly , muscle samples were minced , washed in PBS , and enzymatically digested in 0 . 05% trypsin for 30 min . Muscle was collected by centrifugation , pre-plated for 2 hr , and transferred to a matrigel ( BD Biosciences , San Jose , CA ) coated flask for attachment . Cells were expanded in skeletal muscle growth media containing low glucose DMEM ( Gibco Life Technologies , Grand Island , NY ) , supplements purchased from Lonza , Basel , CH ( EGF , fetuin , dexamethasone , and gentamicin without insulin ) , and supplemented with 10% fetal calf serum as previously described ( Cheng et al . , 2014 ) . A second growth media containing 5 ng/ml bFGF and 20% fetal calf serum was used during optimization as it was previously shown to improve expansion of myogenic cells ( Ham et al . , 1988 ) . Myogenic cells were either cryopreserved in 90% growth medium with 5% fetal calf serum and 5% DMSO at passage 1 or 2 then used at passage 3–5 for the generation of myobundles or staining . A sample of primary human skeletal myoblasts from additional donor was purchased from Lonza for comparison . For calcium imaging studies , expanded myogenic cells were transduced with a lentiviral vector encoding the fluorescent calcium reporter GCaMP6 ( Chen et al . , 2013 ) driven by a myosin heavy chain-creatine kinase promoter MHCK7 ( Salva et al . , 2007 ) for muscle specific expression . For the measurements of myofiber length and nuclei number , 5% of myogenic cells used for myobundle formation were transduced with a lentiviral vector encoding MHCK7 driven GFP ( Li et al . , 2011 ) . This allowed the visualization and measurement of individual GFP+ myotubes within myobundles using immunostaining and confocal microscopy . Myobundles were formed by modifying our previously published methods for engineered rodent muscle tissues ( Hinds et al . , 2011; Juhas et al . , 2014 ) ( Figure 1—figure supplement 2 ) . Expanded myogenic cells were dissociated in 0 . 025% trypsin-EDTA to a single cell suspension and encapsulated in a fibrinogen ( Akron , Boca Raton , FL ) and matrigel solution on laser cut Cerex frames ( 9 . 2 × 9 . 5 mm outer dimensions , 6 . 8 × 8 . 3 mm inner dimensions ) within PDMS molds ( cast from Teflon masters and pretreated with pluronic ) at 15 × 106 cells/ml ( 7 . 5 × 105 cells per myobundle ) . Specifically , a cell solution ( 7 . 5 × 105 cells in 17 . 2 µl media per bundle + 2 µl of 50 unit/ml thrombin in 0 . 1% BSA in PBS [Sigma , St . Louis , MO] ) and a gelling solution ( 11 µl media + 10 µl Matrigel + 10 µl of 20 mg/ml Fibrinogen in DMEM ) were prepared in separate vials on ice for up to six myobundles per vial . Gelling solution was added to the cell solution and mixed thoroughly then each bundle was individual pipetted within the PDMS mold and onto the frame . The cell/hydrogel mixture was polymerized for 30 min at 37°C followed by incubation in growth media containing 1 . 5 mg/ml 6-aminocaproic acid ( ACA , Sigma ) . Myobundles were kept in growth media during gel compaction ( 3–5 days ) and then switched to low glucose DMEM with 2% horse serum ( Hyclone , Logan , UT ) , 2 mg/ml ACA and 10 µg/ml insulin ( Sigma ) . Frames were removed from molds at the time of switch to low serum medium and cultured dynamically in suspension for an additional 1–4 weeks . Starting from a 50 mg donor biopsy , typical cell expansion for 5 passages can allow generation of at least 1000 myobundles with a total mass of >5 g , representing a >100-fold amplification of muscle mass when going from native to engineered tissue system . All drugs were purchased from Sigma . Clenbuterol hydrochloride , chloroquine phosphate , and cerivastatin sodium salt hydrate were prepared at 1000× stock solutions in PBS ( control ) and sterile-filtered for use . Lovastatin was prepared as a 10 , 000× stock solution in DMSO in which case DMSO was used as vehicle control . Drugs studies in myobundles or 2D cultures were initiated after 1 week of differentiation . Myobundles were replenished with fresh media and drug each day to maintain drug concentration . Electrically or chemically stimulated contractile force generation in myobundles was measured using a custom force measurement set-up as previously described ( Hinds et al . , 2011; Juhas et al . , 2014 ) . Briefly , single myobundles on a frame were transferred to the bath of the force measurement set-up , maintained at 37°C . One end of the bundle was secured by a pin to an immobile PDMS block and the other end was attached to a PDMS float connected to the force transducer mounted on a computer-controlled motorized linear actuator ( Thor Labs , Newton , NJ ) . The sides of the frame were cut to allow myobundle stretch by the actuator and isometric measurement of contractile force . Initially , the myobundle was set to its baseline length using the motorized linear actuator . To assess the force-length relationship , myobundle was stretched by 2% of its culture length then stimulated by a 40 V/cm , 10 ms electrical pulse using a pair of platinum electrodes and the twitch force was recorded . At 12% stretch , 1 s long stimulations at 5 , 10 , and 20 Hz were applied and the subsequent contractile force was recorded to assess the force-frequency relationship . Contractile force traces were analyzed for peak twitch or tetanus force , time to peak twitch , and half relaxation time using a custom MATLAB program ( Source code 1 ) . For studies with acetylcholine , 60 µl of drug solution was added to the 6 ml bath at t = 5 s of recording . Myobundles expressing the MHCK7-GcaMP6 reporter were non-destructively monitored for calcium transients following differentiation . A live imaging chamber with heated enclosure was used to maintain cells in physiological conditions during recording . Bundles were placed in sterile tyrode's solution in a custom-designed glass-bottom bath containing electrodes for stimulation . Video-images were acquired using an Andor iXon camera affixed to a Nikon microscope with a FITC filter and either 4× or 10× objective . During studies with caffeine and acetylcholine , 60 µl of drug solution was added to the bath at t = 5 s of recording . Video was analyzed using Andor Solis software and relative changes in fluorescence signal were calculated by ΔF/F = ( Peak-Trough ) / ( Trough-Background ) as previously described ( Juhas et al . , 2014 ) . Cells were fixed in 4% paraformaldehyde in PBS for 10 min and myobundles were fixed in 2% paraformaldehyde in PBS overnight at 4°C . Following fixation , samples were washed in PBS then blocked in 5% chick serum with 0 . 2% Triton-X 100 . The following primary antibodies were used for tissue characterization: desmin ( SCBT , Dallas , TX , 1:200 ) , anti-GFP ( Life Technologies , 1:200 ) , laminin ( Abcam , Cambridge , MA , 1:200 ) , muscle creatine kinase ( SCBT , 1:100 ) , MyoD ( BD , 1:100 ) , myogenin ( SCBT , 1:100 ) , myosin heavy chain 1/2/4/6 ( SCBT , 1:100 ) , Pax7 ( DSHB , Iowa City , IA , 1:50 ) , sarcomeric α-actinin ( Sigma , 1:200 ) , and vimentin ( Sigma , 1:200 ) . Corresponding fluorescently labeled secondary antibodies ( 1:200 ) , α-bungarotoxin ( 1:100 ) , and phalloidin ( 1:200 ) were purchased from Life Technologies . Oil Red O staining was performed using standard protocols on cryosections of myobundles fixed in 4% paraformaldehyde . Hematoxylin and eosin stain was performed on paraffin embedded sections of 2% paraformaldehyde fixed myobundles using Harris modified hematoxylin ( Sigma ) and Eosin Y ( Sigma ) . Images were acquired using a Zeiss 510 inverted confocal microscope and analyzed using LSM Image Software . Mosaic images for fiber length measurements were generated using Mosaic J in FIJI . Cell or myobundle protein was isolated in RIPA lysis and extraction buffer with protease inhibitor ( Sigma ) . Protein concentration was determined using BCA assay ( Pierce of Thermo Scientific , Rockford , IL ) according to manufacturer's instructions . Western blot was performed using Bio-Rad Mini-PROTEAN gels and the Mini-PROTEAN Tetra cell , Mini Trans-blot module ( Bio-Rad , Hercules , CA ) . The following primary antibodies were used for detection: GAPDH ( SCBT , 1:500 ) , LC3 ( Cell Signaling , 1:200 ) , muscle creatine kinase ( SCBT , 1:200 ) , myosin heavy chain 1/2/4/6 ( SCBT , 1:200 ) , and sarcomeric alpha-actinin ( Sigma , 1:200 ) . HRP conjugated anti-mouse ( 1:20 , 000 ) and anti-goat ( 1:5000 ) antibody were purchased from Sigma , and HRP conjugated anti-rabbit was purchased from SCBT ( 1:5000 ) . Chemiluminescence was performed using Clarity Western ECL substrate ( Bio-Rad ) . Images were acquired using a Bio-Rad Chemidoc and analyzed using ImageJ . Results are presents as mean ± SD . Statistical significance was determined by unpaired t-test or one-way ANOVA with post-hoc Bonferroni–Holm test . p < 0 . 05 was considered statistically significant .	Scientists have developed realistic models of the human liver , lung , and heart that allow them to observe living tissue in the laboratory . These models have helped us to better understand how these organs work and what goes wrong in diseases that affect these organs . The models can also be used to test how new drugs may affect a particular organ without the risk of exposing patients to the drug . Efforts to develop a realistic laboratory model of human muscle tissues that can contract like real muscles have not been as successful to date . This shortcoming has potentially hindered the development of drugs to treat numerous disorders that affect muscles and movement in humans—such as muscular dystrophies , which are diseases in which people progressively lose muscle strength . Some important drugs , like cholesterol-lowering statins , have detrimental effects on muscle tissue; one statin was so harmful to muscles that it had to be withdrawn from the market . As such , it would be useful to have experimental models that would allow scientists to test whether potential drugs damage or treat muscle tissue . Madden et al . have now bioengineered a three-dimensional laboratory model of living muscle tissue made of cells taken from biopsies of several different human patients . These tissues were grown into bundles of muscle fibers on special polymer frames in the laboratory . The bioengineered muscle bundles respond to electrical and chemical signals and contract just like normal muscle . They also exhibit the same structure and signaling as healthy muscle tissue in humans . Madden et al . exposed the muscle tissue bundles to three drugs known to affect muscles to determine if the model could be used to test whether drugs have harmful effects . This revealed that the bundles had weaker contractions in response to statins and the malaria drug chloroquine , just like normal muscles do—and that this effect worsened if more of each drug was used . Madden et al . also found that a drug that strengthens muscle contractions at low doses and damages muscle at high doses in humans has similar effects in the model . As well as this model being used to screen for harmful effects of drugs before clinical trials , the technique used to create the model could be used to grow muscle tissue from patients with muscle diseases . This would help researchers and doctors to better understand the patient's condition and potentially develop more efficient therapies . Also , the technique could be eventually developed to grow healthy muscle tissue to implant in patients who have been injured .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine" ]	2015	Bioengineered human myobundles mimic clinical responses of skeletal muscle to drugs
Apoptotic cells can produce signals to instruct cells in their local environment , including ones that stimulate engulfment and proliferation . We identified a novel mode of communication by which apoptotic cells induce additional apoptosis in the same tissue . Strong induction of apoptosis in one compartment of the Drosophila wing disc causes apoptosis of cells in the other compartment , indicating that dying cells can release long-range death factors . We identified Eiger , the Drosophila tumor necrosis factor ( TNF ) homolog , as the signal responsible for apoptosis-induced apoptosis ( AiA ) . Eiger is produced in apoptotic cells and , through activation of the c-Jun N-terminal kinase ( JNK ) pathway , is able to propagate the initial apoptotic stimulus . We also show that during coordinated cell death of hair follicle cells in mice , TNF-α is expressed in apoptotic cells and is required for normal cell death . AiA provides a mechanism to explain cohort behavior of dying cells that is seen both in normal development and under pathological conditions . Apoptosis is a distinct form of programmed cell death in which cells activate an intrinsic suicide program to self-destruct . This process plays a major role in development and tissue homeostasis , and abnormal regulation of apoptosis is associated with a variety of human diseases ( Fuchs and Steller , 2011 ) . Dying cells can secrete signals that will stimulate the recruitment of phagocytes ( find-me signals ) as well as expose signals on their surface to facilitate their engulfment ( eat-me signals ) ( Lauber et al . , 2003; Ravichandran , 2003 ) . However , apoptosis has been traditionally regarded as a silent process that does not affect surrounding tissues . Only more recently has it become clear that apoptotic cells are the source of signals that can have profound effects on their neighbors . Cells that undergo apoptosis in response to stress and injury can secrete mitogenic and morphogenetic signaling proteins to stimulate growth and tissue repair ( Bergmann and Steller , 2010; Morata et al . , 2011; Greco , 2013 ) . These factors include Wnt , Dpp/Bmps and Hedgehog ( Hh ) proteins , which all play major roles in the regulation of growth and patterning during development ( Huh et al . , 2004; Perez-Garijo et al . , 2004; Ryoo et al . , 2004; Fan and Bergmann , 2008 ) . Mitogenic signaling by apoptotic cells has been reported for a diversity of animals , from Hydra , to flat worms , Drosophila and vertebrates , and it has been implicated in regeneration , wound healing and tumor growth ( Tseng et al . , 2007; Chera et al . , 2009; Bergmann and Steller , 2010; Li et al . , 2010; Pellettieri et al . , 2010; Huang et al . , 2011 ) . This mechanism appears well suited to communicate cellular loss to stem and progenitor cells in the tissue environment to stimulate proliferation and tissue repair . On the other hand , large groups of cells often undergo coordinated death during development and under conditions of severe tissue injury ( Glucksmann , 1951; Jacobson et al . , 1997 ) . Classic examples for such group suicide behavior in normal development include the elimination of the tadpod tail during amphibian metamorphosis , and the removal of interdigital membranes during digit individualization in vertebrates . In Drosophila , apoptosis plays a crucial role in several morphogenetic events , sculpting tissues and organs , removing a large number of cells in a coordinated manner and inducing cellular reorganization ( Lohmann et al . , 2002; Link et al . , 2007; Manjon et al . , 2007; Suzanne et al . , 2010; Suzanne and Steller , 2013 ) . In vertebrates , another example of communal death is the regressive phase of the hair follicle ( HF ) , which undergoes cycles of growth ( anagen ) , degeneration ( catagen ) and rest ( telogen ) ( Hardy , 1992; Fuchs , 2007 ) . In catagen , all the cells in the lower portion of the HF are eliminated by apoptosis ( Lindner et al . , 1997; Botchkareva et al . , 2006 ) . In all these cases , cell death takes place in a very rapid and highly synchronized manner . However , it is not known how this cohort behavior is achieved . Likewise , many pathological states are associated with extensive cell death , which leads to severe damage and can have grave consequences for patients . Examples include alcohol-/drug-induced liver failure , viral infection , cardiac infarction , ischemic stroke and cachexia ( Sharma and Anker , 2002; Kang and Izumo , 2003; Guicciardi and Gores , 2005; Yuan , 2009 ) . In all these pathologies , apoptosis accounts for widespread cell loss and is thought to contribute to patient mortality ( Thompson , 1995; Favaloro et al . , 2012 ) . One possible explanation for all these ‘mass suicide’ phenomena is that apoptotic cells may release signals to coordinate their ‘communal death’ . Here we investigated whether apoptotic cells are able to produce signals that can explain the coordinated behavior of groups of dying cells . We observed that massive induction of apoptosis in the posterior compartment of Drosophila wing discs caused non-autonomous apoptosis at a considerable distance in the anterior compartment . Moreover , apoptosis of cells in the anterior compartment requires signaling from apoptotic cells in the posterior compartment , indicating that apoptosis-induced-apoptosis ( AiA ) is an active phenomenon . We next explored the mechanism underlying AiA and found that apoptotic cells produce Eiger , the TNF homolog in Drosophila . Eiger activates the JNK pathway in neighboring cells and induces them to die . Finally , we examined whether AiA also occurs in vertebrates and whether it plays a physiological role for the coordinated death of groups of cells . We found that during the regressive phase ( catagen ) of the HF , apoptotic cells produce TNF-α . Inhibition of TNF-α disrupts the coordinated death of HF cells in catagen , indicating that this mechanism plays a physiological role to maintain synchronicity in the HF cycle . Taken together , these observations reveal a novel mechanism to coordinate cohort behavior of dying cells that is seen both in normal development and under pathological conditions . To reveal novel types of signaling induced by apoptotic cells we made use of ‘undead’ cells: apoptotic cells that are kept alive by the expression of the baculovirus caspase inhibitor p35 ( Hay et al . , 1994 ) . Under these conditions , cells initiate the apoptotic cascade but cannot execute cell death because the activity of effector caspases is blocked . Although apoptotic cells are normally very rapidly cleared in living tissues , undead cells persist for long times and thereby permit analysis of signaling events associated with the induction of apoptosis . We used the Gal4/UAS system for transgene expression of the pro-apoptotic gene hid and the caspase inhibitor p35 in the posterior compartment of wing imaginal discs ( Brand and Perrimon , 1993 ) . As expected , we observed ectopic Wg expression and discs with abnormal and in many cases overgrown posterior compartments due to increased cell proliferation ( Figure 1 ) . Undead cells contain high levels of cleaved caspases and were visualized by staining with activated caspase-3 antibody , which recognizes cleaved effector caspases as well as the activity of the initiator caspase Dronc ( Figure 1B ) ( Fan and Bergmann , 2010 ) . Surprisingly , we also observed large numbers of apoptotic cells in the anterior compartment ( Figure 1B ) . Under these conditions , we typically saw two large clusters of dying cells in the wing pouch . It appears that cells in this region of the wing disc are more susceptible to apoptosis , as indicated by the fact that higher rates of cell death within this region were also observed after X-irradiation and hid over-expression ( Milan et al . , 1997; Moon et al . , 2005 ) . Interestingly , caspase-3 staining of apoptotic cells in the anterior compartment differed significantly from that seen in undead cells . Although active caspase-3 immunoreactivity was cytoplasmatic and diffuse in undead cells , the staining of cells in the anterior compartment was punctate and intense , indicating that these cells are dying ( Figure 1B ) . To confirm this idea , we performed TUNEL labeling ( Figure 1C ) . As expected , undead cells in the posterior compartment did not show TUNEL staining , but caspase-3-positive cells in the anterior compartment also displayed distinct TUNEL labeling ( Figure 1C ) . These findings indicate that undead cells in the posterior compartment of the Drosophila wing disc have the ability to stimulate the induction of apoptosis at a distance in a different compartment . 10 . 7554/eLife . 01004 . 003Figure 1 . Undead cells promote apoptosis in neighboring cells . ( A ) Wing disc of the genotype hh-Gal4>UAS-GFP showing wild-type wg expression ( blue in A , white in A′ ) and normal levels of apoptosis , visualized by cleaved-caspase-3 ( casp ) staining ( red in A , white in A′′ ) . ( B ) Wing discs of the genotype hh-Gal4>UAS-GFP UAS-hid UAS-p35 . Undead cells in the posterior compartment show ectopic wg activation ( blue in B , white in B′ ) and are labeled by diffuse cleaved-caspase-3 staining ( red in B , white in B′′ ) . Non-autonomous apoptosis was observed in the anterior compartment ( arrows ) . ( C ) Apoptotic cells in the anterior compartment ( arrows ) show cleaved-caspase-3 staining ( red ) and TUNEL staining ( blue in C , white in C′′ ) . Scale bars: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01004 . 003 To examine whether the ability of undead cells to induce non-autonomous apoptosis is more general , we used different paradigms to generate undead cells . Expression of the pro-apoptotic gene rpr along with p35 using the same driver also produced extensive cell death in the anterior compartment ( not shown ) . Furthermore , non-autonomous apoptosis is not restricted to the wing imaginal disc , as we also observed apoptosis in the anterior compartment of other discs , such as the haltere or the leg discs ( Figure 2A , B ) . On the other hand , we did not observe apoptosis in the eye-antennal discs , suggesting this is not a general systemic response ( Figure 2C ) . However , this phenomenon is not compartment specific . We used the apterous-Gal4 ( ap-Gal ) driver to express hid and p35 in the dorsal compartment of wing discs . In this case , we observed widespread apoptosis in the ventral compartment ( Figure 2D ) . However , the use of weaker drivers ( such as Ci-Gal4 and en-Gal4 ) produced very little non-autonomous apoptosis . This suggests that a strong apoptotic stimulus is required to induce non-autonomous apoptosis . 10 . 7554/eLife . 01004 . 004Figure 2 . Non-autonomous apoptosis induced in different imaginal discs and with different drivers . ( A and B ) Haltere ( A ) and leg ( B ) discs of the genotype hh-Gal4>UAS-GFP UAS-hid UAS-p35 also show non-autonomous apoptosis in the anterior compartment . ( C ) Eye-antennal discs of the genotype hh-Gal4>UAS-myr-mRFP UAS-hid UAS-p35 ( labeling with UAS-myr-mRFP is shown in green ) . In this case , non-autonomous apoptosis is not observed . ( D ) Wing disc of the genotype ap-Gal4>UAS-GFP UAS-hid UAS-p35 . Undead cells in the dorsal compartment can be visualized by diffuse cleaved-caspase-3 staining ( red in D , white in D′′ ) . Non-autonomous apoptosis was observed in the ventral compartment ( arrows ) . Scale bars: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01004 . 004 We next wanted to confirm that the observed anterior apoptosis is in fact non-autonomous in contrast to undead cells that might have migrated from the posterior compartment and escaped the p35 protection . For this aim , we utilized the Q transgene expression system ( Potter et al . , 2010; Potter and Luo , 2011 ) in combination with the Gal4 system , enabling us to independently control the expression of QF and Gal4 in the posterior and the anterior compartments , respectively ( Figure 3A ) . We then expressed rpr and p35 in the posterior compartment under the control of the QF and use Ci-Gal4>UAS-GFP to label the anterior compartment . Undead cells can be visualized in the posterior compartment by staining with cleaved caspase-3 antibody ( Figure 3B ) . Importantly , we observed that cells that die in the anterior compartment express the anterior marker Ci-Gal4 , demonstrating that they are of anterior origin , and not ‘escaping undead cells’ ( Figure 3B ) . This confirms that the apoptosis taking place in the anterior compartment is indeed non-autonomous apoptosis . 10 . 7554/eLife . 01004 . 005Figure 3 . Apoptosis in the anterior compartment is non-autonomous . ( A ) Combination of Q and Gal4 systems to independently control expression in the anterior and posterior compartments . We make use of the Psc-QF driver , which is expressed ubiquitously in the wing disc ( red ) , and the Ci-Gal4 driver , which is expressed in the anterior compartment ( green ) . By using Ci-Gal4 to drive expression of the QS suppressor ( UAS-QS ) , QF expression can be restricted to the posterior compartment . In this way , we can control transgene expression independently in the anterior compartment ( with the Gal4 system , green ) and in the posterior compartment ( with the Q system , red ) . ( B ) Wing discs of the genotype Psc-QF>QUAS-Tomato QUAS-rpr QUAS-p35 Ci-Gal4>UAS-GFP UAS-QS . QF expression is restricted to the posterior compartment ( red in B and B′ ) , while Gal4 expression can be visualized in the anterior compartment ( green in B and B′′ ) . Expression of rpr and p35 in the posterior compartment using the Q system leads to generation of undead cells in the posterior compartment and the induction of non-autonomous apoptosis in the anterior compartment , as shown by cleaved caspase-3 staining ( blue in B , white in B′′′ ) . Dying cells in the anterior compartment ( arrows ) are of anterior origin , as shown by the expression of Ci-Gal4>UAS-GFP . Scale bar: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01004 . 005 Milan et al . ( 1997 ) reported non-autonomous apoptosis upon ricin overexpression , which they explained as being an indirect consequence of changes in proliferation during compartment size accommodation . Since in our experiments posterior compartments were usually enlarged , we considered the possibility that non-autonomous apoptosis is the consequence of a size compensation mechanism across compartments to maintain the overall size of the disc . To investigate this possibility , we manipulated the size of the posterior compartment by downregulation of dMyc . dMyc is a key regulator of tissue growth , and downregulation of dMyc leads to a reduction of cell and tissue size ( Johnston et al . , 1999 ) . We reasoned that if non-autonomous apoptosis is the consequence of the increased size of the posterior compartment , then this phenomenon should not take place under conditions where the size of the posterior compartment is normal or reduced . Therefore , we generated posterior compartments that express hid and p35 together with dMyc RNAi . Under these conditions , we still observed undead cells , as revealed by cleaved caspase-3 staining , in a compartment with significantly reduced size ( Figure 4A ) . Importantly , despite the size reduction , we still saw large numbers of apoptotic cells in the anterior compartment ( Figure 4A ) . These observations argue strongly against the idea that size compensation causes non-autonomous cell death in this system . 10 . 7554/eLife . 01004 . 006Figure 4 . Signaling from apoptotic cells induces non-autonomous apoptosis . ( A ) Wing disc of the genotype hh-Gal4>UAS-GFP UAS-hid UAS-p35 UAS-RNAi dMyc . The posterior compartment is very reduced in size , but non-autonomous apoptosis is still present in the anterior compartment ( arrows ) , as shown by staining with cleaved caspase-3 ( red in A , white in A′ ) . ( B and C ) Wing discs of the genotype hh-Gal4>UAS-GFP UAS-rpr UAS-p35 UAS-RNAi-dpp ( B ) and hh-Gal4>UAS-GFP UAS-rpr UAS-p35 UAS-RNAi-wg ( C ) . Downregulation of the mitogenic signals Wg and Dpp produced by undead cells does not affect the amount of non-autonomous apoptosis , which is labeled with cleaved caspase-3 antibody ( red in B and C; white in B′ and C′ ) . ( D ) Wing disc of the genotype hh-Gal4>UAS-GFP UAS-rpr UAS-p35 tub-Gal80TS grown at 29°C for the last 72 hr of larval development . Some apoptosis was still observed in the anterior compartment ( arrows ) . Scale bars: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01004 . 006 We also considered the possibility that non-autonomous apoptosis could be a consequence of the production of Dpp and Wg by undead cells , which could distort the morphogen gradient leading to morphogenetic apoptosis ( Adachi-Yamada and O’Connor , 2002 ) . To examine this notion , we made experiments downregulating Wg or Dpp signals in the posterior compartment ( Figure 4B , C ) . Under these conditions we failed to rescue non-autonomous apoptosis , which demonstrates that it is not a consequence of morphogenetic apoptosis or depends on the levels of Wg or Dpp ( Figure 4B , C ) . We also generated undead cells in the posterior compartment for a shorter period of time ( 3–4 days ) to address the possibility that changes in growth , development and patterning could result in non-autonomous apoptosis . For this purpose , we employed the Gal4/Gal80TS system to temporally control transgene expression ( McGuire et al . , 2003 ) . Under these conditions we were able to generate discs that have a normal pattern , size and proliferation rates in the posterior compartment . Nevertheless , we again observed the induction of ectopic apoptosis in the anterior compartment ( Figure 4D ) . Taken together , these results indicate that apoptosis-induced non-autonomous apoptosis is not the consequence of cell competition , abnormal development , morphogenetic apoptosis or size compensation to prune excessive growth . Since the previous experiments utilized undead cells , we investigated whether ‘genuine’ apoptotic cells have the same signaling capacity . For this purpose , we induced apoptosis in the posterior compartment by expressing rpr or hid alone , without p35 . To allow for larval viability , we again used the conditional Gal4/Gal80TS system to express pro-apoptotic proteins for only 48–72 hr . Under these conditions , large numbers of apoptotic cells were generated in the posterior compartment , and again we saw the induction of apoptosis in the anterior compartment ( Figure 5 ) . In this situation , the amount of non-autonomous apoptosis was generally less prominent compared to the use of undead cells , but in some cases it reached very high levels ( Figure 5C ) . Hence , we conclude that genuine apoptotic cells have the capacity to induce non-autonomous apoptosis at a distance . 10 . 7554/eLife . 01004 . 007Figure 5 . Genuine apoptosis also induces non-autonomous apoptosis . ( A ) Wing disc of the genotype hh-Gal4>UAS-GFP UAS-hid tub-Gal80 TS grown at 29°C for 72 hr . ( B ) Wing disc of the genotype hh-Gal4>UAS-GFP UAS-rpr tub-Gal80TS grown at 29°C for 48 hr . ( C ) Wing disc of the genotype hh-Gal4>UAS-GFP UAS-hid tub-Gal80TS grown at 29°C for 96 hr . In all cases , the posterior compartment is labeled by GFP expression ( green ) . In A and B , the posterior compartment is also labeled with the hh-lacZ construct ( blue in A and B , white in A′′′ and B′′′ ) . Genuine apoptosis can be visualized in the posterior compartment by punctate staining with cleaved caspase-3 ( red in A–C , white in A′ and B′ ) . Non-autonomous apoptosis was also induced in the anterior compartment ( arrows ) . Scale bars: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01004 . 007 The preceding results suggest the existence of a signal ( s ) emanating from apoptotic cells that can act at a distance to induce apoptosis across compartment borders , and we term this phenomenon ‘Apoptosis-induced-Apoptosis ( AiA ) ’ . To gain insight into the underlying molecular mechanism , we considered a role of the JNK pathway since it is activated during stress-induced apoptosis in Drosophila and plays a role for the induction of Dpp and Wg in apoptotic cells ( Ryoo et al . , 2004; Perez-Garijo et al . , 2009 ) . For this purpose , we first examined the pattern of JNK activation by generating undead cells in the posterior compartment and used a puckered-lacZ construct to monitor JNK activity . puckered ( puc ) , the sole Drosophila JNK-specific MAPK phosphatase , is a feedback antagonist of the JNK pathway ( Martin-Blanco et al . , 1998 ) . Hence , the puc-lacZ line , which is an insertion in the puckered gene ( pucE69 ) , serves both as a mutant for puckered and as a read-out for JNK activity . As expected , we saw very high puc-lacZ staining in undead cells , where the apoptotic loop keeps JNK constantly activated ( Shlevkov and Morata , 2012 ) . However , we also observed modest activation of JNK in dying cells in the anterior compartment ( Figure 6A ) . Sometimes we also saw a trail of puc-lacZ activity between undead cells and dying cells . We also investigated the effect of increasing the activity of the JNK pathway using the above-mentioned mutant background for puckered ( pucE69 ) . Normally , the generation of undead cells in the posterior compartment for 72 hr causes only a modest induction of apoptosis in the anterior compartment ( Figure 6B ) . However , in a puc+/− background , the amount of apoptosis in the anterior compartment was dramatically increased ( Figure 6C ) . This suggests that JNK is involved in the induction of non-autonomous apoptosis . 10 . 7554/eLife . 01004 . 008Figure 6 . Apoptosis-induced apoptosis depends on JNK signaling . ( A ) Wing disc of the genotype hh-Gal4>UAS-myr-mRFP UAS-rpr UAS-p35 pucE69-lacZ/+ grown at 25°C ( labeling of the posterior compartment with UAS-myr-mRFP is shown in green ) . Non-autonomous apoptosis in the anterior compartment is visualized by staining with cleaved caspase-3 antibody ( red ) . puc-lacZ expression ( blue in A , white in A′ ) reveals very strong activation of JNK pathway in undead cells and modest activation in dying cells in the anterior compartment ( arrows ) . ( B ) Wing disc of the genotype hh-Gal4>UAS-GFP UAS-hid UAS-p35 tub-Gal80TS grown at 29°C for the last 72 hr of larval development . A reduced amount of non-autonomous apoptosis was observed in the anterior compartment ( arrows ) , as shown by staining with cleaved caspase-3 antibody ( red in B , white in B′ ) . ( C ) Wing disc of the genotype hh-Gal4>UAS-GFP UAS-hid UAS-p35 tub-Gal80TS pucE69-lacZ/+ grown at 29°C during the same period of time . The amount of apoptosis in the anterior compartment was greatly increased ( arrows ) . ( D ) Wing disc of the genotype hh-Gal4>UAS-myr-mRFP UAS-hid UAS-p35 grown at 25°C ( labeling of the posterior compartment with UAS-myr-mRFP is shown in green ) . Cleaved caspase-3 staining ( red in D , white in D′ ) reveals a large amount of non-autonomous apoptosis in the anterior compartment ( arrows ) . ( E ) Downregulation of the JNK pathway suppresses non-autonomous apoptosis , even though undead cells are still present in the posterior compartment , as shown by caspase-3 staining ( red in E , white in E′ ) . Genotype: hh-Gal4>UAS-myr-mRFP UAS-hid UAS-p35 hep1/Y ( mRFP is also shown in green ) . ( F ) Wing disc of the genotype Psc-QF>QUAS-Tomato QUAS-rpr QUAS-p35 Ci-Gal4>UAS-QS . Expression of the driver QF is restricted to the posterior compartment ( labeling of the compartment with QUAS-Tomato is shown in green ) . Cleaved caspase-3 ( red in F , white in F′ ) staining labels undead cells in the posterior compartment and apoptotic cells in the anterior compartment ( arrows ) . ( G ) Inhibition of the JNK pathway specifically in the anterior compartment completely suppresses non-autonomous apoptosis . Genotype: Psc-QF>QUAS-Tomato QUAS-rpr QUAS-p35 Ci-Gal4>UAS-QS UAS-RNAi-bsk . Caspase-3 staining labels intact undead cells in the posterior compartment ( red in G , white in G′ ) . Labeling of the compartment with QUAS-Tomato is shown in green . Scale bars: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01004 . 008 To further investigate the role of JNK in AiA , we downregulated the activity of this pathway using a mutant for hemipterous ( hep ) , the Drosophila JNK-Kinase ( dJNKK ) . We used hh-Gal4>UAS-hid UAS-p35 to generate undead cells in the posterior compartment and extensive induction of cell death in the anterior compartment ( Figure 6D ) . Strikingly , apoptosis in the anterior compartment was completely blocked in male larvae hemizygous for hep1 ( Figure 6E ) . Importantly , hep1 hemizygous mutants retained diffuse staining with caspase-3 in the posterior compartment , demonstrating that the generation of undead cells is not suppressed . Collectively , these experiments indicate that the JNK pathway plays an important role in AiA . The previous experiments utilized mutant backgrounds for puc and hep . Under those conditions , both the posterior compartment ( where undead cells are generated ) and the anterior compartment ( where non-autonomous apoptosis takes place ) are mutant for those genes . The caveat of this approach is that reducing JNK activity in undead cells could potentially diminish their signaling capability . To gain insight into the role of JNK in AiA , we decided to specifically block JNK signaling in the anterior compartment , combining once more the Q and Gal4 systems . As previously shown , generation of undead cells in the posterior compartment by expression of rpr and p35 using the Q system leads to non-autonomous apoptosis ( Figure 6F ) . The advantage of this system is that Ci-Gal4 expression in the anterior compartment not only allows us to label the anterior compartment , but also to direct expression of any gene of interest specifically in the anterior compartment using the Gal4 system . To block JNK activity in the anterior compartment , we drove expression of the RNAi of basket ( bsk ) , the Drosophila JNK . Under these conditions , non-autonomous apoptosis is completely abrogated ( Figure 6G ) . Since here we are not affecting JNK activity in undead cells , we can conclude that JNK is required for cells to die in the anterior compartment as a consequence of AiA . One way by which the JNK pathway is activated in Drosophila is through the TNF ligand Eiger . Significantly , over-expression of Eiger can induce cell death through activation of the JNK pathway ( Igaki et al . , 2002; Moreno et al . , 2002; Kauppila et al . , 2003 ) . The molecular basis of TNF-induced cell death in Drosophila has been well studied ( Kanda et al . , 2002; Geuking et al . , 2005 , 2009; Xue et al . , 2007; Narasimamurthy et al . , 2009; Kanda et al . , 2011; Ma et al . , 2012 ) . However , in vivo roles for Eiger-induced cell death have been elusive ( Igaki et al . , 2009; Maezawa et al . , 2009; Keller et al . , 2011 ) . To investigate a possible role of Eiger in AiA , we first examined whether its expression is induced in undead cells ( Figure 7A ) . Indeed , we saw significant up-regulation of Eiger in undead cells , consistent with the idea that Eiger may be produced by dying cells as a signal to induce non-autonomous apoptosis ( Figure 7A′′ ) . 10 . 7554/eLife . 01004 . 009Figure 7 . Eiger is responsible for apoptosis-induced apoptosis . ( A ) Wing disc of the genotype hh-Gal4>UAS-myr-mRFP UAS-hid UAS-p35 . The posterior compartment shows ectopic wg expression ( blue in A and A′′′ , white in A′ ) . Eiger levels are also elevated in undead cells ( red in A and A′′′ , white in A′′ ) . ( B ) Wing disc of the same genotype stained with anti-cleaved caspase-3 antibody ( red in B , white in B′ ) . Non-autonomous apoptosis in the anterior compartment is indicated by arrows . ( C–E ) Apoptosis-induced apoptosis is suppressed in an eiger mutant background , as shown by cleaved caspase-3 staining ( red in C–E , white in C′ and D′ ) . Ectopic expression of Wg is observed in the posterior compartment ( blue in E , white in E′ ) , suggesting that undead cells are not compromised in their signaling capacity under these conditions . Genotypes: hh-Gal4>UAS-myr-mRFP UAS-hid UAS-p35 eiger1/eiger1 ( C ) , hh-Gal4>UAS-myr-mRFP UAS-hid UAS-p35 eiger1/eiger3 ( D ) and hh-Gal4>UAS-myr-mRFP UAS-hid UAS-p35 eiger3/eiger3 ( E ) . The posterior compartment is labeled with UAS-myr-mRFP ( green ) in all cases . Scale bars: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01004 . 009 To critically test this hypothesis , we generated undead cells in the posterior compartment of wing discs homozygous mutant for eiger . Strikingly , the elimination of eiger function completely abrogated the non-autonomous apoptosis in the anterior compartment ( Figure 7B–E ) . Importantly , under these conditions we still observed overgrown posterior compartments and a large amount of ectopic Wg signaling , as well as diffuse caspase staining , in undead cells ( Figure 7E ) . This indicates that loss of Eiger does not impair JNK activation and mitogen production within undead cells . In the previous experiments we used a condition where the entire disc is mutant for Eiger . We next investigated whether Eiger produced by undead cells is required for the induction of non-autonomous apoptosis . To address this question , we specifically downregulated Eiger levels in the posterior compartment with Eiger RNAi . Consistent with our model , we observed that AiA was significantly decreased by inhibiting Eiger specifically in undead cells ( Figure 8A–C ) . Taken together , these experiments demonstrate that Eiger is a key signal that is generated by apoptotic cells to induce apoptosis of other cells at a distance ( Figure 8D ) . 10 . 7554/eLife . 01004 . 010Figure 8 . Downregulation of Eiger in undead cells significantly reduces apoptosis-induced apoptosis . ( A ) Wing disc of the genotype hh-Gal4>UAS-myr-mRFP UAS-hid UAS-p35 grown at 29°C . ( B ) Wing disc of the genotype hh-Gal4>UAS-myr-mRFP UAS-hid UAS-p35 UAS-RNAi-eiger KK108814 UAS-RNAi-eiger KK108814 , also grown at 29°C . Apoptosis is shown in all cases by staining with cleaved caspase-3 antibody ( red in A and B , white in A′ and B′ ) . The posterior compartment is labeled with UAS-myr-mRFP ( green ) . ( C ) Measurement of the levels of apoptosis-induced apoptosis in both conditions . The amount of non-autonomous apoptosis is graded in five different categories , from widespread apoptosis ( ++++ ) to no apoptosis ( − ) . Downregulation of Eiger in the posterior compartment significantly decreases AiA . p<0 . 001 . ( D ) Model for apoptosis-induced apoptosis . Apoptotic cells produce Eiger , the Drosophila TNF homolog , which activates the JNK pathway in neighboring cells , leading to cell death in a non-autonomous manner . Scale bars: 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 01004 . 010 We next sought out to examine whether AiA occurs under a normal physiological setting and whether it can be expanded to mice . The mouse hair follicle ( HF ) is an ideal system for investigating AiA since cohort cell death naturally occurs at precise times in the HF cycle ( Lindner et al . , 1997; Botchkareva et al . , 2006; Tong and Coulombe , 2006 ) . The HF cycles between phases of growth ( anagen ) , destruction ( catagen ) and rest ( telogen ) ( Hardy , 1992; Fuchs , 2007 ) . During catagen , apoptosis leads to degeneration of the lower two-thirds of the HF while the upper part remains intact ( Lindner et al . , 1997; Tong and Coulombe , 2006 ) ( Figure 9A ) . This portion encompasses the bulge , which houses the HF stem cells required for the generation of the new HF ( Rompolas et al . , 2012 ) . The first catagen phase commences at postnatal day 16 ( P16 ) and hence we isolated skin at this time point . At P16 , we could easily detect HFs with a large number of apoptotic cells . These cells displayed apoptotic morphology such as membrane blebbing , condensed nuclei and were positive for cleaved capase-3 ( Figure 9B ) . It was previously shown that the HF cycle is dependent upon TNF signaling ( Hoffmann et al . , 1996; Ruckert et al . , 2000; Tong and Coulombe , 2006 ) , but the cellular source of TNF-α has not been determined . If AiA was involved , TNF-α should be produced by apoptotic cells . We examined TNF-α expression and found that apoptotic cells in the HF expressed high levels of TNF-α ( Figure 9B ) . Significantly , TNF-α was not detected at any other location besides apoptotic cells . These results are consistent with the idea that apoptotic cells are the source of pro-death signals during catagen and that AiA contributes to the destruction of the HF . 10 . 7554/eLife . 01004 . 011Figure 9 . TNF-α is expressed in apoptotic HF cells during catagen and is essential for coordinated apoptosis and hair cycle progression in mice . ( A ) Schematic diagram depicting the HF cycle . HFs cycle between phases of growth ( anagen ) , destruction ( catagen ) and rest ( telogen ) . During catagen , apoptosis leads to the degeneration of the lower two-thirds of the HF . Post-catagen , the HF enters the quiescent telogen phase , which later on enters a new cycle of hair growth ( anagen ) . ( B ) Immunofluorescence staining indicating that apoptotic HF cells express TNF-α during catagen ( P16 ) in wild-type conditions . Apoptotic cells are labeled with cleaved caspase-3 ( red ) and TNF-α staining is shown in green . ( C–F ) Neutralizing TNF-α in vivo impairs the HF cycle progression as a result of decreased apoptosis . ( C and D ) DAPI staining of dorsal skin demonstrates impaired HF cycle in response to TNF-α neutralization ( P17 . 5 ) . ( E and F ) Immunofluorescence staining of tail whole mounts indicates decreased cleaved caspase-3 staining ( red ) and desynchronization of the HF cycle in TNF-α neutralized mice ( P17 . 5 ) . ( G ) Statistical analysis of the HF phase in control and TNF-α inhibited mice . Scale bars: 20 μm in B , 100 μm in C–F . DOI: http://dx . doi . org/10 . 7554/eLife . 01004 . 011 We went on to inhibit the function of TNF-α in vivo by injecting mice with a TNF-α neutralizing antibody ( mp6-xt22 ) ( Abrams et al . , 1992 ) . Injections were commenced on P14 littermates , when HFs were in anagen , and a non-specific IgG was used as control . Antibodies were injected daily from P14-P17 and , as expected , at P17 . 5 control dorsal skin HFs displayed a morphology typical for late catagen ( Figure 9C ) . In contrast , upon inhibition of TNF-α HFs escaped destruction and had a morphology reminiscent of anagen or very early catagen ( Figure 9D ) . We next extended this analysis to tail HFs . We analyzed tailskin whole mounts and observed again an inhibition in the destruction of the HF and also a clear loss of synchronicity ( Figure 9E–G ) . To assess effects on apoptosis , we used an antibody that specifically detects the cleaved form of caspase-3 . Compared to controls , TNF-α inhibited animals had dramatically reduced numbers of cleaved caspase-3-positive cells , indicating that TNF-α promotes apoptosis in this system ( Figure 9E , F ) . Taken together , these data suggest that in mice TNF-α is expressed by apoptotic cells in the skin , and that it plays a physiological role to orchestrate cohort cell death during HF regression . Our experiments demonstrate that induction of apoptosis in one compartment results in induction of non-autonomous apoptosis in the neighboring compartment . This is true under many different conditions: both when we generate undead cells ( expressing rpr/hid and p35 ) or upon induction of genuine apoptosis ( expressing rpr/hid alone ) ; once there is ectopic expression of mitogens that leads to excessive proliferation and growth or while blocking mitogenic production or growth of the compartment . One intriguing observation is that this non-autonomous cell death usually displays a pattern consisting of two groups of cells in the wing pouch . One possible explanation for this is that the affected cells are the most susceptible to the death signal . In fact , the regions of the wing pouch where we observe the non-autonomous cell death are also more prone to cell death as a response to different apoptotic stimuli , such as irradiation or hid over-expression ( Milan et al . , 1997; Moon et al . , 2005 ) . Another possibility to explain why we observe cell death at a distance would be that dying cells are producing other signals that inhibit apoptosis . This protective signal would diffuse only short range , and in this way the distance of cells to the border would determine the ratio between the pro-apoptotic and the protective signal , tipping the balance in favor of death or survival . In fact , it has been shown that cells neighboring apoptotic cells downregulate Hippo pathway and consequently activate Diap1 ( Grusche et al . , 2011; Sun and Irvine , 2011 ) . Another good candidate for such an anti-apoptotic signal would be Wg , as it is expressed in an opposite pattern from the non-autonomous apoptosis and is also diffusing from apoptotic cells in the posterior compartment . However , we tried to modify Wg levels in different ways and we did not observe any change in the apoptosis pattern ( data not shown ) . On the other hand , in physiological conditions such as the coordinated cell death of HF cells observed in mice , it would be expected that signaling between apoptotic cells would occur at a much shorter range , probably affecting the immediate neighbors . In any case , the observation that TNF-α is exclusively detected in apoptotic cells and the fact that its inhibition leads to desynchronization of the HF cycle strongly suggests that AiA can be a mechanism to coordinate cell death within a tissue . In our experimental systems , AiA requires both the TNF and JNK signaling pathways . Eiger is produced by apoptotic cells in the posterior compartment of the wing disc and it activates JNK in cells of the neighboring compartment , inducing them to die . Downregulation of Eiger in the posterior compartment or JNK in the anterior compartment was able to suppress AiA . However , it remains to be elucidated whether Eiger directly diffuses to the cells in the anterior compartment , or if some other mechanism is responsible for the activation of JNK in dying cells in the anterior compartment . Recently , it was shown that , upon wounding , JNK activity can be propagated at a distance through a feed-forward loop ( Wu et al . , 2010 ) . Significantly , AiA is not restricted to the Drosophila wing disc . We obtained evidence for a role of TNF-α-mediated AiA during the destruction of the hair follicle ( HF ) in catagen , the regressive phase of the hair cycle ( Figure 9 ) . TNF-α plays a known role to promote cell death and has been previously implicated in HF progression , wound healing and regeneration ( Werner and Grose , 2003; Tong and Coulombe , 2006; Bohm et al . , 2010 ) . However , the cellular source of TNF-α remained unknown and it was previously not appreciated that apoptotic cells can be the source of these signals . Our results suggest that AiA and at least some of the underlying mechanism have been conserved in evolution to promote coordinated cell death . The observation that apoptotic cells can signal to other cells in their environment and instruct them to die has potentially many important implications . On the one hand , there are situations where propagation of an apoptotic stimulus may be a useful mechanism to achieve the rapid and coordinated death of large cell populations . Our experiments in mice show that this can be the case during the catagen phase of the HF cycle . There are many other examples of cell death being used during development to sculpt tissues and organs , including the removal of structures during metamorphosis ( tadpole tail , larval organs in insects , elimination of inappropriate sex organs in mammals , deletion of the amnio serosa during insect embryogenesis ) and the separation of digits through apoptosis of the interdigital webbing in many vertebrates ( Glucksmann , 1951; Jacobson et al . , 1997; Fuchs and Steller , 2011 ) . In all these cases , AiA may facilitate cohort behavior and contribute to the rapid and complete elimination of large fields of cells . Propagation of cell death may also be an efficient way to prevent infection . It is known that cells respond to viral infection by entering apoptosis and in this way impede the replication of the virus ( Barber , 2001 ) . The process of AiA would extend apoptosis to the neighboring cells , preventing also their infection and thus avoiding the spread of the virus . However , propagation of apoptosis may be detrimental in pathological conditions where excessive cell death underlies the etiology of the disease . This may be the case for neurodegenerative disorders , hepatic diseases , cardiac infarction , etc ( Thompson , 1995; Favaloro et al . , 2012 ) . In all these cases it remains to be studied whether extensive amounts of apoptosis that are observed in the affected tissues are a direct consequence of cell damage in an autonomous manner or if part of the cell loss could be attributed to a process of propagation through AiA . Finally , AiA may play a role in cancer . It is known that radiotherapy in humans can induce biological effects in non-irradiated cells at a considerable distance , a phenomenon called radiation-induced bystander effect ( Hei et al . , 2011; Prise and O’Sullivan , 2009 ) . Our current findings provide a possible explanation for some of these effects . Therefore , large-scale induction of apoptosis by AiA may contribute to successful cancer therapy . TNF family proteins are being used as models for drug development aimed to treat cancer ( Ashkenazi , 2008 ) . Furthermore , Eiger , the only TNF member in Drosophila , has a known role in the elimination of pre-tumoral scrib− clones ( Igaki et al . , 2009; Ohsawa et al . , 2011 ) . In addition , cell competition induces cell death even in aggressive scrib−RasV12 tumors , raising the possibility that AiA is induced during tumor initiation , which may affect the tumor microenvironment and ultimately tumor growth ( Menendez et al . , 2010 ) . It is well known that TNF can play both tumor-promoting and tumor-suppressing roles , but AiA has not been investigated in this context ( Pikarsky and Ben-Neriah , 2006; Vainer et al . , 2008 ) . Future studies will shed new light on the relevance of signaling by apoptotic cells and the implications of this signaling mechanism in different scenarios . All flies were raised in standard fly food at 25°C unless indicated otherwise . To generate undead cells , UAS-rpr UAS-p35 or UAS-hid UAS-p35 flies were crossed to the appropriate drivers ( hh-Gal4 , ap-Gal4 and Ci-Gal4 were a gift from G Morata ) . UAS-GFP and UAS-myr-mRFP ( Bloomington Stock Center ) were used to visualize the compartments . As a control for normal apoptosis we used the hh-Gal4 UAS-GFP line . To generate posterior compartments of reduced size we used the UAS-RNAi dMyc ( P[TRiP . JF01761]attP2 ) stock #25783 from TRiP ( available at Bloomington Stock Center ) . To downregulate Wg and Dpp in the posterior compartment we used the UAS-RNAi wg GD#13352 ( VDRC ) and the UAS-RNAi dpp ( P[TRiP . JF01371]attP2 ) stock #25782 ( TRiP , available at Bloomington Stock Center ) , respectively . For temporal transgene expression we used the Gal4/Gal80 TS system ( McGuire et al . , 2003 ) . We crossed tub-Gal80 TS; hh-Gal4 flies to UAS-hid; UAS-p35 lines ( for transient generation of undead cells ) or UAS-hid or UAS-rpr lines ( for induction of genuine apoptosis ) . Flies were raised at 18°C , where the driver Gal4 is inhibited by the tub-Gal80 , and the larvae were shifted at 29°C at different time points to initiate Gal4 activity . Late third instar larvae were dissected after different periods of time at 29°C , as mentioned in the text . We used a hh-lacZ line to label the posterior compartment ( a gift from G Morata ) . To modify JNK activity we used pucE69 and hep1 mutants ( Martin-Blanco et al . , 1998 ) . To eliminate Eiger function we used eiger1 and eiger3 mutants ( a gift from N Baker ) ( Igaki et al . , 2002 ) . To downregulate Eiger specifically in the posterior compartment we used UAS-RNAi eiger KK#108814 from VDRC . To downregulate JNK specifically in the anterior compartment we used the UAS-RNAi bsk GD#34138 from VDRC . For the experiments using the Q system we used the line QUAS-mtdTomato , Psc-QF and the UAS-QS ( stocks #30043 and #30033 , respectively , from Bloomington Stock Center ) . The reaper and p35 inserts were PCR amplified from genomic DNA from flies containing the UAS-p35 and UAS-rpr transgenes using the primers ATAGAGGCGCTTCGTCTACGG and CCCATTCATCAGTTCCATAGGTTG . PCR products were digested with EcoRI for cloning into pQUAST ( Addgene plasmid 24349 ) . The sequence of the construct was verified by DNA sequencing . For PCR amplifications we used the PfuUltra DNA polymerase ( Agilent Technologies , Inc . , Santa Clara , CA ) . DNA constructs were sent for injection to Bestgene Inc . and one transgenic line for each construct was selected on the third chromosome . Imaginal discs were dissected , fixed and stained as described previously ( Perez-Garijo et al . , 2004 ) using the following antibodies: anti-cleaved caspase-3 ( 1:200 , Cell Signaling Technologies , Danvers , MA ) , anti-Wg 4D4 ( 1:50 , DSHB , Iowa City , IA ) , mouse anti-β-Gal ( 1:50 , DSHB ) and anti-Eiger ( 1:500 , gift from M Miura ) . Secondary antibodies were used 1:200 and purchased from Jackson Laboratories . Discs were then mounted in Vectashield ( Vector Laboratories , Burlingame , CA ) . For TUNEL stainings we used the ApopTag in situ Apoptosis Detection Kit ( Chemicon , Millipore , Billerica , MA ) and followed the instructions provided by the kit . Images were taken with a LSM710 ( Zeiss ) confocal microscope and subsequently processed using Adobe Photoshop . Mice skin harvested at P16 and P17 was embedded in OCT . Cryosections were fixed in 4% PFA for 10 min and blocked with 10% NGS , 0 . 5% and 2% BSA for 2 hr and primary antibodies were placed in blocking solution overnight . The following day secondary antibodies were used and sections were mounted with Vectashield ( Vector Laboratories ) . Images were taken with a LSM710 ( Zeiss ) confocal microscope and subsequently processed using Adobe Photoshop . Tail whole mounts were performed as in Braun et al . ( 2003 ) . Antibodies used: anti-cleaved caspase-3 ( 1:100 , Cell Signaling Technologies ) , LEAF purified anti-mouse TNF-α ( 1:100 , Biolegends , San Diego , CA ) , anti-K15 ( 1:1000 , Abcam , Cambridge , United Kingdom ) . For in vivo neutralization experiments C57B littermate mice were broken into two groups ( n = 7 ) . Each group received either a non-specific Rat LEAF purified IgGI ( 10 µg/gr , Biolegends ) or LEAF purified anti-mouse TNF-α ( 10 µg/gr , Biolegends ) . Antibodies were injected daily from P14-P17 in 200 µl of sterile saline subcutaneously .	The tissues of developing organisms can be shaped by apoptosis , a form of regulated cell killing . Although this process can occur in individual cells , apoptotic signals may also dictate the ‘communal death’ of many cells simultaneously . This occurs frequently in animal development: in human fetuses , for example , cells in the hand are directed to die to remove webbing between the fingers . Apoptosis has been thought to resemble a form of silent suicide by cells , but more recent work suggests that apoptotic cells can also transmit signals . Now , Pérez-Garijo et al . find that these cells can stimulate other cells to die in both fruit flies and mice . In fruit flies , apoptosis is activated by proteins known as Grim , Hid and Reaper . To explore whether apoptotic cells could communicate with other cells , Pérez-Garijo et al . created ‘undead’ cells in which one of these proteins was turned on , but other downstream proteins ( that are responsible for the cellular execution phase of apoptosis ) had been turned off: these cells were undergoing apoptosis , but could not complete the process and die . Strikingly , undead cells in the posterior ( back ) region of the wing imaginal disc—the tissue in the larva that gives rise to the wing in the adult fruit fly—could trigger apoptosis in cells in the anterior ( front ) half . Pérez-Garijo et al . found that the JNK pathway activated apoptosis in anterior cells . In fruit flies , the Eiger protein turns on this pathway; when Eiger was absent from posterior cells in the wing imaginal disc , apoptosis in anterior cells ceased , indicating that Eiger might signal at long range . Eiger is related to a protein called TNF that has been implicated in cycles of destruction and renewal of hair follicles in mice . Pérez-Garijo et al . found that TNF is produced by apoptotic cells in hair follicles , and that blocking TNF inhibits the death of other cells in the same cohort: this suggests that a common mechanism could regulate the communal death of cells in flies and mammals . These studies therefore shed light on a conserved pathway in the modulation of tissue development .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "developmental", "biology" ]	2013	Apoptotic cells can induce non-autonomous apoptosis through the TNF pathway
Although electron cryo-microscopy ( cryo-EM ) single-particle analysis has become an important tool for structural biology of large and flexible macro-molecular assemblies , the technique has not yet reached its full potential . Besides fundamental limits imposed by radiation damage , poor detectors and beam-induced sample movement have been shown to degrade attainable resolutions . A new generation of direct electron detectors may ameliorate both effects . Apart from exhibiting improved signal-to-noise performance , these cameras are also fast enough to follow particle movements during electron irradiation . Here , we assess the potentials of this technology for cryo-EM structure determination . Using a newly developed statistical movie processing approach to compensate for beam-induced movement , we show that ribosome reconstructions with unprecedented resolutions may be calculated from almost two orders of magnitude fewer particles than used previously . Therefore , this methodology may expand the scope of high-resolution cryo-EM to a broad range of biological specimens . For many years , electron cryo-microscopy ( cryo-EM ) has held the promise that macromolecular structure determination should in principle be possible to near-atomic resolution from only several thousand projection images of isolated particles in ice ( Henderson , 1995; Glaeser , 1999 ) . However , cryo-EM single-particle reconstructions in which protein main-chains may be traced reliably or individual side-chains are well resolved have been limited to large icosahedral viruses , and have typically required averaging over millions of asymmetric units ( Grigorieff and Harrison , 2011 ) . These results are in accordance with early observations that the contrast in EM images is not as good as expected from theory ( Henderson , 1995 ) . The main limiting factor of the information content in cryo-EM data is radiation damage: the electron dose needs to be limited to prevent the molecules from disintegrating while they are being imaged . However , radiation damage alone is not enough to account for the loss in expected contrast . Two additional factors that have been attributed to information loss in cryo-EM images are poor detective quantum efficiency ( DQE , a frequency-dependent measure for the signal-to-noise performance ) of conventional recording devices , and beam-induced movement or charging of the specimen during imaging ( Henderson , 1992 ) . Traditionally , EM images have been recorded on photographic film , which has a large field of view and a reasonably good DQE , but is not convenient for high-throughput methods due to laborious steps of film development and digitization . Therefore , over the past decade many electron microscopes have been equipped with charge-coupled device ( CCD ) cameras . These digital detectors allow instant inspection of the images and are much more convenient in automated data collection schemes , so that large quantities of data may be obtained with relatively little effort ( Stagg et al . , 2006 ) . However , the conversion of electrons into visible light that is detected by the CCD comes at the expense of a poorer DQE , in particular at higher voltages where many microscopes operate best . Consequently , despite the advantages of CCDs in terms of ease-of-use and data quantity , most of the near-atomic resolution cryo-EM structures published to date have been recorded on photographic film ( Grigorieff and Harrison , 2011 ) . The second factor that has been attributed to the missing contrast in cryo-EM images is related to the radiation damage itself . For every elastic scattering event , which contributes positively to the phase contrast image , also three to four inelastic scattering events occur ( Henderson , 1995 ) . The inelastic events deposit energy in the sample , which leads to radiation damage by breaking covalent bonds and the generation of free radicals . Besides destroying the very structures one aims to determine , these interactions also generate charges in the sample that could lead to image blurring by deflecting the incoming electrons . In addition , radiolysis products such as hydrogen gas may build up internal pressure , thus leading to mechanical stress in the sample ( Leapman and Sun , 1995 ) . At high electron doses this stress ultimately leads to bubbling of the sample , while at much lower doses it has been postulated to induce movements that cause blurring of the images ( Glaeser , 2008 ) . Recent technological advances may mitigate the image-blurring effects of both poor detectors and beam-induced movements . Building on technological developments that were initiated almost a decade ago ( Prydderch et al . , 2003 ) , three different companies now sell digital cameras that detect electrons directly , that is , without the need to first convert electrons into visible light . Initial characterization of these direct electron detectors indicated that their DQE at high resolution is superior to that of CCD and film ( McMullan et al . , 2009a ) , in particular when backscattering of electrons is reduced by back-thinning of the substrate ( McMullan et al . , 2009c ) . An additional advantage of these devices is that they record images at high rates: ranging from 16 to 400 images per second for the currently available products . This allows the recording of a video during typical exposures , compared to a single image on film or CCD . By processing these videos , Grigorieff and colleagues previously showed that beam-induced movements could be followed for large icosahedral viruses , although the gain in resolution compared to similar data recorded on film was not as large as expected , possibly due to experimental design ( Brilot et al . , 2012; Campbell et al . , 2012 ) . Here , we assess the potential of a back-thinned FEI Falcon direct electron detector for cryo-EM structure determination by recording videos on two well-characterized test samples: the prokaryotic and eukaryotic ribosome . Ribosome samples have been characterized by cryo-EM for more than 20 years ( Frank et al . , 1991 ) , and encompass characteristics that are favorable to cryo-EM structure determination: their large content of dense RNA and their large molecular weight ( 2 . 8 MDa for prokaryotic 70S and ∼4 MDa for eukaryotic 80S ribosomes ) result in images of relatively high contrast compared to many other macromolecular complexes , and RNA is known to be less radiation sensitive than protein . On the other hand , the ribosome is an intrinsically flexible macromolecular machine , which often leads to the presence of more than one structural state in ribosome samples . This represents a challenge , as advanced classification techniques are required to separate the different conformations into structurally homogeneous subsets that are amenable to single-particle reconstruction ( Scheres et al . , 2007; Fischer et al . , 2010 ) . In addition , the absence of symmetry in the ribosome makes it necessary to image many more particles ( and to determine many more orientations ) than is the case for icosahedral viruses with 60-fold symmetry . The combination of structural variability and the lack of symmetry may account for the fact that near-atomic resolution cryo-EM maps of ribosome structures have remained elusive thus far . In the experiments described below , we used a 70S T . thermophilus ribosome sample to develop image-processing procedures that account for beam-induced movement of individual particles . Our detector operates at a rate of 17 frames per second , and we recorded 16-frame videos during 1-s exposures . At a total dose of 16 electrons/Å2 , each individual video frame therefore integrates less than 1 electron/Å2 . We expected the high noise levels in these individual frames to pose severe limitations on the accuracy with which these frames may be aligned . Because the alignment accuracy of individual particles remains unknown in typical single-particle analysis , we collected an entire data set as tilted pairs and used tilt-pair analysis to experimentally assess alignment errors for video frames ( Rosenthal and Henderson , 2003 ) . Guided by the results obtained from this analysis , we then developed a statistical video processing procedure that does not depend on tilted images . Application of this procedure to a second , untilted data set of 80S S . cerevisiae ribosomes serves to illustrate the impact these detectors , combined with suitable video processing procedures , will have on the field of cryo-EM structure determination . We recorded tilt pairs of videos on the 70S T . thermophilus ribosome sample and used 2D and 3D classification to select a structurally homogeneous subpopulation of 15 , 202 particle pairs for 3D reconstruction ( see ‘Materials and methods' ) . In the first instance , we disregarded the time-dimension of the videos , and used 16-frame averages for each particle ( Figure 1 ) . By assessing the consistency between the orientations from independently refined tilt pair particles with the known microscope tilt geometry , we estimated an alignment precision for the tilt pairs of 2° ( Rosenthal and Henderson , 2003 ) ( Figure 2A ) . This is a significant improvement over the 4° precision reported for a similar sample recorded on CCD ( Henderson et al . , 2011 ) . This improvement reflects the improved signal-to-noise ratios ( SNRs ) in the individual particles , which are a direct consequence of the improved DQE of the detector . The benefits of using particles with higher SNRs for structure determination are therefore twofold . On the one hand , averaging images with higher SNRs requires fewer particles to overcome the noise . On the other hand , higher SNRs result in smaller alignment errors , which then lead to decreased blurring in the reconstruction . Both effects will lead to higher-resolution maps from fewer particles , which explains the relatively high resolution of 6 . 5 Å that we obtained from only 15 , 202 particle pairs . 10 . 7554/eLife . 00461 . 003Figure 1 . 70S data collected on a back-thinned FEI Falcon detector . ( A ) The 16-frame ( 1-s ) average of two untilted videos . The scale bar indicates 50 nm . Relative positions for independently aligned four-frame averaged particles are shown with circles connected by white lines . The relative position of the average from the first four frames is shown in green , the relative position of the last four frames in red . The differences between these relative positions are exaggerated 25 times for improved clarity , and only those four-frame averages for which correct alignment was confirmed by tilt-pair analysis are included . Movements in the area on the left are smaller ( up to 1 . 5 Å ) then in the area on the right ( up to 10 Å ) . ( B ) Examples of two individual ribosome particles as averages over a decreasing number of video frames . The scale bar indicates 20 nm . Zoomed-in areas of micrographs , additional individual particles and reference-free 2D class averages for both the 70S and 80S data are shown in Figure 1—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00461 . 00310 . 7554/eLife . 00461 . 004Figure 1—figure supplement 1 . Part of a recorded micrograph and reference-free class averages for the 70S and 80S data sets . The scale bar indicates 50 nm . ( B ) Individual 70S particles . The scale bar indicates 20 nm . ( C ) Reference-free 2D class averages showing distinct views of the 70S ribosome . ( D–F ) As in ( A–C ) , but for the 80S data . The scale bar indicates 50 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 00461 . 00410 . 7554/eLife . 00461 . 005Figure 2 . Development of video processing procedures on the 70S data set . ( A ) Histogram of tilt-pair alignment errors for particles that were calculated as 16-frame averages . The width of the first peak at half its height is 2° . This value is plotted as the tilt-pair alignment precision in ( B ) . ( B ) Tilt-pair alignment precision ( top ) and the number of incorrectly aligned particle pairs ( bottom ) for independent refinements of 16-frame , 8-frame , 6-frame , 4-frame and 2-frame averages ( ranges in frame numbers are indicated on the x-axis ) ; the total number of particle pairs was 15 , 202 . Particle pairs with alignment errors larger than three times the reported precision were considered as aligned incorrectly . ( C ) Gold-standard FSC curves . The same blue colors are used as in ( B ) ; orange lines indicate the results of the statistical video-processing approach described in the main text . ( D ) FSC-curves between a rigid-body fitted atomic model and the cryo-EM maps ( using the same color scheme as in [C] ) . ( E–F ) Illustrative density and atomic model for the reconstructions obtained from 16-frame averaged particles ( light blue ) , independently refined four-frame averages ( dark blue ) and the statistical video processing procedure ( orange ) . The density maps were sharpened with B-factors of −211 , −185 and −178 Å2 , respectively . Complete density maps for these three reconstructions are shown in Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00461 . 00510 . 7554/eLife . 00461 . 006Figure 2—figure supplement 1 . Overall views of the 70S ribosome reconstructions obtained from 16-frame averaged particles ( left ) , independently refined four-frame averages ( middle ) and the statistical video processing procedure ( right ) . The large subunit is shown in blue; the small subunit in yellow . DOI: http://dx . doi . org/10 . 7554/eLife . 00461 . 006 We then assessed how accurately particle movement may be followed during the electron exposure . To this purpose , we partitioned each original 16-frame average particle into multiple independent particles that were calculated as the average of a varying number of individual video frames . In this manner , we artificially created four additional data sets of increasing size , comprising averages of 2 × 8 , 3 × 6 , 4 × 4 or 8 × 2 video frames for each original particle . Tilt-pair analysis for these data showed that , as expected , the numbers of correctly aligned particles as well as the alignment precision decrease with averaging over fewer video frames ( Figure 2B ) . In addition , we noted a trend for particles that were calculated from later video frames to align less accurately , most likely as a result of accumulated radiation damage . To assess the effects of these alignment errors on reconstruction quality , we also inspected the Fourier Shell Correlation ( FSC ) curves that were reported for the four refinements ( Figure 2C ) . The best reconstructions , at 5 . 4 Å resolution , were obtained when using averages of six or four individual video frames , which apparently represent the optimum between accurately describing particle movement and not introducing too large alignment errors for these data . Analysis of the orientations for the four-frame averaged particles that were aligned correctly according to the tilt-pair analysis revealed that overall rotations and translations during the exposure are in the order of 1 . 7 ± 1 . 2° and 4 . 2 ± 2 . 3 Å , respectively . Complicated patterns of movement for particles on some of the illuminated areas ( e . g . , see white lines in Figure 1A ) indicate that the observed movements were not merely the result of a drifting stage , but more likely to be attributed to , as yet poorly understood , beam-induced effects . Although useful for tilt-pair analysis , treating the averages of multiple video frames as completely independent also left valuable information unused . The observations that beam-induced movements are relatively small , and that the 16-frame averages may be aligned most accurately , represent valuable prior knowledge that may be efficiently expressed using statistical refinement methods . In the ‘Materials and methods' section we describe the implementation of a probabilistic prior on the orientations of the video frames inside a Bayesian refinement approach ( Scheres , 2012a ) . This prior down-weights orientations for the video frames in a continuous manner according to their distance from the orientation determined for the 16-frame averages . Application of this procedure to the 70S ribosome data yielded a reconstruction with a resolution of 5 . 1 Å ( orange line in Figure 2C ) . Comparison of the FSCs between the reconstructed maps and a rigid-body fitted crystal structure ( Selmer et al . , 2006 ) , decreasing estimated B-factors , and visual inspection of the maps all confirmed that reconstruction quality may be improved by following the movement of individual particles , and that the statistical approach outperforms the straightforward use of multi-frame averaged particles in this respect ( Figure 2D–F , Figure 2–figure supplement 1 ) . Finally , the developed procedure was applied to a second , untilted data set that was collected on an 80S S . cerevisiae ribosome sample during a single , manual microscopy session . Classification identified two major conformations , comprising 35 , 813 and 22 , 638 particles , respectively . A difference of ∼2° in ratchet-like movement represents the largest difference between the two classes ( also see Video 1 ) . Using the statistical video processing procedure , refinement of class 1 yielded a map with an overall resolution of 4 . 5 Å , while class 2 yielded a 4 . 6 Å map ( Figure 3A ) . However , both maps still exhibited signs of unresolved structural heterogeneity , in particular in the small subunit and most notably for class 1 ( Figure 3B ) . Therefore , the overall resolution estimates are too optimistic for the disordered parts , whereas the most stable parts of the structures still contain significant information beyond the overall estimated resolutions . The density for the 60S subunit clearly shows separation of β-strands , the pitch of α-helices , density for many side chains , and the separation of RNA bases , indicating that useful information to ∼4 Å resolution is present in the best parts of the map ( Figure 3C–G , Video 2 ) . The level of detail in the density for the 40S subunit varies , but is generally lower than for the 60S subunit ( Figure 3H , I , Video 2 ) . Calculation of FSC curves between the cryo-EM maps and rigid-body fitted crystal structures for the 60S and 40S subunits ( Ben-Shem et al . , 2011 ) confirmed the estimated resolutions , as well as the variation in quality of the density across the maps ( Figure 3J , K ) . 10 . 7554/eLife . 00461 . 007Video 1 . Shown are the density maps for the two classes of the 80S data set after 3D classification , which display a difference of ∼2° in ratchet-like rotation . The resolution is 6 . 5 Å for class 1 and 6 . 7 Å for class 2 . However , the maps appear to be of lower resolution as they have not been sharpened . DOI: http://dx . doi . org/10 . 7554/eLife . 00461 . 00710 . 7554/eLife . 00461 . 008Figure 3 . Application of the statistical video processing procedure to an 80S ribosome data set . ( A ) Gold-standard FSC curves for class 1 ( red ) and class 2 ( green ) . ( B ) Slices through the reconstructions of class 1 ( left ) and class 2 ( right ) . The fuzzy appearance for the density of the 40S subunits ( green and red lines ) is an indication of unresolved structural heterogeneity . ( C–G ) Densities for the 60S subunit of class 1 showing a protein loop interacting with a flipped-out RNA base ( C ) , a short stretch of an RNA helix ( D ) , a β-strand ( E ) , a β-sheet ( F ) , and an α-helix ( G ) . ( H–I ) Density for the 40S subunit of class 1 showing a well-resolved α-helix ( H ) and a poorly-resolved one ( I ) . The density map of class 1 was sharpened with a B-factor of −160 Å2 . ( J ) FSC curves between the map of class 1 and the rigid-body fitted atomic models of the entire 80S particle , and for the 40S and 60S subunits separately . ( K ) As in J , but for class 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 00461 . 00810 . 7554/eLife . 00461 . 009Video 2 . Shown is the cryo-EM density map for class 1 of the 80S data set together with the atomic models that are also shown in Figure 3C–I . Density for the 60S subunit is shown in blue , density for the 40S subunit in yellow . The density for the 40S subunit in the overall view is filtered to 5 . 0 Å resolution for improved clarity , all other densities are filtered at 4 . 1 Å resolution . DOI: http://dx . doi . org/10 . 7554/eLife . 00461 . 009 The reconstructions presented in this paper are of significantly higher resolution than any prokaryotic or eukaryotic ribosome structures that were calculated from cryo-EM data on photographic film or CCD . Moreover , they were obtained from unprecedented small numbers of particles . For comparison , currently the highest resolution ribosome map in the EMDB ( an 80S structure from the plant Triticum aestivum ) was calculated from almost 1 . 4 million particles and was reported to be at 5 . 5 Å resolution ( Armache et al . , 2010 ) . We obtained reconstructions with useful information up to ∼4 Å resolution from nearly two orders of magnitude fewer particles . The observation that we could calculate maps with useful information up to 90% of the Nyquist frequency surpassed our initial expectations and clearly illustrates the potential of direct electron detectors combined with adequate video processing for cryo-EM structure determination . Incorporation of the new detectors into automated data collection schemes will result in much larger amounts of data than we have used in our study . For relatively high molecular-weight complexes that display sufficient contrast for accurate alignment and classification , this may then allow high-resolution reconstruction of many more structural states , or of states that are adopted by only a small fraction of the particles . Moreover , if the sample is sufficiently rigid to exist in only one or a few distinct states , then larger amounts of data are expected to lead to even higher resolutions than the ones reported here . For example , following the SNR considerations proposed by ( Rosenthal and Henderson , 2003 ) , our overall resolution of 4 . 5 Å would increase to ∼3 . 3 Å if we would apply our approach to 1 million particles of similar quality ( provided no other resolution-degrading factors play a role ) . Therefore , we foresee that the combination of direct electron detectors , video processing approaches , and automated data collection schemes will significantly increase the number of specimens for which de novo building of atomic models into cryo-EM maps becomes feasible . The advantages of direct electron detectors and video processing procedures are not limited to relatively large complexes alone . Current alignment procedures are often severely limited by low SNRs in images of complexes that are smaller than ∼300 to 500 kDa . Also in this respect , the ability to record videos may provide significant benefits . Because SNRs in individual particles drop quickly with increasing spatial frequency image alignments are typically mainly driven by the lower-resolution components in the images ( Henderson et al . , 2011; Scheres and Chen , 2012 ) . These components have been observed to remain intact after much higher electron doses than the high-resolution information ( Hayward and Glaeser , 1979; Baker and Rubinstein , 2010 ) . Therefore , relatively long videos with a high dose could be used to obtain reliable low-resolution components , while a dose-dependent weighting scheme to account for radiation damage may be used to optimize the SNR for higher spatial frequencies at the same time ( Campbell et al . , 2012 ) . Such procedures are expected to significantly lower the size limit of particles that may still be aligned correctly and are thus amenable to cryo-EM structure determination . Still , accurately following the movement of small particles during videos will be more challenging than for the high-contrast ribosome particles presented here . Therefore , in order to also obtain near-atomic resolutions for smaller particles , further investigations into the nature of beam-induced image blurring and how to stop it will still be necessary . Also these investigations are expected to benefit significantly from the possibility to analyze beam-induced movements during videos recorded by direct electron detectors . Based on our observations and the potential for further improvements in the technique , we believe the future for cryo-EM structure determination to be bright . Currently available detectors , combined with video processing algorithms like the one presented here , will result in higher-resolution cryo-EM structures for many more samples than those reported until now . Meanwhile , cryo-EM image quality will keep improving as ongoing technological developments such as single-electron counting detectors ( McMullan et al . , 2009b ) , phase-plate imaging ( Nagayama , 2011 ) and better sample preparation techniques mature . Therefore , we are optimistic that the field will continue to progress towards fulfilling the promise of providing near-atomic resolution reconstructions , and thereby more detailed biological insights , for a wide range of specimens . 70S T . thermophilus ribosome samples with tRNAs and mRNA were produced as described previously ( Selmer et al . , 2006 ) . 80S ribosomes were purified from S . cereviseae strain YAS-2488 . Cells were grown to OD600 = 2–4 and starved for 10 min at 4°C in buffer without sucrose ( 20 mM Hepes-KOH pH 7 . 45 , 150 mM KCl , 150 mM K-acetate , 10 mM Mg-acetate , 1 mg/ml heparin , 0 . 1 mM PMFS , 0 . 1 mM benzamidine , 2 mM DTT ) . After starvation , cells were frozen in liquid nitrogen and mechanically disrupted by a blender machine operated under liquid nitrogen conditions . The lysate was allowed to defrost at 4°C and then clarified by a 20 min centrifugation at 14 , 500×g . The 80S ribosomes from the supernatant were pelleted through a sucrose cushion for 4 hr at 45 , 000 rpm in a Ti45 Beckman Coulter rotor ( Palo Alto , California , USA ) , in buffer that was supplemented with 1 M sucrose . The pellets were re-suspended in a sucrose gradient buffer without sucrose ( 20 mM Hepes-KOH pH 7 . 45 , 50 mM KCl , 5 mM Mg-acetate , 0 . 1 mM PMFS , 0 . 1 mM benzamidine , 2 mM DTT ) and incubated for 15 min with 1 mM puromycin . The sample was loaded on a 10–40% sucrose gradient and centrifuged for 18 hr at 28 , 000 rpm in a Ti25 zonal rotor . A single peak after gradient fractionation was confirmed to correspond to 80S ribosome particles . All ribosomal proteins , as well as the protein Stm1 , were identified using mass spectrometry . For storage in liquid nitrogen , the buffer was exchanged to buffer M ( 3 mM Hepes-KOH pH 7 . 45 , 6 . 6 mM Tris-acetate pH 7 . 2 , 3 mM NH4Cl , 6 . 6 mM NH4-acetate , 48 mM K-acetate , 4 mM Mg-acetate , 2 . 4 mM DTT ) and the sample was concentrated to 6 μM . For both the 70S and 80S samples , aliquots of 3 μl at a concentration of ∼80 nM were incubated for 30 s on glow-discharged holey carbon grids ( Quantifoil R2/2 ) , on which a home-made continuous carbon film ( estimated to be ∼30-Å thick ) had previously been deposited . Grids were blotted for 2 . 5 s and plunge-frozen in liquid ethane using an FEI Vitrobot . Grids were transferred to an FEI Polara G2 microscope that was operated at 300 kV . A C2 aperture of 70 μm and an objective aperture of 100 μm were used . Defocus was varied from 1 . 3 to 3 . 8 μm . Using an extraction voltage of 3900 V , a gun lens setting of 2 and a spotsize of 4 or 5 , an estimated dose of 16 electrons/Å2 was applied during 1-s exposures . The beam used was larger than the Quantifoil hole , illuminating the carbon all around the hole . Images were recorded at the approximate center of the hole on a back-thinned FEI Falcon detector at a calibrated magnification of 79 , 096 ( yielding a pixel size of 1 . 77 Å ) . The small area that was imaged relative to the area that was illuminated resulted in a beam-tilt that was much smaller than the one expected from the relatively large C2 aperture that we used , but we cannot exclude that our final resolution was limited by beam-tilt . An in-house built system was used to intercept the videos from the detector ( we were capable of recording 16 frames during a 1-s exposure ) . All data were collected manually during two half-day sessions for the 70S sample , and a single half-day session for the 80S sample . Tilt pairs were collected at tilt angles of 0° and 10° . Electron micrographs were evaluated for astigmatism and drift . For the 70S sample , 159 out of 285 micrograph pairs were selected for further analysis; 260 out of 291 micrographs were selected for the 80S sample . A total of 24 , 044 70S particle pairs and 72 , 447 80S particles were selected using the swarm tool in the e2boxer . py program of EMAN2 ( Tang et al . , 2007 ) , contrast transfer function parameters were estimated using CTFFIND3 ( Mindell and Grigorieff , 2003 ) , and all 2D and 3D classifications and refinements were performed in RELION ( Scheres , 2012b ) . Prior to 3D refinement , both data sets were subjected to reference-free 2D class averaging and 3D classification to identify structurally homogeneous subsets . Initial 3D classifications were run for 25 iterations , with four classes , at the original image size , with an angular sampling of 7 . 5° , and a regularization parameter T = 4 . For the 70S data set , discarded particles included 50S subunits and 70S ribosomes with tRNAs in the E- and P-sites , and 15 , 202 particles were selected to correspond to 70S ribosomes with a single tRNA in the P-site . For the 80S data , dissociated subunits could again be identified in the 2D class averages , but initial 3D classification did not yield structurally distinct classes . However , after 3D refinement of a single model , blurry density for the 40S subunit indicated that significant structural heterogeneity was still present . In a subsequent 3D classification run with four classes , an angular sampling of 1 . 8° was combined with local angular searches around the refined orientations , and the refined single model was used as a starting model . This calculation separated two conformations with different degrees of ratchet-like movement , comprising 35 , 813 and 22 , 638 particles , respectively . Initial 3D classifications and all 3D refinements were started from ribosome maps that were downloaded from the Electron Microscopy Data Bank ( EMDB-1657 for the 70S; EMDB-1780 for the 80S ribosome; Seidelt et al . , 2009; Armache et al . , 2010 ) and subsequently low-pass filtered to 60 Å . All 3D refinements used gold-standard FSC calculations to avoid overfitting and reported resolutions were based on the FSC = 0 . 143 criterion ( Scheres and Chen , 2012 ) . Final FSC curves were calculated using a soft spherical mask ( with a 5-pixel fall-off ) on the two independent reconstructions . Prior to visualization , all density maps were corrected for the modulation transfer function ( MTF ) of the detector , and then sharpened by applying a negative B-factor that was estimated using automated procedures ( Rosenthal and Henderson , 2003 ) . Rigid body fitting of 70S and 80S crystal structures ( PDB-IDs: 2WH1-2 , Weixlbaumer et al . , 2008 , and 3U5B-E , Ben-Shem et al . , 2011 , respectively ) was performed using UCSF Chimera ( Pettersen et al . , 2004 ) . In order to describe the distinct degrees of ratchet rotation in the two 80S conformations , the 40S and 60S subunits were fitted separately . The resulting models merely served to illustrate the quality of our maps; we did not analyze our maps in terms of differences with the crystal structures . To allow others to mine our structures for additional information ( e . g . , we see continuous density for RNA nucleotides 440–499 of expansion segment 7 and nucleotides 1954–2093 of expansion segment 27 in 25S , which were not modelled in the 80S crystal structure ) , we have deposited our maps at the EMDB ( accession numbers 2275 , 2276 and 2277 ) . A novel procedure for video-frame alignment was developed that exploits the relatively high accuracy of aligning 16-frame average particles ( Figure 2B ) , as well as the prior knowledge that particles are unlikely to undergo very large rotations or translations during the 1-s exposure . To this purpose , we defined Gaussian prior distributions on the rotations and translations of the video frames , and centered these distributions at the observed orientations for alignments with the corresponding 16-frame average particles . The priors were then incorporated as prior probability distributions , P ( φ\|Θ , Y ) in Eq . 7 of Scheres ( 2012a ) , in the Bayesian refinement approach of the RELION program . This open-source program may be downloaded from http://www2 . mrc-lmb . cam . ac . uk/relion; the video-processing procedures described here are available in version 1 . 2 . The widths of the priors were implemented as user-controlled parameters that may be tuned to express the expected amount of movement during the video . In addition to the Gaussian priors , we also implemented an option to use running averages of a user-defined number of frames for the alignments . This allows more precise sampling of the particle movement than the four discrete 4-frame averages along the 16-frame videos shown in Figure 1 . The corresponding orientations ( or probability distributions in the Bayesian approach ) are only applied to the single , middle video frame of the running average window . For all calculations shown in this paper , we used running averages of five video frames; a standard deviation of 1° for the priors on the Euler angles; and the standard deviation that was estimated for the picked particle positions in the 16-frame alignment for the prior on the translations . Rotational searches ( or integrations in the Bayesian approach ) were then limited to ±3° with a step size of 0 . 45° , while translational searches were performed up to ±2 pixels with a step size of 0 . 5 pixels . We note that in this case we used the observed translations and rotations of the four-frame average refinements to choose the widths of our priors . Alternatively , one may optimize these values based on their effects on the gold-standard FSC curve . Although not pursued in this study , the Bayesian approach also provides a convenient framework to estimate these widths from the data themselves . In addition , as the field gains a better understanding of beam-induced movements , more detailed priors ( e . g . , with widths that depend on the accumulated dose for each frame ) may easily be incorporated .	Determining the structure of proteins and other biomolecules at the atomic level is central to understanding many aspects of biology . X-ray crystallography is the best-known technique for structural biology but , as the name suggests , it works only with samples that can be crystallized . Electron cryo-microscopy ( cryo-EM ) could , potentially , be used to determine the atomic structures of biomolecules that cannot be crystallized , but at present the resolution that can be achieved with this approach is sufficient only for imaging certain types of viruses . In cryo-EM , a solution of the biomolecule of interest is frozen in a thin layer of ice , and this layer is imaged in an electron microscope . By combining images of many identical biomolecules in many different orientations , it is possible to work backwards and determine their 3D structure . However , in order to determine this structure at high resolution , it is necessary to make repeated measurements to reduce high levels of noise in the images . Cryo-EM images are usually recorded on a photographic film or a CCD ( charge-coupled device ) camera . However , photographic film is unsuitable for high-throughput methods because it has to be handled manually , while the efficiency of CCD cameras is limited because the electrons have to be converted into visible light to be detected . Digital cameras that can detect electrons directly have become available recently , and these are more efficient than both film and CCD cameras . They are also much faster , which means that it is possible to record videos of the sample during the time ( typically ∼1 s ) it is being exposed to the electron beam . Processing these videos could then—in theory—compensate for any movements of the biomolecules that are induced by the electron beam . Along with radiation damage caused by the electrons , these beam-induced movements have been a major limitation on the resolution that can be achieved with cryo-EM . Bai et al . demonstrate the potential of direct-electron detectors in cryo-EM by determining the structures of two ribosomes . Using a novel statistical algorithm to accurately follow the movements of the ribosomes during the time they are exposed to the electron beam , they are able to compensate for these movements , and this makes it possible to determine the structures of the ribosomes with near-atomic precision . Moreover , the resolution they achieve with just ∼30 , 000 ribosomes is better than that previously achieved with more than a million ribosomes , allowing small details inside the ribosome – such as ß-strands and bulky amino-acid side chains – to be resolved with cryo-EM for the first time . The work of Bai et al . could , therefore , allow researchers to use cryo-EM to determine the structure of many more biomolecules with atomic precision .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2013	Ribosome structures to near-atomic resolution from thirty thousand cryo-EM particles
Symbiotic bacteria assist in maintaining homeostasis of the animal immune system . However , the molecular mechanisms that underlie symbiont-mediated host immunity are largely unknown . Tsetse flies ( Glossina spp . ) house maternally transmitted symbionts that regulate the development and function of their host’s immune system . Herein we demonstrate that the obligate mutualist , Wigglesworthia , up-regulates expression of odorant binding protein six in the gut of intrauterine tsetse larvae . This process is necessary and sufficient to induce systemic expression of the hematopoietic RUNX transcription factor lozenge and the subsequent production of crystal cells , which actuate the melanotic immune response in adult tsetse . Larval Drosophila’s indigenous microbiota , which is acquired from the environment , regulates an orthologous hematopoietic pathway in their host . These findings provide insight into the molecular mechanisms that underlie enteric symbiont-stimulated systemic immune system development , and indicate that these processes are evolutionarily conserved despite the divergent nature of host-symbiont interactions in these model systems . Mutualistic bacteria are functionally critical to the physiological well-being of their animal hosts . These microbes benefit their hosts by providing essential nutrients , aiding in digestion and maintaining intestinal equilibrium ( Douglas , 2015; Marchesi et al . , 2016 ) . Additionally , mutualistic symbionts promote the development , differentiation and proper function of their host’s immune system ( Chu and Mazmanian , 2013; Khosravi et al . , 2014; Gomez de Agüero et al . , 2016 ) . Despite such an important role , the molecular mechanisms that underlie symbiont-mediated homeostasis of animal immunity remain largely unknown . Insect models are useful for studying host-microbe interactions because , relative to their mammalian counterparts , they generally house taxonomically simple bacterial communities that can be easily manipulated during their host’s development . Tsetse flies ( Glossina spp . ) house two gut-associated bacterial symbionts , obligate Wigglesworthia glossinidia and commensal Sodalis glossinidius ( Maltz et al . , 2012; Wang et al . , 2013 ) . In adult flies Wigglesworthia resides within cells that collectively form a bacteriome organ that is attached to the anterior midgut . Sodalis can be found extracellularly in the gut lumen , or intracellularly within gut epithelial cells ( Wang et al . , 2013 ) . Tsetse reproduce via adenotrophic viviparity , during which pregnant females give birth to one larva each reproductive , or ‘gonotrophic’ ( GC ) , cycle . Individual larvae mature through three developmental instars within the uterus , all the while receiving nourishment in the form of a milk-like substance produced by a modified accessory gland ( milk gland; Benoit et al . , 2015 ) . Both Wigglesworthia and Sodalis are also found extracellularly in tsetse milk , and these bacteria colonize the gut of developing intrauterine larvae as they imbibe this nutrient source ( Attardo et al . , 2008 ) . Tsetse that undergo larvagenesis in the absence of their indigenous microbiota are highly immuno-compromised as adults ( Wang et al . , 2013 ) . These ‘aposymbiotic’ ( hereafter referred to as ‘GmmApo’ ) flies exhibit a dysfunctional cellular immune system that is characterized by the conspicuous absence of hemocytes . This phenotype results from the disruption of hematopoiesis ( blood cell differentiation ) during larval development ( Weiss et al . , 2011 , 2012 ) . In insects , distinct hemocyte lineages mediate essential immune-related functions , including the phagocytosis and encapsulation of foreign invaders and the closing of cuticular wounds via the deposition of melanin at the site of injury ( Lemaitre and Hoffmann , 2007; Hillyer and Strand , 2014; Lee and Miura , 2014 ) . These immune mechanisms serve as the first line of defense following systemic challenge with exogenous organisms . Actuation of host immune system development represents an evolution driven mechanism that steadfastly links the tsetse fly with its symbiotic bacterial partners . In this study we identify symbiont regulated genes and pathways in tsetse larvae using RNA-seq . Functional studies revealed that one symbiont induced gene , which encodes an odorant binding protein ( OBP ) , regulates hematopoietic pathways during tsetse’s larval development . We also demonstrate that Drosophila’s indigenous microbiota regulates expression of an orthologous , functionally conserved OBP-encoding gene . Our findings detail a newly characterized , evolutionarily conserved component of a blood cell differentiation regulatory pathway that occurs in response to the presence of enteric symbionts . Drosophila hematopoiesis occurs primarily during early larval development ( Evans and Banerjee , 2003 ) . This process is likely similarly timed in wild-type tsetse , but fails to occur when larvae develop in the absence of their indigenous symbiotic bacteria ( Weiss et al . , 2011 , 2012 ) . We sequenced RNA transcripts from age-matched ( first and second instar ) GmmWT and GmmApo ( generated as described in Materials and methods , Fly lines and bacteria ) larvae in an effort to identify genes and pathways associated with hematopoiesis during tsetse larvagenesis . RNA-seq analysis revealed that 1166 genes exhibited a differential expression profile in GmmWT compared to GmmApo larvae , and approximately 76% of these genes were expressed at higher levels in GmmWT individuals ( Figure 1A; Supplementary file 1 ) . Gene ontogeny analysis revealed significant enrichment of genes functionally associated with B vitamin metabolism , larval development , organismal growth and chitin synthesis in the GmmWT compared to GmmApo larvae ( Figure 1B ) . These genes likely underlie previously observed phenotypes associated with dysfunctional chitin generation and B vitamin metabolism in GmmApo flies ( Weiss et al . , 2013; Michalkova et al . , 2014 ) . Herein we found that specific orthologues putatively clustering within the ‘hematopoiesis’ gene ontology category ( GOC ) were enriched in GmmWT larvae compared to adult female and male flies ( Figure 1—figure supplement 1 ) . However , genes associated with hematopoiesis were not significantly enriched in GmmWT compared to GmmApo larvae ( Figure 1—figure supplement 1; Supplementary file 2 ) . Because our previous studies demonstrate that GmmApo larvae fail to develop hemocytes ( Weiss et al . , 2011 , 2012 ) , we hypothesized that factors not typically grouped within the hematopoiesis GOC likely induce hemocyte differentiation in GmmWT larvae . 10 . 7554/eLife . 19535 . 003Figure 1 . Symbiont-mediated differential expression of odorant binding protein 6 in tsetse larvae . ( A ) Number of genes exhibiting significant differential expression , and a relative transcript abundance [in transcripts per million ( TPM ) ] over 3 , in GmmWT compared to GmmApo larvae . Significance is based on a Baggerly’s test followed by a false detection rate correction ( p<0 . 01 ) . ( B ) Significantly enriched gene ontology categories , determined using a Fisher’s exact test . ( C ) Genes exhibiting significant differential expression ( measure as fold-change in gene expression ) in GmmWT compared to GmmApo larvae , and a minimum TPM value of 1000 . Significance was determined as in ( A ) . ( D ) Enrichment analysis of odorant binding protein-encoding genes expressed in GmmWT adult males ( purple ) and females ( yellow ) , and GmmWT larvae ( green ) . ( E ) Relative expression ( TPM ) of tsetse odorant binding protein-encoding genes in GmmWT larvae , and their differential expression ( measure as fold-change in gene expression ) in GmmWT compared to GmmApo larvae . Significance was determined as in ( A ) . ( F ) Relative obp6 expression in GmmWT larvae , as well as larvae derived from symbiont-cured moms fed a diet supplemented with yeast and Wigglesworthia-containing bacteriome extracts ( Gmmbact/Wgm+ ) , Wigglesworthia-free bacteriome extracts ( Gmmbact/Wgm- ) , Sodalis cell extracts ( GmmSgm+ ) , and bacteriome extracts harvested from GmmApo females ( Gmmbact/Apo ) . GmmWT and GmmApo flies served as controls . n = 6 biological replicates for groups GmmWT , Gmmbact/Wgm+ and GmmSgm+ samples , and n = 5 biological replicates for Gmmbact/Wgm- , Gmmbact/Apo and GmmApo samples . Replicates for all groups contain a mixture of four first and second instar larvae . Data are presented as mean ± SEM . Bars with different letters indicate a statistically significant difference ( specific p values are listed in Figure 1—source data 1 ) between samples . Statistical analysis = ANOVA followed by Tukey’s HSD post-hoc analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 00310 . 7554/eLife . 19535 . 004Figure 1—source data 1 . Obp6 expression in aposymbiotic tsetse larvae following supplementation . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 00410 . 7554/eLife . 19535 . 005Figure 1—figure supplement 1 . Developmental stage-specific enrichment analysis of tsetse orthologues that putatively cluster within the ‘hematopoiesis’ COG . Ontology enrichment analyses were performed as described in the Materials and methods , under the ‘Transcriptomics’ sub-heading . Refer to Supplementary file 2 for a description of the enriched hematopoiesis-associated gene expressed specifically in GmmWT and GmmApo tsetse larvae . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 005 When our RNA-seq libraries were screened to identify highly abundant ( TPM ≥ 103 ) and differentially transcribed genes , we observed that tsetse odorant binding protein 6 ( obp6 ) exhibited the ninth highest level of differential expression between all annotated genes present in the larval GmmWT and GmmApo libraries ( Supplementary file 1 ) . Specifically , GmmWT larvae expressed 22x more obp6 transcripts than did their GmmApo counterparts ( Figure 1C; Supplementary file 1 ) . Because chemosensory-related genes exhibit hematopoietic properties and immune system-associated expression profiles in other insects ( Thomas et al . , 2016; Shim et al . , 2013a; Aguilar et al . , 2005; Bartholomay et al . , 2004; Sabatier et al . , 2003 ) , we investigated the functional relationship between obp6 and immune system maturation processes during tsetse larvagenesis . Obp6 , which encodes a 145 amino acid protein ( 16kD ) with an N-terminal secretion signal ( Liu et al . , 2010 ) , is larvae-enriched ( Figure 1D ) and the only OBP-encoding gene expressed at significantly different levels between GmmWT and GmmApo individuals ( Figure 1E; Supplementary file 1 ) . Obp6 expression can be restored in GmmApo larvae when their symbiont-cured moms are fed a diet supplemented specifically with Wigglesworthia cell extracts ( Figure 1F ) , thus demonstrating that expression of this gene is stimulated by a Wigglesworthia derived molecule ( s ) . Furthermore , this stimulus is likely not a bacterium generated nutrient , as the vitamin rich yeast extract included with the supplements fails to elicit the same response in individuals of the other treatment and control groups . GmmApo larvae express 22-fold fewer obp6 transcripts than do their GmmWT counterparts ( Supplementary file 1 ) , and GmmApo adults present a highly depleted population of hemocytes ( Weiss et al . , 2012 ) . Equipped with this information , we set out to determine if obp6 influences larval hemocyte differentiation processes and the subsequent function of these cells during adulthood . To do so we experimentally reduced obp6 expression in intrauterine GmmWT larvae using a novel RNAi-based trans-generational gene knock down approach ( a graphical representation of the experimental design is presented in Supplementary file 3 ) . To coincide with larval eclosion and subsequent milk uptake , pregnant GmmWT females were injected with either anti-obp6 ( two anti-obp6 siRNAs were used , one of which was conjugated to a Cy3 dye ) or anti-gfp siRNAs on days 8 and 11 post-mating . siRNA-administered treatment ( anti-obp6 ) and control ( anti-gfp ) moms , and their larval and adult offspring , are hereafter designated ‘siOBP6’ and ‘siGFP’ , respectively . Three days after the second treatment , pregnant females were viewed under a fluorescent microscope and siRNA was observed to have diffused throughout their hemocoel ( Supplementary file 4A , top left panel ) . Additionally , siRNAs were taken up by the maternal milk gland and imbibed by developing ( First gonotrophic cycle , GC1 ) intrauterine larvae , which subsequently fluoresced orange ( Supplementary file 4A , bottom left panel ) . Obp6 expression was reduced in GmmWT larvae by an average of 68% when they acquired corresponding siRNAs trans-generationally from their mother’s milk ( Supplementary file 4B ) . Finally , by the third gonotrophic cycle ( 26 days post-siRNA treatment ) , anti-obp6 siRNAs were no longer visible in treated moms or their offspring ( Supplementary file 4A , top and bottom right panels , respectively ) , and larval obp6 expression had rebounded to levels equivalent to that found in GC1 control ( siGFP ) larvae ( Supplementary file 4B ) . These recovered flies are hereafter designated ‘siOBP6R’ . The capacity of an adult insect to survive systemic challenge with exogenous microbes depends largely on the efficacy of its cellular immune system , and more specifically , hemocyte-mediated phagocytosis ( Hillyer and Strand , 2014; Vlisidou and Wood , 2015 ) . We thus investigated the ability of siOBP6 , siGFP and siOBP6R adults to survive following systemic challenge with 103 CFU of E . coli K12 . We observed that 88% of siOBP6 adults , 12% siGFP adults and 4% of siOBP6R adults perished over the course of the experiment ( Figure 2A ) . The lethal effect of E . coli K12 on siOBP6 adults , which is similar to that observed in GmmApo adults following exposure to the same challenge ( Weiss et al . , 2012 ) , indicates that tsetse must express obp6 during larvagenesis in order for subsequent adults to survive following thoracic exposure to a needle-inoculated E . coli K12 challenge . 10 . 7554/eLife . 19535 . 006Figure 2 . Tsetse odorant binding protein 6 does not mediate the development and function of phagocytic hemocytes . ( A ) Survival following systemic challenge of siOBP6 and siGFP adults with 5 × 102 CFU of E . coli K12 . Fly survival was monitored every other day for the duration of the 14 day experimental period . Survival assays were performed in triplicate , using 25 flies per replicate . Red curve depicts a statistically significant difference in infection outcome ( p<0 . 0001 , log-rank test ) . ( B ) Hemocyte abundance in siOBP6 and siGFP adults was quantified microscopically using a hemocytometer ( Figure 2—source data 1 ) . ( C ) A representative micrograph of hemocyte-engulfed recE . coliGFP from siOBP6 , siGFP and siOBP6R adults . Experiment was performed using hemolymph collected from four distinct flies per ( Figure 2—source data 2 ) . Hemolymph was collected 12 hpc and fixed on glass slides using 2% paraformaldehyde . Magnification is x400 . ( D ) E . coli densities ( CFU/μl of hemolymph ) in the hemolymph of siOBP6 , siGFP and siOBP6R adults at 2 and 6 dpc ( Figure 2—source data 3 ) . In ( B ) and ( D ) , symbols represent one hemolymph sample per group , and bars represent the median hemocyte quantity ( B ) or bacterial density ( D ) per sample . Statistical analysis = ANOVA followed by Tukey’s HSD post-hoc analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 00610 . 7554/eLife . 19535 . 007Figure 2—source data 1 . Circulating hemocytes per microliter of hemolymph . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 00710 . 7554/eLife . 19535 . 008Figure 2—source data 2 . Phagocytosis by tsetse hemocytes . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 00810 . 7554/eLife . 19535 . 009Figure 2—source data 3 . Colony forming units ( CFU ) per microliter of hemolymph . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 009 siOBP6 adults perish unusually fast following subjection to a hemocoelic challenge with E . coli K12 . We thus investigated whether these flies house a depleted population of phagocytic hemocytes , which compromise the majority of the collective hemocyte population ( Kurucz et al . , 2007 ) . We discovered no significant difference in the number of circulating hemocytes present in the hemocoel of siOBP6 compared to siGFP and siOBP6R adults ( 1349 ± 56 , 1365 ± 33 and 1413 ± 31 hemocytes per μl of hemolymph , respectively; Figure 2B ) , and microscopic examination of hemolymph revealed that hemocytes from adult individuals of all three groups actively engulfed E . coli cells ( recE . coliGFP; Figure 2C ) . Finally , we observed that E . coli density initially increased during the first two days following injection into the hemocoel of siOBP6 , siGFP and siOBP6R tsetse ( 3278 ± 806 , 3530 ± 482 and 4085 ± 442 CFU per μl of hemolymph , respectively ) , but by four days later , had decreased to levels below that of the initial inoculate ( 104 ± 19 , 115 ± 29 and 111 ± 28 CFU per μl of hemolymph , respectively; Figure 2D ) . These findings indicate that adult siOBP6 tsetse perish following systemic challenge with E . coli K12 , but that this phenotype is not due to a reduced or dysfunctional population of phagocytic hemocytes . Additionally , E . coli growth does not appear to cause death in these flies , as bacterial density within both siOBP6 and siGFP ( which survive this systemic challenge ) individuals is maintained at similar densities during the course of infection . We found that adult siOBP6 tsetse perish following systemic challenge with E . coli K12 , but that this outcome surprisingly does not result from either a lack of functional phagocytic hemocytes or unimpeded bacterial replication . The insect cellular immune response also includes the synthesis of melanin , which is involved in the encapsulation of foreign organisms in the hemocoel as well as the deposition of melanin at the site of cuticular wounds ( Babcock et al . , 2008; Tang , 2009 ) . We next investigated whether adult siOBP6 tsetse present a defective melanization cascade , and as such are unable to deposit melanin at the wound site inflicted during the systemic E . coli challenge procedure . We used a heat-sterilized glass needle to puncture the cuticle of siOBP6 , siGFP and siOBP6R individuals , and monitored percent survival over time . Similar to their counterparts that were challenged with E . coli , the majority ( 92% ) of siOBP6 adults perished after receiving a ‘clean wound’ to their thorax . Conversely , significantly fewer siGFP ( 8% ) and siOBP6R ( 16% ) controls died following this treatment ( Figure 3A ) . 10 . 7554/eLife . 19535 . 010Figure 3 . Obp6 mediates the melanization cascade in adult tsetse . ( A ) Survival following administration of clean wounds to the thoracic cuticle of siOBP6 , siGFP and siOBP6R adults . Survival assays were performed in triplicate , using 25 flies per replicate . Red curve depicts a statistically significant difference in infection outcome ( p<0 . 0001 , log-rank test ) . ( B ) A representative micrograph of the cuticle of siRNA treated adults 3 hr post-wounding ( hpw ) with a clean needle . Melanin deposited at the wound site of siGFP and siOBP6R controls , and hemolymph exudate from a siOBP6 treatment individual , are identified by black and red arrowheads , respectively . Scale bar = 500 μm . Experiment was performed using four distinct flies per group ( Figure 3—source data 1 ) . ( C ) Quantitation of PPO1 and PPO2 in the hemolymph of siOBP6 , siGFP and siOBP6R adults three hpw with a clean needle . Shown is a representative Western blot analysis using Drosophila anti-PPO1 and anti-PPO2 antibodies . 8 μl of pooled hemolymph was run per gel lane . Hemolymph was collected and pooled from four individuals from each group . Western blots were repeated in triplicate [Figure 3—source data 2 ( for PPO1 westerns ) and Figure 3—source data 3 ( for PPO2 westerns ) ] . ( D ) PO activity in the hemolymph of siOBP6 , siGFP and siOBP6R adults at 0 and 3 hpw with a clean needle . n = 5 biological replicates per group per time point for pre-wound readings , and n = 8 biological replicates per group per time point for post-wound readings . Data are presented as mean ± SEM . Bars with different letters indicate a statistically significant difference between pre- and post-wound values ( specific p values are listed in the Figure 3—source data 4 ) . Statistical test = 2 way ANOVA followed by Tukey’s HSD post-hoc analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 01010 . 7554/eLife . 19535 . 011Figure 3—source data 1 . Melanin deposition at tsetse cuticular wound sites . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 01110 . 7554/eLife . 19535 . 012Figure 3—source data 2 . Tsetse prophenoloxidase 1 ( PPO1 ) western blots . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 01210 . 7554/eLife . 19535 . 013Figure 3—source data 3 . Tsetse prophenoloxidase 2 ( PPO2 ) western blots . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 01310 . 7554/eLife . 19535 . 014Figure 3—source data 4 . Tsetse phenoloxidase ( PO ) assays . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 014 Adult siOBP6 tsetse perished after they were pricked with a clean needle while their siGFP and siOBP6R counterparts survived . We monitored the wound site of a select number of flies from each of these groups to determine if melanin was deposited at this location . Three hours post-treatment , melanin was observed at the wound site of siGFP and siOBP6R tsetse . Conversely , no melanin was present at the wound site of siOBP6 individuals at this time ( Figure 3B ) , and , as the wound never fully healed , hemolymph continued to slowly exude from these flies for the entirety of the 2-week experimental period . Melanin deposition represents the end product of a complex biochemical cascade , several steps of which are catalyzed by the enzyme phenol oxidase ( PO; Eleftherianos and Revenis , 2011 ) . Because toxic intermediates are produced as a byproduct of melanin production , catalytic PO is usually synthesized as an inactive zymogen called prophenoloxidase ( PPO; Tang , 2009 ) . We quantified PPO levels in hemolymph collected from siOBP6 , siGFP and siOBP6R flies to determine if the different wound melanization phenotypes we observed reflected different quantities of this enzyme in the hemolymph of treatment versus control individuals . Western blots using anti-PPO1 and anti-PPO2 antibodies revealed that siOBP6 adults produced significantly less of these proteins than did age-matched siGFP and siOBP6R individuals ( Figure 3C ) . We next employed a L-DOPA assay to measure PO activity in hemolymph collected from siOBP6 , siGFP and siOBP6R adults at 0 and 3 hr after subjection to a clean needle wound . We observed a 9 . 0-fold and 13 . 7-fold increase in PO activity in siGFP and siOBP6R adults , respectively , 3 hr after cuticle penetration . Conversely , PO activity only increased 2-fold in clean wounded siOBP6 adults over the same time frame ( Figure 3D ) . Taken together these results indicate that when intrauterine tsetse larvae express reduced levels of obp6 they present a dysfunctional melanization cascade during adulthood . In Drosophila larvae , a specific subset of hemocytes called crystal cells produce the majority of PPO . Upon immunological stimulation , crystals cells rupture and release PPO into the hemolymph where enzymes convert it into PO that subsequently catalyzes melanin synthesis ( Honti et al . , 2014 ) . Drosophila hemocytes originate in the fly’s embryonic and larval lymph gland . This tissue is resorbed during metamorphosis , and evidence of prolific and prolonged de novo production of hemocytes in adult flies does not exist ( Grigorian and Hartenstein , 2013 ) . Assuming a similar situation occurs in tsetse , the absence of cuticular melanization in siOBP6 adults following wounding with a clean needle may reflect abnormal crystal cell development during embryogenesis and/or intrauterine larvagenesis . To investigate this hypothesis , we quantified this cell subtype in siRNA treated larvae by subjecting third instar individuals to a 65°C heat shock for 10 min . In Drosophila this treatment induces spontaneous activation of PPO in crystal cells , which are then visible as black melanotic spots on the larval cuticle ( Binggeli et al . , 2014 ) . Following this treatment , we counted 7 ( ±2 . 0 ) , 58 ( ±4 . 7 ) and 52 ( ±5 . 7 ) melanized spots on siOBP6 , siGFP and siOBP6R larvae , respectively ( Figure 4A ) . These findings suggest that siOBP6 larvae house either a reduced number of crystal cells , or that these cells exhibit a dysfunctional PPO pathway . 10 . 7554/eLife . 19535 . 015Figure 4 . Obp6 expression in the gut of larval tsetse is an integral component of the systemic pathway that actuates crystal cell production . ( A ) Representative micrograph depicting spontaneous PPO activation in early third instar siGFP , siOBP6 and siOBP6R tsetse larvae following subjection to a 10 min heat shock at 65°C . Experiment was repeated using one larvae from five distinct moms from each group . Melanotic spots were quantitated microscopically . Statistical analysis = Kruskal-Wallis test followed by Dunn’s post-hoc analysis ( Figure 4—source data 1 ) . ( B ) RT-qPCR analysis of obp6 , serpent and lozenge expression in embryos prior to maternal treatment with siRNA , and in siOBP6 , siGFP and siOBP6R tsetse larvae from siRNA treated moms . Embryo replicates ( n = 5 ) contain three embryos , larval replicates ( n = 7 for siOBP6 , n = 5 for siGFP and n = 6 for siOBP6R ) contain a mixture of four first and second instar larvae . ND , not detectable . Data are presented as mean ± SEM . Bars with different letters indicate a statistically significant difference between samples ( specific p values for larval samples are listed in the Figure 4—source data 2 ) . Statistical analysis = 2 way ANOVA followed by Tukey’s HSD post-hoc analysis . ( C ) Representative image of obp6 and lozenge spatial expression patterns , determined using semi-quantitative RT-PCR , in the gut and carcass of second instar GmmWT larvae . Experiment was repeated using guts and carcasses from five distinct individuals ( Figure 4—source data 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 01510 . 7554/eLife . 19535 . 016Figure 4—source data 1 . Sessile crystal abundance in larval tsetse . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 01610 . 7554/eLife . 19535 . 017Figure 4—source data 2 . Relative obp6 , serpent and lozenge gene expression in tsetse embryoes and larvae . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 01710 . 7554/eLife . 19535 . 018Figure 4—source data 3 . Tissue distribution of obp6 and lozenge expression in tsetse larvae . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 018 In Drosophila’s lymph gland , prohemocytes express the GATA factor serpent , and these cells subsequently differentiate into either functionally mature plasmatocyte ( phagocytic hemocytes ) or crystal cell lineages via induction of the hematopoietic transcription factors glial cells missing or lozenge , respectively ( Lebestky et al . , 2000; Ferjoux et al . , 2007 ) . Thus , serpent serves as an efficacious marker for determining the presence of hemocyte progenitors , while lozenge expression reflects differentiation of a pool of these precursors into functional members of the crystal cell lineage ( Binggeli et al . , 2014 ) . In an effort to determine why siOBP6 larvae present significantly fewer melanotic spots than do their siGFP and siOBP6R counterparts , we quantified transcript abundance of obp6 , serpent and lozenge in tsetse embryos prior to maternal inoculation with siRNA , and then in first and second instar larvae from the three siRNA treated groups . We observed that obp6 and lozenge transcripts were undetectable during embryonic development , while serpent expression remained unchanged across groups ( Figure 4B ) . This serpent expression profile presented by tsetse embryos mirrors the presence of prohemocytes , while the absence of lozenge transcripts in individuals of this developmental stage suggests that crystal cell differentiation has yet to commence . Subsequent analysis of larval offspring from siRNA treated moms revealed that obp6 and lozenge expression was significantly reduced in siOBP6 compared to siGFP and siOBP6R larvae , while serpent expression was similar in all individuals tested ( Figure 5B ) . Serpent expression by siOBP6 larvae indicates maintenance of the hemocyte precursor pool throughout development of immature stages . However , fewer of these progenitors likely become crystal cells because siOBP6 larvae express significantly reduced levels of the lozenge transcription factor that actuates the differentiation process . Intestinal microbiota can exert their physiological influence at local epithelial surfaces or peripheral tissues ( Round and Mazmanian , 2009; Clarke , 2014a ) . To determine the spatial dynamics of obp6 mediated induction of tsetse hematopoiesis , we analyzed the expression pattern of this gene , and lozenge , in gut and carcass tissues of second instar larvae . We detected obp6 and lozenge transcripts only in the larval gut and carcass , respectively ( Figure 4C ) . These findings indicate that Wigglesworthia stimulates local expression of obp6 , and that tsetse’s hematopoietic niche is likely not attached to the larval gut . Thus , symbiont-induced obp6 regulates tsetse hematopoiesis on a systemic level . Hematopoietic signaling pathways and their transcriptional regulators within the niche are functionally conserved across many animal taxa ( Evans et al . , 2003; Hartenstein , 2006; Makhijani and Brückner , 2012 ) . However , little is known about the evolutionary conservation of upstream , extra-niche factors that induce blood cell lineage commitment . Thus , for comparative purposes , we investigated whether Drosophila’s indigenous microbiota regulate crystal cell differentiation , and thus the melanization response , within their host . To do so we measured expression levels of obp28a ( which is orthologous to tsetse obp6 based on sequence similarity; OrthoDB; Kriventseva et al . , 2015 ) and lozenge in conventionally reared and axenic ( reared in the absence of their indigenous microbiota ) Drosophila larvae . We noted that conventionally reared w1118 and Oregon-R larvae expressed significantly more obp28a and lozenge transcripts than did their axenic counterparts ( Figure 5A ) . Additionally , conventionally reared larvae and adults presented more cuticular sessile crystal cells ( following heat shock ) and produced more PPO , respectively , than did age-matched axenic individuals ( Figure 5B , Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 19535 . 019Figure 5 . Drosophila’s indigenous microbiota actuates larval hematopoietic pathways and thus functionality of the adult melanization response . ( A ) Relative expression of obp28a and lozenge in conventionally reared ( CR ) and axenic ( AX ) Oregon-R and w1118 Drosophila larvae . n = 9 ( Oregon-R ) and 6 ( w1118 ) biological replicates per group , each containing a mixture of thirty second and early-3rd instar larvae . Data are presented as mean ± SEM . Asterisks indicate statistical significance ( specific p values are listed in the Figure 5—source data 1 ) . Statistical analysis = unpaired t-tests , corrected for multiple comparisons using the Holm-Sidak method . ( B ) AX larvae house fewer sessile crystal cells , and produce less PPO , than do and CR individuals . Top panels are representative micrographs depicting spontaneous PPO activation in AX and CR w1118 and Oregon-R larvae following subjection to a 10 min heat shock at 65°C . n = 5 larvae per group [Figure 5—source data 2 ( for Oregon-R larvae ) and Figure 5—source data 3 ( for w1118 larvae ) ] . Bottom panels are representative Western blots using Drosophila anti-PPO1 antibodies . 8 μl of pooled hemolymph was run per gel lane . Hemolymph was collected and pooled from 15 individuals from each group . Western blots were repeated in triplicate ( Figure 5—source data 4 ) . ( C ) Survival of obp28a RNAi knockdown Drosophila ( tub-GAL4/UAS-obp28a RNAi ) and knockout ( Obp28a- ) adults compared to controls ( tub-GAL4 , UAS-obp28a RNAi and wCs ) following wounding with a clean needle . Experiment was performed in triplicate , n = 50 ( RNAi ) and n = 45 ( knockout ) per group per replicate . Red curve depicts a statistically significant difference in infection outcomes [p<0 . 0001 ( RNAi ) and p=0 . 0002 ( knockout ) , log-rank test] . ( D ) A representative micrograph of the cuticle of obp28a knockdown , knockout and control Drosophila adults six hpw with a clean needle . Experiment was performed using two distinct experimental fly cohorts [n = 4 flies per group per experiment; Figure 5—source data 5 ( for RNAi flies ) and Figure 5—source data 6 ( for deletion mutants ) ] . Wounds on the cuticle of control ( melanized ) and obp28a knockdown and knockout individuals ( not melanized ) are identified with black and red ovals , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 01910 . 7554/eLife . 19535 . 020Figure 5—source data 1 . Obp28a and lozenge expression in conventionally reared and axenic w1118 and Oregon-R Drosophila . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 02010 . 7554/eLife . 19535 . 021Figure 5—source data 2 . Sessile crystal cells in conventionally reared and axenic Oregon-R Drosophila larvae . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 02110 . 7554/eLife . 19535 . 022Figure 5—source data 3 . Sessile crystal cells in conventionally reared and axenic w1118 Drosophila larvae . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 02210 . 7554/eLife . 19535 . 023Figure 5—source data 4 . Drosophila prophenoloxidase 1 ( PPO1 ) western blots . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 02310 . 7554/eLife . 19535 . 024Figure 5—source data 5 . Melanin deposition at Drosophila cuticular wound sites following RNAi-mediated knockdown of obp28a . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 02410 . 7554/eLife . 19535 . 025Figure 5—source data 6 . Melanin deposition at Drosophila cuticular wound sites in obp28a deletion mutants . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 02510 . 7554/eLife . 19535 . 026Figure 5—figure supplement 1 . Axenic and conventionally reared w1118 and Oregon-R larvae following subjection to a 10 min heat shock at 65°C . White ellipses demarcate an area of the larval cuticle that contains a high density of ruptured crystal cells . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 026 We next investigated the wound healing phenotype of both a Drosophila RNAi line ( GAL4/UAS-obp28a RNAi ) that expresses reduced levels of obp28a and an obp28a mutant line ( Obp28a- ) . Following thoracic injury with a clean needle , we observed that Drosophila tub-GAL4/UAS-obp28a RNAi and Obp28a- adults perished significantly faster than did control individuals [tub-GAL4 driver , UAS-obp28a RNAi and wCs ( mutant ) progenitor lines; Figure 5C] . Additionally , at 6 hr post-injury , melanin had deposited at the wound site of control Drosophila but not their obp28a knockdown or knockout counterparts ( Figure 5D ) . Collectively these findings suggest that OBP-mediated hematopoiesis represents an evolutionarily conserved mechanism that benefits these flies by preventing dehydration and/or exposure to opportunistic infections with environmental microbes following cuticular injury . Tsetse flies must house their maternally transmitted enteric symbionts during larval development in order to present a functional cellular immune system as adults . In the absence of these microbes , tsetse larvae express reduced levels of the GATA and RUNX transcription factors serpent and lozenge . This inhibition prevents the production of blood cell progenitors and thus the differentiation of phagocyte and crystal cell lineages ( Weiss et al . , 2011 , 2012 ) . Herein , we characterize an obligate symbiont regulated tsetse gene that actuates a distinct component of the fly’s hematopoietic program . Specifically , these microbes regulate obp6 transcript abundance in developing intrauterine larvae . The encoded protein subsequently induces expression of lozenge , which drives a pool of larval hemocyte precursors to differentiate into functional crystal cells . When adult tsetse house a depleted population of this hemocyte subtype , which initiates the melanization cascade via the release of prophenoloxidase , cuticular wounds fail to clot , thus leaving the fly exposed to dehydration and/or infection with opportunistic environmental microbes . This phenotype is similar to that observed when Wigglesworthia-free flies are exposed to cuticular wounds ( Weiss et al . , 2011; Figure 6 ) . We further discovered that Drosophila’s indigenous microbiota regulate orthologous components of their host’s hematopoietic program . Taken together , these findings accentuate the functional relevance of symbiotic bacteria as they relate to hematopoiesis , and detail an evolutionarily conserved component of the insect innate immune system that coordinates a cellular process essential for survival . 10 . 7554/eLife . 19535 . 027Figure 6 . Model illustrating the functional relationship between maternally-transmitted enteric symbionts and melanization in tsetse . GmmWT larvae imbibe enteric symbiotic-containing milk gland secretions throughout their intrauterine developmental program . These bacteria colonize larval gut-associated tissues , including the bacteriome , and in doing so , induce the expression of obp6 . OBP6 is either secreted directly into the hemolymph , or acts locally to induce expression of another unknown , ( also secreted ) protein . One of these molecules then acts systemically in the larval hematopoietic niche ( hn ) to stimulate lozenge ( lz ) expression in a small proportion of serpent ( srp ) expressing prohemocytes . These cells then become PPO-producing crystal cells [remaining prohemocytes become phagocytes after expressing glial cells missing ( gcm ) ] . Finally , crystal cells are expelled from the hn , where they circulate in the hemolymph and are available to produce wound-healing melanin . Larvae that develop in the absence of symbiotic bacteria ( GmmApo ) fail to produce any hemocytes , while those that develop in the presence of reduced obp6 transcript abundance ( GmmOBP6- ) fail to express lozenge and thus likely fail to generate crystal cells . dv , dorsal vessel; hc , hemocoel; w , wound; ep , epithelial cells of midgut; bc , bacteriome; pm , peritrophic matrix; gl , gut lumen . DOI: http://dx . doi . org/10 . 7554/eLife . 19535 . 027 When tsetse and Drosophila larvae express reduced levels of orthologous obp6 and obp28a , respectively , subsequent adults exhibit defective melanization cascades and thus perish unusually fast following wounding with a clean needle . These phenotypes result from reduced expression of hematopoietic lozenge and a consequently depleted population of PPO-producing crystal cells . In tsetse , obp6 expression is tightly linked with intrauterine larval recognition of obligate Wigglesworthia . Drosophila larvae are free-living and thus do not obtain bacteria trans-generationally . Instead , female Drosophila lay their eggs in decaying organic matter , and immediately following eclosion , larvae begin feeding on this substrate ( Markow , 2015; Broderick and Lemaitre , 2012 ) . Thus , the gut of larval Drosophila is colonized by a relatively diverse population of environmentally acquired bacteria ( Chandler et al . , 2011;Wong et al . , 2013 ) . Axenic Drosophila larvae develop slower and weigh less than wild-type individuals ( Shin et al . , 2011; Storelli et al . , 2011 ) , and adults exhibit reduced stem cell activity and epithelial turnover ( Buchon et al . , 2009; Broderick et al . , 2014 ) and can be more susceptible to enteric viral and bacterial pathogens ( Blum et al . , 2013; Sansone et al . , 2015 ) . The immune-compromised phenotypes exhibited by axenic adult Drosophila may reflect a developmental deficiency that , similar to the situation in tsetse , occurs when larval stages mature in the absence of symbiotic bacteria . This scenario would suggest that Drosophila’s indigenous microbiota also influences hematopoietic processes via a mechanism homologous to Wigglesworthia’s influence on tsetse . These convergent mechanisms , which control crucial immune phenotypes that are coordinated by evolutionarily conserved genes and regulatory pathways , exist despite the divergent nature of these host-symbiont model systems . Specifically , the tsetse-Wigglesworthia symbiosis originated 50–80 million years ago ( Chen et al . , 1999 ) , and due to the bacterium’s strict vertical route of transmission , is highly steadfast in nature . Conversely , Drosophila’s relationship with its microbiota is relatively transient and highly dependent on diet and local environmental factors ( Blum et al . , 2013; Broderick et al . , 2014 ) . The extent to which obp6 and obp28a orthologues are functionally conserved in other arthropods remains to be determined . Furthermore , although vertebrates lack direct obp6 and obp28a orthologues ( Kriventseva et al . , 2015 ) , unrelated odorant binding proteins may present similar roles in these systems . Innate immunity is germ-line encoded and regulated by highly conserved pathways ( Rubin et al . , 2000; Lemaitre and Hoffmann , 2007 ) , some of which are actuated by systemically derived signals . This type of molecular coordination is well characterized in Drosophila , where insulin secreted by insulin-producing cells in the brain , and essential amino acids emanating from the gut , promote Wingless signaling that maintains blood cell progenitors in the hematopoietic niche ( Shim et al . , 2012 ) . Smell also contributes to progenitor maintenance in larval Drosophila . Specifically , olfactory receptor neurons , stimulated by the detection of small food-derived volatile molecules , secrete GABA into the larval hemocoel where it subsequently binds to blood cell progenitors in the lymph gland . This interaction increases cytosolic calcium concentrations that are necessary and sufficient to maintain the progenitor population . When larval Drosophila are reared on an odor-free diet , they fail to retain a pool of these cells ( Shim et al . , 2013a ) . These findings highlight the intriguing association between smell and homeostasis of innate immune-related activities . For the experiments performed herein , Drosophila and tsetse were reared in the absence of indigenous microbes , but on diets ( sterilized ) and in environments that emitted normal food odors . Under these conditions both flies presented dysfunctional hematopoietic programs , indicating that microbe-derived factors may also influence hematopoiesis . While the exact chemical structure of these molecules is currently unknown , they could take the form of microbe-associated molecular patterns ( MAMPs ) , including bacterial cell wall components such as polysaccharide A , peptidoglycan and Nod-like receptor ligand , which in mammals actuate differentiation of T cell lineages ( Mazmanian et al . , 2005 ) , enhance bone marrow-derived neutrophil killing ( Clarke et al . , 2010 ) and stimulate macrophage activity in lung tissues ( Clarke , 2014b; Gauguet et al . , 2015 ) , respectively . Additionally , microbial metabolites mediate regulatory T cell abundance in colonic tissues , as germ-free mice that lack gut microbiota-derived fatty acids present fewer of this immune cell type ( Smith et al . , 2013; Furusawa et al . , 2013 ) . These findings accentuate the concept that general metabolic defects associated with dysbiosis can give rise to cellular immunity-related pathologies ( Norata et al . , 2015 ) , including impairment of hematopoietic programs ( Shim et al . , 2012 , 2013b ) . Bacteria also release a wide variety of volatile compounds ( Audrain et al . , 2015 ) , some of which , following chemosensory detection , influence animal immune phenotypes . A situation of this nature occurs when adult Drosophila detect geosmin , which is a microbial odorant that signals the presence of harmful microbes . This stimulus induces expression of olfactory receptor Or56a in olfactory sensory neurons , which leads to an aversion behavior , reduction in the activity of other olfactory pathways , and inhibition of positive chemotaxis , oviposition and feeding behavior ( Stensmyr et al . , 2012 ) . A similar phenomenon occurs in the nematode Caenorhabditis elegans , which executes a protective avoidance behavior following chemosensory detection of the aromatic compounds phenazine-1-carboxamide and pyochelin that are produced by pathogenic Pseudomonas aeruginosa ( Meisel et al . , 2014 ) . Drosophila detects and prefers specific odors produced by members of its microbiome ( Fischer et al . , 2016 ) . As such , odorant molecules derived from tsetse and Drosophila enteric microbes could modulate the expression of obp6 and obp28a in the gut of larval flies , thus serving as the signal that activates hematopoiesis . In larval tsetse Wigglesworthia is found exclusively within the gut-bacteriome axis ( Balmand et al . , 2013 ) , and obp6 expression is restricted to this same environment . Conversely , lozenge transcripts are found solely in the carcass , indicating that the fly’s hematopoietic niche is likely not attached to the gut . The spatial expression pattern of these genes indicates that OBP6 exerts a systemic effect on tsetse hematopoiesis , and this process could occur in one of two ways . First , this protein could act locally at the mucosal interface to induce the production and secretion of a distinct immuno-stimulatory molecule that subsequently acts in tsetse’s hematopoietic niche . A mechanism of this nature occurs in mice , where enteric symbiont-derived molecules induce group three innate lymphoid cells in the intestine to increase production of the cytokine interleukin-17 ( IL-17 ) . Circulating IL-17 subsequently induces production of peripheral granulocyte colony-stimulating factor , which actuates neutrophil differentiation in bone marrow tissues ( Deshmukh et al . , 2014 ) . Insect OBPs are widely believed to serve as vehicles that carry odorant molecules from sensory sensillum pores to corresponding dendritic odorant receptors ( Leal , 2013 ) . Based on this definition , tsetse OBP6 could also serve a carrier-like function following translocation from the fly’s gut into its circulatory system . More specifically , this protein could complex with another hematopoietic molecule produced either in tsetse’s gut or extra-intestinally , and then both together , or the latter molecule alone , stimulate hemocyte development in the hematopoietic niche . Notably , the function of insect OBPs was defined based on experiments performed in the chemosensory apparatus , and the role of these proteins in other insect tissue types is largely unknown . Interestingly , a recent study exploring the roles of OBPs in Drosophila sensilla found that deletion of obp28a did not reduce the magnitude of fly olfactory responses , suggesting a novel role for the encoded protein ( Larter et al . , 2016 ) . As such , tsetse OBP6 could also function unconventionally by directly stimulating hematopoiesis following secretion into the larval hemocoel . Further studies are required to determine the specific cell type within larval tsetse’s gut and/or bacteriome that express obp6 . Additionally , identification of OBP6 target tissues will provide valuable insight into the immuno-stimulatory function of this protein . Finally , we speculate that our tsetse fly related findings reported herein are directly relevant to the transmission of vector-borne pathogens . Melanization is a cellular immune response that occurs across invertebrate taxa ( Cerenius and Söderhäll , 2011 ) , and this mechanism influences the transmission of several pathogens through their respective insect vectors ( Collins et al . , 1986; Zou et al . , 2008; Bartholomay , 2014 ) . Thus , an increased understanding of the physiological processes that regulate melanization may have translational implications pertinent to the development of pathogen-refractory insects . G . morsitans morsitans were maintained in Yale’s insectary at 24°C with 50–55% relative humidity . All flies received defibrinated bovine blood ( Hemostat Laboratories ) every 48 hr through an artificial membrane feeding system . Aposymbiotic tsetse larvae ( GmmApo ) were derived from females fed a diet supplemented with tetracycline ( 20 μg per ml of blood ) to clear their indigenous microbiota , and yeast extract ( 1% w/v ) to rescue the sterile phenotype associated with the absence of Wigglesworthia ( Alam et al . , 2011 ) . Thus , GmmApo offspring developed in the absence of all symbiotic bacteria . Embryos and 1st and second instar larvae for all experimental groups were age-matched by taking individual samples from pregnant females undergoing their second gonotrophic cycle ( Attardo et al . , 2014 ) . Axenic Drosophila larvae ( Oregon-R and w1118 strains ) were generated as described previously ( Broderick et al . , 2014 ) . Drosophila tub-GAL4 driver and UAS-obp28a RNAi progenitor lines were crossed , and resulting tub-GAL4/UAS-obp28a RNAi F1 offspring were used for experiments ( Dietzl et al . , 2007 ) . Anti-obp28a RNAi target gene specificity , knockdown efficacy and accompanying phenotypic characterization , were determined previously ( Swarup et al . , 2011 ) . Obp28a deletion mutants ( Obp28a- ) were generated from progenitor flies ( CAS-0003; Kondo and Ueda , 2013 ) using the CRISPR-Cas9 system , and backcrossed to w Canton-S ( wCs ) for five generations ( Larter et al . , 2016 ) . Flies were maintained on a cornmeal-yeast-agar medium ( per liter of water: 50 g inactivated yeast , 70 g maize flour , 6 g agar , and 40 g of dextrose ) at 25°C in ambient humidity . GFP-expressing E . coli K12 ( recE . coliGFP ) were produced via electroporation with pGFP-UV plasmid DNA ( Clontech ) . First and second instar tsetse larvae ( n = 5 of each ) were collected from two distinct cohorts of pregnant GmmWT and GmmApo females 48 hr post-feeding . Total RNA from the above-mentioned tsetse larvae was extracted , DNAse treated and purified as described previously ( Benoit et al . , 2014 ) . RNA-seq libraries were constructed using polyadenylated RNA and standard Illumina RNA-seq protocols . Libraries were sequenced at the McDonnel Genome Institute ( Washington University ) . Read files have been deposited in the NCBI BioProject database ( ID# PRJNA309164 ) . FastQC analyses were performed on the RNA-seq datasets to assess read quality . Low quality reads and sequencing adaptors were removed with Trimmomatic ( Bolger et al . , 2014 ) . Transcript expression levels were determined using CLC Genomics Workbench ( CLC Bio , Cambridge , MA ) . Briefly , RNA-seq datasets were mapped directly to the tsetse fly genome ( International Glossina Genome Initiative , 2014 ) with an algorithm that allowed only two mismatches and a maximum of 10 hits per read . Transcripts per million ( TPM ) was used as a proxy for gene expression . The predicted gene set associated with the genome was version 1 . 1 obtained from Vectorbase ( Giraldo-Calderón et al . , 2015 ) . The following samples were compared: GmmWT larvae against GmmApo larvae [Sequence Read Archive ( SRA ) IDs SRR3107831-SRR3107834 ( BioProject ID PRJNA309164 ) ] , and GmmWT larvae against GmmWT male and female adults ( BioProject IDs PRJNA295435 and PRJNA205861 , respectively; Scolari et al . , 2016; Benoit et al . , 2014 ) . SRA and BioProject sequence data is available at the NCBI website ( http://www . ncbi . nlm . nih . gov/sra and http://www . ncbi . nlm . nih . gov/bioproject , respectively ) . Relative fold differences in gene expression between samples were determined as a ratio of each TPM . Significance was determined via Baggerly’s test followed by a false detection rate at p<0 . 01 ( Baggerly et al . , 2003 ) . In conjunction with the above-described analyses , we conducted a de novo assembly of the larval transcriptomes using Trinity ( Grabherr et al . , 2011 ) , followed by RNA-seq analyses to identify additional differentially expressed genes of interest between GmmWT and GmmApo individuals . No additional targets were identified by this secondary analysis . Thus , subsequent functional studies focused on results obtained from the predicted genomic gene set . Predicted genes were annotated using tblastx , with an E-value cut-off of 1e−10 and bit score of 200 , to a previously annotated Glossina transcriptome ( Benoit et al . , 2014 ) . Another comparative analysis , using the same parameters , was performed with annotated protein sequences from D . melanogaster and Pediculus humanus from FlyBase and Vectorbase , respectively . Blast2GO was utilized to identify specific gene ontology ( GO ) terms that were enriched between treatments based on a Fisher’s Exact Test ( Conesa et al . , 2005 ) . Specific GO-based functional categories were developed based on comparison with associated D . melanogaster genes acquired from Flybase ( Marygold et al . , 2013; Attrill et al . , 2016 ) . These categories included those involved in B vitamin metabolism , hematopoiesis , midgut development , larval development , immunity , organismal growth and chitin associated . For these category assignments Drosophila and G . morsitans gene sets were compared , and a functional match was considered valid if the E-value was below 10−40 and the bit score was over 200 . Enrichment for a specific GOC associated with each sample was determined with a Fisher’s Exact Test . Specific analyses of the odorant binding protein genes were accomplished by obtaining predicted models for these genes from the G . morsitans genome . A cartoon summarizing temporal aspects of RNAi and subsequent functional experiments is shown in Supplementary file 3 . Two cohorts ( n = 150 individuals per group ) of virgin female tsetse were mated three days post-eclosion ( dpe ) , and embryos ( n = 3 in each of seven biological replicates ) were collected five days later from a subset of individuals to obtain baseline serpent ( srp ) and lozenge ( lz ) expression values . Subsequently , pregnant female flies were subjected to thoracic microinjection ( using glass needles and a Narashige IM300 micro-injector ) with either anti-obp6 ( treatment ) or anti-gfp ( control ) siRNAs ( siRNA sequences listed in Supplementary file 5 ) on days 8 and 11 post-mating . This window of time post-mating was chosen in an effort to expose feeding first and/or second instar larval tsetse to siRNAs ( generated by Integrated DNA Technologies , Coralville , IA ) taken up by the milk gland following inoculation into the maternal hemocoel . Anti-obp6 siRNA was coupled to a Cy3 dye ( siOBP6Cy3 ) to track nucleic acid dissemination through the maternal hemocoel and into the developing intrauterine larvae . siOBP6Cy3 larvae were rigorously washed in PBS prior to stimulation with UV light to ensure Cy3 labeled siRNA was removed from the cuticular surface . All anti-obp6 and anti-gfp siRNA treated moms and their larval and adult offspring are designated ‘siOBP6’ and ‘siGFP’ , respectively , throughout this study . siRNA target specificity was confirmed in silico at VectorBase via BLAST analysis against a G . morsitans RNA-seq library , and a complete set of tsetse genomic scaffolds ( both available on the VectorBase website; www . vectorbase . org ) . Real time quantitative real-time PCR ( RT-qPCR ) was performed as described previously ( Weiss et al . , 2012 ) . All RT-qPCR results were normalized relative to tsetse's constitutively expressed β-tubulin or Drosophila’s Rpl32 gene ( determined from each corresponding sample ) . Replicate numbers and sample sizes are presented on figures or in their corresponding legends . For semi-quantitative reverse transcription PCR ( RT-PCR ) analysis , second instar individuals were removed from pregnant females and an incision was made the length of the larval cuticle . The larval gut , which rapidly exudes following cuticular incision , and the corresponding carcass , were collected separately in PBS . RNA and cDNA were made from larval gut and carcass tissues as described previously ( Weiss et al . , 2012 ) . Tsetse β-tubulin was used as a loading control . RT-qPCR and RT-PCR primers are listed in Supplementary file 6 . Primer target specificity was confirmed in silico at VectorBase and FlyBase via BLAST analysis against G . morsitans and D . melanogaster RNA-seq libraries . siGFP and siOBP6 adult tsetse were subjected to systemic challenge with recE . coliGFP during adulthood . Percent survival was subsequently monitored over a two week period . For E . coli infections , tsetse were anesthetized on ice and subsequently injected with 5 × 102 colony forming units ( CFU ) of live bacterial cells using glass needles and a Narashige IM300 micro-injector . ‘Clean’ wounds were administered to siGFP , siOBP6 and siOBP6R adult tsetse , and tub-GAL4 , UAS-obp28a RNAi , tub-GAL4/UAS-obp28a RNAi , Obp28- and wCs adult Drosophila , by pricking individual flies in the thorax with a heat sterilized glass needle . Injured tsetse were housed under normal insectary conditions , while injured Drosophila were maintained in a desiccated environment with no access to food or water . All tsetse survival experiments were performed in triplicate , using 25 flies per replicate . All Drosophila survival experiments were performed in triplicate using 50 flies per replicate . Hemolymph was collected by removing a fly leg with forceps and exerting gentle pressure on the abdomen , thus causing a hemolymph droplet to exude from the neck . Determination of circulating hemocyte abundance was performed using a Bright-Line hemocytometer . Hemocyte phagocytic capacity of siOBP6 and siGFP adults was determined by injecting individuals with 5 × 102 CFU of live recE . coliGFP . Six hours post-inoculation , hemolymph was collected from three individuals and hemocytes monitored for the presence of engulfed GFP-expressing bacterial cells . Hemolymph samples were fixed on glass microscope slides via a 2 min incubation in 2% paraformaldehyde and then overlaid with VectaShield hard set mounting medium containing DAPI ( Vector Laboratories , Burlingame , CA ) . Cells were visualized using a Zeiss Axioscope microscope . To quantify rec E . coliGFP in siOBP6 and siGFP individuals at 2 and 6 days post-challenge ( n = 5 per siRNA treatment per time point ) , 3 μl of hemolymph was serially diluted in 0 . 85% NaCl and plated on LB/agar supplemented with ampicillin ( 50 μg/ml ) . CFU per plate were counted manually . Western blots were performed in triplicate , with each replicate containing 8 μl of hemolymph . For tsetse , 2 μl of hemolymph ( collected by removing a leg at the proximal joint nearest the thorax ) was pooled ( and then immediately frozen ) from four flies per group 3 hr post-wounding ( hpw ) with a clean glass needle . Adult Drosophila ( n = 75 per group ) were thoracically wounded with a sterilized tungsten needle . Three hpw , flies were chilled on ice and placed into Zymo-Spin IV columns ( n = 15 flies per column; Zymo Research , Irvine , CA ) preloaded with 0 . 5 mm glass beads ( Scientific Industries , Bohemia , NY ) . Columns were centrifuged at 4°C for 15 min and hemolymph pooled in the column collection tube was frozen . Denatured ( 100°C for 5 min in protein loading buffer ) hemolymph extracts were separated on a 10% polyacrylamide gel , transferred to nitrocellulose membranes , blocked with 3% bovine serum albumin ( prepared in PBST buffer ) for 1 hr at room temperature and incubated overnight at 4°C with rabbit anti-PPO1 or anti-PPO2 ( generated against recombinant Drosophila PPO1 and PPO2 , respectively; Nam et al . , 2012 ) antibodies at a 1:1500 dilution . Blots were subsequently probed with an HRP conjugated goat anti-rabbit 2° antibody ( BioRad , Hercules , CA ) , and PPO protein bands were visualized using a SuperSignal West Pico Chemiluminescent Substrate kit according to the manufacturer’s ( Thermo Scientific , Waltham , MA ) protocol . L-3 , 4-dihydroxyphenylalanine ( L-DOPA; Sigma-Aldrich , St . Louis , MO ) assays ( performed as described in Perdomo-Morales et al . , 2007; Binggeli et al . , 2014 ) were used to quantify enzymatic phenoloxidase ( PO ) activity in 3 μl of hemolymph collected from siOBP6 , siGFP and siOBP6R individuals immediately ( 0 hr; n = 5 individuals per group at this time point ) and 3 hr post-wounding ( n = 8 individuals per group at this time point ) with a clean needle . Enzymatic activity of tsetse’s melanization cascade was arrested by adding protease inhibitor at the time of hemolymph collection . Thus , values reflect in vivo PO at this time point . Values are represented as the mean ( ±SEM ) . In Drosophila , sessile crystal cells attached to the hemocoelic side of the larval cuticle can be visualized as dark spots following spontaneous activation of PPO ( Binggeli et al . , 2014 ) . This phenotype was induced in tsetse and Drosophila larvae by heating individuals to 65°C for 10 min . PPO spots were quantified visually using a dissecting microscope ( Zeiss Discovery ) equipped with a digital camera ( Zeiss AxioCam MRc 5 ) . Three cohorts ( n = 25 individuals/group ) of pregnant female tsetse were fed a diet containing tetracycline ( 100 μg/ml of blood ) every other day for 10 days to eliminate all indigenous bacteria . All blood meals ( three per week ) , throughout the course of the entire experiment , also contained vitamin-rich yeast extract ( 1% w/v ) to restore fertility associated with the absence of Wigglesworthia ( Weiss et al . , 2012 ) . Ten days post-copulation , 4 cohorts of symbiont-cured females were regularly fed a diet supplemented with either 1 ) Wigglesworthia-containing bacteriome extracts ( obtained by dissecting bacteriomes from GmmWT females ) , 2 ) Wigglesworthia-free bacteriome extracts ( derived from the offspring of females fed a diet supplemented with ampicillin , which results in the production of progeny that lack Wigglesworthia but still harbor Sodalis; Pais et al . , 2008; Weiss et al . , 2011 ) , 3 ) Sodalis cell extracts ( derived from Sodalis maintained in culture; Hrusa et al . , 2015 ) , and 4 ) bacteriome extracts harvested from aposymbiotic females . Offspring ( symbiont-free ) of these extract supplemented females were designated Gmmbact/Wgm+ , Gmmbact/Wgm- , GmmSgm+ and Gmmbact/Apo , respectively . Bacteriome supplemented females were fed one tissue equivalent per four females , and GmmSgm+ females were fed 4 × 107 Sodalis per ml of blood ( these flies thus ingested ~1 × 106 Sodalis each time they fed ) . Control cohorts consisted of wild-type ( GmmWT ) and aposymbiotic ( GmmApo ) offspring . RT-qPCR was used to determine if complementation with bacterial extracts plus yeast altered the expression pattern of obp6 in aposymbiotic larvae ( First and second instar ) from the second gonotrophic cycle ( these offspring were used to ensure that antibiotic treatment had cleared all maternal symbionts such that none were present for transmission to larvae ) of symbiont-cured moms . Throughout the manuscript , all replicates are ‘biological’ , implying that data were obtained by repeating experiments using the indicated number of distinct samples . Replicates and sample sizes for all experiments are provided in the legend that corresponds to each representative figure ( except for Figure 1 , for which sample size is indicated in the ‘Materials and methods section , subheading ‘Transcriptomics’ ) . Statistical significance between treatments and controls is indicated on figures or in the corresponding figure legends . Tests used to determine statistical significance are indicated in figure legends . All statistical analyses were performed using GraphPad Prism software ( v . 6 ) .	Bacteria live within all animals . While a small number of these microbes can cause disease , most promote the health and wellbeing of their host . Microbes that support their host and benefit from the close association are often referred to as symbionts . Animals can be negatively affected and even become diseased if their symbionts are disrupted . As a result , a more complete understanding of the molecular interactions between animal hosts and their beneficial microbes will lead to better treatments for a number of diseases . Tsetse flies are insects that harbor two bacterial symbionts , which are transferred from pregnant females to their larval offspring . If the offspring mature without these microbes , they fail to develop cells called hemocytes . These cells are normally found in the insect’s equivalent of blood – a fluid called hemolymph – and they comprise an important component of the insect’s immune system . Adult tsetse flies that lack hemocytes are susceptible to certain infections . These findings indicate that the bacterial symbionts induce the production of hemocytes in tsetse fly larvae via an unknown mechanism . Benoit et al . now reveal that the bacterial symbionts trigger tsetse flies to produce a small protein called “odorant binding protein 6” . This protein controls the generation of one specific type of hemocyte called crystal cells in developing larvae . Crystal cells are largely responsible for triggering the production of melanin , a protein involved in killing disease-causing microbes and inhibiting the loss of hemolymph from wound sites in the insect’s exoskeleton . Benoit et al . discovered that bacterial symbionts associated with the larvae of fruit flies also support the development of their host’s immune system . Although these symbionts are acquired from the external environment rather than from the insect’s parent , they too control the production of an odorant binding protein and crystal cells in their larval host . Collectively , these findings confirm that bacterial symbionts are critically important for the development of the immune systems of insects , and they show that this process has been conserved throughout evolution . Future studies are likely to focus on identifying which molecules from the symbionts stimulate their hosts to produce new hemolymph cells . Furthermore , identifying which tissues and cell types in the animal hosts are targets for these molecules will provide a more complete picture of the pathways that lead to the production of new hemolymph cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "immunology", "and", "inflammation" ]	2017	Symbiont-induced odorant binding proteins mediate insect host hematopoiesis
Mammalian pluripotent stem cells ( PSCs ) represent an important venue for understanding basic principles regulating tissue-specific differentiation and discovering new tools that may facilitate clinical applications . Mechanisms that direct neural differentiation of PSCs involve growth factor signaling and transcription regulation . However , it is unknown whether and how electrical activity influences this process . Here we report a high throughput imaging-based screen , which uncovers that selamectin , an anti-helminthic therapeutic compound with reported activity on invertebrate glutamate-gated chloride channels , promotes neural differentiation of PSCs . We show that selamectin’s pro-neurogenic activity is mediated by γ2-containing GABAA receptors in subsets of neural rosette progenitors , accompanied by increased proneural and lineage-specific transcription factor expression and cell cycle exit . In vivo , selamectin promotes neurogenesis in developing zebrafish . Our results establish a chemical screening platform that reveals activity-dependent neural differentiation from PSCs . Compounds identified in this and future screening might prove therapeutically beneficial for treating neurodevelopmental or neurodegenerative disorders . Mouse embryonic stem cells ( mESCs ) , capable of generating most cell types that constitute the entire organism , have made important contributions to our understanding of mammalian biology ( Smith , 2001 ) . How mESCs differentiate into neural lineages is a fascinating question that remains incompletely understood ( Okano and Temple , 2009; Gaspard and Vanderhaeghen , 2010 ) . Neural induction , the first step in neural differentiation of mESCs , requires active FGF signaling ( Ying et al . , 2003 ) and inhibition of the BMP/TGF-beta pathway ( Chambers et al . , 2009 ) . Subsequent regional identity and lineage-guided differentiation are further regulated by the presence or absence of various morphogens or transcription factors ( Lee et al . , 2000; Wichterle et al . , 2002; Andersson et al . , 2006; Martinat et al . , 2006 ) . However , it is unknown whether mechanisms additional to growth factors and transcription regulators direct the differentiation of mESCs into neural lineages . Effective means for perturbing a complex biological system are key to gaining new insights into the underlying molecular and cellular mechanisms . Small organic molecules have proven to be invaluable tools for probing biological mechanisms , owing to their versatile nature and ease of application and removal from the system under study ( Stockwell , 2004; Zon and Peterson , 2005 ) . These features also make bioactive small molecules highly attractive for therapeutic applications . One critical challenge in small molecule discovery is that the chemical space is infinite , thereby requiring high throughput screening for speed and bioassays that are of sufficient specificity and sensitivity to distinguish active small molecules from background noise . Here we report an imaging-based screen of ∼2000 bioactive compounds in mESC monolayer cultures labeled with the anti-tyrosine hydroxylase ( TH ) antibody ( a marker for dopaminergic , noradrenergic , and adrenergic neurons ) . We identified small molecules that increased the appearance of TH+ neurons in the assay , including those with known neurotrophic activity and those that are functionally novel . Notably , we show that the anti-parasitic compound selamectin , with reported activity on invertebrate glutamate-gated chloride channels , increased not only TH+ neurons , but also multiple other neural types including the serotonergic ( 5-HT ) , GABAergic , and Islet+ motor neurons as well as Olig2+ oligodendrocytes . We further reveal , through pharmacology , genetics , single-cell electrophysiological recordings and clonal analyses , that selamectin acts by enhancing GABAA receptor signaling , increasing the expression of proneural and lineage-specific transcription factors , and promoting cell cycle exit and differentiation of neural progenitors . We also demonstrate that selamectin can increase neuronal differentiation in human ESCs and induced pluripotent stem cells ( iPSCs ) as well as in vivo in the developing zebrafish . In order to apply chemistry to probe the basic biology of neural differentiation from pluripotent stem cells ( PSCs ) , we designed a high content screen to isolate small molecules that can increase the total number of TH+ neurons derived from mESC monolayer cultures . The system was chosen for several reasons: First , the mESC culture system is an established model for understanding neural development , with much insight gained in recent years ( Okano and Temple , 2009; Gaspard and Vanderhaeghen , 2010 ) . Second , mESCs can be cultured in large quantities and in multi-well plates in a high throughput manner . Finally , our adaptation of the mESC monolayer culture and differentiation method ( Ying and Smith , 2003 ) showed that a relatively low and consistent number of TH+ neurons were detected in the culture system ( Figure 1 ) . 10 . 7554/eLife . 00508 . 003Figure 1 . High-throughput screening . ( A ) Scheme of the three-stage mESC neuronal differentiation-based chemical screening . ( B ) E14 mESC cultures express neural progenitor makers ( Sox2 , Lmx1A , and Nestin ) after 7-day stage one culture . 46C mESC cultures , in which GFP is driven by the sox1 promoter , are GFP positive after 7-day stage one culture . ( C ) The quantification analysis of TH+ neurons among total cells using the INCell Developer software . ( D ) Summary of coefficient of variation ( C . V . ) of the DMSO control . ( E ) Representative images of immunostaining in control ( left , treated with DMSO ) and a hit compound ( right ) . ( F ) A schematic summary of the chemical screen . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00508 . 003 A three-stage protocol was devised ( ‘Materials and methods’ for details ) ( Figure 1A ) . The mESCs of both E14 and 46C lines ( Ying and Smith , 2003 ) were used , the latter of which expresses GFP reporter under the control of sox1 promoter . During stage one , undifferentiated mESCs were cultured on a gelatin-coated surface and in the media without LIF , resulting in neural progenitors that express Sox2 , Lmx1a , Nestin , and Sox1 ( Figure 1B ) . At stage two , neural progenitors were plated into multi-well plates and treated with chemicals for three days . Finally , chemical treatment was withdrawn and cells were cultured for additional three days before immunostaining with anti-TH antibody ( Stage three ) . This protocol was further subjected to automation at multiple steps , including cell dispensing into 96-well plate using Thermo Matrix Well Plate , compound distribution into wells using Biomek FXP Laboratory Automation Workstation , immunostaining using Thermo Matrix PlateMate Plus , image capture using GE INCell 1000/2000 , and image quantification using INCell Developer software ( ‘Materials and methods’ for details ) . We then screened a library containing 2080 biologically active and structurally diverse compounds , including many FDA approved and currently marketed drugs . Compounds were screened at a final concentration of 1 μM in a volume of 120 μl per well containing 0 . 67% DMSO ( vol/vol ) . After automated immunostaining , image acquisition , and image analysis , the percentage of TH+ cells in each well was calculated ( Figure 1C ) . We did not use actual cell count ( as cells in the well are not well separated , making ‘cell count’ inaccurate ) ; instead , we calculated the area of each segmented target . The percentage of TH signal in each well was expressed as a ratio of TH-covered area over DNA-covered area . The final readout was calculated as fold change compared to the DMSO-treated control . The cut-off for selecting primary hits was set as fold change > mean + 3 S . D . relative to DMSO control , which is a rather stringent selection criteria based on previous studies ( Borowiak et al . , 2009 ) . To assess assay performance , the coefficient of variation ( C . V . ) of DMSO control was calculated for each of the twenty-six 96-well plates screened , and all C . V . s but one were smaller than 20% , suggesting an acceptable variation during this cell-based screen ( Figure 1D ) . Out of 2080 chemicals screened , 26 led to a fold change of TH+ cells larger than mean + 3 S . D . ( 1 . 16% ) ( Figure 1E for an example ) , and 20 out of the 26 were neither cytotoxic nor auto-fluorescent ( Figure 1F ) . After two rounds of validation , two compounds were selected as hits , yielding an overall hit rate of 0 . 09% . One identified molecule is Dihydrodeoxygedunin ( DOG ) , which is a natural product with known neurotrophic activity via activating the TrkB receptor and its downstream signaling cascades ( Jang et al . , 2010a ) . Both DOG and 7 , 8-dihydroxyflavone ( DHF , another selective TrkB agonist [Jang et al . , 2010b] ) increased TH+ cells in mESC cultures , albeit modestly ( Figure 2 ) . This data suggest that our screen is capable of identifying compounds with neuronal promoting activity . 10 . 7554/eLife . 00508 . 004Figure 2 . The neurotrophin receptor TrkB agonists [Dihydrodeoxygedunin ( DOG ) and 7 , 8-dihydroxyflavone ( DHF ) ] increases TH+ cells in mESC cultures . ( A ) Structure of DHF and DOG . ( B ) DHF and DOG increase TH% in mESCs ( t test , p<0 . 05 , n = 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00508 . 004 The other hit from our screen is selamectin , whose role in promoting ESC differentiation into TH+ neurons is novel , and was selected for further study . We first determined whether selamectin-induced increase of TH+ neurons is selective for these subtypes by immunocytochemistry with the pan-neuronal marker NeuN . Treatment with selamectin increased the percentage of total neurons , compared to the DMSO-treated control ( Figure 3A–B ) . This result suggests that the effect of selamectin is not specific to TH+ neuronal subtypes . Further analysis showed that selamectin also significantly increased the production of 5-HT neurons ( Figure 3C ) , GABAergic neurons ( Figure 3D ) , and Islet+ motor neurons ( Figure 3E ) . The increase of 5-HT neurons was remarkably high ( ∼sevenfold ) , suggesting that selamectin might have a preferential activity for inducing 5-HT neurons . 10 . 7554/eLife . 00508 . 005Figure 3 . Selamectin increases mESC differentiation into multiple neural lineages . ( A ) Representative images of TH and NeuN staining in control ( DMSO ) and selamectin-treated cultures . ( B ) Quantification shows increased production of both TH+ and total neurons by selamectin ( t-test , n = 4 , p<0 . 001 ) . ( C–F ) Selamectin treatment also increases the production of 5-HT neurons ( C ) , GABAergic neurons ( D ) , islet+ motor neurons ( E ) , and Olig2+ oligodendocyte precursors ( F ) ( t-test , n = 4 , p<0 . 05 , p<0 . 01 , p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00508 . 005 Since neuronal differentiation precedes that of gliogenesis in the vertebrate central nervous system ( Alvarez-Buylla et al . , 2001; Mehler , 2002 ) , we wondered whether the increased neuronal production by selamectin is at a cost of later-born glial cells . As our culturing protocol largely favors the differentiation of neurons , we were able to observe only a small numbers of Oligo2+ cells at a much later stage ( Day 18 ) , suggesting that they were likely to be oligodendrocytes or their precursors . After treatment with selamectin , the percentage of Oligo2+ cells also increased compared with the DMSO-treated control ( Figure 3F ) , suggesting that the increase of neuronal production is not at the expense of glial production . Taken together , selamectin appears capable of promoting mESC differentiation into both neuronal and oligodendrocyte lineages . To determine the dose response and time course of selamectin action , we tested a wide range of selamectin concentrations ( from 32 nM to 500 nM ) . The result showed that selamectin functions in a dose-dependent manner with an effector concentration for half-maximum response EC50 = 293 nM ( Figure 4A ) . When the concentration of selamectin was higher than 500 nM , it became toxic to cells . 10 . 7554/eLife . 00508 . 006Figure 4 . Dose response and time course of selamectin’s action in mESC cultures . ( A ) Dose response curve of selamectin ( based on TH immunostaining ) . Cells were treated from Day 8 to Day 11 . Selamectin is toxic above 500 nM . The EC50 value and curve fitting were performed with GraphPad Prism using a Sigmoidal dose-response ( variable slope ) . Data are presented as mean ± S . D . , n = 4 . ( B ) Time course of selamectin’s action . Only cells treated with selamectin from Day 7 to Day 10 show significant increase of TH+ neurons ( t-test , n = 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00508 . 006 Importantly , analysis of the temporal response showed that when cells were treated with selamectin at different stages , its effect was quite different . Treatment during Stage one ( from Day 4 to Day 7 , mainly composed of undifferentiated ESCs ) suppressed cell proliferation , resulting in insufficient number of cells on Day 7 for further evaluation . Treatment during Stage two ( from Day 7 to Day 10 , mainly composed of neural progenitors ) led to a significant increase of neurons evaluated on Day 13 ( Figure 4B ) . Shortening the treatment regimen to one day ( i . e . , from Day 7 to Day 8 ) or treatment during Stage three ( from Day 10 to Day 13 , mainly composed of differentiating neurons ) showed no significant effect of selamectin ( Figure 4B ) . These results suggest that selamectin’s proneurogenic activity is likely due to its action on mESC-derived neural progenitors . Selamectin belongs to a chemical family of macrocyclic lactones used to treat nematode infections that cause onchocerciasis ( also known as the river blindness ) in humans ( Goa et al . , 1991 ) and as topical or oral parasiticide and antihelminthic on dogs and cats ( Bishop et al . , 2000 ) . This class of drugs disables parasites by displacing glutamate in their muscle synapses through acting on the glutamate-gated chloride ( GluCl ) channels ( Bloomquist , 2003 ) , which have no orthologues in vertebrates . A better-studied example of this family is avermectin ( for the structures of selamectin and avermectin , Figure 5A ) , which , when tested for its activity in promoting neuronal differentiation , showed a less potent effect than selamectin with a marginal significance achieved at 0 . 25 μM ( Figure 5B ) . 10 . 7554/eLife . 00508 . 007Figure 5 . Pharmacological evidence indicates that selamectin’s proneurogenic activity is mediated by GABAA receptors . ( A ) Selamectin and avermectin belong to a chemical family of macrocyclic lactones and have the same structural backbone . ( B ) Avermectin has less potent proneurogenic activity than selamectin . Only 250 nM avermectin shows a significant effect , which was much less potent compared to that of 250 nM selamectin ( t-test , n = 4 , p=0 . 034 for Avermetin vs p=0 . 002 for selamectin ) . ( C ) Taurine , the most abundant endogenous ligand for glycine receptors during neocortical development has no proneurogenic activity in mESC cultures . ( D ) Muscimol , a GABAA receptor agonist , has a significant proneurogenic activity ( t-test , n = 4 , p<0 . 001 ) . ( E ) Chlordiazepoxide ( CDZ ) , a positive allosteric modulator of GABAA receptor , also has a significant proneurogenic activity ( t-test , p<0 . 05 ) , but there was no obvious additive effect when cells were treated with both selamectin and CDZ ( t-test , p=0 . 480 ) . ( F ) The GABAA receptor antagonist bicucullin and non-competitive blockers picrotoxin and pentylenetetrazol alone had no effect on neuronal production ( white columns , control group were normalized to fold change = 1 , displayed as the red dot line ) . However , when tested together with selamectin , the effect of selamectin was blocked ( gray columns ) . In contrast , the glycine receptor inhibitor strychnine does not block the effect of selamectin . Final concentration: Bicuculline ( Bicu ) = 100 μM; Picrotoxin ( PTX ) = 500 μM; Pentylenetetrazol ( PTZ ) = 5 mM; Strychnine ( STY ) = 100 μM; Selamectin ( Sela ) = 0 . 3 μM; Muscimol ( Musci ) = 10 μM ( t-test , n = 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00508 . 007 Since GluCl channels are not found in vertebrates , selamectin must act on other protein targets in mESC cultures . Based on gene structure and phylogenetic analyses , vertebrate glycine channels are thought to be orthologous to the invertebrate GluCl channels ( Vassilatis et al . , 1997 ) . It has been reported that ivermectin ( an avermectin derivative ) can act as an agonist of the glycine receptor ( Shan et al . , 2001 ) . Moreover , non-synaptically released taurine can activate the endogenous glycine receptor during neocortical development ( Flint et al . , 1998 ) . Therefore , we first tested whether taurine , the most abundant endogenous ligand for glycine receptors during neocortical development ( Agrawal et al . , 1971 ) , the deficiency of which affects cortical development in kittens ( Palackal et al . , 1986 ) , has the same proneurogenic effect as selamectin . We treated cells from Day 7 to Day 11 with a wide range of taurine concentrations , none of which , however , showed any effect in increasing TH% among total cells ( Figure 5C ) . This result suggests that selamectin does not act through glycine receptors to promote neuronal differentiation from mESCs . Next we turned to another potential candidate , the GABAA receptor , since some reports in the literature suggest that avermectin can bind to the GABAA receptor in rat brain membranes or cultured cerebellar neuronal assays ( Pong et al . , 1982; Huang and Casida , 1997 ) , and avermectin exhibits an anticonvulsant action in a mouse seizure model ( Dawson et al . , 2000 ) . We first tested the effect of the GABAA receptor agonist Muscimol at a wide range of concentrations ( from 1 μM to 100 μM ) . Remarkably , all of them had proneurogenic activity like selamectin ( Figure 5D ) . We also tested the effect of Chlordiazepoxide ( CDZ ) , a positive allosteric modulator of the GABAA receptor . When cells were treated alone with 10 μM Chlordiazepoxide or 0 . 3 μM Selamectin , significant proneurogenic activity was observed as compared to the DMSO control ( p<0 . 05 , t-test , Figure 5E ) . When cells were treated with both Selamectin ( 0 . 3 μM ) and CDZ ( 10 μM ) , no significant additive effect was observed ( p=0 . 480 , t-test , Figure 5F ) , suggesting a lack of synergistic or cooperative action between these two compounds . To determine whether the proneurogenic activity of selamectin is indeed mediated through the GABAA receptor , we asked whether it could be blocked by bicuculline ( Bicu , GABAA receptor antagonist ) , picrotoxin ( PTX ) , or pentylenetetrazol ( PTZ ) ( non-competitive blockers of the GABAA receptor ) . When cells were treated alone with these chemicals , no effect on baseline differentiation of TH+ neurons was observed , suggesting that at this stage of the mESC culture , GABA might not be released in sufficient quantities to affect neuronal differentiation ( Figure 5F , white columns , all control were normalized to fold change = 1 , displayed as the red dotted line ) . When cells were treated with selamectin together with these antagonists or blockers ( 500 μM PTX , 5 mM PTZ , or 100 μM Bicu ) , the effect of selamectin was blocked , leading to no significant difference between the treated group and control ( Figure 5F , gray columns ) . In contrast , the glycine receptor inhibitor strychnine ( STY , 100 μM ) failed to block the effect of selamectin ( Figure 5F ) . Together , these pharmacological data suggest that selamectin promotes multi-lineage neuronal differentiation from mESCs through the activation of GABAA receptor signaling . To determine whether GABAA receptor is required genetically to mediate the effects of selamectin , we used an EsiRNA approach to perturb the activity of genes encoding various GABAA receptor subunits . The mammalian CNS expresses nineteen GABAA receptor subunits ( α1-6 , β1-3 , γ1-3 , δ , ε , θ , π , ρ1−3 ) , the combinatorial co-assembly of which enables a potentially enormous molecular and functional heterogeneity of GABAA receptor subtypes ( Farrant and Nusser , 2005 ) . Through analyzing the expression profiles of all nineteen receptor subunits in mESC cultures , six genes were identified to display high expression level in mESC-derived neural progenitors ( data not shown ) and were therefore chosen for esiRNA knockdown . Transfection of gene-specific esiRNAs on Day 6/9 and qRT-PCR analysis on Day 8/11 ( Figure 6A ) showed that esiRNAs targeting the α1 ( encoded by the gabara1 gene ) , β2 ( gabarb2 gene ) , γ2 ( gabarg2 ) , and π ( gabarp gene ) subunits were highly effective in reducing the transcript levels of respective genes ( Figure 6B ) . Furthermore , treatment with the esiRNA targeting gabarg2 abolished the effect of selamectin in inducing NeuN+ and TH+ neurons ( Figure 6C–H ) . Treatment with the esiRNAs targeting gabara1 , gabarb2 , and gabarp genes did not noticeably block the effect of selamectin , suggesting that either the reduction of transcript levels is not sufficient to abrogate their gene activity or these subunits do not mediate the effect of selamectin . Together , these results provide genetic evidence that selamectin acts through the γ2-containing GABAA receptor to promote neuronal differentiation . 10 . 7554/eLife . 00508 . 008Figure 6 . Genetic evidence indicates that selamectin’s proneurogenic activity is mediated by the γ2 subunit-containing GABAA receptor . ( A ) Scheme of the EsiRNA transfection and cell harvest for qRT-PCR . ( B ) qRT-PCR shows fold change of the expression of different GABAA receptor subunits after EsiRNA transfection . p<0 . 001 vs non-transfection control , $$$p<0 . 001 vs GFP RNA transfection control . ( C ) Scheme of the EsiRNA knockdown experiment to identify the GABAA receptor subunit that mediates selamectin’s activity . ( D–F ) Representative images of NeuN and TH staining in DMSO and selamectin-treated cultures of non-transfected ( D ) , α1 EsiRNA transfected ( E ) , and γ2 EsiRNA transfected ( F ) . ( G ) Quantification shows knockdown of γ2 subunit but not α1β1 and π subunits abolishes the effect of selamectin in increasing neurons ( n = 4 , p<0 . 01 , p<0 . 001 vs DMSO ) . ( H ) Quantification shows knockdown of γ2 subunit but not α1β1 and π subunits abolishes the effect of selamectin in increasing TH+ neurons neurons ( n = 4 , p<0 . 01 , *p<0 . 001 vs DMSO ) . Scale bar , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00508 . 008 Opposite to its inhibitory roles in adult neurons , GABA signaling mediated by the activation of GABAA receptors can depolarize cells in the ventricular zone of the rat embryonic cortex ( LoTurco et al . , 1995 ) , neural progenitor/stem cells in the postnatal mouse sub-ventricular zone ( Liu et al . , 2005 ) or sub-granular zone ( Tozuka et al . , 2005 ) , due to the elevated internal Cl− concentrations in progenitor cells and young neurons ( Owens and Kriegstein , 2002; Spitzer , 2006; Ben-Ari et al . , 2007; Ge et al . , 2007 ) . Based on this information , we sought to identify the cell types in mESC cultures that might be responsive to GABA and selamectin by performing single-cell electrophysiological recordings of neural activity . The mESC cell line 46C that expresses GFP driven by Sox1 promoter was used . Based on the observed temporal effect of selamectin in mESC cultures , we focused our attention on neural rosettes , a functionally distinct type of neural stem cells suggested to represent the earliest NSC stage in vivo ( Elkabetz et al . , 2008 ) . Neural rosettes are easily recognizable because of their characteristic bipolar morphology and radial floral-like arrangement . For the majority of neural rosettes , we also verified their GFP signals ( Figure 7A ) . In addition , we examined some non-rosette cells with the morphology of young neurons ( Figure 7E ) . For a majority of cells recorded , we tested both GABA and selamectin-induced currents . 10 . 7554/eLife . 00508 . 009Figure 7 . GABA and selamectin ( Sela ) -induced currents in neural rosettes and non-rosette cells . ( A ) Bright field ( BF ) image ( left ) and green fluorescent protein ( GFP ) signal ( right ) of neural rosette cells . Majority of the cells in the view are neural rosette cells , and one of them with typical morphology is indicated by a red arrow . ( B ) An example of reduction of GABA ( 500 μM ) induced currents by bicuculline ( BICC , 200 μM ) . ( C ) Upper trace , an example of neural rosette cell displaying GABA ( 100 μM ) induced currents that also displayed selamectin induced current . Lower trace , an example of neural rosette cell displaying GABA ( 100 μM ) induced currents that did not display selamectin induced current . 16 μM Selamectin solution contains 0 . 4% DMSO , therefore 0 . 4% DMSO containing bath solution was used as the control solution as indicated in the figure . ( D ) A pie chart of the numbers of the four groups of neural rosette cell based on whether they displayed GABA and selamectin induced currents . The four groups are: 1 ) GABA+; Sela+ , 2 ) GABA+; Sela− , 3 ) GABA−; Sela+ , 4 ) GABA−; Sela− . The number of the cells in each group is indicated in the figure , except the group GABA−; Sela+ , which is zero . ( E ) Bright field ( BF ) image ( left ) and green fluorescent protein ( GFP ) signal ( right ) of non-rosette cells . One cell with typical young neuron morphology is indicated by a red arrow . ( F ) Example of the inhibition of GABA ( 100 μM ) induced currents by bicuculline ( BICC , 100 μM ) . ( G ) Upper trace , an example of non-rosette cell displaying GABA ( 4 mM ) induced current that also displayed salemectin ( 8 μM ) induced current . Lower trace , an example of non-rosette cell displaying GABA ( 400 μM ) responsive induced current that did not display salemectin ( 16 μM ) induced current . ( H ) A pie chart of the numbers of four groups of non-rosette cells based on whether they displayed GABA and selamectin induced currents . The four groups are: 1 ) GABA+; Sela+ , 2 ) GABA+; Sela− , 3 ) GABA−; Sela+ , 4 ) GABA−; Sela− . The number of the cells in each group is indicated in the figure , except the group GABA−; Sela+ , which is zero . The application time courses of the control solution ( BATH ) , GABA , bicuculline ( BICC ) and selamectin ( Sela ) are indicated by horizontal bars in the figure . The membrane potential was holding at −70 mV in all recordings . The scales for time ( horizontal ) and currents ( vertical ) are indicated for each recording in the figure . DOI: http://dx . doi . org/10 . 7554/eLife . 00508 . 009 Both neural rosettes and non-rosette cells displayed similar passive membrane properties . For neural rosettes , we observed membrane capacitance Cm = 9 . 4 ± 12 . 5 pF ( mean ± S . D . , n = 106 ) , input resistance Rm = 772 . 5 ± 1037 . 1 MΩ ( mean ± S . D . , n = 106 ) , and resting membrane potential ERev = − 20 . 9 ± 8 . 3 mV ( mean ± S . D . , n = 71 ) . For non-rosette cells , we observed membrane capacitance Cm = 9 . 5 ± 6 . 2 pF ( mean ± S . D . , n = 74 ) , input resistance Rm = 1 . 2 ± 1 . 2 GΩ ( mean ± S . D . , n = 74 ) , and resting membrane potential ERev = − 22 . 4 ± 9 . 2 mV ( mean ± S . D . , n = 63 ) . The resting membrane potentials of both groups of cells are similar to what has been observed in neural progenitors using similar internal and external recording solutions , either in acute slices ( Wang et al . , 2003 ) or in vitro ( Stewart et al . , 2002 ) . To test whether neural progenitors in neural rosettes respond to GABA , we measured the whole cell currents upon GABA applications when membrane potentials of the cells were held at −70 mV . In approximately 21% of the cells ( 24 out of 111 ) , applications of GABA induced inward currents . The GABA induced currents are sensitive to bicucculin ( Bicu , 200 μM ) , but in some cases the application of 200 μM Bicu did not completely block the GABA induced currents ( Figure 7B ) . To determine whether selamectin regulates membrane electrophysiological properties in neural rosette cells , we analyzed responses of the whole-cell currents to the applications of selamectin . Selamectin indeed induced inward currents in neural rosette cells ( Figure 7C ) , but the response was heterogeneous . We grouped the cells into four groups based on whether they displayed detectable GABA and selamectin-induced currents . They were GABA+;Sela+ ( the cells that displayed both GABA and selamectin induced currents ) , GABA+;Sela− ( the cells that displayed only GABA induced currents ) , GABA−;Sela+ ( the cells that displayed only selamectin induced currents ) , and GABA−;Sela− ( the cells that displayed neither GABA nor selamectin induced currents ) . The numbers of cells belonging to each group were shown in Figure 7D . Among the 77 recorded cells , the largest group ( 43 cells , 64 . 2% ) was those that did not display induced currents by either GABA or selamectin . The second largest group ( 14 cells , 20 . 9% ) is those that displayed induced currents only by GABA . The third group ( 10 cells , 14 . 9% ) was those that displayed both GABA and selamectin induced currents . No cells displayed only selamectin-induced currents . Slightly higher percentage ( 27 out of 73 cells , 37% ) of non-rosette cells ( Figure 7E ) displayed GABA induced currents , which could be blocked by Bicu ( Figure 7F ) . Examples of the cells that displayed ( upper trace ) or did not display ( lower trace ) selamectin-induced currents were shown in Figure 7G and their frequency distribution was shown in Figure 7H . Out of 44 non-rosette cells that we tested both GABA and selamectin induced currents , the largest group ( 20 cells , 45 . 5% ) was the one that displayed only GABA-induced currents . The second group ( 17 cells , 38 . 6% ) was the one that displayed neither GABA nor selamectin induced currents . The third group ( 7 cells , 15 . 9% ) displayed both GABA- and selamectin-induced currents . Similar to neural rosette cells , no non-rosette cells displayed only selamectin-induced currents . Taken together , our results uncover considerable heterogeneity in the response of neural rosettes and young neurons to GABA or selamectin and indicate that a subset of GABA-responsive neural rosette cells in mESC cultures respond to selamectin . Although GABA-induced depolarization in postnatal neural progenitors causes reduced proliferation , its effect on neurogenesis has been controversial . While one study shows that GABA could promote neuronal differentiation in adult hippocampal progenitor cells ( Tozuka et al . , 2005 ) , another found an inhibitory role on neuronal production in the postnatal sub-ventricular zone ( Liu et al . , 2005 ) . To further probe the mechanisms by which selamectin promotes neuronal differentiation , we performed BrdU incorporation experiment in mESC cultures on Day 11 after the treatment with selamectin for 4 days . Significantly fewer BrdU+ cells were detected in selamectin-treated groups , suggesting decreased proliferation ( Figure 8A–B ) . 10 . 7554/eLife . 00508 . 010Figure 8 . Selamectin decreases proliferation and increases the expression of proneural and lineage-associated transcription factors . ( A ) Representative fields show the BrdU incorporation on Day 11 after cells were treated with selamectin ( right panel ) or DMSO ( left panel ) for 4 days . Significantly fewer BrdU+ cells were detected in the selamectin-treated group . ( B ) Quantification as the percentage of BrdU+ cells among total cells ( t-test , n = 4 , p=0 . 008 ) . ( C ) qRT-PCR detects increased expression of proneural ( Ascl1 , NeurD ) and lineage-associated transcription factors ( Lmx1a , Lmx1b and Nurr1 ) in selamectin-treated group . β-actin was used as an input control and data was normalized to expression level on Day 8 . ( D ) Representative fields show the TUNEL staining on Day 14 in cell cultures treated with DMSO or selamectin , with NeuN staining as a control to confirm selamectin efficacy in this experiment . ( E ) Quantification shows significant difference in NeuN% ( p<0 . 001 ) and no significant difference in TUNEL% ( p=0 . 058 ) ( t-test , n = 4 ) . Scale bar , 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00508 . 010 Since electrical activity in neural progenitor cells can lead to neurotransmitter re-specification through regulating transcription factor expression ( Spitzer , 2012 ) , we asked whether selamectin exerts an effect on the expression of transcription factors that regulate neuronal specification . At Day 8 , cells were sampled for qRT-PCR while the rest were split into two groups for further treatment with either DMSO or selamectin . The gene expression levels were examined by qRT-PCR on Day 11 and the results were compared to that of Day 8 . We detected an increased expression of the proneural genes ascl1 and neurod , as well as the DA lineage- associated genes lmx1a , lmx1b , and nurr1 in Day 11 culture compared to Day 8 culture , and such increase was further enhanced by selamectin ( Figure 8C ) . These results are consistent with the idea that selamectin activates GABAA receptor and enhances the electrical activity of neural progenitors , thereby increasing the expression of proneural and lineage-associated transcription factors , leading to increased neuronal differentiation . We also tested whether decreased cell death ( or improved neuronal survival ) might also contribute to increased neurons detected in selamectin-treated mESC cultures . Cells were treated with selamectin or DMSO from Day 8 to Day 11 and cultured for 4 more days . On Day 14 , TUNEL labeling was carried out . The result showed no significant difference in cell death between control and selamectin-treated cultures ( Figure 8D–E ) , suggesting that alteration in cell death contributes little to selamectin’s pro-neurogenic activity . Selamectin-induced decrease of proliferation in mESC cultures could be due to either prolonged cell cycle length or increased cell cycle exit . To further probe the underlying cellular mechanisms , we carried out clonal analysis . Day 6 mESC cultures were transfected with pCAG-GFP that resulted in sparse labeling of clonally related cells ( Figure 9A ) . Cultured cells were treated with DMSO or 0 . 3 μM selamectin from Day 8 to Day 11 . Time-lapse live imaging was performed on sparsely labeled neural progenitors from Day 8 to Day 14 , followed by fixation and immunostaining ( Figure 9A ) . We found that the cell cycle length exhibited no significant difference between DMSO- ( 25 . 61 ± 6 . 99 hr , n = 31 ) and selamectin-treated cells ( 24 ± 6 . 53 hr , n = 28 ) ( Figure 9B , left panel , Videos 1 , 2 ) . We also quantified the number of cells within single clones and detected no difference between DMSO- ( 13 . 46 ± 4 . 14 cells per clone , n = 56 clones ) and selamectin- ( 15 . 15 ± 4 . 67 per clone , n = 59 clones ) treated groups ( Figure 9B , right panel ) . The results suggest that selamectin does not change the cell cycle length or the proliferation rate of mESC-derived neural progenitors . 10 . 7554/eLife . 00508 . 011Figure 9 . Clonal analysis reveals that selamectin promotes progenitor cell cycle exit toward terminal differentiation . ( A ) Scheme of clonal culture analyses . ( B ) Quantification of cell cycle length and number of cells within single clones . ( C–E ) Representative images of double immunostaining of GFP with NeuN ( C ) , TH ( D ) , or the proliferation marker Ki67 ( E ) within single clones . ( F ) Quantification of the percentage of NeuN+ ( left , DMSO n = 49 , selamectin n = 51 ) , TH+ ( middle , DMSO n = 6 , selamectin n = 7 ) , and Ki67+ cells ( right , DMSO n = 46 , selamectin n = 41 ) within single clones . p<0 . 001 vs DMSO ) . Scale bar , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00508 . 01110 . 7554/eLife . 00508 . 012Video 1 . Time-lapse of a single GFP labeled neural progenitor derived from mESC . The progenitors were treated with DMSO from Day 8 to Day 11 . The interval between each frame is 2 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 00508 . 01210 . 7554/eLife . 00508 . 013Video 2 . Time-lapse of a single GFP labeled neural progenitor derived from mESC . The progenitors were treated with 0 . 3 μM Selamectin from Day 8 to Day 11 . The interval between each frame is 2 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 00508 . 013 To determine the cell fates within single clones , we performed co-immunostaining of GFP and NeuN ( or TH ) as well as the proliferation marker Ki67 ( Figure 9C–E ) . Clonal quantification of cell fates showed a significant increase of NeuN+ ( or TH+ ) neurons and a concurrent decrease of Ki67+ progenitor cells within single clones treated with selamectin ( Figure 9F ) . Taken together , these findings suggest that selamectin acts to promote progenitor cell cycle exit toward terminal differentiation . In order to test the effects of selamectin on neuronal differentiation from hPSCs , we devised a three-stage neuronal differentiation protocol and used the H9 line of hESC ( Thomson et al . , 1998 ) and a human induced pluripotent stem cell line ( hiPSC ) ( Kreitzer et al . , 2013 ) ( Figure 10A ) . In agreement with the results from mESCs , treatment with selamectin ( 0 . 25 μM , 0 . 5 μM , 0 . 75 μM ) significantly increased the percentage of total neurons , TH and 5-HT neurons compared to the DMSO-treated control in both H9 hESCs ( Figure 10B–C ) and hiPSCs ( Figure 10D–E ) . 10 . 7554/eLife . 00508 . 014Figure 10 . Selamectin increases the differentiation of multiple neuronal lineages from human pluripotent stem cells . ( A ) Scheme of the three-stage neuronal differentiation protocol for H9 hESCs and hiPSCs . ( B ) Representative images of NeuN and TH staining in control ( DMSO ) and selamectin-treated cultures of H9 hESCs . ( C ) Quantification shows increased production of both TH+ and total neurons by selamectin in H9 hESCs . ( n = 4 , p<0 . 05 , p<0 . 01 , p<0 . 001 vs DMSO ) . ( D ) Representative images of NeuN and TH staining in control ( DMSO ) and selamectin-treated cultures of hiPSCs . ( E ) Quantification shows increased production of both TH+ and total neurons by selamectin in hiPSCs . ( n = 4 , p<0 . 05 , p<0 . 01 , *p<0 . 001 vs DMSO ) . Scale bar , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 00508 . 014 To determine whether selamectin has a pro-neurogenic activity in developing embryos in vivo , we tested its effect on zebrafish . A transgenic line carrying the HuC:GFP transgene ( Park et al . , 2000 ) was used . HuC/ELAVL3 is a neuron-specific RNA binding protein that is expressed in most differentiating neurons . We treated embryos with selamectin starting at ∼75% epiboly ( 8 hr post fertilization , hpf ) , when the neuro-ectodermal fate has been determined . The treatment of selamectin lasted until 22 hpf , when nascent neurons began to emerge but not yet became too numerous to quantify ( Figure 11A ) . We noted an obvious increase of the overall HuC-GFP+ neurons in the selamectin-treated embryos as compared to the DMSO control ( Figure 11B ) . Quantification of the midbrain cluster ( boxed ) showed a highly significant difference between the selamectin-treated and the control ( Figure 11C , n =20 , t-test , p<0 . 001 ) . 10 . 7554/eLife . 00508 . 015Figure 11 . Selamectin promotes neurogenesis in vivo in the developing zebrafish brain . ( A ) Scheme of the selamectin treatment on HuC:GFP transgenic zebrafish embryos . ( B ) Representative images show that selamectin ( 2 μM , 14 hr from 8 hpf to 22 hpf ) increases Huc-GFP signal . ( C ) Quantification of the midbrain cluster ( boxed ) shows a significant difference between two groups ( t-test , n = 20 , p<0 . 001 ) . ( D ) Representative images show increased DA neurons in selamectin-treated embryos ( 2 μM for 40 hr from 8 hpf to 48 hpf ) . ( E ) Quantification of the ventral forebrain DA neurons shows a significant difference between two groups ( t-test , n = 10 , p<0 . 001 ) . ( F ) A schematic model shows the effect of selamectin on neuronal differentiation from mESCs . DOI: http://dx . doi . org/10 . 7554/eLife . 00508 . 015 We also examined the effect of selamectin on DA neurons , by treating embryos with selamectin from 8 hpf to 48 hpf followed by immunostaining with the anti-TH antibody ( Figure 11A ) . The number of ventral forebrain DA neurons was significantly increased in selamectin-treated embryos as compared to the DMSO control ( Figure 11D–E ) . These results suggest that selamectin has pro-neurogenic activity in vivo . In this study , we have undertaken a chemical genetic approach to identify new mechanisms that regulate neuronal differentiation from PSCs . Three significant advancements are reported here: First , we have established an imaging-based high content small molecule screening method for identifying chemicals that increase the production of TH+ DA neurons from mESCs . We have further shown that compounds identified through mESC-based screening are applicable to hPSCs . Second , It has been previously reported that mild electrical stimulation influences mESCs to assume a neuronal fate ( Yamada et al . , 2007 ) . Through the identification of selamectin , we reveal a novel mechanism underlying the activity-dependent regulation of neuronal differentiation from PSCs . Third , by employing single-cell electrophysiological recordings of mESC-derived neural rosette cells , we uncover for the first time the heterogeneity of neural progenitor responses to GABA and selamectin , which has provided a plausible explanation for paradoxical observations of GABA’s effects on in vivo neurogenesis ( Liu et al . , 2005; Tozuka et al . , 2005; Song et al . , 2012 ) . Compared to previous high content chemical screening for neuronal enhancers in embryonic stem/progenitor cultures ( Ding et al . , 2003; Saxe et al . , 2007; Desbordes et al . , 2008; Zhou et al . , 2010 ) , the advances of our method are the following: First , Instead of using culture systems that require either feeder cells or neurosphere formation , our assay uses the monolayer differentiation method ( Ying and Smith , 2003 ) , thus is simple and easy to carry out . Second , the use of monolayer differentiation also renders the background noise fairly low , with baseline TH+ cells of 0 . 5–3% in control ( DMSO ) conditions . Third , compared to previous screening that uses either pluripotency markers ( e . g . , Oct4 ) or general neuronal differentiation markers ( e . g . , a-tubulin or TuJ1 ) , we used the neuronal subtype specific marker TH . The small number of TH+ cells makes the quantification convenient and accurate , thereby significantly increasing the sensitivity of our assay . Finally , our assay is the first that offers the prospect for identifying compounds that regulate neuronal subtype differentiation , since previous assays use only general pluripotency or neuronal markers . This screening platform has enabled us to identify selamectin and show that it increases the production of multiple neuronal lineage types including DA , 5-HT , GABA , and Islet+ neurons as well as olig2+ oligodendrocytes . Selamectin is known to target invertebrate glutamate-guided chloride channels that have no orthologues in vertebrates . Our pharmacological , genetic and clonal studies provide first evidence that selamectin targets the γ2 subunit-containing GABAA receptor to promote progenitor cell cycle exit toward terminal differentiation . Single-cell electrophysiological recordings further show that a subset of neural rosette cells responds to GABA and selamectin via functional GABAA receptors . The blocking of GABA-induced current by bicuculline ( the GABAA receptor antagonist ) , however , is incomplete , suggesting that GABA receptors other than GABAA are also expressed in neural rosette progenitors . Expression of functional GABAB receptors has been demonstrated in ES cells ( Schwirtlich et al . , 2010 ) . Interestingly , many recorded neural rosette cells display very low membrane input resistance ( less than 100 MΩ ) , and high membrane capacitances ( up to 50 pF ) . Furthermore , applications of 100 μM meclofenamic acid ( MFA ) ( Liu et al . , 2005 ) could reversibly reduce the inward currents when the membrane potential of the cells are held at −70 mV . These observations suggest that neural rosette cells may be electrically connected , similar to neural progenitor cells in the ventricular zone of embryonic brains ( LoTurco et al . , 1995 ) and those in the sub-ventricular zone of adult brains ( Liu et al . , 2005 ) . Thus , it is possible that functional GABA receptors expressed in one cell may render other electrically connected cells to display apparent GABA-induced currents . Our data reveal that selamectin increases the expression of proneural and lineage-specific transcription factors while reducing proliferation in mESC cultures . Clonal analysis and time-lapse live imaging further uncover the role of selamectin ( hence , likely GABA ) in promoting progenitor cell cycle exit toward terminal differentiation . These molecular and cellular findings , together with the pharmacogenetic and electrophysiological studies , lead us to propose that selamectin-potentiated activation of the γ2-containing GABAA receptor in mESC-derived neural progenitor cells causes Cl− outflow ( due to the high internal chloride levels in these progenitor cells ) , thereby leading to the depolarization of neural progenitors and calcium influx . This further activates proneural and lineage-specific transcription factors , which have established roles in promoting cell cycle exit toward terminal neuronal differentiation ( Bertrand et al . , 2002 ) ( Figure 8F ) . Such neural activity-dependent regulation of transcription factor expression has been reported in developing Xenopus embryos ( Demarque and Spitzer , 2010; Marek et al . , 2010 ) . Our findings suggest that this is an evolutionarily conserved phenomenon . GABA signaling influences embryonic cortical neural progenitor proliferation ( Owens and Kriegstein , 2002 ) and regulates adult neurogenesis ( Ge et al . , 2007 ) . Paradoxically , either an increase or a decrease of neurogenesis by GABA activation has been observed ( Liu et al . , 2005; Tozuka et al . , 2005 ) . Recently , in the context of adult neurogenesis , GABA released by local interneurons has been shown to promote the exit of adult neural stem cells from quiescence ( thereby promote their proliferation ) ( Song et al . , 2012 ) . Our single neural rosette cell recordings , which reveal the heterogeneous responses among progenitors , provide an explanation as to why the influence of GABA signaling on neurogenesis appears cell type- and context-dependent . It is worth pointing out that our screen of 2000 compounds was not successful in identifying chemicals that specifically increase DA neuronal production , suggesting that these compounds may be rare , and large compound libraries need to be screened in order to find them . The high throughput capability of our assay will enable such screen to be carried out , and is thus an important future direction . The mouse ESC lines E14Tg2a and 46C were used . 46C was a generous gift from Dr Austin Smith , in which GFP was knocked into the sox1 locus ( Ying and Smith , 2003 ) . mESCs were cultured in GMEM media ( G5154 , Sigma , St . Louis , MO ) supplemented with glutamine , sodium pyruvate , 0 . 1 mM MEM non-essential amino acids , 10% ( vol/vol ) fetal bovine serum ( characterized , Hyclone , Thermo Scientific , Waltham , MA ) , beta-mercaptoethanol , and 500–1000 units per ml of leukocyte inhibitory factor ( ESG1107 , Chemicon , Billerica , MA ) , on gelatinized cell culture surface without feeder cells . To induce neuronal differentiation , the monolayer differentiation protocol developed by Ying et al . ( Ying et al . , 2003 ) was used . Briefly , E14 cells was dissociated with Trypsin-EDTA ( TE ) into single cells and plated onto gelatinized cell culture dish at a density of 1 . 0 × 104 cell/cm2 in N2B27 media . Cells were cultured in N2B27 media for 7 days , with media change every other day . On day 7 , cells were dissociated with TE again and re-plated onto poly-L-ornithine-laminin coated 96-well plate in N2B27 media , at a density of 2∼5 × 104/cm2 . Media was changed every 2 days after re-plating . For high throughput screening , on Day 7 , cells were dissociated in TE and re-suspended in fresh N2B27 media . The cell suspension was then dispensed into 96-well micro-clear imaging plates ( Greiner cat . no 655956 ) with the WellMate liquid dispenser ( Thermo Matrix ) , at a density of 1 . 5 × 104/ well . These assay plates were incubated in 37°C overnight for cell settling and adherence to the surface . On Day 8 , screening compounds were dispensed into assay plates with Biomek FXP Laboratory Automation Workstation ( Beckman Coulter , Brea , CA ) , at a final concentration of 1 μM . Chemical treatment lasted for 3 days from Day 8 to Day 11 . On Day 11 , chemical treatment was withdrawn via change of media . Cells were cultured in assay plates for additional 3 days until Day 14 before automated immunostaining using PlateMate Plus ( Thermo Matrix ) and image acquisitioning with INCell 1000 or 2000 ( G . E . Healthcare , Little Chalfont , UK ) . The chemical library was obtained from the UCSF SMDC , which is composed of FDA-approved drugs , bioactive compounds , and natural products ( Microsource Spectrum Collection ) . Primary antibodies for immunocytochemistry include: Rabbit-anti-TH ( AB152 , Millipore , Billerica , MA ) ; Mouse-anti-NeuN , ( MAB377 , Millipore ) ; mouse anti-sox2 ( MAB2018 , R&D ) ; Rabbit-anti-Lmx1 ( a generous gift from Dr . German , UCSF ) ; mouse-anti-Nestin ( MAB353 , Chemicon ) ; mouse-anti-islet ( DSHB 39 . 4D5 ) ; Rabbit anti-GABA ( A2052 , Sigma ) ; Rabbit-anti-Olig2 ( AB9610 , Millipore ) . After immunostaining , images were taken using the automatic system INCell 1000 or 2000 ( GE ) . 20 field of views on three different channels ( For TH , NeuN and DAPI ) were taken for each well . Images were analyzed using the INCell Developer software ( G . E . Healthcare ) . The percentage of TH in each well was expressed as a ratio . Fold change of chemical-treated well was calculated relative to the average of DMSO control wells . The percentage of other neuronal types was calculated similarly . The following pharmacological compounds were used in this study: selamectin ( 01503720 , Microsource ) , avermectin ( 31732-100 MG , Sigma ) , taurine ( T8691-25 G , Sigma ) , muscimol ( M1523-5MG , Sigma ) , Chlordiazepoxide ( C2517 , Sigma ) , picrotoxin ( R284556-50 MG , Sigma ) , Pentylenetetrazole ( P6500-25 G , Sigma ) , bicuculine ( 14340-25 MG , Sigma ) , STY ( Strychnine , S0532-5G , sigma ) . Drugs were prepared as 10 mM stock and diluted to appropriate concentrations as indicated in the text . The algorithm Deqor is used to design esiRNA , which can be found at http://www . mpi-cbg . de/esiRNA/ . Two rounds of PCR were done to obtain the template for in vitro synthesis of double-stranded RNAs . cDNAs from day 3 mESC-derived neural progenitors was used as template for the first round PCR . Primers for first round PCR begin with T7 ‘anchor’ sequence: 5′ GGGCGGGT 3′ , to which the T7 Anchor primer will anneal in the second round PCR . T7 promoter was incorporated in the primers for the second round PCR . The product from the second round PCR was used as template for in vitro transcription with T7 RNA polymerase . Annealing is done in the same program immediately after in vitro transcription: 1 ) 37°C , 5 . 5 hr; 2 ) 90°C , 3 min; 3 ) Ramp ( 0 . 1°C/s ) to 70°C; 4 ) 70°C , 3 min; 5 ) Ramp ( 0 . 1°C/s ) to 50°C; 6 ) 50°C , 3 min; 7 ) Ramp ( 0 . 1°C/s ) to 25°C; 8 ) 25°C , 3 min . Double-stranded RNA was digested with Shortcut RNAse III ( NEB ) and purified for transfection . The EsiRNAs targeting different GABAA receptor subunits were transfected into mESC-derived neural progenitors on day 6 and day 9 using Lipofectamine 2000 Reagent ( Invitrogen ) following the manufacturer’s protocol . Cells were labeled with 10 μM BrdU for 6 hr before immunostaining . Rat-anti-BrdU antibody ( ab6326 , Abcam , Cambridge , MA ) was diluted 1:2000 . An in situ cell death detection kit from Roche ( Cat . # 12 , 156 792 910 ) was used . Staining was performed following manufacturer’s instruction . Total RNA was isolated using TRIzol reagent ( Invitrogen ) and qPCR was carried out following manufacturer’s instructions ( Applied Bio-systems ) . Primer sequences are: ascl1 ( GenBank accession number NM_008553 . 4 ) , forward , 5′-GAAGCAGGATGGCAGCAGAT-3′ , reverse , 5′-TCGGGCTTAGGTTCAGACAC-3′; neuroD1 ( GenBank accession number NM_010894 . 2 ) , forward , 5′-AGCCACGGATCAATCTTCTC-3′ , reverse , 5′-ACTGTACGCACAGTGGATTC-3′; lmx1a ( GenBank accession number NM_033652 ) , forward , 5′- ACCCCTATGGTGCTGAACCT- 3′ , reverse , 5′- CAGCAACCCTTCACACAGTA -3′; lmx1b ( GenBank accession number NM_010725 ) , forward , 5′-CTGGGCCAAGAGGTTCTGTC-3′ , reverse , 5′-GAAGAGCCGAGGAAGCAGTC-3′; nurr1 ( GenBank accession number NM_013613 ) , forward , 5′-CTGGCTATGGTCACAGAGAGACAC-3′ , reverse , 5′-GGTACCAAGTCTTCCAATTTCAGG-3′; β-actin ( GenBank accession number NM_007393 ) , forward , 5′-TCCTTCTTGGGTATGGAATCCTG-3′ reverse , 5′-GGAGGAGCAATGATCTTGATCTTC-3′ . Ct values were the means of triplicate replicates . Each sample was normalized with loading reference β-actin ( ΔCt ) , and then normalized with expression on Day 8 . For relative expression level comparison , the difference in cycle threshold ( ΔΔCt ) between D11 and D8 was evaluated . Day 5 mESC culture was transfected with low concentration pCAG-GFP plasmid to sparsely label mESC-derived neural progenitors . Transfection was done with Lipofectamine 2000 Reagent ( Invitrogen ) following the manufacturer’s protocol . Briefly , 0 . 8-μg pCAG-GFP plasmid DNA was used for a single confluent well of a 6-well plate . On day 7 , cells were dissociated with Trypsin-EDTA and re-plated onto poly-L-ornithine-laminin coated 96-well plate in N2B27 media , at a density of 2∼5 × 104/cm2 . Time-lapse live imaging was performed with a 2-hr interval from Day 8 to Day 14 , using a third generation automated robotic microscopy system that incorporated several advances over earlier systems ( Arrasate et al . , 2004; Arrasate and Finkbeiner , 2005; Sharma et al . , 2012 ) . Multiple images were taken for each condition and stitched together using a custom designed plugin for the open source image processing package Fiji ( Schindelin et al . , 2012 ) . The cells were fixed for immunostaining on day 14 . Neural rosettes are recognized because of the cells’ characteristic bipolar morphology and their radial floral-like arrangement . We used cell line that expresses GFP under SOX1 promoter . For majority of neural rosettes , we also verified their GFP signals . There are also non-rosette GFP positive cells , among which are new born neurons . New born neurons were identified based on their characteristic morphology , round-shaped cell body . The whole-cell recordings were performed at room temperature . Pipette electrodes ( Sutter , Novato , CA ) were fabricated using a Sutter P-97 horizontal puller and fire-polished and had final tip resistances of 2–4 MΩ . All recording have been performed using gap free protocol while the membrane potential was holding at −70 mV . The bath solution contained ( in mM ) NaCl 110 , KCl 30 , CaCl2 1 . 8 , MgCl2 0 . 5 , HEPES 5 , and glucose 10 , pH adjusted to 7 . 4 with NaOH . The internal solution for patch recordings contained ( in mM ) NaCl 10 , KCl 130 , MgCl2 0 . 5 , HEPES 5 , EGTA 1 , and MgATP 5 , pH adjusted to 7 . 3 with KOH . The applications of the activation and inhibition reagents were performed by a pressurized micro-perfusion system . The pressure was typically 7–10 kPa . The stock solutions were made by dissolving the reagents in the bath solution ( GABA ) or DMSO ( bicuculline and selamectin ) . The stock solutions were kept at −80°C and were diluted to the working concentration using bath solution before each experiment . Unless otherwise indicated , we used 100 μM GABA , 100 μM bicuculline and 8 μM selamectin . The working solutions of bicuculline and selamectin contained up to 0 . 5% DMSO , therefore the bath solutions containing the corresponding concentration of DMSO were routinely used as control solution prior to the applications of bicuculline or selamectin . The human ESC lines H9 ( Thomson et al . , 1998 ) and a human iPSC line ( a gift from Dr Bruce Conklin ) ( Kreitzer et al . , 2013 ) were used . Stem Cells were cultured on growth factor reduced Matrigel ( BD Biosciences , Franklin Lakes , NJ ) in mTeSR1 media ( Stemcell Technologies , Vancouver , Canada ) with the media changed daily . To initiate differentiation , H9 and WTC-10 cells were plated at a density of 2 × 104 cells/cm2 in N2B27 media ( DMEM/F12:Neurobasal [1:1] , N2 supplement ( 1:100 ) , B27 supplement without vitamin A ( 1:50 ) , Glutamax , Insulin ( 20 μg/ml ) , beta-mercaptoethanol ( 110 μM ) , BSA Fraction V ( 20 μg/ml ) , bFGF ( 20 ng/ml ) supplemented with Rock Inhibitor Y-27632 ( 10 μM , Millipore ) ) . Media was changed with fresh N2B27 media every other day until Day 11 . Cells were plated at 3 × 104 cells/cm2 in N2B27 media supplemented with Rock inhibitor on Day 11 . Media was changed on Day 12 to neuronal differentiation media ( Neurobasal , B27 without vitamin A ( 1:50 ) , BDNF ( 20 ng/ml , Peprotech , Rocky Hill , NJ ) , GDNF ( 10 ng/ml , Peprotech ) , cAMP ( 500 μM , Sigma Aldrich ) , Ascorbic Acid ( 200 μM , Sigma Aldrich ) ) . Cells were treated with DMSO or Selamectin from Day 12 to Day 19 , and were fixed on Day 26 for immunostaining . All reagents were purchased from LIfe Technologies , unless otherwise stated . A transgenic line ( Hu-GFP ) marking nascent neurons was used . At 10 hpf , embryos were de-chorionated with forceps and transferred to a glass vial with 3 ml Embryo Solution , and Selamectin was added into solution to a final concentration of 2 μM . Embryos were incubated at 28°C until 22 hpf and then fixed with 4% PFA and mounted onto slides for confocal imaging . Embryos were also incubated at 28°C until 48 hpf to evaluate the effect of selamectin on TH differentiation ( selamectin treatment lasted from 8 hpf to 48 hpf ) . At 48 hpf , embryos were stained with anti-TH antibody ( custom made , 1:1000 ) .	Pluripotent stem cells have the potential to become most of the cell types that make up an organism . However , the signals that trigger these cells to turn into neurons rather than lung cells or muscle cells , for example , are not fully understood . Proteins called growth factors are known to have a role in this process , as are transcription factors , but it is not clear if other factors are also involved . In an attempt to identify additional mechanisms that could contribute to the formation of neurons , Sun et al . screened more than 2 , 000 small molecules for their ability to transform mouse pluripotent stem cells into neurons in cell culture . Surprisingly , they found that a compound called selamectin , which is used to treat parasitic flatworm infections , also triggered stem cells to turn into neurons . Selamectin works by blocking a particular type of ion channel in flatworms , but this ion channel is not found in vertebrates , which means that selamectin must be promoting the formation of neurons in mice via a different mechanism . Given that a drug related to selamectin is known to act on a subtype of receptors for the neurotransmitter GABA , Sun et al . wondered whether these receptors—known as GABAA receptors—might also underlie the effects of selamectin . Consistent with this idea , drugs that increased GABAA activity stimulated the formation of neurons , whereas drugs that reduced GABAA function blocked the effects of selamectin . In addition , Sun et al . showed that selamectin triggers human embryonic stem cells to become neurons , and that it also promotes the formation of new neurons in developing zebrafish in vivo . As well as revealing an additional mechanism for the formation of neurons from stem cells , the screening technique introduced by Sun et al . could help to identify further pro-neuronal molecules , which could aid the treatment of neurodevelopmental and neurodegenerative disorders .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "neuroscience" ]	2013	Imaging-based chemical screening reveals activity-dependent neural differentiation of pluripotent stem cells
Middle East respiratory syndrome coronavirus ( MERS-CoV ) is a zoonotic virus from camels causing significant mortality and morbidity in humans in the Arabian Peninsula . The epidemiology of the virus remains poorly understood , and while case-based and seroepidemiological studies have been employed extensively throughout the epidemic , viral sequence data have not been utilised to their full potential . Here , we use existing MERS-CoV sequence data to explore its phylodynamics in two of its known major hosts , humans and camels . We employ structured coalescent models to show that long-term MERS-CoV evolution occurs exclusively in camels , whereas humans act as a transient , and ultimately terminal host . By analysing the distribution of human outbreak cluster sizes and zoonotic introduction times , we show that human outbreaks in the Arabian peninsula are driven by seasonally varying zoonotic transfer of viruses from camels . Without heretofore unseen evolution of host tropism , MERS-CoV is unlikely to become endemic in humans . Middle East respiratory syndrome coronavirus ( MERS-CoV ) , endemic in camels in the Arabian Peninsula , is the causative agent of zoonotic infections and limited outbreaks in humans . The virus , first discovered in 2012 ( Zaki et al . , 2012; van Boheemen et al . , 2012 ) , has caused more than 2000 infections and over 700 deaths , according to the World Health Organization ( WHO ) ( World Health Organization , 2017 ) . Its epidemiology remains obscure , largely because infections are observed among the most severely affected individuals , such as older males with comorbidities ( Assiri et al . , 2013a; WHO MERS-Cov Research Group , 2013 ) . While contact with camels is often reported , other patients do not recall contact with any livestock , suggesting an unobserved community contribution to the outbreak ( WHO MERS-Cov Research Group , 2013 ) . Previous studies on MERS-CoV epidemiology have used serology to identify factors associated with MERS-CoV exposure in potential risk groups ( Reusken et al . , 2015; Reusken et al . , 2013 ) . Such data have shown high seroprevalence in camels ( Müller et al . , 2014; Corman et al . , 2014; Chu et al . , 2014; Reusken et al . , 2013; Reusken et al . , 2014 ) and evidence of contact with MERS-CoV in workers with occupational exposure to camels ( Reusken et al . , 2015; Müller et al . , 2015 ) . Separately , epidemiological modelling approaches have been used to look at incidence reports through time , space and across hosts ( Cauchemez et al . , 2016 ) . Although such epidemiological approaches yield important clues about exposure patterns and potential for larger outbreaks , much inevitably remains opaque to such approaches due to difficulties in linking cases into transmission clusters in the absence of detailed information . Where sequence data are relatively cheap to produce , genomic epidemiological approaches can fill this critical gap in outbreak scenarios ( Liu et al . , 2013; Gire et al . , 2014; Grubaugh et al . , 2017 ) . These data often yield a highly detailed picture of an epidemic when complete genome sequencing is performed consistently and appropriate metadata collected ( Dudas et al . , 2017 ) . Sequence data can help discriminate between multiple and single source scenarios ( Gire et al . , 2014; Quick et al . , 2015 ) , which are fundamental to quantifying risk ( Grubaugh et al . , 2017 ) . Sequencing MERS-CoV has been performed as part of initial attempts to link human infections with the camel reservoir ( Memish et al . , 2014 ) , nosocomial outbreak investigations ( Assiri et al . , 2013b ) and routine surveillance ( Wernery et al . , 2015 ) . A large portion of MERS-CoV sequences come from outbreaks within hospitals , where sequence data have been used to determine whether infections were isolated introductions or were part of a larger hospital-associated outbreak ( Fagbo et al . , 2015 ) . Similar studies on MERS-CoV have taken place at broader geographic scales , such as cities ( Cotten et al . , 2013 ) . It is widely accepted that recorded human MERS-CoV infections are a result of at least several introductions of the virus into humans ( Cotten et al . , 2013 ) and that contact with camels is a major risk factor for developing MERS , per WHO guidelines ( World Health Organization , 2016 ) . Previous studies attempting to quantify the actual number of spillover infections have either relied on case-based epidemiological approaches ( Cauchemez et al . , 2016 ) or employed methods agnostic to signals of population structure within sequence data ( Zhang et al . , 2016 ) . Here , we use a dataset of 274 MERS-CoV genomes to investigate transmission patterns of the virus between humans and camels . Here , we use an explicit model of metapopulation structure and migration between discrete subpopulations , referred to here as demes ( Vaughan et al . , 2014 ) , derived from the structured coalescent ( Notohara , 1990 ) . Unlike approaches that model host species as a discrete phylogenetic trait of the virus using continuous-time Markov processes ( or simpler , parsimony based , approaches ) ( Faria et al . , 2013; Lycett et al . , 2016 ) , population structure models explicitly incorporate contrasts in deme effective population sizes and migration between demes . By estimating independent coalescence rates for MERS-CoV in humans and camels , as well as migration patterns between the two demes , we show that long-term viral evolution of MERS-CoV occurs exclusively in camels . Our results suggest that spillover events into humans are seasonal and might be associated with the calving season in camels . However , we find that MERS-CoV , once introduced into humans , follows transient transmission chains that soon abate . Using Monte Carlo simulations we show that R0 for MERS-CoV circulating in humans is much lower than the epidemic threshold of 1 . 0 and that correspondingly the virus has been introduced into humans hundreds of times . The structured coalescent approach we employ ( see Materials and methods ) identifies camels as a reservoir host where most of MERS-CoV evolution takes place ( Figure 1 ) , while human MERS outbreaks are transient and terminal with respect to long-term evolution of the virus ( Figure 1—figure supplement 1 ) . Across 174 MERS-CoV genomes collected from humans , we estimate a median of 56 separate camel-to-human cross-species transmissions ( 95% highest posterior density ( HPD ) : 48–63 ) . While we estimate a median of 3 ( 95% HPD: 0–12 ) human-to-camel migrations , the 95% HPD interval includes zero and we find that no such migrations are found in the maximum clade credibility tree ( Figure 1 ) . Correspondingly , we observe substantially higher camel-to-human migration rate estimates than human-to-camel migration rate estimates ( Figure 1—figure supplement 2 ) . This inference derives from the tree structure wherein human viruses appear as clusters of highly related sequences nested within the diversity seen in camel viruses , which themselves show significantly higher diversity and less clustering . This manifests as different rates of coalescence with camel viruses showing a scaled effective population size Ne⁢τ of 3 . 49 years ( 95% HPD: 2 . 71–4 . 38 ) and human viruses showing a scaled effective population of 0 . 24 years ( 95% HPD: 0 . 14–0 . 34 ) . We believe that the small number of inferred human-to-camel migrations are induced by the migration rate prior , while some are derived from phylogenetic proximity of human sequences to the apparent ‘backbone’ of the phylogenetic tree . This is most apparent in lineages in early-mid 2013 that lead up to sequences comprising the MERS-CoV clade dominant in 2015 , where owing to poor sampling of MERS-CoV genetic diversity from camels the model cannot completely dismiss humans as a potential alternative host . The first sequences of MERS-CoV from camels do not appear in our data until November 2013 . Our finding of negligible human-to-camel transmission is robust to choice of prior ( Figure 1—figure supplement 2 ) . The repeated and asymmetric introductions of short-lived clusters of MERS-CoV sequences from camels into humans leads us to conclude that MERS-CoV epidemiology in humans is dominated by zoonotic transmission ( Figure 1 and Figure 1—figure supplement 1 ) . We observe dense terminal clusters of MERS-CoV circulating in humans that are of no subsequent relevance to the evolution of the virus . These clusters of presumed human-to-human transmission are then embedded within extensive diversity of MERS-CoV lineages inferred to be circulating in camels , a classic pattern of source-sink dynamics . Our findings suggest that instances of human infection with MERS-CoV are more common than currently thought , with exceedingly short transmission chains mostly limited to primary cases that might be mild and ultimately not detected by surveillance or sequencing . Structured coalescent analyses recover the camel-centered picture of MERS-CoV evolution despite sequence data heavily skewed towards non-uniformly sampled human cases and are robust to choice of prior . Comparing these results with a currently standard discrete trait analysis ( Lemey et al . , 2009 ) approach for ancestral state reconstruction shows dramatic differences in host reconstruction at internal nodes ( Figure 1—figure supplement 3 ) . Discrete trait analysis reconstruction identifies both camels and humans as important hosts for MERS-CoV persistence , but with humans as the ultimate source of camel infections . A similar approach has been attempted previously ( Zhang et al . , 2016 ) , but this interpretation of MERS-CoV evolution disagrees with lack of continuing human transmission chains outside of Arabian peninsula , low seroprevalence in humans and very high seroprevalence in camels across Saudi Arabia . We suspect that this particular discrete trait analysis reconstruction is false due to biased data , that is , having nearly twice as many MERS-CoV sequences from humans ( n=174 ) than from camels ( n=100 ) and the inability of the model to account for and quantify vastly different rates of coalescence in the phylogenetic vicinity of both types of sequences . We can replicate these results by either applying the same model to current data ( Figure 1—figure supplement 3 ) or by enforcing equal coalescence rates within each deme in the structured coalescent model ( Figure 1—figure supplement 4 ) . We use the posterior distribution of MERS-CoV introduction events from camels to humans ( Figure 1 ) to model seasonal variation in zoonotic transfer of viruses . We identify four months ( April , May , June , July ) when the odds of MERS-CoV introductions are increased ( Figure 2 ) and four when the odds are decreased ( August , September , November , December ) . Camel calving is reported to occur from October to February ( Almutairi et al . , 2010 ) , with rapidly declining maternal antibody levels in calves within the first weeks after birth ( Wernery , 2001 ) . It is possible that MERS-CoV sweeps through each new camel generation once critical mass of susceptibles is reached ( Martinez-Bakker et al . , 2014 ) , leading to a sharp rise in prevalence of the virus in camels and resulting in increased force of infection into the human population . Strong influx of susceptibles and subsequent sweeping outbreaks in camels may explain evidence of widespread exposure to MERS-CoV in camels from seroepidemiology ( Müller et al . , 2014; Corman et al . , 2014; Chu et al . , 2014; Reusken et al . , 2013; Reusken et al . , 2014 ) . Although we find evidence of seasonality in zoonotic spillover timing , no such relationship exists for sizes of human sequence clusters ( Figure 2B ) . This is entirely expected , since little seasonality in human behaviour that could facilitate MERS-CoV transmission is expected following an introduction . Similarly , we do not observe any trend in human sequence cluster sizes over time ( Figure 2B , Spearman ρ=0 . 06 , p=0 . 68 ) , suggesting that MERS-CoV outbreaks in humans are neither growing nor shrinking in size . This is not surprising either , since MERS-CoV is a camel virus that has to date , experienced little-to-no selective pressure to improve transmissibility between humans . Structured coalescent approaches clearly show humans to be a terminal host for MERS-CoV , implying poor transmissibility . However , we wanted to translate this observation into an estimate of the basic reproductive number R0 to provide an intuitive epidemic behaviour metric that does not require expertise in reading phylogenies . The parameter R0 determines expected number of secondary cases in a single infections as well as the distribution of total cases that result from a single introduction event into the human population ( Equation 1 , Materials and methods ) . We estimate R0 along with other relevant parameters via Monte Carlo simulation in two steps . First , we simulate case counts across multiple outbreaks totaling 2000 cases using Equation 1 and then we subsample from each case cluster to simulate sequencing of a fraction of cases . Sequencing simulations are performed via a multivariate hypergeometric distribution , where the probability of sequencing from a particular cluster depends on available sequencing capacity ( number of trials ) , numbers of cases in the cluster ( number of successes ) and number of cases outside the cluster ( number of failures ) . In addition , each hypergeometric distribution used to simulate sequencing is concentrated via a bias parameter , that enriches for large sequence clusters at the expense of smaller ones ( for its effects see Figure 3—figure supplement 1 ) . This is a particularly pressing issue , since a priori we expect large hospital outbreaks of MERS to be overrepresented in sequence data , whereas sequences from primary cases will be sampled exceedingly rarely . We record the number , mean , standard deviation and skewness ( third standardised moment ) of sequence cluster sizes in each simulation ( left-hand panel in Figure 3 ) and extract the subset of Monte Carlo simulations in which these summary statistics fall within the 95% highest posterior density observed in the empirical MERS-CoV data from structured coalescent analyses . We record R0 values , as well as the number of case clusters ( equivalent to number of zoonotic introductions ) , for these empirically matched simulations . A schematic of this Monte Carlo procedure is shown in Figure 3—figure supplement 2 . Since the total number of cases is fixed at 2000 , higher R0 results in fewer larger transmission clusters , while lower R0 results in many smaller transmission clusters . We find that observed phylogenetic patterns of sequence clustering strongly support R0 below 1 . 0 ( middle panel in Figure 3 ) . Mean R0 value observed in matching simulations is 0 . 72 ( 95% percentiles 0 . 57–0 . 90 ) , suggesting the inability of the virus to sustain transmission in humans . Lower values for R0 in turn suggest relatively large numbers of zoonotic transfers of viruses into humans ( right-hand panel in Figure 3 ) . Median number of cross-species introductions observed in matching simulations is 592 ( 95% percentiles 311–811 ) . Our results suggest a large number of unobserved MERS primary cases . Given this , we also performed simulations where the total number of cases is doubled to 4000 to explore the impact of dramatic underestimation of MERS cases . In these analyses , R0 values tend to decrease even further as numbers of introductions go up , although very few simulations match currently observed MERS-CoV sequence data ( Figure 3—figure supplement 3 ) . Overall , our analyses indicate that MERS-CoV is poorly suited for human-to-human transmission , with an estimated R0<0 . 90 and sequence sampling likely to be biased towards large hospital outbreaks ( Figure 3—figure supplement 1 ) . All matching simulations exhibit highly skewed distributions of case cluster sizes with long tails ( Figure 3—figure supplement 4 ) , which is qualitatively similar to the results of ( Cauchemez et al . , 2016 ) . We find that simulated sequence cluster sizes resemble observed sequence cluster sizes in the posterior distribution , with a slight reduction in mid-sized clusters in simulated data ( Figure 3—figure supplement 5 ) . Given these findings , and the fact that large outbreaks of MERS occurred in hospitals , the combination of frequent spillover of MERS-CoV into humans and occasional outbreak amplification via poor hygiene practices ( Assiri et al . , 2013b; Chen et al . , 2017 ) appear sufficient to explain observed epidemiological patterns of MERS-CoV . Recombination has been shown to occur in all genera of coronaviruses , including MERS-CoV ( Lai et al . , 1985; Makino et al . , 1986; Keck et al . , 1988; Kottier et al . , 1995; Herrewegh et al . , 1998 ) . In order to quantify the degree to recombination has shaped MERS-CoV genetic diversity , we used two recombination detection approaches across partitions of taxa corresponding to inferred MERS-CoV clades . Both methods rely on sampling parental and recombinant alleles within the alignment , although each quantifies different signals of recombination . One hallmark of recombination is the ability to carry alleles derived via mutation from one lineage to another , which appear as repeated mutations taking place in the recipient lineage , somewhere else in the tree . The PHI ( pairwise homoplasy index ) test quantifies the appearance of these excessive repeat mutations ( homoplasies ) within an alignment ( Bruen et al . , 2006 ) . Another hallmark of recombination is clustering of alleles along the genome , due to how template switching , the primary mechanism of recombination in RNA viruses , occurs . 3Seq relies on the clustering of nucleotide similarities along the genome between sequence triplets – two potential parent-donors and one candidate offspring-recipient sequences ( Boni et al . , 2007 ) . Both tests can give spurious results in cases of extreme rate heterogeneity and sampling over time ( Dudas and Rambaut , 2016 ) , but both tests have not been reported to fail simultaneously . PHI and 3Seq methods consistently identify most of the apparent ‘backbone’ of the MERS-CoV phylogeny as encompassing sequences with evidence of recombination ( Figure 4—figure supplement 1 ) . Neither method can identify where in the tree recombination occurred , but each full asterisk in Figure 4—figure supplement 1 should be interpreted as the minimum partition of data that still captures both donor and recipient alleles involved in a recombination event . This suggests a non-negligible contribution of recombination in shaping existing MERS-CoV diversity . As done previously ( Dudas and Rambaut , 2016 ) , we show large numbers of homoplasies in MERS-CoV data ( Figure 4—figure supplement 2 ) with some evidence of genomic clustering of such alleles . These results are consistent with high incidence of MERS-CoV in camels ( Müller et al . , 2014; Corman et al . , 2014; Chu et al . , 2014; Reusken et al . , 2014; Ali et al . , 2017 ) , allowing for co-infection with distinct genotypes and thus recombination to occur ( Sabir et al . , 2016 ) . Owing to these findings , we performed a sensitivity analysis in which we partitioned the MERS-CoV genome into two fragments and identified human outbreak clusters within each fragment . We find strong similarity in the grouping of human cases into outbreak clusters between fragments ( Figure 4A ) . Between the two trees in Figure 4B four ( out of 54 ) ‘human’ clades are expanded where either singleton introductions or two-taxon clades in fragment 2 join other clades in fragment 1 . For the reverse comparison , there are five ‘human’ clades ( out of 53 ) in fragment 2 that are expanded . All such clades have low divergence ( Figure 4B ) and thus incongruences in human clades are more likely to be caused by differences in deme assignment rather than actual recombination . And while we observe evidence of distinct phylogenetic trees from different parts of the MERS-CoV genome ( Figure 4B ) , human clades are minimally affected and large portions of the posterior probability density in both parts of the genome are concentrated in shared clades ( Figure 4—figure supplement 3 ) . Additionally , we observe the same source-sink dynamics between camel and human populations in trees constructed from separate genomic fragments as were observed in the original full genome tree ( Figures 1 and 4B ) . Observed departures from strictly clonal evolution suggest that while recombination is an issue for inferring MERS-CoV phylogenies , its effect on the human side of MERS outbreaks is minimal , as expected if humans represent a transient host with little opportunity for co-infection . MERS-CoV evolution on the reservoir side is complicated by recombination , although is nonetheless still largely amenable to phylogenetic methods . Amongst other parameters of interest , recombination is expected to interfere with molecular clocks , where transferred genomic regions can give the impression of branches undergoing rapid evolution , or branches where recombination results in reversions appearing to evolve slow . In addition to its potential to influence tree topology , recombination in molecular sequence data is an erratic force with unpredictable effects . We suspect that the effects of recombination in MERS-CoV data are reigned in by a relatively small effective population size of the virus in Saudi Arabia ( see next section ) where haplotypes are fixed or nearly fixed , thus preventing an accumulation of genetic diversity that would then be reshuffled via recombination . Nevertheless the evolutionary rate of the virus appears to fall firmly within the expected range for RNA viruses: regression of nucleotide differences to Jordan-N3/2012 genome against sequence collection dates yields a rate of 4 . 59×10-4 subs/site/year , Bayesian structured coalescent estimate from primary analysis 9 . 57×10-4 ( 95% HPDs: 8 . 28-10 . 9×10-4 ) subs/site/year . Here , we attempt to investigate MERS-CoV demographic patterns in the camel reservoir . We supplement camel sequence data with a single earliest sequence from each human cluster , treating viral diversity present in humans as a sentinel sample of MERS-CoV diversity circulating in camels . This removes conflicting demographic signals sampled during human outbreaks , where densely sampled closely related sequences from humans could be misconstrued as evidence of demographic crash in the viral population . Despite lack of convergence , neither of the two demographic reconstructions show evidence of fluctuations in the scaled effective population size ( Ne⁢τ ) of MERS-CoV over time ( Figure 5 ) . We recover a similar demographic trajectory when estimating Ne⁢τ over time with a skygrid tree prior across the genome split into ten fragments with independent phylogenetic trees to account for confounding effects of recombination ( Figure 5—figure supplement 1 ) . However , we do note that coalescence rate estimates are high relative to the sampling time period , with a mean estimate of Ne⁢τ at 3 . 49 years ( 95% HPD: 2 . 71–4 . 38 ) , and consequently MERS-CoV phylogeny resembles a ladder , as often seen in human influenza A virus phylogenies ( Bedford et al . , 2011 ) . This empirically estimated effectived population can be compared to the expected effective population size in a simple epidemiological model . At endemic equilibrium , we expect scaled effective population size Ne⁢τ to follow I/ 2⁢β , where β is the equilibrium rate of transmission and I is the equilibrium number of infecteds ( Frost and Volz , 2010 ) . We assume that β is constant and is equal to the rate of recovery . Given a 20 day duration of infection in camels ( Adney et al . , 2014 ) , we arrive at β=365/20=18 . 25 infections per year . Given extremely high seroprevalence estimates within camels in Saudi Arabia ( Müller et al . , 2014; Corman et al . , 2014; Chu et al . , 2014; Reusken et al . , 2013; Reusken et al . , 2014 ) , we expect camels to usually be infected within their first year of life . Therefore , we can estimate the rough number of camel infections per year as the number of calves produced each year . We find there are 830 , 000 camels in Saudi Arabia ( Abdallah and Faye , 2013 ) and that female camels in Saudi Arabia have an average fecundity of 45% ( Abdallah and Faye , 2013 ) . Thus , we expect 830 000×0 . 50×0 . 45=186 750 new calves produced yearly and correspondingly 186 , 750 new infections every year , which spread over 20 day intervals gives an average prevalence of I=10 233 infections . This results in an expected scaled effective population size Ne⁢τ=280 . 4 years . Comparing expected Ne⁢τ to empirical Ne⁢τ we arrive at a ratio of 80 . 3 ( 64 . 0–103 . 5 ) . This is less than the equivalent ratio for human measles virus ( Bedford et al . , 2011 ) and is in line with the expectation from neutral evolutionary dynamics plus some degree of transmission heterogeneity ( Volz et al . , 2013 ) and seasonal troughs in prevalence . Thus , we believe that the ladder-like appearance of the MERS-CoV tree can likely be explained by entirely demographic factors . In this study we aimed to understand the drivers of MERS coronavirus transmission in humans and what role the camel reservoir plays in perpetuating the epidemic in the Arabian peninsula by using sequence data collected from both hosts ( 174 from humans and 100 from camels ) . We showed that currently existing models of population structure ( Vaughan et al . , 2014 ) can identify distinct demographic modes in MERS-CoV genomic data , where viruses continuously circulating in camels repeatedly jump into humans and cause small outbreaks doomed to extinction ( Figure 1—figure supplement 1 ) . This inference succeeds under different choices of priors for unknown demographic parameters ( Figure 1—figure supplement 2 ) and in the presence of strong biases in sequence sampling schemes ( Figure 3 ) . When rapid coalescence in the human deme is not allowed ( Figure 1—figure supplement 4 ) structured coalescent inference loses power and ancestral state reconstruction is nearly identical to that of discrete trait analysis ( Figure 1—figure supplement 3 ) . When allowed different deme-specific population sizes , the structured coalescent model succeeds because a large proportion of human sequences fall into tightly connected clusters , which informs a low estimate for the population size of the human deme . This in turn informs the inferred state of long ancestral branches in the phylogeny , that is , because these long branches are not immediately coalescing , they are most likely in camels . From sequence data , we identify at least 50 zoonotic introductions of MERS-CoV into humans from the reservoir ( Figure 1 ) , from which we extrapolate that hundreds more such introductions must have taken place ( Figure 3 ) . Although we recover migration rates from our model ( Figure 1—figure supplement 2 ) , these only pertain to sequences and in no way reflect the epidemiologically relevant per capita rates of zoonotic spillover events . We also looked at potential seasonality in MERS-CoV spillover into humans . Our analyses indicated a period of three months where the odds of a sequenced spillover event are increased , with timing consistent with an enzootic amongst camel calves ( Figure 2 ) . As a result of our identification of large and asymmetric flow of viral lineages into humans we also find that the basic reproduction number for MERS-CoV in humans is well below the epidemic threshold ( Figure 3 ) . Having said that , there are highly customisable coalescent methods available that extend the methods used here to accommodate time varying migration rates and population sizes , integrate alternative sources of information and fit to stochastic nonlinear models ( Rasmussen et al . , 2014 ) , which would be more appropriate for MERS-CoV . Some distinct aspects of MERS-CoV epidemiology could not be captured in our methodology , such as hospital outbreaks where R0 is expected to be consistently closer to 1 . 0 compared to community transmission of MERS-CoV . Outside of coalescent-based models , there are population structure models that explicitly relate epidemiological parameters to the branching process observed in sequence data ( Kühnert et al . , 2016 ) , but often rely on specifying numerous informative priors and can suffer from MCMC convergence issues . Strong population structure in viruses often arises through limited gene flow , either due to geography ( Dudas et al . , 2017 ) , ecology ( Smith et al . , 2009 ) or evolutionary forces ( Turner et al . , 2005; Dudas et al . , 2015 ) . On a smaller scale , population structure can unveil important details about transmission patterns , such as identifying reservoirs and understanding spillover trends and risk , much as we have done here . When properly understood naturally arising barriers to gene flow can be exploited for more efficient disease control and prevention , as well as risk management . Severe acute respiratory syndrome ( SARS ) coronavirus , a Betacoronavirus like MERS-CoV , caused a serious epidemic in humans in 2003 , with over 8000 cases and nearly 800 deaths . Since MERS-CoV was also able to cause significant pathogenicity in the human host it was inevitable that parallels would be drawn between MERS-CoV and SARS-CoV at the time of MERS discovery in 2012 . Although we describe the epidemiology of MERS-CoV from sequence data , indications that MERS-CoV has poor capacity to spread human-to-human existed prior to any sequence data . SARS-CoV swept through the world in a short period of time , but MERS cases trickled slowly and were restricted to the Arabian Peninsula or resulted in self-limiting outbreaks outside of the region , a pattern strongly indicative of repeat zoonotic spillover . Infectious disease surveillance and control measures remain limited , so much like the SARS epidemic in 2003 or the H1N1 pandemic in 2009 , zoonotic pathogens with R0>1 . 0 are probably going to be discovered after spreading beyond the original location of spillover . Even though our results show that MERS-CoV does not appear to present an imminent global threat , we would like to highlight that its epidemiology is nonetheless concerning . Pathogens Bacillus anthracis , Andes hantavirus ( Martinez et al . , 2005 ) , monkeypox ( Reed et al . , 2004 ) and influenza A are able to jump species barriers but only influenza A viruses have historically resulted in pandemics ( Lipsitch et al . , 2016 ) . MERS-CoV may join the list of pathogens able to jump species barriers but not spread efficiently in the new host . Since its emergence in 2012 , MERS-CoV has caused self-limiting outbreaks in humans with no evidence of worsening outbreaks over time . However , sustained evolution of the virus in the reservoir and continued flow of viral lineages into humans provides the substrate for a more transmissible variant of MERS-CoV to possibly emerge . Previous modelling studies ( Antia et al . , 2003; Park et al . , 2013 ) suggest a positive relationship between initial R0 in the human host and probability of evolutionary emergence of a novel strain which passes the supercritical threshold of R0>1 . 0 . This leaves MERS-CoV in a precarious position; on one hand its current R0 of ∼0 . 7 is certainly less than 1 , but its proximity to the supercritical threshold raises alarm . With very little known about the fitness landscape or adaptive potential of MERS-CoV , it is incredibly challenging to predict the likelihood of the emergence more transmissible variants . In light of these difficulties , we encourage continued genomic surveillance of MERS-CoV in the camel reservoir and from sporadic human cases to rapidly identify a supercritical variant , if one does emerge . All MERS-CoV sequences were downloaded from GenBank and accession numbers are given in Supplementary file 1 ( Assiri et al . , 2016a , 2016b; Azhar et al . , 2014; van Boheemen et al . , 2012; Briese et al . , 2014; Chu et al . , 2014; Cotten et al . , 2013 , 2014; Drosten et al . , 2013 , 2015; Fagbo et al . , 2015; Haagmans et al . , 2014; Hemida et al . , 2014; Hunter et al . , 2016; Kandeil et al . , 2016; Kapoor et al . , 2014; Kim et al . , 2015 , 2016; Lamers et al . , 2016; Lau et al . , 2016; Lu et al . , 2017; Park et al . , 2016a , 2016b; Plipat et al . , 2017; Raj et al . , 2014; Sabir et al . , 2016; Seong et al . , 2016; Xie et al . , 2015 ) . Sequences without accessions were kindly shared by Ali M . Somily , Mazin Barry , Sarah S . Al Subaie , Abdulaziz A . BinSaeed , Fahad A . Alzamil , Waleed Zaher , Theeb Al Qahtani , Khaldoon Al Jerian , Scott J . N . McNabb , Imad A . Al-Jahdali , Ahmed M . Alotaibi , Nahid A . Batarfi , Matthew Cotten , Simon J . Watson , Spela Binter , and Paul Kellam prior to publication . Sequences available on GenBank but not associated with publications were shared by Meriadeg Ar Gouilh , Emad M . Elassal , Monica Galiano , Krista Queen and Ying Tao . Fragments of some strains submitted to GenBank as separate accessions were assembled into a single sequence . Only sequences covering at least 50% of MERS-CoV genome were kept , to facilitate later analyses where the alignment is split into two parts in order to account for effects of recombination ( Dudas and Rambaut , 2016 ) . Sequences were annotated with available collection dates and hosts , designated as camel or human , aligned with MAFFT ( Katoh and Standley , 2013 ) , and edited manually . Protein coding sequences were extracted and concatenated , reducing alignment length from 30 , 130 down to 29 , 364 , which allowed for codon-partitioned substitution models to be used . The final dataset consisted of 174 genomes from human infections and 100 genomes from camel infections ( Supplementary file 1 ) . Here , we employ a Monte Carlo simulation approach to identify parameters consistent with observed patterns of sequence clustering ( Figure 3—figure supplement 2 ) . Our structured coalescent analyses indicate that every MERS outbreak is a contained cross-species spillover of the virus from camels into humans . The distribution of the number of these cross-species transmissions and their sizes thus contain information about the underlying transmission process . At heart , we expect fewer larger clusters if fundamental reproductive number R0 is large and more smaller clusters if R0 is small . A similar approach was used in Grubaugh et al . ( 2017 ) to estimate R0 for Zika introductions into Florida . Branching process theory provides an analytical distribution for the number of eventual cases j in a transmission chain resulting from a single introduction event with R0 and dispersion parameter ω ( Blumberg and Lloyd-Smith , 2013 ) . This distribution follows ( 1 ) Pr ( j\|R0 , ω ) =Γ⁢ ( ω⁢j+j-1 ) Γ⁢ ( ω⁢j ) ⁢Γ⁢ ( j+1 ) ( R0ω ) j-1 ( 1+R0ω ) ω⁢j+j-1 . Here , R0 represents the expected number of secondary cases following a single infection and ω represents the dispersion parameter assuming secondary cases follow a negative binomial distribution ( Lloyd-Smith et al . , 2005 ) , so that smaller values represent larger degrees of heterogeneity in the transmission process . As of 10 May 2017 , the World Health Organization has been notified of 1952 cases of MERS-CoV ( World Health Organization , 2017 ) . We thus simulated final transmission chain sizes using Equation 1 until we reached an epidemic comprised of N=2000 cases . 10 , 000 simulations were run for 121 uniformly spaced values of R0 across the range [0 . 5–1 . 1] with dispersion parameter ω fixed to 0 . 1 following expectations from previous studies of coronavirus behavior ( Lloyd-Smith et al . , 2005 ) . Each simulation results in a vector of outbreak sizes 𝐜 , where ci is the size of the ith transmission cluster and ∑i=1Kci=2000 and K is the number of clusters generated . Following the underlying transmission process generating case clusters 𝐜 , we simulate a secondary process of sampling some fraction of cases and sequencing them to generate data analogous to what we empirically observe . We sample from the case cluster size vector 𝐜 without replacement according to a multivariate hypergeometric distribution ( see Algorithm 1: Multivariate hypergeometric sampling scheme ) . The resulting sequence cluster size vector 𝐬 contains K entries , some of which are zero ( i . e . case clusters not sequenced ) , but ∑i=1Ksi=174 which reflects the number of human MERS-CoV sequences used in this study . Note that this ‘sequencing capacity’ parameter does not vary over time , even though MERS-CoV sequencing efforts have varied in intensity , starting out slow due to lack of awareness , methods and materials and increasing in response to hospital outbreaks later . As the default sampling scheme operates under equiprobable sequencing , we also simulated biased sequencing by using concentrated hypergeometric distributions where the probability mass function is squared ( bias = 2 ) or cubed ( bias = 3 ) and then normalized . Here , bias enriches the hypergeometric distribution so that sequences are sampled with weights proportional to ( c1bias , c2bias , … , ckbias ) . Thus , bias makes larger clusters more likely to be ‘sequenced’ . After simulations were completed , we identified simulations in which the recovered distribution of sequence cluster sizes 𝐬 fell within the 95% highest posterior density intervals for four summary statistics of empirical MERS-CoV sequence cluster sizes recovered via structured coalescent analysis: number of sequence clusters , mean , standard deviation and skewness ( third central moment ) . These values were 48–61 for number of sequence clusters , 2 . 87–3 . 65 for mean sequence cluster size , 4 . 84–6 . 02 for standard deviation of sequence cluster sizes , and 415 . 40–621 . 06 for skewness of sequence cluster sizes . We performed a smaller set of simulations with 2500 replicates and twice the number of cases , that is , ∑i=1KCi=4000 , to explore a dramatically underreported epidemic . Additionally , we performed additional smaller set of simulations on a rougher grid of R0 values ( 23 values , 0 . 50–1 . 05 ) , with 5 values of dispersion parameter ω ( 0 . 002 , 0 . 04 , 0 . 1 , 0 . 5 , 1 . 0 ) and 3 levels of bias ( 1 , 2 , 3 ) to justify our choice of dispersion parameter ω that was fixed to 0 . 1 in the main analyses ( Figure 3—figure supplement 6 ) . Pseudocode describes the multivariate hypergeometric sampling scheme that simulates sequencing . Probability of sequencing a given number of cases from a case cluster depends on cluster size and sequences left ( i . e . ‘sequencing capacity’ ) . The bias parameter determines how probability mass function of the hypergeometric distribution is concentrated . Data: Array of case cluster sizes in outbreak 𝐜= ( c1 , c2 , … , cK ) , sequences available M , total outbreak size N , where N=∑i=1Kci . Result: Array of sequence cluster sizes sampled: 𝐬= ( s1 , s2 , … , sK ) . Draw si from a hypergeometric distribution with ci successes , N-ci failures after M trials; while i<Ki=i+1 do i=i+1; M=M-si-1; Compute the probability mass function ( pmf ) for all possible values of si , 𝐩= ( p⁢ ( 0 ) bias , p⁢ ( 1 ) bias , … , p⁢ ( ci ) bias ) × ( ∑ipibias ) -1 , where p⁢ ( ⋅ ) is the pmf for a hypergeometric distribution with ci successes , N-ci failures after M trials; Draw a sequence cluster size si from array of potential sequence cluster sizes ( 0 , 1 , … , ci ) according to 𝒑; end Add remaining sequences to last sequence cluster cK=M-sK-1; In order to infer the demographic history of MERS-CoV in camels we used the results of structured coalescent analyses to identify introductions of the virus into humans . The oldest sequence from each cluster introduced into humans was kept for further analysis . This procedure removes lineages coalescing rapidly in humans , which would otherwise introduce a strong signal of low effective population size . These subsampled MERS-CoV sequences from humans were combined with existing sequence data from camels to give us a dataset with minimal demographic signal coming from epidemiological processes in humans . Sequences belonging to the outgroup clade where most of MERS-CoV sequences from Egypt fall were removed out of concern that MERS epidemics in Saudi Arabia and Egypt are distinct epidemics with relatively poor sampling in the latter . Were more sequences of MERS-CoV available from other parts of Africa we speculate they would fall outside of the diversity that has been sampled in Saudi Arabia and cluster with early MERS-CoV sequences from Jordan and sequences from Egyptian camels . However , currently there are no indications of what MERS-CoV diversity looks like in camels east of Saudi Arabia . A flexible skygrid tree prior ( Gill et al . , 2013 ) was used to recover estimates of scaled effective population size ( Ne⁢τ ) at 50 evenly spaced grid points across six years , ending at the most recent tip in the tree ( 2015 August ) in BEAST v1 . 8 . 4 ( Drummond et al . , 2012 ) , under a relaxed molecular clock with rates drawn from a lognormal distribution ( Drummond et al . , 2006 ) and codon position partitioned ( positions 1+2 and 3 ) HKY +Γ4 ( Hasegawa et al . , 1985; Yang , 1994 ) nucleotide substitution models . At time of writing advanced flexible coalescent tree priors from the skyline family , such as skygrid ( Gill et al . , 2013 ) are available in BEAST v1 ( Drummond et al . , 2012 ) but not in BEAST v2 ( Bouckaert et al . , 2014 ) . We set up five independent MCMC chains to run for 500 million states , sampling every 50 000 states . This analysis suffered from poor convergence , where two chains converged onto one stationary distribution , two to another and the last chain onto a third stationary distribution , with high effective sample sizes . Demographic trajectories recovered by the two main stationary distributions are very similar and differences between the two appear to be caused by convergence onto subtly different tree topologies . This non-convergence effect may have been masked previously by the use of all available MERS-CoV sequences from humans which may have lead MCMC towards one of the multiple stationary distributions . To ensure that recombination was not interfering with the skygrid reconstruction , we also split our MERS-CoV alignment into ten parts 2937 nucleotides long . These were then used as separate partitions with independent trees and clock rates in BEAST v1 . 8 . 4 ( Drummond et al . , 2012 ) . Nucleotide substitution and relaxed clock models were set up identically to the whole genome skygrid analysis described above ( Drummond et al . , 2006; Hasegawa et al . , 1985; Yang , 1994 ) . Skygrid coalescent tree prior ( Gill et al . , 2013 ) was used jointly across all ten partitions for demographic inference . Five MCMC chains were set up , each running for 200 million states , sampling every 20 , 000 states . Sequence data and all analytical code is publicly available at https://github . com/blab/mers-structure ( Dudas , 2017 ) . A copy is archived at https://github . com/elifesciences-publications/mers-structure .	Coronaviruses are one of many groups of viruses that cause the common cold , though some members of the group can cause more serious illnesses . The SARS coronavirus , for example , caused a widespread epidemic of pneumonia in 2003 that killed 774 people . In 2012 , a new coronavirus was detected in patients from the Arabian Peninsula with severe respiratory symptoms known as Middle East respiratory syndrome ( or MERS for short ) . To date the MERS coronavirus has also killed over 700 people ( albeit over a number of years rather than months ) . Yet unlike the SARS coronavirus that spreads efficiently between humans , cases of MERS were rarely linked to each other or to contact with animals , with the exception of hospital outbreaks . Though camels were later identified as the original source of MERS coronavirus infections in humans , the role of these animals in the epidemic was not well understood . Throughout the epidemic nearly 300 genomes of the MERS coronavirus had been sequenced , from both camels and humans . Previous attempts to understand the MERS epidemic had either relied on these data or reports of case numbers but led to conflicting results , at odds with other sources of evidence . Dudas et al . wanted to work out how many times the MERS coronavirus had been introduced into humans from camels . If it happened once , this would indicate that the virus is good at spreading between humans and that treating human cases should be a priority . However , if every human case occurred as a new introduction of the MERS coronavirus from camels , this would mean that the human epidemic would not stop until the virus is controlled at the source , that is , in camels . Many scientists had argued that the second of these scenarios was most likely , but this had not been strongly demonstrated with data . By looking at the already sequenced genomes , Dudas et al . worked out how the MERS coronaviruses were related to each other , and reconstructed their family tree . Information about the host from which each sequence was collected was then mapped on the tree . Unlike previous attempts to complete this kind of analysis , Dudas et al . took an approach that could deal with the viruses in camels being more diverse than those in humans . Consistent with the scenario where human cases occurred as new introductions from camels , the analysis showed that the MERS coronavirus populations is maintained exclusively in camels and the viruses seen in humans are evolutionary dead-ends . This suggests that MERS coronavirus spreads poorly between humans , and has instead jumped from camels to humans hundreds of times since 2012 . As well as providing data to confirm a previously suspected hypothesis , these findings provide more support to the current plans to mitigate infections with MERS coronavirus in the Arabian Peninsula by focusing control efforts on camels .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health", "microbiology", "and", "infectious", "disease" ]	2018	MERS-CoV spillover at the camel-human interface
The unfolded protein response ( UPR ) monitors the protein folding capacity of the endoplasmic reticulum ( ER ) . In all organisms analyzed to date , the UPR drives transcriptional programs that allow cells to cope with ER stress . The non-conventional splicing of Hac1 ( yeasts ) and XBP1 ( metazoans ) mRNA , encoding orthologous UPR transcription activators , is conserved and dependent on Ire1 , an ER membrane-resident kinase/endoribonuclease . We found that the fission yeast Schizosaccharomyces pombe lacks both a Hac1/XBP1 ortholog and a UPR-dependent-transcriptional-program . Instead , Ire1 initiates the selective decay of a subset of ER-localized-mRNAs that is required to survive ER stress . We identified Bip1 mRNA , encoding a major ER-chaperone , as the sole mRNA cleaved upon Ire1 activation that escapes decay . Instead , truncation of its 3′ UTR , including loss of its polyA tail , stabilized Bip1 mRNA , resulting in increased Bip1 translation . Thus , S . pombe uses a universally conserved stress-sensing machinery in novel ways to maintain homeostasis in the ER . Homeostatic control mechanisms are essential to life , allowing cells to balance capacity and demand of numerous physiological processes . One such mechanism , the unfolded protein response ( UPR ) , operates in all eukaryotic cells to adjust the protein folding capacity of the endoplasmic reticulum ( ER ) according to need . Environmental or physiological demands can lead to an imbalance between the protein folding load and the protein folding capacity in the ER lumen , resulting in an accumulation of unfolded or misfolded proteins , a condition termed ‘ER stress’ ( Walter and Ron , 2011 ) . When unmitigated , ER stress is toxic to cells and triggers cell death ( Shore et al . , 2011; Tabas and Ron , 2011; Hetz , 2012 ) . The UPR is a network of evolutionarily conserved signal transduction pathways that monitors the conditions in the ER lumen to induce a transcriptional response . In metazoan cells , three ER-resident transmembrane sensors , Ire1 , PERK , ATF6 transmit information into the cytosol . Each sensor activates transcription factors that collaborate to drive expression of UPR target genes ( Walter and Ron , 2011 ) , including genes encoding ER-lumenal chaperones , such as BiP , an abundantly expressed Hsp70 family member . Ire1 is a bifunctional transmembrane kinase/endoribonuclease that controls expression of the transcription factor XBP1 by a non-conventional splicing of its mRNA . Ire1 uses its ER-lumenal domain to detect unfolded proteins , and in response activates by homo-oligomerization , trans-autophosphorylation , and allosteric activation of its cytosolic nuclease modality ( Korennykh et al . , 2009; Gardner and Walter , 2011 ) . Activated Ire1 cleaves the XBP1 mRNA at two discrete stem-loop structures , excising a short intron . The two severed exons are then ligated to produce spliced XBP1 mRNA , which because of a frame-shift induced by the splicing event , are translated to produce active XBP1 ( Yoshida et al . , 2001; Calfon et al . , 2002 ) . Ire1 was first discovered in the budding yeast S . cerevisiae , where it constitutes the core machinery of the cells' only UPR signaling pathway ( Cox et al . , 1993; Mori et al . , 1993 ) . S . cerevisiae Ire1 splices Hac1 mRNA , encoding the yeast ortholog of XBP1 , by a mechanism that was later found conserved in all metazoan cells ( Cox and Walter , 1996; Sidrauski et al . , 1996 ) . Ire1-mediated mRNA splicing therefore is considered to be the most evolutionary ancient branch of the UPR . By first approximation , the three UPR branches collaborate to effect comprehensive transcriptional outputs , thereby enhancing the capacity of the ER according to need . PERK superimposes another layer of control by reducing the load of proteins entering the ER through translational control ( Pavitt and Ron , 2012 ) . Similarly , Ire1 is thought to play a dual role in UPR regulation . In particular , Hollien and Weissman ( 2006 ) first discovered in Drosophila cells that Ire1 induction not only results in splicing of XBP1 mRNA but also mediates enhanced mRNA breakdown . This output of Ire1 activation , termed ‘regulated Ire1-dependent decay’ ( RIDD ) , is conserved in mammalian cells , but not in S . cerevisiae , where transcriptional control via Hac1 mRNA splicing remains the only known route of UPR signaling ( Niwa et al . , 2005; Han et al . , 2009; Hollien et al . , 2009 ) . All identified RIDD target mRNAs are translated by membrane-bound ribosomes at the ER surface , where they are cleaved , most likely by Ire1 directly ( Han et al . , 2009; Hollien et al . , 2009; Cross et al . , 2012 ) . Once nicked and no longer protected by their polyA tails and 5′ caps , mRNA fragments are quickly degraded by the RNA surveillance machinery ( Hollien and Weissman , 2006; Garneau et al . , 2007 ) . By contrast to the strictly conserved stem/loop structures found at Hac1/XBP1 mRNA splice sites ( Gonzalez et al . , 1999 ) , RIDD target mRNAs do not contain easily recognizable features in common . Consequently , RIDD is thought to arise by a more promiscuous cleavage mode of Ire1 . It is unclear whether RIDD is mediated by an alternate conformation of activated Ire1 , or whether it arises in a specific Ire1 oligomerization state , as high-order oligomerization may serve to locally enhance low affinity interactions through avidity effects . RIDD cleavage reactions have been reconstituted in vitro with recombinantly expressed purified Ire1 , lending support to the notion that Ire1's endoribonuclease actvity , rather than another enzyme recruited to it , carries out the initial cleavage reaction ( Lee et al . , 2011; Cross et al . , 2012 ) . Because of Ire1's dual output , the physiological consequences of RIDD have been difficult to decipher . RIDD has been suggested to play cytoprotective roles , such as contributing to important feedback control on proinsulin expression in pancreatic beta-cells or protecting liver cells from acetaminophen toxicity by degrading the mRNAs encoding the cytochrome P450 variants responsible for the drug's toxification ( Lipson et al . , 2008; Hur et al . , 2012 ) . RIDD has also been suggested to play cytotoxic roles as a major contributor driving cells into apoptosis after prolonged and unmitigated exposure to ER stress ( Han et al . , 2009 ) . Surprisingly , in the work presented here we found no evidence that Ire1 controls transcription in the UPR of Schizosaccharomyces pombe . Instead , in S . pombe Ire1 maintains ER homeostasis through two post-transcriptional mechanisms: it initiates RIDD of a large , select set of ER-targeted mRNAs and processes Bip1 mRNA in an unprecedented way , thereby stabilizing it . Our studies reveal an unforeseen evolutionary plasticity in maintaining ER homeostasis . UPR induction in all eukaryotic cells analyzed to date involves the Ire1-mediated , non-conventional splicing of Hac1/XBP1 mRNA . The splice sites at which Ire1 cleaves the mRNA to initiate splicing lie in well-conserved stem/loop structures that are readily identified ( Gonzalez et al . , 1999 ) . We and others were therefore perplexed when bioinformatic analyses failed to identify Hac1/XBP1 orthologs in S . pombe and other yeasts of the same genus ( Figure 1a ) ( Hooks and Griffiths-Jones , 2011; Frost et al . , 2012 ) . The Hac1/XBP1 transcription factors are well conserved between species and are easily recognized by sequence alignment among the superfamily of bZIP transcription factors ( Figure 1—figure supplement 1 ) . By contrast , Ire1 is well conserved in S . pombe , with all of the functionally important hallmarks identified in other eukaryotes , including its ER lumenal unfolded protein sensing domain and its cytosolic kinase and RNase domains . Moreover , Ire1 was essential for S . pombe growth on tunicamycin ( Tm ) ( Figure 1b ) , which induces ER stress by blocking N-linked glycosylation , indicating that Ire1 serves an essential function in allowing cells to cope with ER stress . This function required Ire1's RNase activity , as Ire1 ( H1018N ) carrying a single amino acid substitution of a catalytic residue in Ire1's RNase active site failed to support cell growth on tunicamycin ( Figure 1b ) . 10 . 7554/eLife . 00048 . 003Figure 1 . The UPR in fission selectively down-regulates ER-targeted mRNAs . ( a ) Phylogenetic tree showing the components of the UPR in yeasts . The presence of recognizable orthologs of Ire1 and Hac1 is indicated . ( b ) Viability assay by serial dilution of wild type , Ire1Δ and Ire1 ( H1018N ) cells spotted on solid media with or without 0 . 03 µg/ml of the ER stress inducer tunicamycin ( Tm ) . Plates were photographed after 3 day of growth at 30°C . ( c ) Strand-specific polyA+ enriched mRNA-Seq analysis of annotated ORFs . The plot indicates the fold change ( log2 ) of transcript abundance in DTT-stressed Ire1Δ cells ( 2 mM DTT , 1 hr ) compared to DTT-stressed wild type cells ( 2 mM DTT , 1 hr ) in the x-axis , and transcript abundance in unstressed wild type cells compared to DTT-stressed wild type cells ( 2 mM DTT , 1 hr ) in the y-axis . Symbol sizes indicate abundance classes for each mRNA ( reads per kilobase ) . Transcripts encoding proteins with a signal sequence or transmembrane segment are colored red , all other transcripts are colored blue ( Figure 1—source data ) . ( d ) DTT-dependent and Ire1-dependent expression changes of transcripts displaying a signal sequence or a transmembrane domain . The skew of the left tail of the distribution indicates an enrichment ( p<1×10−20 ) of down-regulated mRNAs . Coloring is as in Figure 1d . ( e ) Distribution of gene-ontology ( GO ) annotations for Ire1-dependent down-regulated mRNAs . Percentages indicate genes within a particular GO category in relation to the total number of genes that have a GO annotation ( N=39 ) ( see Figure 1—figure supplement 3 for annotated list of genes ) . http://dx . doi . org/10 . 7554/eLife . 00048 . 00310 . 7554/eLife . 00048 . 004Figure 1—source data 1 . Gene expression and fold changehttp://dx . doi . org/10 . 7554/eLife . 00048 . 00410 . 7554/eLife . 00048 . 005Figure 1—figure supplement 1 . Alignment of DNA binding domain of Hac1 ( bZIP ) homologues in different yeast species . DOI: http://dx . doi . org/10 . 7554/eLife . 00048 . 00510 . 7554/eLife . 00048 . 006Figure 1—figure supplement 2 . Plot depicting ER-targeted mRNAs abundance [log10] ( reads per million ) versus DTT-dependent expression changes [log2] for wild type cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00048 . 00610 . 7554/eLife . 00048 . 007Figure 1—figure supplement 3 . Ontology of genes down-regulated more than twofold Ire1-and DTT-dependent . DOI: http://dx . doi . org/10 . 7554/eLife . 00048 . 007 To address the conundrum posed by the missing Ire1 splicing substrate in S . pombe , we first explored the scope of UPR-dependent changes in gene expression . To this end , we isolated polyA+ RNA from wild type and Ire1Δ cells , in which the UPR was induced with the reducing agent dithiothreitol ( DTT ) . DTT causes ER stress by impairing disulfide bond formation in the ER . The purified mRNA population was reverse-transcribed and subjected to deep-sequencing . Unexpectedly , we observed widespread Ire1-dependent mRNA down-regulation , but virtually no mRNA up-regulation ( Figure 1c ) . Thirty-nine mRNA species were reduced by more than twofold in a DTT- and Ire1-dependent manner ( Figure 1c , bottom left grayed area ) . Most members of this set of down-regulated mRNAs were abundantly expressed , as depicted by the size of the plotted circles . Down-regulation , however , did not correlate with mRNA abundance ( Figure 1—figure supplement 2 ) . Intriguingly , the set of down-regulated genes exclusively encoded proteins targeted to the ER ( identified by signal sequences and/or transmembrane segments ) ( Figure 1c , red circles ) . As shown in Figure 1d , the genome-wide profile of Ire1- and ER stress-dependent mRNA changes of genes encoding ER-targeted proteins is skewed to a significantly greater extent toward down-regulation than that of other mRNAs ( p<1×10−20 ) . More than half of the most down-regulated mRNAs encode proteins with annotated functions in the secretory pathway , in particular proteins involved in lipid metabolism , trafficking , and ER functions ( Figure 1e ) . As the reduction in mRNA abundance was ER stress- and Ire1-dependent , we next explored if Ire1 could be directly involved in destabilizing ER-bound mRNAs . To this end , we sought to trap any putative primary Ire1-cleavage products prior to degradation by deleting Ski2 , which encodes a helicase component of the cytosolic Ski complex ( cytosolic exosome ) that mediates 3′ → 5′ RNA decay . Northern blot analysis of Ski2Δ cells revealed that Gas2 mRNA ( which is down-regulated 2 . 5-fold in an ER stress and Ire1-dependent manner ) yielded two discrete cleavage products upon ER stress ( Figure 2a ) . Gas2 mRNA cleavage was dependent on Ire1 , as no mRNA reduction and no cleavage products were observed in Ire1Δ Ski2Δ double deletion cells ( Figure 2a ) . Another target , Yop1 , behaved similarly ( Figure 2—figure supplement 1 ) . In time-course experiments , reduction of Gas2 mRNA and accumulation of the cleavage products peaked at 30 min after UPR induction ( Figure 2b ) ; at later time points the abundance of intact full-length mRNA increased , suggesting that newly transcribed mRNA is not cleaved if the Ire1-dependent cleavage products are not further degraded . Indeed , Ski2Δ cells failed to grow on plates containing tunicamycin ( Figure 2c ) , indicating that an intact mRNA decay is important for S . pombe cells to cope with ER stress . 10 . 7554/eLife . 00048 . 008Figure 2 . Ire1 cleaves down-regulated mRNAs at specific sequences . ( a ) Northern blot of total RNA extracted from wild type , Ire1Δ , Ski2Δ and double mutant ER stressed Ire1Δ Ski2Δ cells ( 2 mM DTT , 1 hr ) . A probe complementary to the 5′ UTR of Gas2 was used to detect cleavage products . The triangle and asterisk indicate two different mRNA cleavage products . ( b ) Northern blot of total RNA extracted from ER-stressed Ski2Δ cells ( 2 mM DTT ) . ( c ) Viability assay by serial dilution of wild type , Ire1Δ , Ski2Δ and Ire1Δ Ski2Δ cells spotted on solid media with or without ER stress as in Figure 1b . ( d ) RNA-sequence read density map of the Gas2 locus derived from 3′ end deep-sequencing data . Library was generated by ligating a DNA-linker using tRNA ligase to 3′ end mRNAs with 2′ , 3′-cyclic phosphates in Ski2Δ and Ire1Δ Ski2Δ ER-stressed cells ( 2 mM DTT , 30 min ) . The arrows indicate two Ire1-depedent cleavage sites . The U6 snRNA locus was used as a positive control ( Figure 2—source data ) . ( e ) Ire1 RNA sequence recognition motifs generated by deep-sequencing analysis of tRNA ligase-generated RNA libraries of 39 mRNA targets down-regulated twofold or more in an Ire1-dependent manner . The resulting position weight matrices are illustrated as a logo . The dotted line indicates the cleavage site . ( f ) Real-time qPCR of a chromosomally integrated reporter containing the coding sequence of Gas2 under the control of the Nda2 ( tubulin ) promoter and including the UTRs of Nda2 . A time course after DTT addition ( 2 mM ) is shown . Endogenous Nda2 was used as a normalization control . Error bars: standard deviation . ( g ) Northern blot analysis of total RNA extracted from Ski2Δ and Ire1Δ Ski2Δ cells carrying a mutant version of the reporter indicated in ( f ) were the putative Ire1 cleavage site ( ▲ , UG\CU→ UC\UU ) was mutated . Note that the band labeled × migrates distinctly faster , as shown by scan on the right . DOI: http://dx . doi . org/10 . 7554/eLife . 00048 . 00810 . 7554/eLife . 00048 . 009Figure 2—source data 1 . 2′ , 3′ cyclic-phosphate 3′ end mappinghttp://dx . doi . org/10 . 7554/eLife . 00048 . 00910 . 7554/eLife . 00048 . 010Figure 2—figure supplement 1 . Northern blot of total RNA isolated from wild type , Ire1Δ , Ski2Δ and double mutant Ire1Δ Ski2Δ ER stressed cells ( 2 mM DTT , 1 hr ) . A probe complementary to the 5′ UTR of Yop1 was used to detect cleavage products . The arrows indicate Ire1-dependent mRNA cleavage products . DOI: http://dx . doi . org/10 . 7554/eLife . 00048 . 01010 . 7554/eLife . 00048 . 011Figure 2—figure supplement 2 . Comparison of ER stress-dependent down-regulation of endogenous Gas2 mRNA ( 2 mM DTT , 1 hr: deep-sequencing ) and reporter Gas2 mRNA ( from qPCR: 2 mM DTT , 1 hr; see also Figure 2f ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00048 . 011 To determine the mRNA cleavage sites genome-wide , we prepared total RNA fractions from Ski2Δ and from Ire1Δ Ski2Δ cells . We used tRNA ligase to attach linker sequences specifically to those RNA fragments terminating in a 2′ , 3′-cyclic phosphate , which is the expected product of Ire1-catalyzed RNA cleavage ( Schutz et al . , 2010 ) . We then amplified the cleavage products in 3′ RACE reactions priming at the linker sequence . Alignment of the sequencing data to the S . pombe genome identified the 3′ ends of Ire1-dependent fragments . In particular , we identified 39 Ire1-dependent fragments mapping to 24 of the most down-regulated genes , as shown in Figure 2d for Gas2 mRNA ( left panel ) . By size estimation , the major Ire1-dependent peak corresponded to the smaller , more abundant Gas2 mRNA cleavage product ( labeled ▲ in Figures 2a and 2b ) . A second , less abundant short fragment was also observed in the sequencing data ( labeled × in Figure 2d ) . ( Fragment × was absent or below the detection limit on the Northern blot unless the primary cleavage site was mutated ( see Figure 2g , discussed below ) . ) Spliceosomal U6 RNA normally terminates in a 2′ , 3′-cyclic phosphate and thus provided a valuable control for the ligation reaction ( Figure 2d , right panel ) . Alignment of the experimentally determined Ire1-dependent cleavage sites revealed a core motif with a signature of three conserved nucleotides ( UG\C ) that flank the Ire1-dependent cleavage sites at positions −2 , −1 , and +1 with an additional strong bias against G in position +2 ( Figure 2e ) . Most mapped mRNA cleavage sites ( 34 of 39 ) , including those in Gas2 mRNA , localized within the open reading frames . Indeed , a Gas2 reporter construct transcribed off a heterologous alpha-tubulin ( Nda2 ) promoter and containing only the Gas2 ORF flanked by heterologous 5′ and 3′ tubulin untranslated regions ( UTRs ) , was down-regulated upon ER stress in an Ire1-dependent manner ( Figure 2f ) . This degradation was quantitatively comparable to that of the native Gas2 transcript ( Figure 2—figure supplement 2 ) , indicating that the information contained within the Gas2 ORF is sufficient to confer susceptibility to Ire1-dependent cleavage . To assess the functional importance of the identified Gas2 mRNA cleavage site experimentally , we mutated the UGC-residues of the ▲-site ( UG\CU ) . As expected , ER stress-dependent cleavage of the Gas2 reporter mRNA at the mapped site was abolished ( Figure 2g ) . In its place , however , we observed two new Ire1-dependent fragments ( labeled × and * ) . Scanning gel densitometry revealed that fragment × is distinctly smaller than fragment ▲ , and hence represents a cryptic site that is only utilized when site ▲ is mutated . Fragment * likely corresponds to the lower abundance cleavage product observed in Figures 2a and 2b , which becomes more prominent in the mutant construct . Taken together , we conclude that Ire1-dependent mRNA cleavage in S . pombe is sequence dependent . The data presented so far suggest that homeostatic control of ER protein folding is regulated differently in S . pombe than S . cerevisiae . Rather than relying on a transcriptional program to upregulate genes that enhance ER protein folding capacity as in S . cerevisiae , S . pombe cells reduce the amount of specific proteins entering the organelle by decreasing the level of ER-targeted mRNAs using Ire1-dependent mRNA degradation . In all species analyzed to date , Bip1 is a major UPR target gene that is upregulated when cells experience ER stress . Paradoxically , we found S . pombe Bip1 mRNA among the 39 down-regulated mRNAs identified by the analyses shown in Figure 1c . Analysis by Northern blotting yielded seemingly conflicting results: by this analysis , Bip1 mRNA was fourfold more abundant in ER stressed cells ( Figure 3a ) . Intriguingly , the appearance of a faster migrating mRNA species ( ‘tBip1 mRNA’ ) indicates that Bip1 mRNA changes size in cells experiencing ER stress ( Figure 3a , lanes 3–4 ) . Appearance of the tBip1 mRNA species was Ire1-dependent and in wild type cells accounted for the increase in overall mRNA abundance . The increase did not result from augmented transcription . We measured the activity of a heterologous reporter in which the Bip1 promoter was fused to GFP and showed no Ire1-dependent change in mRNA abundance with ER stress ( Figure 3b ) . In agreement with this result , we found that the stability of an mRNA bearing the Bip1 ORF and 3′ UTR showed a more than threefold increase in half-life from T1/2=20 min for the unprocessed form present in unstressed cells to T1/2=70 min for the processed form present in ER-stressed cells ( Figure 3c ) . Furthermore , the Bip1 3′ UTR and the presence of a signal sequence were sufficient to a heterologous mRNA construct to confer Ire1-dependent processing ( Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 00048 . 012Figure 3 . Ire1 truncates Bip1 mRNA within the 3′ UTR . ( a ) Northern blot analysis of total RNA extracted from wild type and Ire1Δ cells untreated or treated with tunicamycin ( 1 µg/ml ) , and hybridized with a probe complementary to the ORF of Bip1 mRNA . Right panel: quantitation normalized to Pgk1 mRNA . ( b ) The abundance of a GFP mRNA driven by the Bip1 promoter ( black ) compared to endogenous Bip1 mRNA was determined as a time course after DTT ( 2 mM ) addition by quantitative Northern blotting . ( c ) Wild type cells bearing a construct encoding the Nmt1 5′ UTR , Bip1 ORF and Bip1 3′ UTR driven by the Nmt1 promoter were pre-treated with tunicamycin ( 0 . 25 µg/ml , 1 hr ) . At different time points after thiamine ( 15 µM ) addition ( to effect transcriptional shut-off of the Nmt1 promoter ) , RNA was extracted and analyzed by Northern hybridization . Blots were probed for the Nmt1 5′ UTR . Nmt1-Bip1 mRNA and Nmt-tBip1 mRNA were quantitated and normalized to the unspecific band ( asterisk ) . ( d ) RNA-sequence read density map of the Bip1 locus derived from mRNA-enriched ( ribosome depleted ) RNA in wild type cells untreated or treated with DTT ( 2 mM DTT , 1 hr; left panels ) . Data are representative of one of two biological replicates . Single nucleotide resolution of the 3′ terminus of Bip1 mRNA determined by 3′ RACE ( right panels ) . ( e ) Mutational analysis of the Bip1 mRNA cleavage site by Northern blotting . Total RNA was extracted from wild type , Ire1Δ or cells carrying mutations in the Bip1 3′ UTR mRNA . Cells were treated with 2 mM DTT , 1 hr or left untreated as indicated . The fold-changes indicate Bip1 mRNA abundance relative to that of Pgk1 mRNA . ( f ) Strand-specific , mRNA enriched ( after removal of ribosomal RNA ) deep-sequence analysis of annotated ORFs ( y-axis ) compared to strand-specific polyA+ enriched mRNA deep-sequence analysis of annotated ORFs ( x-axis ) ( see Figure 1—source data ) . The plot indicates the ratio of transcript abundance in unstressed vs DTT-stressed ( 2 mM DTT , 1 hr ) wild type cells . Symbol sizes and colors are as described in Figure 1c . DOI: http://dx . doi . org/10 . 7554/eLife . 00048 . 01210 . 7554/eLife . 00048 . 013Figure 3—figure supplement 1 . Northern blot analysis of wild type cells bearing a construct expressing a fusion protein of GFP preceded by the Bip1 signal sequence . The constructs includes the Bip1 3' UTR . Expression was driven by the Nmt1 promoter . Cells were untreated or treated with DTT ( 2 mM DTT , 1 hr ) as indicated . To abolish targeting of the mRNA to the ER , three hydrophobic leucine residues in the signal sequence were replaced by three charged arginine residues . MKKFQLFSILSYFVALFLLPMAFA ( WT ) to MKKFQRFSIRSYFVARFLLPMAFA . Silent mutations were created by changing one nucleotide in each three leucine residues ( above ) without changing the amino acid . DOI: http://dx . doi . org/10 . 7554/eLife . 00048 . 01310 . 7554/eLife . 00048 . 014Figure 3—figure supplement 2 . Sequencing read coverage of the 3′; end nucleotide positions in tBip1 mRNA from derived from mRNA enriched by subtractive hybridization against rRNA ( Ribominus kit , Invitrogen kit ) . Cells were treated with DTT ( 2 mM , 1 hr ) . DOI: http://dx . doi . org/10 . 7554/eLife . 00048 . 014 Sequencing of the expressed genome in UPR-induced and uninduced cells revealed the molecular difference between Bip1 and tBip1 mRNA ( Figure 3d ) . For these experiments , we extracted total RNA and then , without selecting for polyA+ RNA , removed rRNA by subtractive hybridization . After reverse transcription , deep-sequencing of the cDNA pool from uninduced cells revealed good coverage of reads spanning the entire Bip1 mRNA including its 5′ and 3′ UTR ( Figure 3d , left , blue profile ) . By contrast , cDNA isolated from DTT-treated cells revealed a precipitous drop in reads mapping to the 3′ end ( Figure 3d , left , red profile and Figure 3—figure supplement 2 ) . We also performed 3′-RACE to determine the 3′ end of tBip1 mRNA . The sequence of the amplified DNA confirmed that tBip1 mRNA lacks a polyA tail and terminates at G373 in the 3′ UTR ( Figure 3d , right panel ) . In seven independently isolated clones , we found no sequence variations in the tBip1 linker junction . The sequences flanking G373 align with the UG\CU motif ( Figure 2e ) , suggesting that tBip1 mRNA is produced by truncation of Bip1 mRNA in an Ire1-dependent RNA cleavage reaction that resembles those of the Ire1-dependently down-regulated mRNAs described above . Mutational analysis of the cleavage site confirmed that specific sequences are required . Mutation of G373 to C or U and its deletion together with preceding nucleotides abolished Ire1-dependent Bip1 mRNA processing ( Figure 3f ) . By contrast , a mutation of the preceding G370 to C diminished cleavage only marginally ( less than twofold ) . In all analyzed mutants of Bip1 mRNA , UPR-induction increased abundance of the transcript approximately twofold ( a level comparable to that observed in Ire1Δ cells ) ( Figure 3a , right panel ) , whether processing took place or not , perhaps due to compensatory transcriptional regulation that is independent of Ire1 . For all mutants , however , the increased abundance stayed shy of the fourfold increase observed for wild type Bip1 mRNA . As Bip1 mRNA truncation resulted in a loss of the polyA tail , this result resolves the paradox of why Bip1 mRNA appeared to be down-regulated in the polyA+ mRNA pool analyzed in Figure 1c . Indeed , directly comparing the UPR-dependent fold-change in mRNA abundance of polyA+ RNA and rRNA-depleted total RNA uniquely positioned Bip1 sequences as an anti-correlated outlier , whereas all other mRNAs were well correlated between the samples ( Figure 3e ) . From these data we conclude that , remarkably , Bip1 mRNA is the only stable mRNA in the cell that loses its polyA tail upon UPR induction . It was unexpected to find an mRNA that had lost its polyA tail to be more stable in cells . To determine the translation proficiency of tBip1 mRNA , we subjected UPR-induced cells to polysome profiling . These experiments confirmed that despite lacking its polyA tail , tBip1 mRNA sedimented in the polyribosome fractions in sucrose gradients ( Figure 4a ) . Moreover , ribosome footprinting demonstrated that Bip1 mRNA in uninduced cells and tBip1 mRNA in UPR-induced cells were engaged with actively translating ribosomes , mapping throughout the Bip1 ORF ( Figure 4b ) . The larger number of reads obtained upon UPR induction correlated with the higher abundance of tBip1 mRNA . We note a significant ribosome occupancy preceding the translation start site in both Bip1 and tBip1 mRNA most likely presenting previously unrecognized small uORFs ( see Figure 4—figure supplement 1 for a zoomed-in view ) . The relative ribosome occupancy of these putative uORFs did not change with UPR induction . Translation of the processed mRNA resulted in an enhanced steady-state concentration of Bip1 protein , as shown by quantitative Western blotting ( Figure 4—figure supplement 2 ) . 10 . 7554/eLife . 00048 . 015Figure 4 . tBip1 mRNA is translated and is important for fitness during ER stress . ( a ) Northern blot analysis of the distribution of total or tBip1 mRNA in polyribosomes from extracts of unstressed or ER-stressed ( 2 mM DTT , 1 hr ) cells . Fractions 11 , 12 , 13 of the sucrose gradients ( lower panels ) were analyzed by Northern blotting ( upper panels ) . ( b ) Ribosome footprints ( as described in Materials and methods ) of Bip1 mRNA in unstressed or ER stressed cells . The region that depicts ribosome density preceding the Bip1 ORF is shown enlarged in Figure 4—figure supplement 1 ( Figure 4—source data 1 ) . ( c ) Cell growth of wild type cells and cells carrying a deletion of Bip1 mRNA cleavage sites ( Bip1 ( Δ TTAACTGGTGC ) ) . Cells were treated with tunicamycin ( 0 . 5 µg/ml ) for 3 hr and then recovered from ER stress by washing out the drug and re-seeding in warm fresh media . Optical density ( OD ) at 660 nm was measured immediately afterward in 10 min intervals . ( d ) Viability assay of the same cells as in ( c ) . The percentage of viable cells was determined by counting the number of colony-forming units ( CFU ) after growth for 3 hr at varying tunicamycin concentration . DOI: http://dx . doi . org/10 . 7554/eLife . 00048 . 01510 . 7554/eLife . 00048 . 016Figure 4—source data 1 . Ribosome footprint reads for Bip1http://dx . doi . org/10 . 7554/eLife . 00048 . 01610 . 7554/eLife . 00048 . 017Figure 4—figure supplement 1 . Zoomed-in ribosome occupancy profile of Bip1 mRNA around the start AUG codon of wild type cells . Putative non-canonical uORFs are highlighted in both Bip1 and tBip1 mRNA derived from untreated and DTT-treated ( 2 mM , 1 hr ) cells . DOI: http://dx . doi . org/10 . 7554/eLife . 00048 . 01710 . 7554/eLife . 00048 . 018Figure 4—figure supplement 2 . Upper panel: Western blot of Bip1 and , as a loading control , RNA polymerase II CTD repeat ( RNAPII ) . Wild type and Ire1Δ cells were treated with tunicamycin 0 . 5 µg/ml and samples were taken at indicated time points . Lower panel: Quantification of Western blotting with values normalized to RNAPII . DOI: http://dx . doi . org/10 . 7554/eLife . 00048 . 01810 . 7554/eLife . 00048 . 019Figure 4—figure supplement 3 . Viability assay by serial dilution of wild type , Ire1Δ and different Bip1 mRNA cleavage mutants spotted on solid media with or without 0 . 03 µg/ml of the ER stress inducer tunicamycin . Plates were photographed after 3 day of growth at 30°C . DOI: http://dx . doi . org/10 . 7554/eLife . 00048 . 019 To assess the physiological consequences of this unique regulatory mechanism of Bip1 expression , we explored the growth of strains carrying a mutation of the Bip1 mRNA processing site ( ΔTTAACTGGTG\C ) . Liquid cultures of Bip1 mRNA mutant , that were exposed to a pulse of ER stress ( tunicamycin ) and allowed to recover after washout of the drug showed a marked growth delay in early log phase ( Figure 4c ) and enhanced cell death ( Figure 4d ) , indicating that Bip1 mRNA processing is important for maintaining cell fitness in the face of ER stress . By contrast to cell growth in liquid culture , Bip1 mutant cells grew on UPR-inducing tunicamycin plates only marginally worse that wild type cells ( Figure 4—figure supplement 3 ) . The importance of Bip1 mRNA processing , therefore , varies with growth conditions . ER stress arises from insufficiencies in handling the protein-folding load in the ER lumen . Homeostasis , therefore , can be reestablished in two principal ways: increasing the capacity of the ER to handle the load—or decreasing the load to meet the capacity . Here we show that mRNA decay can serve as the sole means of resolving ER stress without transcriptional up-regulation of classical UPR target genes . The identified transcripts targeted for RIDD compose a subset of mRNAs , all encoding proteins that reside in or traverse the secretory pathway . Being translated by membrane-bound ribosomes , these mRNAs are therefore in an appropriate cellular location to meet activated Ire1 . As all proteins encoded by RIDD target mRNAs enter the ER lumen , their synthesis by definition contributes to the burden of the ER protein folding machinery . RIDD therefore helps reduce the protein-folding load . It is less clear however whether such reduction would have a major impact . Indeed , a back-of-the-envelope calculation indicates that in S . pombe RIDD reduces the total protein influx into the ER by only 15% , even under the severe ER stress conditions explored here experimentally . This estimate derives from our ribosome footprinting data in normal vs ER-stressed cells: We scored the relative translational engagement of all mRNAs encoding proteins displaying signal sequences or transmembrane regions to estimate flux of newly synthesized polypeptides in to the ER and calculated the impact of RIDD on this set ( see Materials and methods ) . It is difficult to envision how a mere 15% reduction of bulk protein flux into the ER would suffice to alleviate an otherwise lethal ER stress . It is possible that RIDD preferentially targets proteins that are particularly difficult to fold and hence might have a disproportional impact on the protein folding load in the ER lumen . Indeed , Ire1 may be localized to the vicinity of mRNAs encoding such proteins by interactions of its ER lumenal unfolded protein sensing domain with portions of the nascent polypeptide chains that have entered the ER lumen , as previously proposed ( Hollien and Weissman , 2006; Ron , 2006 ) . An alternative and not mutually exclusive view poses that RIDD qualitatively changes the gene expression profile . In support of this notion , we notice that the population of RIDD target mRNAs is highly selective . mRNAs encoding proteins involved in lipid metabolism are highly enriched ( 31% of RIDD target mRNA as compared to 6 . 7% in all ER-targeted mRNAs ) . Moreover , we find that RIDD targets encoding proteins involved in sterol metabolism are particularly enriched ( 13% as compared to 1 . 3% ) . How reduced sterol synthesis would counteract the toxic effects of ER stress remains unclear . One possible explanation would be that ER stress limits sterol exit through vesicular transport and a compensatory reduction in sterol synthesis becomes important to sustain basic ER functions , perhaps by maintaining appropriate membrane fluidity ( Nilsson et al . , 2001; Feng et al . , 2003 ) . In this way , RIDD ( akin to other degradative pathways ) could adjust basic metabolic parameters in the cell ( Bernasconi et al . , 2012 ) . Previous work strongly suggests that Ire1 is the nuclease that initiates RIDD in metazoan cells ( Han et al . , 2009; Hollien et al . , 2009; Cross et al . , 2012 ) . Our results provide two further lines of evidence in support of this view: first , we show that an Ire1 RNase active site mutant , Ire1 ( H1018N ) , is unable to sustain cell growth on ER stress-inducing media . This mutation was designed to block catalysis while retaining hydrogen-bonding interactions of the amino acid side chain . Indeed , the equivalent single amino acid substitution in S . cerevisiae Ire1 , Ire1 ( H1080N ) , reduces catalytic activity by >105-fold . Otherwise , Ire1 ( H1080N ) is indistinguishable from wild type Ire1 , both in its oligomerization and structural properties as determined by crystallography ( Korennykh et al . , 2011 ) . Second , we showed that cleaved RIDD target mRNAs carry a 2′ , 3′-cyclic phosphate group , which is a prerequisite for the ligation reaction ( tRNA ligase ) used in the genome-wide mapping of mRNA ends created upon ER stress . Ire1 and tRNA endonuclease ( and the Ire1-family member RNase L found in mammalian cells ) are the only cytoplasmic nucleases known to produce such products . In S . cerevisiae , Ire1 has a single known substrate , Hac1 mRNA ( Niwa et al . , 2005 ) . Hac1 and XBP1 mRNA have highly conserved and readily recognizable stem loop structures that demarcate the splice sites . Cleavage occurs at a universally conserved G always found at position 3 in the seven-base loop ( Gonzalez et al . , 1999 ) . This information is interchangeable between species: constructs derived from S . cerevisiae Hac1 mRNA are properly spliced in mammalian cells , and yeast Ire1 recognizes and precisely cleaves XBP1 mRNA-derived substrates ( Niwa et al . , 1999 ) . We show that RIDD target mRNAs contain a short three-base UGC consensus at the Ire1 cleavage site where cleavage occurs after the G , consistent with cleavage specificity previously assigned to Ire1 ( Gonzalez et al . , 1999 ) . Thus by contrast to our understanding of the RNA-elements directing Ire1-cleavage that initiates Hac1 and XBP1 mRNA splicing , the information that directs mRNAs into RIDD remains vastly underspecified . By chance , UGC triplets occur much more frequently in mRNAs than Ire1 cleavage sites . Therefore , the information that specifies an mRNA as an Ire1 substrate must require additional elements . Potential determinants may lie in sequence or secondary structure determinants that to date have escaped bioinformatics identification . We and others note that many of the identified cleavage sites lie in loops of potential hairpin structures ( Hur et al . , 2012; Oikawa et al . , 2010 ) ; however , the position of the scissile G in the loops is not conserved . These structures therefore do not provide a structurally plausible explanation . Alternatively , Ire1's lumenal domain may become preferentially engaged with nascent polypeptide chains that display higher affinity and/or longer exposed peptide sequences , thereby selecting mRNAs co-translationally by recognizing features in the encoded protein . The concept of Ire1 recruitment , whether through interactions via the nascent chain or elements in the mRNA per se , is supported by our finding that mutation of one cleavage site ( UG\C → UC\U ) in Gas2 mRNA gives rise to cleavage at alternative sites . These data indicate that local proximity rather than the RNA sequence surrounding the immediate cleavage site may guide substrate selection . In order to stabilize the primary Ire1 cleavage products , we used mutant cells impaired in exosome function catalyzing 3′ → 5′ RNA degradation ( Anderson and Parker , 1998; Gatfield and Izaurralde , 2004 ) . Intriguingly , we observed that Gas2 mRNA decay in Ski2Δ mutant strains was observed only transiently , peaking at 30 min after ER stress induction . This kinetic behavior suggests that clearance of Ire1-cleavage products by Ski2 may be a requirement for continued Ire1 activity . Thus , initiation of RIDD by Ire1 and further decay may be obligatorily coupled . The observation that S . pombe Bip1 mRNA changes size upon ER stress dates back to 1992 ( Pidoux and Armstrong , 1992 ) . To date , Bip1 mRNA processing has not been observed in any other species . The phenotypic observation of the mRNA size shift has been deployed many times as an ER stress indicator ( D'Alessio et al . , 1999 ) , but , surprisingly , its origin has not been investigated . As we suggest here , Bip1 mRNA is also cleaved by Ire1 , yet escapes decay . The Ire1 cleavage site shares the same features described above for RIDD target mRNAs . During ER stress-induced processing , Bip1 mRNA loses a portion of its 3′ UTR and polyA tail . The resulting processed tBip1 mRNA is more stable and hence is present at an increased steady-state concentration . A plausible explanation for the increased stability of tBip1 mRNA is the loss of an RNA degron located in the severed portion of the 3′ UTR . tBip1 mRNA is actively translated with its ribosome density paralleling its increased abundance . Although polyA tails are generally linked to stability and translational efficiency , histone mRNAs , which likewise lack polyA tails , provide precedence for such an exception to the rule ( Marzluff et al . , 2008 ) . Histone mRNAs terminate in a well-conserved 3′ stem-loop structure , which protects from exonucleolytic degradation . Proteins binding there functions akin to the polyA binding proteins found on other mRNAs to enhance histone mRNA translation by looping back to the 5′ cap structure ( Sanchez and Marzluff , 2002 ) . Why tBip1 mRNA escapes decay remains to be explored . Possible explanations include the presence of secondary structure elements . Indeed , we find a predicted conserved hairpin structure at the 3′ termini of tBip1 RNAs in some fission yeasts ( S . pombe , S . octosporus and S . cryophilus ) ; however preliminary mutational analysis failed to validate its importance for mRNA stability in S . pombe . An alternative possibility is that the 3′ end of tBip1 mRNA may be covalently modified . Such modification would need to be restricted to the 2′-OH , because the 3′-OH group is still accessible for modification by RNA ligase ( Figure 3d ) . In this regard , tBip1 mRNA would resemble miRNAs , which are 2′-O-methylated , conferring resistance to degradation ( Yu et al . , 2005 ) . Our data show that Bip1 mRNA is the only mRNA in S . pombe in the expressed genome that is subject to this unique regulation . Why would such a singular mechanism have evolved exclusively for Bip1 ? From work in other species , Bip1 emerges as the most pleiotropically important and precisely controlled ER chaperone ( Gulow et al . , 2002 ) . Moreover , Bip1 holds a unique position in S . pombe , where it is glycosylated ( Pidoux and Armstrong , 1993 ) . A recent comprehensive gene interaction map revealed that Ire1 in S . pombe clusters tightly with enzymes involved in the quality control cycle of glycosylated proteins , pointing toward a unique connection between glycosylation and ER stress ( Frost et al . , 2012 ) . By contrast , corresponding E-maps in S . cerevisiae succinctly confirm the long-appreciated linear relationship between Ire1 and Hac1 ( Schuldiner et al . , 2005 ) . One may speculate that ER stress in S . pombe , as in other species , enhances turnover of glycosylated proteins and that the regulation of Bip1 mRNA is beneficial to its stability by compensating for such loss ( Bernasconi et al . , 2012 ) . From an evolutionary angle , the UPR in S . pombe provides an intriguing example of how molecular machines can be repurposed . While input ( unfolded proteins , ER stress ) and output ( RNA cleavage ) have been conserved , both detail and global consequences of downstream processes have been adapted to serve different needs . The UPR in both S . cerevisae and S . pombe fulfills a cytoprotective role , yet the mechanisms of executing this task are opposed . In S . cerevisiae , folding capacity is increased via transcriptional up-regulation; in S . pombe the folding load is decreased and the ER is restructured . In metazoan cells , both modes of Ire1 activity are merged , and depending on condition and cell type can serve different purposes . RIDD can protect cells by removing major secretory protein loads , as it is the case in insulin secreting cells ( Lipson et al . , 2008 ) , or it can serve to activate apopototic pathways , as it is the case in cells experiencing prolonged and unmitigated ER stress ( Han et al . , 2009 ) . It has always been puzzling how a strictly cognate system , such as the Ire1- and Hac1-mediated UPR regulatory pathway , would have evolved . While we cannot ascertain what represents the ancestral state , it is tempting to speculate that a broader mRNA degradation pathway preceded the development of the more specialized splicing mechanism . The prevalence of a much broader scope of Ire1 targets in S . pombe suggests that a primitive UPR may have served primarily as an ER-localized yet promiscuous RNA degradation system . Individual mRNA substrates would then have evolved appropriate affinities for the enzyme , rendering them more or less susceptible substrates . In this way , the stem/loop splice sites of HAC1/XBP1 mRNAs could be the result of a long time optimization process: duplication of the cleavage site with concomitant recruitment and repurposing of tRNA ligase would have culminated the UPR splicing reaction . In this view , S . cerevisiae emerges as the endpoint of an optimization process rather than an evolutionary precursor . By losing the more ancient RIDD function of the UPR , S . cerevisiae cell would have developed to rely exclusively on a more refined and more powerful transcriptional regulation program . The unified convention used in this manuscript for genes and proteins is based on the treatment by Alberts , et al . in Molecular Biology of the Cell ( 5th edition , Garland Publishing ) , page xxxii . In brief , all genes are denoted in italics with a capitalized first letter . Mutant alleles are indicated by appending a descriptor to the gene name . Proteins are indicated in Roman letters . Standard media and genome engineering methods for fission yeast were used as described previously ( Moreno et al . , 1991 ) . Strains and plasmids used in this study are listed in Table 1 . Briefly , reporter constructs were integrated into the Leu1 locus using plasmid pJK148 . Mutant alleles were integrated by the pop-in/pop-out method using the integrative plasmid pJK210 ( ura4 ) ( Guthrie , 2004 ) . All experiments were carried out in yeast extract complex media ( YE5S ) or in Edingburgh minimal media ( EMM ) , supplemented with 0 . 225 mg/ml of l-histidine , l-leucine , l-lysine , adenine and uracil at 30°C , unless otherwise described . For pop-in/pop-out experiments , 5-FOA media containing 1 g/l 5-fluoro-orotic acid was used . 10 . 7554/eLife . 00048 . 020Table 1 . Strains and plasmids used in this studyDOI: http://dx . doi . org/10 . 7554/eLife . 00048 . 020StrainDescription yPK001WTyPK002Ire1ΔyPK003Ire1 ( H1018N ) yPK004Ski2ΔyPK005Ski2Δ Ire1ΔyPK006Bip1 ( CTCGTG\C ) yPK007Bip1 ( CTGGTC\C ) yPK008Bip1 ( CTGGTT\C ) yPK009Bip1 ( ΔACTGGTG\C ) yPK010Bip1 ( ΔTTAACTGGTG\C ) PlasmidDescriptionMarkerpPK001pJK148::pNda2::Gas2::Nda2 3′UTRLeu1pPK002pJK148::pBip1::GFP::Nmt1 3′UTRLeu1pPK003pJK148::pNmt1+5′UTR::Bip1 ( ORF ) ::Bip1 3′UTRLeu1pPK004pJK148::pNmt1::ss ( Bip1 ) ::GFP::Bip1 3′UTRLeu1 Total RNA was purified by standard hot-phenol extraction ( Kohrer and Domdey , 1991 ) . After precipitation RNA samples were re-suspended in DEPC-treated water and quantified by spectrophotometry . Northern blotting , electrophoresis , labeling , analysis and quantification were performed as described ( Ruegsegger et al . , 2001 ) . Cells were cultured in YE5S media . Between 5 and 10 OD units were collected by centrifugation and snap-frozen . Cell pellets were thawed on ice , re-suspended in 200 µl of lysis buffer ( 8 M urea , 50 mM HEPES pH 7 . 4 ) and lysed in a glass bead mill ( 5 min at 4°C ) . After adding 20 µl of a 25% SDS solution , samples were incubated at 65°C for 5 min . The lysates were collected by piercing the bottom of the tubes with a syringe needle and clarified by re-centrifugation ( 1000 rpm for 10 s ) . Total protein concentration was determined by a standard bichromic acid ( BCA ) assay ( manual instruction , Pierce Biotechnology ) . 10 µg of lysate per lane were electrophoresed on SDS-polyacrylamide gradient gels ( 4–15% , BioRad ) . The separated proteins were subsequently transferred to PDVF membranes at 200 mA for 1 hr . Blots were blocked with 5% milk in Tris-buffered saline ( 10 mM Tris , 150 mM NaCl , 0 . 1% Tween-20 ) and incubated with primary antibodies overnight at 4°C . Antibodies and dilutions: rabbit polyclonal anti-Kar2 ( 1:5000 ) , mouse monoclonal anti-RNA polymerase II carboxy-terminal domain ( CTD ) repeat ( Abcam ab817 ) ( 1:8000 dilution ) . Blots were washed and incubated with Li-Cor fluorescently-coupled secondary antibodies ( 1:5000 ) for 30 min . Immunoreactive bands were identified using a Li-Cor infrared scanner , and processed with the Odyssey software package . Total RNA ( 10 µg ) from treated or untreated ( 2 mM DTT , 1 hr ) WT cells were incubated with 25 units of polynucleotide kinase ( PNK ) to remove 2′ , 3′-cyclic phosphates for 1 hr at 37°C . After deactivating the enzyme ( 75°C for 10 min ) , dephosphorylated RNA was precipitated and re-suspended in 6 µl DEPC-treated water . After denaturing the RNA at 80°C for 5 min , a linker-ligation reaction was performed in the presence of dephosporylated , denatured RNA , T4RNA Ligase II ( NEB ) , RNase inhibitor ( 40 Units ) , DMSO , 50% PEG and 1 µg of 5′ pre-adenylated linker ( 5′AppCTGTAGGCACCATCAAT/3ddC3′ , as described ( Ingolia et al . , 2011 ) . Reverse transcription was performed using a reverse complement DNA-oligonucleotide to the linker sequence . The 3′ RACE reaction was performed as described ( Scotto-Lavino et al . , 2006 ) . Nested Bip1-PCR products were purified from ethidium-agarose gels and sequenced . Total RNA ( 1 µg ) was reverse-transcribed with random hexamers . 1% of the resulting cDNAs was employed for real-time PCR reactions utilizing SYBR green . The reactions were run and analyzed in a DNA Engine OPTICON 2 using the BioRad Opticon Monitor 3 . 0 software . WT and Bip1 ( ΔTTAACTGGTGC ) mutant cells were grown overnight in ( YE5S ) media . The next morning cultures were diluted and grown until they reached an OD of 0 . 25 . ER-stress was induced with tunicamycin as indicated . After 3 hr , stressed cells were washed four times with pre-warmed YES5 media to remove the drug , the culture was readjusted to an OD of 0 . 25 , and incubated in a 24 well plate ( 1 ml per well ) over 32 hr at 32°C . The OD was measured every 10 min in a microplate reader ( Synergy 4 BioTeK ) . The colony forming unit assay was performed by plating washed cells at different dilutions on solid media ( YE5S ) and incubated for 3 days at 30°C . Colonies were counted from the dilutions series . Untreated cells served as a control . To determine the half-life of Bip1 mRNA , we constructed an integrative plasmid ( pJK148 ) containing the open reading frame and 3′ UTR of Bip1 mRNA under the control of a thiamine-regulated Nmt1 promoter . The construct included the Nmt1 5′ UTR . Cells were grown in EMM complete media ( without any thiamine ) for 24 hr . Cells were then diluted into fresh EMM complete media and re-grown to an OD of 0 . 3 . After inducing ER stress with tunicamycin ( 0 . 25 µg/ml ) for 1 hr , thiamine was added to 15 µM to block transcription of the Nmt1 promoter ( Maundrell , 1990 ) . Samples were processed at indicated time points and subjected to Northern blot analysis . A DNA probe complementary to the 5′ UTR of nmt1 was used for detection . Figure 1—source data 1:oligo-dT-enriched mRNA set I ( including: WT , WT +DTT ( 2 mM , 1 hr ) and Ire1∆+DTT ( 2 mM , 1 hr ) ) oligo-dT-enriched mRNA set II ( including: WT , WT +DTT ( 2 mM , 1 hr ) and ( 2 mM , 1 hr ) ) total ( -depleted rRNA ) RNA set I ( including: WT , WT +DTT ( 2 mM , 1 hr ) total ( -depleted rRNA ) RNA set II ( including: WT , WT +DTT ( 2 mM , 1 hr ) and Ire1∆+DTT ( 2 mM , 1 hr ) ) Figure 2—source data 1:2′ , 3′-cyclic phosphate 3′end mapping setI ( Ski2∆ ( 2 mM DTT , 30 min ) Ire1∆ Ski2∆ ( 2 mM DTT , 30 min ) ) As indicated , one of three methods was utilized to isolate RNA with specific chemical properties: polyA+ tail enrichment , rRNA depletion , or 3′ end 2′ , 3′-cyclic phosphate enrichment . Total RNA was prepared from cells by the hot acid phenol method ( Kohrer and Domdey , 1991 ) , and subsequent enrichment performed . PolyA+ mRNA was purified by two sequential rounds of enrichment using oligo-dT DynaBeads ( Invitrogen ) according to the manufacturer's instructions . rRNA depletion was performed by first depleting all abundant RNAs smaller than 200 nt using the modified protocol for isolating only large RNAs provided in the mirVana miRNA Purification Kit ( Ambion ) followed by two rounds of subtractive hybridization using the Ribominus Eukaryote Kit for RNA-Seq ( Invitrogen ) , according to the manufacturer's instructions . To sequence 2′ , 3′-cyclic-phosphate cleavage products purified tRNA ligase ( a kind gift from J . R . Hesselberth ) was used to selectively ligate an RNA linker to all 2′ , 3′-cyclic phosphates in total RNA as previously described ( Schutz et al . , 2010 ) . PolyA+-enriched and rRNA-depleted samples were randomly fragmented under basic conditions , precipitated by standard methods , and ∼50 nt fragments were size-selected by polyacrylamide gel electrophoresis as described in Ingolia et al . ( 2009 ) . Ribosome footprints were isolated as described in Ingolia et al . ( 2009 ) , with minor modifications . Briefly , 750 ml mid-log yeast cultures ( ± 30 min 2 mM DTT treatment ) were harvested by filtration without the addition of cycloheximide , and were immediately flash frozen . Frozen cells were cryogenically lysed in the presence of 3 ml of frozen polysome lysis buffer ( 20 mM Tris pH 8 . 0 , 140 mM KCl , 1 . 5 mM MgCl2 , 100 μg/ml cycloheximide , 1% Triton ) on a Retsch MM301 mixer mill , and thawed lysates were subsequently clarified by centrifugation as described . ∼50 A260 units of clarified lysate was treated with 750 U of E . coli RNase I ( Ambion ) for 1 hr on ice to minimize 80S degradation . Monosomes were collected from sucrose density gradients ( 10–50% wt/vol , prepared in polysome lysis buffer: 20 mM Tris pH 8 . 0 , 140 mM KCl , 5 mM MgCl2 , 100 μg/ml , cycloheximide , 0 . 5 mM DTT , 20 U/ml SUPERase·In ) as described , and undigested control samples were loaded to generate polysome profiles shown in Figure 4b . RNA from monosome or polysome fractions was isolated using the hot acid phenol method . For ribosome profiling , 28–34 nt RNA fragments from the monosome fraction were size-selected by gel electrophoresis as described above . Deep sequencing libraries were constructed as described in Ingolia et al . ( 2011 ) . Briefly , size-selected mRNA ( polyA+ and -rRNA ) or ribosome footprints were 3′-dephosphorylated with T4 polynucleotide kinase ( NEB ) . 3′-dephosphorylated RNA was ligated to a preadenylated miRNA cloning linker ( IDT , Linker 1 ) using T4 RNl2 ( tr ) ( NEB ) rather than enzymatically polyadenylated . Subtractive hybridization of rRNA contaminants was not performed . Ligated samples were directly reverse transcribed using SuperScriptIII ( Invitrogen ) , circularized using CircLigase ( Epicenter ) , and PCR amplified . Data from deep-sequencing analyses of total ( -ribosomal RNA ) RNA , polyA+ enriched mRNA and ribosome foot-printing were collected and the resulting sequences were aligned using the following method: the linker sequences at the 3′ ends were removed prior to alignment using SOAP2 . 20 , allowing a maximum of two total mismatches ( Li et al . , 2009 ) . Ribosome footprint reads were assigned to a specific A-site nucleotide by an offset of +15 from 5′ end of the read ( only for ribosome foot-printing data set ) . All reads aligned to rRNA and tRNA were removed . To align intron-exon junction reads , all reads with no alignment against S . pombe genomics sequences were re-aligned against a sequence library of S . pombe processed protein-coding transcripts . All the alignments were performed against the most recent version of the S . pombe genome ( the www . genedb . org/genedb/pombe/ ) . The raw sequencing data will be available for download at NCBI GEO . Biological replicates ( set I and set II , where available ) were combined to increase read coverage . After combining sets , mRNA ( ORF ) transcripts with fewer than 100 reads before normalization were excluded . Next , the passed mRNAs ( ORF ) were normalized to reads per million total reads ( rpM ) . Because the total number of reads may not reflect the total RNA production correctly ( Robinson and Oshlack , 2010 ) , the observed count for gene g in condition k need to be normalized to the total RNA production to calculate the fold-change between two conditions . We used the following method to estimate the correct RNA abundance for given RPM values . Given conditions k and r , we calculated the expected RNA abundance of condition k given the RPM value in condition r with the following equation: xk , r , g=2 ( ak , rlog2 ( xr , g ) +bk , r ) . xr , g is the RPM value for gene g in the reference condition , xk , r , g is expected RPM estimated with linear regression between the RPM values from condition k and r . ak , r and bk , r are the coefficients of the linear regression where ak , r=∑ ( Yk , g−Y¯k ) ( Yr , g−Y¯r ) ∑ ( Yr , g−Y¯r ) ( Yr , g−Y¯r ) , bk , r=Y¯k−ak , rY¯r in which Yk , g=log2 ( xk , g ) , Yr , g=log2 ( xr , g ) . For the above linear regression , genes with Mg=xk , gxr , g≥10 or≥0 . 1 were removed from the data set to estimated ak , r and bk , r . The fold change between condition k and r is then calculated as: Fg , k , r=xk , gxk , r , g , which is the RPM value for gene g under condition k divided by the expected RPM estimated above . Sequence read alignments from the 2′ , 3′-cyclic phosphate mapping data were performed as for the RNA-Seq reads described above . The design of the library ( 3′ end mapping ) made it necessary to align the reads to the opposed strand of the gene . To map putative Ire1-dependent cleavage sites with high stringency , we identified the positions within each transcript containing more than four reads in RNA derived from the Ski2∆ sample and zero reads in RNA derived from the Ire1∆ Ski2∆ sample . We identified by this method 4027 putative Ire1-dependent cleavage sites in the genome . By using a less stringent criterion ( by taken the ration between Ski2∆ sample and Ire1∆ Ski2∆ sample: we allowed a ratio of 5 or more ) . By this criterion we identified 4134 putative Ire1-dependent cleavage sites . The overlap between the two methods is 97% . We continued our analysis with the more stringent criteria of only allowing zero reads in Ire1∆ Ski2∆ sample . To identify an overrepresented consensus sequence , we first extended the sequences 9 nt upstream and downstream of the potential cleavage sites . Furthermore , we used a position weight matrix ( PWM ) , which was generated from these sequences by weighting each sequence with the reads from the Ski2∆ sample . By using all annotated genes and the corresponding putative cleavage sites , we could not identify overrepresented sequences . The same held true for the ER-target mRNAs set ( N=1014 ) . By contrast , a strong consensus motif ( Figure 2e ) emerged when we mapped the putative cleavage sites in our set of 39 Ire1- and ER stress-dependently twofold down-regulated transcripts . We scored ribosome footprint reads per kilobase ( to normalize for length of open reading frames ) per million reads ( to compare different conditions ) for the set of mRNAs , encoding proteins predicted to enter the ER as described above ( see Table 2 ) . 10 . 7554/eLife . 00048 . 021Table 2 . Ribosomes footprints of mRNAs encoding protein predicted to enter the ERDOI: http://dx . doi . org/10 . 7554/eLife . 00048 . 021All ER-targeted mRNAs ribosome occupancies ( N=1014 mRNAs ) rpkmRation normalized to WT-DTT%WT ( -DTT ) 83 , 5401100WT ( +DTT ) 71 , 9960 . 861814786 . 18146995Ire1∆ ( +DTT ) 90 , 1661 . 079315298107 . 9315298rpkm: reads per kilo-base of transcript per million reads .	Protein folding—the process by which a sequence of amino acids adopts the precise shape that is needed to perform a specific biological function—is one of the most important processes in all of biology . Any sequence of amino acids has the potential to fold into a large number of different shapes , and misfolded proteins can lead to toxicity and other problems . For example , all cells rely on signaling proteins in the membranes that enclose them to monitor their environment so that they can adapt to changing conditions and , in multicellular organisms , communicate with neighboring cells: without properly folded signaling proteins , chaos would ensue . Moreover , many diseases—including diabetes , cancer , viral infection and neurodegenerative disease—have been linked to protein folding processes . It is not surprising , therefore , that cells have evolved elaborate mechanisms to exert exquisite quality control over protein folding . One of these mechanisms , called the unfolded protein response ( UPR ) , operates in a compartment within the cell known as the endoplasmic reticulum ( ER ) . The ER is a labyrinthine network of tubes and sacs within all eukaryotic cells , and most proteins destined for the cell surface or outside the cell adopt their properly folded shapes within this compartment . If the ER does not have enough capacity to fold all of the proteins that are delivered there , the UPR switches on to increase the protein folding capacity , to expand the surface area and volume of the compartment , and to degrade misfolded proteins . If the UPR cannot adequately adjust the folding capacity of the ER to meet the demands of the cell , the UPR triggers a program that kills the cell to prevent putting the whole organism at risk . Researchers have identified the cellular components that monitor the protein folding conditions inside the ER . All eukaryotic cells , from unicellular yeasts to mammalian cells , contain a highly conserved protein-folding sensor called Ire1 . In all species analyzed to date , Ire1 is known to activate the UPR through an messenger RNA ( mRNA ) splicing mechanism . This splicing event provides the switch that drives a gene expression program in which the production of ER components is increased to boost the protein folding capacity of the compartment . Kimmig , Diaz et al . now report the first instance of an organism in which the UPR does not involve mRNA splicing or the initiation of a gene expression program . Rather , the yeast Schizosaccharomyces pombe utilizes Ire1 to an entirely different end . The authors find that the activation of Ire1 in S . pombe leads to the selective decay of a specific class of mRNAs that all encode proteins entering the ER . Thus , rather than increasing the protein folding capacity of the ER when faced with an increased protein folding load , S . pombe cells correct the imbalance by decreasing the load . The authors also show that a lone mRNA—the mRNA that encodes the molecular chaperone BiP , which is one of the major protein-folding components in the ER—uniquely escapes this decay . Rather than being degraded , Ire1 truncates BiP mRNA and renders it more stable . By studying the UPR in a divergent organism , the authors shed new light on the evolution of a universally important process and illustrate how conserved machinery has been repurposed .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2012	The unfolded protein response in fission yeast modulates stability of select mRNAs to maintain protein homeostasis
Precise patterning of dendritic fields is essential for the formation and function of neuronal circuits . During development , dendrites acquire their morphology by exuberant branching . How neurons cope with the increased load of protein production required for this rapid growth is poorly understood . Here we show that the physiological unfolded protein response ( UPR ) is induced in the highly branched Caenorhabditis elegans sensory neuron PVD during dendrite morphogenesis . Perturbation of the IRE1 arm of the UPR pathway causes loss of dendritic branches , a phenotype that can be rescued by overexpression of the ER chaperone HSP-4 ( a homolog of mammalian BiP/ grp78 ) . Surprisingly , a single transmembrane leucine-rich repeat protein , DMA-1 , plays a major role in the induction of the UPR and the dendritic phenotype in the UPR mutants . These findings reveal a significant role for the physiological UPR in the maintenance of ER homeostasis during morphogenesis of large dendritic arbors . The organization of dendritic arbors is fundamental to the shape and connectivity of the nervous system ( Ramón y Cajal , 1911; Wassle et al . , 1981 ) . Complex and type specific dendritic arbors are pivotal for many neurons to receive appropriate inputs from their receptive fields and to function properly in a neural circuit ( MacNeil and Masland , 1998 ) . During development , dendrites acquire their morphology by precisely regulated branch morphogenesis , which requires extracellular interactions and intracellular signaling pathways ( Jan and Jan , 2010 ) . For example , several diffusive or cell-surface molecules play instructive roles in guiding the growth and patterning of dendritic arbors . The diffusible chemoattractant Semaphorin 3A instructs the dendritic extension of cortical pyramidal neurons toward the pial surface ( Polleux et al . , 2000 ) while the graded expression of transmembrane Semaphorin 1A regulates the precise targeting of the dendrites of projection neurons in the Drosophila olfactory system ( Komiyama et al . , 2007 ) . In the mammalian retina , a number of neuronal homotypic adhesion molecules , including Sdk1 , Sdk2 and Cntn2 , restrict dendritic arbors of amacrine cells and retinal ganglion cells in specific sublaminae in the inner plexiform layer ( Yamagata and Sanes , 2008 , 2012; Sanes and Zipursky , 2010 ) . Moreover , one common feature for dendrite development is that the sister branches from the same neuron avoid each other , while coexist with the branches of their neighboring neurons . This self-avoidance phenomenon has been elegantly elucidated by the function of two classes of highly diversified , contact-mediated repulsive molecules: Down syndrome cell adhesion molecules in Drosophila and protocadherins in vertebrates ( Schmucker et al . , 2000; Wojtowicz et al . , 2004; Matthews et al . , 2007; Lefebvre et al . , 2012 ) . These extrinsic cues must trigger intracellular signaling transduction that leads to cytoskeletal rearrangement as well as membrane biogenesis and trafficking ( Hanus and Ehlers , 2008 ) . For example , early endosome small G-protein RAB5 facilitates dendrite branching in Drosophila class IV da neurons ( Satoh et al . , 2008 ) . Large cells with highly branched dendrites such as Purkinje cells accommodate the biosynthesis demand with a large soma containing extensive Golgi apparatus and abundant mitochondria ( Herndon , 1963 ) . Molecularly , the secretory pathway components including Sec23 , Sar1 , and Rab1 are particularly required for dendrite growth compared with axon development in the highly branched mammalian and Drosophila neurons ( Ye et al . , 2007 ) . As part of the biosynthetic pathway , the production of membrane proteins requires protein folding in the endoplasmic reticulum ( ER ) . It is currently unclear whether protein folding pathways play a role in the increased protein production required for dendrite development . In the ER , a highly conserved protein quality control pathway , the unfolded protein response ( UPR ) , maintains the ER homeostasis by adjusting the ER folding capacity upon detection of unfolded proteins ( Schroder and Kaufman , 2005; Ron and Walter , 2007; Walter and Ron , 2011; Worby and Dixon , 2014 ) . In higher eukaryotes , three proteins sense the ER stress and activate the UPR: the protein kinase ( PKR ) -like ER kinase ( PERK ) , the activating transcription factor 6 ( ATF6 ) and the inositol-requiring enzyme 1 ( IRE1 ) . Conserved in all eukaryotes , IRE1 contains an ER luminal domain , which is involved in the recognition of misfolded proteins in the ER , and cytoplasmic kinase and endoribonuclease domains , which can activate downstream pathways ( Credle et al . , 2005; Gardner and Walter , 2011 ) ( Figure 1—figure supplement 1A ) . Activated IRE1 mediates the non-conventional splicing of an intron from the X box binding protein 1 ( XBP1/HAC1 ) mRNA ( Cox and Walter , 1996 ) , and the IRE1-spliced XBP1 acts as a transcription factor to up-regulate the expression of ER chaperones such as BiP and other target genes to relieve the ER stress ( Travers et al . , 2000; Lee et al . , 2003 ) . In the nematode Caenorhabditis elegans , the multidendritic polymodal nociceptive neuron PVD has an elaborate and organized dendritic arbor ( Figure 1A , B ) . PVD's largely orthogonally arranged secondary ( 2° ) , tertiary ( 3° ) and quaternary ( 4° ) branches form repeated structural units resembling menorahs ( Smith et al . , 2012 ) . During development , PVD grows its entire dendrite arbor that spans 800 μm along the body of the animal in just 24 hr ( Smith et al . , 2010 ) , suggestive of a high level of biosynthesis during the growth phase of this cell . The formation of PVD dendrites requires a single transmembrane leucine rich repeat ( LRR ) protein DMA-1 , which acts cell autonomously in PVD to promote dendrite branching and stabilization ( Liu and Shen , 2012 ) . The elaborate dendritic branch pattern is instructed by hypodermal derived ligands SAX-7/L1CAM and MNR-1 . Subcellularly localized stripes of SAX-7/L1CAM , together with MNR-1 form a tripartite receptor–ligand complex and guide the growth and branching of the PVD dendritic arbor ( Dong et al . , 2013; Salzberg et al . , 2013 ) . 10 . 7554/eLife . 06963 . 003Figure 1 . ire-1 is required for PVD dendritic morphogenesis . ( A ) Cartoon showing the PVD dendritic arbor . The dash-boxed region is magnified to show the PVD soma ( asterisk ) , axon , primary dendrite ( 1° ) , secondary dendrite ( 2° ) , tertiary dendrite ( 3° ) and quaternary dendrite ( 4° ) . ( B ) Representative wild type ( WT ) dendritic morphology of PVD neuron expressing membrane associated mCherry ( wyIs581 ) . Starting from the cell body , the anterior and the posterior sections of the primary dendrite are divided into 8 and 4 equal segments , respectively , indicated by dashed lines . Anterior , left; dorsal , top . Asterisk , cell body; arrowhead , quaternary ( 4° ) dendrite . Scale bars , 50 μm . ( C ) Quantification of the number of quaternary ( 4° ) branches in each segment in WT . The position of cell body is indicated by the black line . Error bars show mean ± s . e . m . , n = 10 . ( D to G ) Defective PVD dendritic morphogenesis in ire-1 ( ok799 ) mutants ( D and E ) is rescued by expressing ire-1 cell-autonomously ( F and G ) . ( H and I ) Representative dendritic morphology of FLP neurons labeled by cytoplasmic GFP in wild-type ( H ) and ire-1 ( ok799 ) mutants ( I ) . Asterisks , cell bodies; arrows , secondary branches; arrowheads , tertiary branches . Scale bar , 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 00310 . 7554/eLife . 06963 . 004Figure 1—figure supplement 1 . Schematic diagrams of IRE-1 dependent UPR pathway and of C . elegans IRE-1 protein showing three mutations . ( A ) Cartoon showing the IRE-1 dependent UPR pathway . Two missense mutations in wy762 and wy782 are shown by asterisks ( B ) IRE-1 contains a luminal domain ( blue ) , trans-membrane ( TM ) domain ( black line ) , a linker ( yellow ) , a S/T kinase domain ( red ) and an endoribonuclease domain ( green ) . Two missense mutations ( wy762 and wy782 ) are shown by asterisks and the deleted fragment in the null mutation ok799 is illustrated by the gray bar . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 00410 . 7554/eLife . 06963 . 005Figure 1—figure supplement 2 . The ire-1 mutants showing PVD dendritic morphogenesis defect . Representative dendritic morphology of PVD neuron expressing membrane associated GFP ( wyIs378 ) in two ire-1 alleles wy762 ( A ) and wy782 ( B ) isolated from forward genetic screen . Asterisks , cell bodies; scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 00510 . 7554/eLife . 06963 . 006Figure 1—figure supplement 3 . Compared with wild type animal ( A ) , expressing ire-1 cDNA ( B ) or spliced xbp-1 cDNA ( C ) did not cause overbranching of PVD in wild type animals . Asterisks , cell bodies . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 006 Using two ire-1 mutant alleles that we isolated from a dendrite morphology screen , we reveal that the physiological UPR is induced and required in the PVD neuron during dendrite morphogenesis . The IRE-1/XBP-1/BiP molecular cascade of the UPR pathway governs dendritic branching by regulating the folding and processing of DMA-1 . Surprisingly , our evidence indicates that among many cell surface molecules required for dendrite formation , DMA-1 is largely responsible for the induction of the UPR . We visualized the PVD neurons using a membrane associated mCherry or GFP marker expressed under the control of a cell-specific promoter ( ser2prom3::myr-mCherry or ser2prom3::myr-GFP ) . From a forward genetic screen for mutations that alter the PVD dendritic morphology , we identified two loss-of-function mutations , wy762 and wy782 . Both alleles cause dramatic loss of dendritic branches , especially in the distal dendrites of PVD ( Figure 1—figure supplement 2 ) . Mapping and cloning of these two alleles showed that each allele contains a single point mutation in the ire-1 ( inositol-requiring 1 protein kinase ) gene . In addition , a known null deletion allele of ire-1 , ok799 ( Henis-Korenblit et al . , 2010 ) showed indistinguishable phenotype in PVD compared with that of wy762 and wy782 ( Figure 1D ) . The complexity of the menorahs nearest to the cell body appeared unaffected in these mutants ( Figure 1E ) , as did the morphology of PVD axon ( data not shown ) . Interestingly , in the entire nervous system of C . elegans , the only other pair of highly branched neurons in the head region , FLP also showed severe dendritic arbor defects in ire-1 mutants ( Figure 1H , I ) . Other neurons with fewer dendritic or axonal branches such as IL2 , VC and ADL did not show branching defects in ire-1 mutants ( data not shown ) . Together , these results suggest that ire-1 is required for establishing highly branched dendrites . To investigate where IRE-1 functions to regulate dendritic development , we generated transgenic mosaic animals . In the ire-1 mutant background , expression of IRE-1 with a PVD-specific promoter ( ser2prom3 ) fully restored the distal branch number and complexity of the whole dendritic arbor ( Figure 1F , G ) indicating that IRE-1 functions cell-autonomously in PVD to regulate dendrite morphogenesis . Expressing ire-1 cDNA did not cause overbranching in wild-type animals ( Figure 1—figure supplement 3B ) . IRE1 is conserved in all eukaryotes and contains an ER luminal domain for recognizing misfolded proteins in the ER , and a cytoplasmic kinase and an endoribonuclease domain , which lead to the non-conventional cytoplasmic splicing of xbp-1 ( Figure 1—figure supplement 1A ) . One missense mutation ( wy782 ) of ire-1 is a substitution of a conserved residue in the kinase domain while another missense mutation ( wy762 ) is a substitution of a conserved residue in the endoribonuclease domain ( Figure 1—figure supplement 1 ) , indicating both domains might be required for dendrite morphogenesis . Since these two domains are required for splicing of the xbp-1 mRNA , we reasoned that the neurons should be able to bypass the requirement of IRE-1 if a spliced form xbp-1 was provided in PVD . Consistent with this hypothesis , PVD-specific expression of spliced xbp-1 cDNA in ire-1 mutants rescued the loss of distal dendrite branches phenotype . In contrast , expression of unspliced xbp-1 genomic DNA at the same concentration did not rescue branching defect ( Figure 2A–D ) . Expressing spliced xbp-1 cDNA did not cause overbranching in wild type animals ( Figure 1—figure supplement 3C ) . These data offer compelling evidence that XBP-1 functions downstream of IRE-1 to establish complex dendritic arbor in PVD . Hence , the IRE-1 arm of the UPR pathway is likely involved in dendrite morphogenesis . 10 . 7554/eLife . 06963 . 007Figure 2 . The UPR is required for PVD dendritic morphogenesis . Expressing xbp-1 cDNA in PVD rescues the defective dendritic morphogenesis in ire-1 mutants ( A and B ) while expressing xbp-1 genomic DNA in PVD does not ( C and D ) . ( E and F ) Expressing ER chaperone HSP-4 in PVD rescues the dendritic defect in ire-1 mutants . Anterior , left; dorsal , top . Asterisks , cell bodies . Scale bar , 50 μm . Error bars show mean ± s . e . m . , n = 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 00710 . 7554/eLife . 06963 . 008Figure 2—figure supplement 1 . Dendritic morphogenesis defects of PVD is likely due to lack of specific ER chaperones . Defective dendritic branching in ire-1 mutants ( A ) cannot be rescued by overexpressing another ER chaperone HSP-3 ( B ) or a cytoplasmic chaperone DAF-21 in PVD ( C ) . No detectable PVD branching defect in hsp-4 ( gk514 ) single mutant ( D ) . Asterisks , cell bodies . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 00810 . 7554/eLife . 06963 . 009Figure 2—figure supplement 2 . The RIDD pathway in parallel with XBP-1 to regulate PVD dendritic arbor development . No PVD dendritic arbor defects are shown in xbp-1 null mutant ( B ) compared to ire-1 mutant ( A ) while in xbp-1 background , conditional knockout xrn-1 induced by somatic CRISPR phenocopied the subtle ire-1 like phenotype ( C ) . Asterisks , cell bodies . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 00910 . 7554/eLife . 06963 . 010Figure 2—figure supplement 3 . Other UPR arms also contribute to dendrite morphogenesis of PVD . No PVD dendritic arbor defects was shown in atf-6 ( A ) or pek-1 ( B ) null mutants while xbp-1 pek-1 double mutant showed ire-1 like phenotype with low-penetrance ( C ) . Asterisks , cell bodies . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 010 Because of the well-established role of the IRE-1/XBP-1 pathway in enhancing protein folding capacity in the ER , we hypothesized that IRE-1/XBP-1 upregulates specific ER chaperones to promote dendrite morphogenesis . We searched the PVD-specific gene profiling data ( Smith et al . , 2010 ) and found that two abundant ER chaperones of the Hsp70 family ( homologous to mammalian grp78/BiP ) , HSP-3 and HSP-4 , are enriched in PVD and therefore might be the targets of XBP-1 in PVD neurons ( Urano et al . , 2002 ) . Consistent with this idea , overexpression of hsp-4 in PVD restored normal dendritic branches in ire-1 mutants ( Figure 2E , F ) . However , overexpression of hsp-3 or daf-21 ( a cytoplasmic chaperone of the Hsp90 family ) did not rescue the phenotype ( Figure 2—figure supplement 1B , C ) . Furthermore , hsp-4 single mutant did not show the dendritic arbor defects ( Figure 2—figure supplement 1D ) , indicating other ER chaperones or co-chaperones functioning in parallel with HSP-4 . These results indicate that the dendritic defect in the ire-1 mutants is likely due to lack of specific chaperones in the ER . Importantly , xbp-1 mutant ( Figure 2—figure supplement 2B ) did not show the dendritic arbor defects . This suggests that other pathways downstream of IRE-1 but independent of XBP-1 can play redundant roles in dendrite morphogenesis . During ER stress , Ire1 can promote the degradation of mRNAs encoding some ER proteins to maintain homeostasis through regulated Ire1-dependent decay ( RIDD ) ( Hollien and Weissman , 2006; Hollien et al . , 2009 ) . The RIDD pathway has been shown to affect cell fate in various organisms , such as photoreceptor development in Drosophila ( Coelho et al . , 2013; Maurel et al . , 2014 ) . We next investigated whether the RIDD pathway functions in parallel with XBP-1 to regulate dendrite morphogenesis . mRNA degradation is initiated by internal cleavage mediated by RIDD , and the resulting RNA fragments would be subject to degradation by cytoplasmic 5′-3′ mRNA degradation machinery . However , all null mutants of the RIDD pathway components are lethal and difficult to examine dendrite phenotypes . Therefore , we used somatic clustered regularly interspaced short palindromic repeat ( CRISPR ) to create mosaic viable and conditional knock out of various genes ( Jinek et al . , 2012; Cong et al . , 2013; Shen et al . , 2014 ) . Using this method , we found that in the xbp-1 mutant background , somatic knockout of xrn-1 , which encodes a 5′-3′ exoribonuclease and is a key component of the 5′-3′ mRNA degradation pathway ( Newbury and Woollard , 2004 ) , phenocopied the ire-1 dendritic phenotype in PVD neurons in about 10% of the animals ( Figure 2—figure supplement 2C ) . Somatic CRISPR is intrinsically mosaic and often generates low-penetrance phenotypes compared with viable null alleles . These results indicate that the RIDD pathway functions in parallel to the XBP-1 to regulate dendrite branching of PVD . We also examined mutations in the other two arms of the UPR pathway , ATF-6 and PERK/PEK-1 , and found that they did not show any dendrite morphogenesis phenotype in PVD ( Figure 2—figure supplement 3A , B ) . However , xbp-1 pek-1 double mutant showed a low-penetrance ( about 25% ) ire-1-like phenotype ( Figure 2—figure supplement 3C ) . This suggests that the ER homeostasis mediated by other UPR pathways also contribute to dendrite morphogenesis . We next asked which protein ( s ) are potential targets of the IRE-1 UPR pathway in PVD executing dendrite morphogenesis . The severe decrease of distal dendritic branches of PVD in ire-1 mutants is reminiscent of dma-1 mutants . DMA-1 is a single transmembrane leucine-rich repeat ( LRR ) protein prominently expressed in PVD , and mutations in dma-1 result in severely reduced dendritic branching and complexity ( Liu and Shen , 2012 ) ( Figure 3—figure supplement 1A ) . DMA-1 acts in PVD as a receptor to recognize the SAX-7/L1CAM and MNR-1 ligand complex in the surrounding skin cell to promote branching and precisely pattern the dendritic arbor ( Dong et al . , 2013; Salzberg et al . , 2013 ) . We reasoned that the folding of DMA-1 might require IRE-1 . Consistent with this hypothesis , ire-1 dma-1 double mutants showed a phenotype that is indistinguishable from the dma-1 single mutant phenotype , suggesting that these two molecules function in the same genetic pathway in dendrite morphogenesis ( Figure 3—figure supplement 1B ) . Furthermore , hsp-4 overexpression in ire-1 dma-1 double mutants was not able to rescue the dendritic arbor defect ( Figure 3—figure supplement 1C ) , suggesting that DMA-1 might be a target of HSP-4 . As a single transmembrane protein , DMA-1 is synthesized in the ER and delivered to the plasma membrane through the secretory pathway . In wild type animals , GFP-tagged DMA-1 was detected on all the PVD dendritic processes and at the cortex of the cell body as diffusive fluorescence . In addition , numerous discrete intracellular puncta were found in the cell body and along the dendrites , which presumably represent the membrane trafficking organelles that carry DMA-1 ( Figure 3B ) ( Liu and Shen , 2012 ) . In ire-1 mutants , the punctate DMA-1::GFP in the cell body was lost ( Figure 3E ) . Instead , the somatic DMA-1::GFP in the ire-1 mutants co-localized with an general ER marker , cytochrome b5 ( cb5 ) ( Rolls et al . , 2002 ) ( Figure 3G ) . Moreover , the diffuse DMA-1::GFP signal on the distal dendrites was dramatically reduced in the ire-1 mutant while the signal on the proximal dendrites in ire-1 mutants was the same as in wild type ( Figure 3H , J , K ) . These observations suggest that DMA-1 is trapped in the ER and is not delivered to the distal dendrite plasma membrane , leading to the distal dendritic phenotype . Consistent with this hypothesis , overexpression of the ER chaperone HSP-4 , restored the DMA-1::GFP subcellular localization in ire-1 mutants to the normal distribution ( Figure 3J , K ) . Taken together , these data suggest that ER chaperones such as HSP-4 help to fold DMA-1 , which is required for the plasma membrane localization of DMA-1 and dendrite branching . 10 . 7554/eLife . 06963 . 011Figure 3 . DMA-1 is stuck in the somatic ER in ire-1 mutants . ( A ) Diagram of PVD in a young adult animal showing three representative subcellular regions: CB ( cell body ) , P ( proximal ) , and D ( distal ) dendrites . ( B to G ) Subcellular localization of DMA-1::GFP and general ER marker cb5::mCherry in PVD cell bodies in wild-type animals ( B to D ) and ire-1 mutants ( E to G ) . ( H to J ) DMA-1::GFP subcellular localization in WT ( H ) , ire-1 ( I ) and ire-1 mutants expressing HSP-4 ( J ) . Top panels: cell body; middle panels: proximal menorahs; bottom panels: distal menorahs . The morphology of the dendritic menorah is shown by cytoplasmic mCherry . Arrowheads , DMA-1::GFP puncta; brackets , tertiary branches ( without puncta ) used for measuring diffuse DMA-1:GFP . ( K ) Quantification of diffuse DMA-1:GFP ( normalized to cytoplasmic mCherry ) on tertiary branches in ( H to J ) . Error bars show mean ± s . e . m . , n = 50–60 . ns , not significant; p < 0 . 01 ( two-way ANOVA and post hoc Sidak's multiple comparisons test ) . Scale bars , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 01110 . 7554/eLife . 06963 . 012Figure 3—figure supplement 1 . DMA-1 is required for PVD dendrite morphogenesis and may be the downstream of HSP-4 . ( A ) Severe PVD dendritic branching defect in dma-1 mutants . ( B ) ire-1;dma-1 double mutants show a similar dendritic phenotype to dma-1 single mutants . ( C ) Overexpression of hsp-4 in ire-1;dma-1 double mutants cannot rescue the dendritic arbor defect . Asterisks , cell bodies . Scale bar , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 01210 . 7554/eLife . 06963 . 013Figure 3—figure supplement 2 . Subcellular localization of HSP-4::GFP , general ER marker cb5::mCherry and rough ER marker BFP::TRAM in PVD cell body ( CB ) region ( A to C and G to I ) and in distal ( D ) dendritic region ( D to F ) in wild type animals . Arrows indicating the general ER marker in PVD distal branches . Scale bar , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 013 To further understand why the dendrite loss in the ire-1 mutants was restricted to the distal dendrites , we investigated where the synthesis and folding of membrane proteins took place in PVD . This is an important question because the existence of local translation in dendrites might provide a source of DMA-1 production ( Holt and Schuman , 2013; Tom Dieck et al . , 2014 ) . Since HSP-4 is capable of folding DMA-1 , we first examined the subcellular localization of HSP-4 and found that HSP-4::GFP was exclusively localized in the PVD soma ( Figure 3—figure supplement 2A , D ) , co-localizing with a rough ER marker TRAM ( Figure 3—figure supplement 2G–I ) , HSP-4's ER localization pattern is consistent with the observation that its mammalian homolog BiP is localized in rough ER ( Bole et al . , 1989; Lai et al . , 2010 ) . These data suggests that the main protein synthesis and folding capacity for DMA-1 is likely in the cell body . In ire-1 mutants , lack of the upregulation of hsp-4 by spliced XBP-1 results in less DMA-1 in the secretory pathway and insufficient diffusion of DMA-1 to the distal region might be responsible for the specific loss of distal dendrites . If the ire-1 phenotype was the result of diminished DMA-1 levels in the distal dendrites , we reasoned two potential outcomes of DMA-1 overexpression in ire-1 mutants . The increased expression of DMA-1 might reach the plasma membrane and rescue the ire-1 phenotype . Alternatively , the DMA-1 overexpression might increase the protein-folding load and exacerbate the already strained protein folding machinery and lead to a more severe dendrite defect . Interestingly , we observed both effects: about 70% of animals showed efficient rescue of the dendritic arbor ( Figure 4B , D ) , while about 25% of animals showed a more severe phenotype , with the loss of proximal branches in addition to the distal ones ( Figure 4C , D ) . We hypothesized that in the absence of IRE-1 , the remaining protein folding capacity is at a critical level where overexpression of DMA-1 can produce functional or misfolded proteins , possibly depending on the slightly variable levels of endogenous chaperones in individual animals ( Burga et al . , 2011 ) . Consistent with the hypothesis , high level of chaperon HSP-4 expression together with DMA-1 decreased the percentage of dma-1-like phenotype , in a dose-dependent manner ( Figure 4D ) . To further test this hypothesis , we separated the transgenic animals into the phenotypically rescued animals and the severely defective animals based on their dendrite morphology , we found that there was much less accumulation or aggregation of DMA-1::GFP in the PVD cell bodies with the rescued morphology compared to more severely defective animals ( Figure 4—figure supplement 1A–G ) . Together , these rescuing results argue that the insufficient level of functional DMA-1 due to decreased protein folding capacity accounts for a large part of PVD dendritic defect in the ire-1 mutants . 10 . 7554/eLife . 06963 . 014Figure 4 . Overexpressing DMA-1 in ire-1 mutants can either rescue dendritic defects or cause more severe dendrite branching defects . ( A ) Representative defective PVD dendritic arbor in ire-1 mutants . ( B and C ) Overexpressing DMA-1 in PVD in ire-1 mutants rescues the dendritic defect ( WT-like ) ( B ) or causes a more severe phenotype with fewer branches ( dma-1-like ) ( C ) . Asterisks , cell bodies; Scale bar , 50 μm . ( D ) Proportions of different phenotypes in different transgenic rescue strains with overexpression of dma-1 and/or supplementation of different doses of HSP-4 chaperon . n > 120 . p < 0 . 01 , χ2 test with Sidak's multiple comparison correction . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 01410 . 7554/eLife . 06963 . 015Figure 4—figure supplement 1 . Accumulation or aggregation of DMA-1::GFP in the PVD cell bodies is tightly correlated PVD dendrite morphology . In the same trangene ( wyEx7338 ) bearing dma-1::GFP expression in ire-1 mutants , there is much less accumulation or aggregation of DMA-1::GFP in the PVD cell bodies with the rescued ‘WT-like’ PVD morphology ( A to C ) compared with more severely defective ‘dma-1 like’ animals ( D to F ) . ( G ) Quantification of the DMA-1::GFP intensity in the cell bodies ( normalized to cytoplasmic mCherry ) in ( A to F ) . Error bars show mean ± s . e . m . n = 20 . p < 0 . 01 , t-test . Scale bar , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 015 We have shown that the UPR machinery is required for dendrite morphogenesis in PVD . However , it is not clear whether the dendritic branching activates the UPR in PVD during development . To answer this question , we designed a UPR activity reporter which contains the genomic fragment of xbp-1 fused with a GFP in frame followed by an SL2::mCherry cassette ( Figure 5A ) . Upon UPR activation , the intron in genomic xbp-1 DNA will be spliced out by IRE-1 , leading to the production of XBP-1::GFP ( Iwawaki et al . , 2004 ) . The SL2::mCherry cassette permits the bicistronic expression of XBP-1::GFP and mCherry ( Spieth et al . , 1993 ) , and its function is similar to the viral IRES sequence in mammalian system . The whole reporter is driven by a PVD specific promoter ( Pdes-2 ) . The XBP-1::GFP intensity indicates the endogenous UPR activity in PVD while the intensity of mCherry is used to normalize to transgene expression levels among different animals . Consistent with the requirement of IRE-1 to activate XBP-1 , the XBP-1::GFP intensity in PVD neurons was diminished in ire-1 mutants compared with wild-type animals ( Figure 6—figure supplement 1A , D ) . 10 . 7554/eLife . 06963 . 016Figure 5 . The UPR activity is correlated with dendritic branching during development in the PVD neuron . ( A ) Design of the PVD-specific UPR activity reporter . The xbp-1 genomic DNA is fused with GFP followed by an SL2::mCherry cassette , which permits the bicistronic expression of XBP-1::GFP and mCherry . The reporter is driven by a PVD specific promoter . The XBP-1::GFP brightness indicates the UPR activity while the intensity of mCherry is used to normalize to transgene expression levels . ( B to J ) The PVD UPR activity in L3 stage ( B to D ) , L4 stage ( E to G ) and adult stage ( H to J ) . Arrowheads , nuclei of PVD . Scale bar , 5 μm . ( K ) Quantification of the normalized UPR activity in PVD in ( B to J ) . Error bars show mean ± s . e . m . n = 30–50 . p < 0 . 01 , Kruskal–Wallis one-way test and post hoc Dunn's test . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 01610 . 7554/eLife . 06963 . 017Figure 5—figure supplement 1 . The UPR activity is correlated with dendritic branching during development in the PVD neuron with another UPR reporter . The UPR activity is indicated by nuclear HIS-24::GFP brightness ( driven by the hsp-4 promoter ) while the PVD neuron is labeled and identified by cytoplasmic mCherry . The UPR activity of PVD is shown in L3 stage ( A to C ) , L4 stage ( D to F ) and adult stage ( G to I ) . Arrowheads , nuclei of PVD . Scale bar , 5 μm . ( J ) Quantification of the UPR activity in PVD in ( A to I ) . a . u . , arbitrary unit . Error bars show mean ± s . e . m . p < 0 . 01 , Kruskal–Wallis one-way test and post hoc Dunn's test . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 017 PVD neurons are derived postembryonically during the mid-L2 larval stage ( Sulston and Horvitz , 1977 ) , and starting from the late L2/early L3 , 2° branches begin to form followed by extension of 3° branches in the L3 stage . Dendrite morphogenesis is completed in the early L4 stage after 4° branches have sprouted from the 3° branches to form a network of menorah-shaped processes ( Smith et al . , 2010 ) . Using this PVD specific UPR activity reporter , we observed XBP-1::GFP in the nucleus of PVD starting at the L3 stage . The normalized XBP-1::GFP intensity increased between L3 and late L4 animals , coincidental with the stage of rapid dendrite branch addition . The XBP-1::GFP intensity subsequently decreased in mid-adult animals ( Figure 5B–K ) . We verified this result by using another UPR activity reporter ( Phsp-4::HIS-24::GFP ) . As an ER chaperone , HSP-4 is a transcriptional target of activated XBP-1 ( Calfon et al . , 2002; Urano et al . , 2002 ) , and its transcriptional level shows tight correlation with activation of the UPR with high sensitivity ( Iwawaki et al . , 2004 ) . We used the hsp-4 promoter to drive the expression of the C . elegans H1 histone , HIS-24 fused with GFP to detect the UPR activity in PVD neurons labeled with cytoplasmic mCherry . We found that the HIS-24::GFP signal became clearly detectable in L3 and further increased in L4 animals during which the menorahs form . The GFP fluorescence is dramatically downregulated in adult animals , ( Figure 5—figure supplemental 1 ) . Taken together , these results suggest that the UPR activity occurs most strongly during the time of PVD dendritic branching . The next question we wanted to address was how the UPR in PVD is induced during dendrite morphogenesis . The rapid dendritic growth of PVD requires high level of biosynthesis of plasma membrane proteins and efficient folding of them in the ER . PVDs are one of the only two pairs of highly branched neurons in C . elegans . Several transcription factors have been implicated in the cell fate determination of PVD . We considered two possibilities for the induction of the UPR activity in PVD neuron . In a ‘top down’ cell fate model , the enhanced UPR might be part of the cell fate decision controlled by transcription factors . Alternatively , the UPR might be induced because of the protein folding demand , in particular , maybe due to the translation of DMA-1 , an essential membrane molecule for PVD dendrite branching ( Figure 6A ) . 10 . 7554/eLife . 06963 . 018Figure 6 . The induction of the UPR in PVD depends on the DMA-1 . ( A ) Diagrams showing two possible models for the activation of the UPR in PVD . ( B to J ) The PVD UPR activity in WT ( B to D ) , dma-1 mutants ( E to G ) and WT with overexpession of dma-1 ( H to J ) . Scale bar , 5 μm . ( K ) Quantification of the normalized UPR activity in PVD in ( B to J ) and other mutants in ( Figure 6—figure supplement 1D–L ) . Error bars show mean ± s . e . m . ns , not significant , p < 0 . 01 , p < 0 . 05 , Kruskal–Wallis one-way test and post hoc Dunn's test . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 01810 . 7554/eLife . 06963 . 019Figure 6—figure supplement 1 . The UPR activity in PVD does not depend on other known proteins that are processed in the ER and are required for PVD dendrite morphogenesis . The PVD UPR activity in WT ( A to C ) , ire-1 ( D to F ) , hpo-30 ( G to I ) and kpc-1 mutants ( J to L ) . UPR activity is indicated by XBP-1::GFP brightness while the intensity of mCherry is used to normalize for transgene expression level . Scale bar , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 01910 . 7554/eLife . 06963 . 020Figure 6—figure supplement 2 . Another UPR reporter in PVD also showed dramatic decrease in dma-1 mutants . The UPR activity is indicated by nuclear HIS-24::GFP brightness ( regulated by hsp-4 promoter ) while the intensity of cytoplasmic mCherry is used to normalize for transgene expression level . The UPR activity of PVD is shown in WT ( A to C ) , ire-1 ( D to F ) and dma-1 ( G to I ) animals . Arrowheads , nuclei of PVD . Arrow , PDE cell body occasionally labeled in this strain . Scale bar , 5 μm . ( J ) Quantification of the normalized UPR activity in PVD in ( A to I ) . Error bars show mean ± s . e . m . p < 0 . 01 , Kruskal–Wallis one-way test and post hoc Dunn's test . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 02010 . 7554/eLife . 06963 . 021Figure 6—figure supplement 3 . The UPR in unbranched PVC neurons does not depend on DMA-1 . The UPR activity in WT ( A to C ) and dma-1 mutants ( D to F ) . UPR activity is measured by XBP-1::GFP brightness while the intensity of mCherry is used to normalize for transgene expression level . Scale bar , 5 μm . ( G ) Quantification of the normalized UPR activity in PVC in ( A to F ) . Error bars show mean ± s . e . m . n = 20 . NS , not significant , t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 021 To distinguish these models , we first tested this PVD specific UPR activity reporter in dma-1 knockout mutants and we found that the normalized XBP-1::GFP fluorescence level was significantly lower compared with wild type ( Figure 6B , E , K ) , suggesting that a functional dma-1 gene is required to turn on the UPR activity in PVD . Conversely , overexpression of dma-1 cDNA in PVD leads to an increase of UPR activity ( Figure 6H , K ) . Consistently , in dma-1 mutants , another UPR activity reporter , Phsp4::HIS-24::GFP also showed dramatic decrease in PVD neurons ( Figure 6—figure supplement 2 ) . These surprising results argue that DMA-1 is necessary for UPR induction in PVD , despite the fact that there must be many membrane proteins necessary to build dendrites . To test whether DMA-1's role in UPR induction is specific , we asked if mutations in other membrane proteins required for PVD dendrite morphogenesis also result in a decrease in UPR activation . Deletion mutations in kpc-1 ( a Kex2/subtilisin-like proprotein convertase and a Furin homolog ) ( Schroeder et al . , 2013 ) and hpo-30 ( a claudin homolog ) ( Smith et al . , 2013 ) cause severe reduction of PVD dendrites . While both of these gene products are processed in ER , neither mutation causes reduced UPR reporter activity in PVD ( Figure 6K and Figure 6—figure supplement 1G–L ) . In addition , this reporter also showed activity in the unbranched neuron PVC , indicating that there might be UPR activity that is unrelated to branched dendrite morphogenesis . Nevertheless , the dma-1 mutation did not change the UPR activity in PVC ( Figure 6—figure supplement 3 ) . These results demonstrate that the activation of UPR in PVD specifically depends on DMA-1 production . Surveying morphological phenotypes of other types of neurons , we found that the dendritic arbor defects in ire-1 mutants were restricted to PVD and FLP the only two pairs of highly branched neurons in the C . elegans nervous system ( Figure 1I ) . Coincidentally , only PVD and FLP showed sustained expression of DMA-1 ( Liu and Shen , 2012 ) . These observations suggest that the establishment of a complex dendritic arbor not only requires instructive cell surface molecules but also physiological UPR to increase the protein folding capacity and maintain cellular homeostasis . Since PVD and FLP are also the largest neurons in worms with complicated dendritic arbor , we wondered if the UPR is particularly activated in these large cells . To directly test this idea , we examine the PVD morphology in dpy-5 ire-1 double mutants . dpy-5 mutants have reduced body length ( about two third of that of wild type ) due to bearing a deletion in the cuticle procollagen DPY-5 gene ( Thacker et al . , 2006 ) and correspondingly reduced PVD size ( Figure 7—figure supplement 1B ) . Interestingly , the defective PVD phenotype of ire-1 was dramatically rescued with some animals showing wild type morphology ( Figure 7—figure supplement 1C–E ) , indicating the UPR is particularly required for neurons with large and complicated dendritic arbors . To further test this hypothesis , we determined the sufficiency of UPR activation to induce ectopic branches in neurons that normally do not branch extensively . The sensory neuron PDE shares the same lineage with PVD and does not express detectable levels of dma-1 . The cell body of PDE is positioned close to the PVD's and has a single processes running adjacent to the PVD dendrites . Consequently , the extracellular environment for PDE including the molecular ligands for DMA-1 is similar to that of PVD ( Figure 7A , B ) . 10 . 7554/eLife . 06963 . 022Figure 7 . Expression of dma-1 and hsp-4 together in the morphologically simple PDE neurons can induce ectopic branching more dramatically . ( A ) Diagram showing the PDE neuron ( in red ) which is located close to the PVD cell body and has a simple processes running adjacent to the PVD dendrites . Orthogonal ectopic branching ( in orange ) at a stereotyped position in the PDE commissure , reminiscent in location and direction to PVD tertiary branching . ( B to D ) Representative dendritic morphology of PDE neurons expressing cytoplasmic GFP in wild-type ( B ) , a strain with expression of dma-1 ( C ) and a strain with co-expression of dma-1 and hsp-4 in PDE ( D ) . Scale bar , 20 μm . Arrows , orthogonal ectopic branches . ( E ) More dramatic ectopic branching in the same strain co-expressing dma-1 and hsp-4 in PDE . ( F ) No ectopic branching in the same strain co-expressing hpo-30 and hsp-4 in PDE . ( G ) Percentages of PDE ectopic branching in strains expressing hsp-4 only , dma-1 only , and dma-1 with hsp-4 . n > 100 . p < 0 . 01 , χ2 test with Sidak correction . ( H ) Length of PDE ectopic branches in the strains expressing dma-1 only , and dma-1 with hsp-4 . p < 0 . 01 , Mann Whitney U-test . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 02210 . 7554/eLife . 06963 . 023Figure 7—figure supplement 1 . The defective PVD phenotype of ire-1 is suppressed by reduced size of PVD dendritic arbor . ( A ) . Defective dendritic branch in ire-1 mutants ( B ) . Reduced body length and smaller PVD size in dpy-5 mutants . ( C ) . dpy-5 ire-1 double mutants showing partially rescued PVD morphology . ( D ) . Some dpy-5 ire-1 double mutants showing wild-type PVD morphology . ( E ) . Quantification of numbers of secondary branches per 100 μm in ( A–D ) . Asterisks , cell bodies . Scale bar , 50 μm . p < 0 . 01 , one-way ANOVA and post hoc Sidak's multiple comparisons test . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 02310 . 7554/eLife . 06963 . 024Figure 7—figure supplement 2 . Overexpression of dma-1 in PDE induces ectopic branches and increases UPR activity . Representative morphology of PDE neurons expressing cytoplasmic mCherry in wild-type ( A ) and a strain with ectopic expression of dma-1 in PDE ( D ) . In these animals , the UPR activity is indicated by Phsp-4::HIS-24::GFP reporter in ( B ) and ( E ) . Asterisks , PDE nuclei . Arrow , orthogonal ectopic branch of PDE . Scale bar , 5 μm . ( G ) Quantification of the UPR activity in PDE in ( A to F ) . a . u . , arbitrary unit . Error bars show mean ± s . e . m . *p < 0 . 01 , Mann Whitney U-test . DOI: http://dx . doi . org/10 . 7554/eLife . 06963 . 024 Overexpression of dma-1 in PDE resulted in ectopic orthogonal branches that were similar to the PVD tertiary level branches ( Figure 7C ) ( Liu and Shen , 2012 ) . However , the low efficiency of ectopic branch induction ( 21% of animals bearing transgene ) suggests there might be other intrinsic mechanisms that are necessary to establish exuberant branches . Notably , increasing the ER folding capacity by expressing HSP-4 together with DMA-1 in PDE induced ectopic branches more efficiently ( 47% of animals ) while expressing other PVD-branching cell surface molecules such as HPO-30 with HSP-4 did not induce any ectopic branches ( Figure 7D–G ) . Further , these ectopic branches were more branched and significantly longer than overexpressing DMA-1 alone ( Figure 7H ) , and the Phsp-4::HIS-24::gfp reporter also showed increased UPR activity in PVD ( Figure 7—figure supplement 2 ) . These data support our model that the UPR is required for highly branched dendrites . Conserved in all eukaryotes , the UPR pathway plays significant roles in dealing with cellular stress and balancing homeostasis and apoptosis ( Walter and Ron , 2011 ) . Failure to mitigate the ER stress and reestablish homeostasis correlates with cell death , playing a central role in numerous human diseases such as pancreatic β-cell loss in diabetes ( Fonseca et al . , 2011 ) , retinal degeneration triggered by misfolded proteins in retinal dystrophies ( Lin and Lavail , 2010 ) and dopaminergic neuron degeneration in Parkinson disease models ( Valdes et al . , 2014 ) . In addition , under other biological conditions involving intense ER functions such as viral infection ( Dimcheff et al . , 2003 ) or pathogen defense ( Richardson et al . , 2010 ) , the UPR is activated to relieve the ER stress . In the nervous system , several reports indicate normal regulatory roles of the IRE-1 . For example , IRE-1 was shown to be involved in trafficking cell surface molecules such as AMPA receptor in cultured cells ( Vandenberghe et al . , 2005 ) or glutamate receptor GLR-1 in C . elegans interneurons ( Shim et al . , 2004 ) , and rhodopsins in Drosophila photoreceptors ( Coelho et al . , 2013 ) . However , to the best of our knowledge , the role of the UPR pathway in the development of the nervous system is poorly understood except in a few isolated cases . For example , a different arm of the UPR , PERK1 , was recently shown to be required for the olfactory receptor choice in mammalian olfactory sensory neurons through a feedback loop ( Dalton et al . , 2013 ) . Our results on the IRE-1/XBP-1/BiP/DMA-1 molecular cascade regulating dendrite morphogenesis in PVD neurons provide another clear example that UPR is directly involved in the development of neuronal cell morphology . More importantly , these results provide a link between dendrite morphogenesis and cellular homeostasis . First , only highly branched neurons such as PVD and FLP require the UPR to establish their dendritic arbors; second , the UPR activity in PVD is correlated with the development of dendrites; finally the induction of UPR is dependent on the expression of a pivotal molecule for dendritic branching DMA-1 in a homeostatic manner . Thus , in addition to the instructive extracellular cues required for complex branch formation and guidance , intrinsic mechanisms are also required . The secretory and endocytic pathways constitute the main membrane trafficking pathway in cell ( Sallese et al . , 2006; Brandizzi and Barlowe , 2013 ) . In order to form dendrites , membrane and transmembrane proteins need to be synthesized and delivered to the growing dendritic arbor . It is therefore not surprising that molecular components , in these pathways such as the early endosome protein RAB-5 are involved in dendrite morphogenesis ( Satoh et al . , 2008 ) . Interestingly , one previous study showed that secretory pathway mutants preferentially alter dendrite morphology and not axon extension ( Ye et al . , 2007 ) . Similarly , we found that the dendrite but not axon morphogenesis is specifically compromised in the ire-1 mutants , suggesting that specific molecular program and membrane trafficking pathways are required for dendrite development . Thus , our studies add weight to the idea that different molecular and trafficking pathways are utilized during dendrite morphogenesis and axon extension . The physiological functions of the UPR have been best demonstrated in highly differentiated cells that produce specific types of proteins in large amounts . One best-characterized example is the requirement of IRE1 and XBP1 for differentiation of B cells into plasma cells , where the UPR is activated to accommodate the secretion of large amounts of immunoglobulins ( Reimold et al . , 2001 ) . In these cells , the ER and secretory system are highly specialized for antibody biosynthesis , which accounts for half of the total protein production in these cells ( Askonas , 1975 ) . In Ig heavy chain knockout mice , the UPR activity was diminished in B cells , indicating that the production of immunoglobulins in B cells is required for induction of the UPR ( Iwakoshi et al . , 2003 ) . Another example is in the mammalian olfactory sensory neurons . Olfactory receptor ( OR ) genes are among the most highly transcribed G-protein-coupled receptors ( GPCRs ) in these neurons and trigger the UPR feedback loop required for specific OR choice ( Dalton et al . , 2013 ) . In contrast to these specialized cell types , a large number of diverse proteins and lipids are required in a developing neuron to establish its dendritic arbor . Many of these proteins are folded and processed in the ER . Surprisingly , our results suggest that among these proteins , a single transmembrane protein , DMA-1 , appears to be largely responsible for the activation of the UPR pathway in PVD during dendritic development . The characteristics of the DMA-1-like LRR proteins including their non-globular flexible solenoid like structure , repetitive amino acid sequences and high content of hydrophobic leucines , might make them particularly challenging to fold and assemble properly ( Freiberg et al . , 2004 ) . These characteristics may trigger the UPR in these cells and thus make this system required for dendrite morphogenesis . Other evidence also supports the notion that specific proteins have higher folding demands . For example , although IRE-1 has been shown to function in the normal secretory pathway ( Safra et al . , 2013 ) , the protein trafficking defect in ire-1 mutants is not general . It has been shown that for several different membrane-spanning proteins including Golgi-resident mannosidase , a TWK-type potassium channel , a single transmembrane synaptic vesicle protein synaptobrevin and a vesicular monoamine transporter CAT-1 , their subcellular localizations in ire-1 mutants were unaffected in interneurons ( Shim et al . , 2004 ) . These results suggest that DMA-1 is specifically regulated by the UPR pathway in the PVD neuron . Together , these findings indicate that certain proteins are intrinsically more challenging for folding and specific cell types have to employ the UPR pathway to accommodate the influx of these proteins and maintain homeostasis in the ER . Strains were grown at 20°C on NGM agar plates seeded with Escherichia coli OP50 except the UPR mutants and UPR reporter strains growing at 16°C . The wild-type strain was C . elegans N2 Bristol . The following mutant alleles and transgenes were used in this study: LGI: dma-1 ( wy686 ) , kpc-1 ( gk8 ) dpy-5 ( e907 ) ; LGII: ire-1 ( ok799 ) , hsp-4 ( gk514 ) ; LGIII: xbp-1 ( tm2482 ) wyIs592 [ser2prom3::myrGFP , odr-1::dsRed]; LGIV: wyIs581[ser2prom3::myr-mCherry , odr-1::dsRed]; LGV: hpo-30 ( ok2047 ) ; LGX: atf-6 ( ok551 ) , pek-1 ( ok275 ) , qyIs369[ser2prom3::dma-1::GFP , unc-119+] ) , wyIs378[ser2prom3::myrGFP , Prab3::mCherry , odr-1::dsRed] . The wy762 and wy782 alleles were isolated from an F2 semiclonal screen of 3000 haploid genomes in the strain containing wyIs378 ( Dong et al . , 2013 ) . Based on SNIP-SNP mapping and whole genome sequencing ( Sarin et al . , 2008 ) , we got the missense mutation information on ire-1 locus and verified by Sanger sequencing and complementation test with the null allele . Expression clones were made in the pSM vector , a derivative of pPD49 . 26 ( A Fire ) with extra cloning sites ( a kind gift from S McCarroll and CI Bargmann ) . The ser2prom3 ( PVD ) and Pdat-1 ( PDE ) promoters were used for cell-specific expression . cDNAs of ire-1 , xbp-1 ( long isoform ) , hsp-3 and his-24 were amplified from cDNA library while genomic DNAs of xbp-1 , hsp-4 , cb5 ( C31E10 . 7 ) and tram-1 were amplified from genomic templates . The XBP-1 UPR reporter construct was driven by Pdes-2 ( PVD ) , containing xbp-1 genomic DNA fused with GFPnovo2 ( Arakawa et al . , 2008 ) followed by gpd-2 SL2::mCherry ( from pBALU12 ) ( Tursun et al . , 2009 ) . For hsp-4 transcriptional activity reporter , the 1 . 1 kb 5′ upstream of hsp-4 ATG was cloned , driving HIS-24 fused with GFPnovo2 . For HSP-4::GFP , GFPnovo2 was inserted right before the C-terminus HDEL sequence of genomic HSP-4 . For somatic CRISPR , two DNA templates of xrn-1 sgRNA were 5′- GATATCGCTCCGATGTCCAT-3′ and 5′- AACGTGACGTCATCGTCATT-3′ , under the control of U6 promoter as in ( Chen et al . , 2013 ) . The transgenic extrachromosomal arrays were generated via injection using standard microinjection techniques ( Mello and Fire , 1995 ) . For rescue experiments , wyEx7329[ser2prom3::ire-1 ( 40 ng/μl ) , pBluescript ( 60 ng/μl ) , odr-1::dsRed ( 90 ng/μl ) ]; wyEx7332[ser2prom3::xbp-1 ( cDNA ) ( 20 ng/μl ) , pBluescript ( 60 ng/μl ) , odr-1::dsRed ( 90 ng/μl ) ]; wyEx6502[ser2prom3::xbp-1 ( genomic DNA ) ( 40 ng/μl ) , pBluescript ( 60 ng/μl ) , odr-1::dsRed ( 90 ng/μl ) ]; wyEx6816[ser2prom3::hsp-4 ( 50 ng/μl ) , pBluescript ( 60 ng/μl ) , odr-1::dsRed ( 90 ng/μl ) ]; wyEx7333[ser2prom3::hsp-3 ( 50 ng/μl ) , pBluescript ( 60 ng/μl ) , odr-1::dsRed ( 90 ng/μl ) ]; wyEx7335[ser2prom3::daf-21 ( 50 ng/μl ) , pBluescript ( 60 ng/μl ) , odr-1::dsRed ( 90 ng/μl ) ] . For ER markers and chaperone co-labeling , wyEx8074[ser2prom3::cb5::mCherry PCR fusion product ( 20 ng/μl ) , ser2prom3::hsp-4::GFPnovo2::HDEL ( 10 ng/μl ) , pBluescript ( 30 ng/μl ) , odr-1::dsRed ( 90 ng/μl ) ]; wyEx8075[Pdes-2::tagBFP::TRAM ( 15 ng/μl ) , ser2prom3::hsp-4::GFPnovo2::HDEL ( 10 ng/μl ) , pBluescript ( 30 ng/μl ) , odr-1::dsRed ( 90 ng/μl ) ] . For DMA-1 overexpression experiment , in wyIs581 background , wyEx7338[ser2prom3::dma-1::GFP ( 50 ng/μl ) , pBluescript ( 60 ng/μl ) , Pmyo-2::mCherry ( 1 . 5 ng/μl ) ]; For HSP-4 dose-dependent rescue experiments , with wyEx7338 and wyIs581 , wyEx7859[ser2prom3::hsp-4 ( 30 ng/μl ) , pBluescript ( 60 ng/μl ) , odr-1::dsRed ( 60 ng/μl ) , pBluescript ( 60 ng/μl ) ]; wyEx7770[ser2prom3::hsp-4 ( 60 ng/μl ) , pBluescript ( 30 ng/μl ) , odr-1::dsRed ( 60 ng/μl ) ] . For the UPR activity reporter , wyEx6766[Pdes-2::xbp-1 ( genomic ) ::GFPnovo2::SL2-mCherry ( 80 ng ) , Punc-122::dsRed ( 30 ng/μl ) , pBluescript ( 30 ng/μl ) ]; wyEx6812[ser2prom3::dma-1 ( 50 ng/μl ) , odr-1::dsRed ( 60 ng/μl ) , pBluescript ( 60 ng/μl ) ] For UPR activation experiment with hsp-4 transcriptional reporter , with wyIs581 , wyEx7820[Phsp-4::HIS-24::GFPnovo2 ( 20 ng/μl ) , pBluescript ( 60 ng/μl ) , odr-1::dsRed ( 70 ng/μl ) ] . For somatic CRISPR , in xbp-1 ( tm2482 ) background , wyEx7862[Phsp-16 . 2::Cas9 ( 50 ng/μl ) , PU6::xrn-1-sgRNA1 temp ( 30 ng/μl ) , PU6::xrn-1-sgRNA2 temp ( 30 ng/μl ) , odr-1::GFP ( 40 ng/μl ) ] . For PDE ectopic branching experiments , wyEx7035 [Pdat-1::hsp-4 ( 40 ng/μl ) , Pdat-1::GFP ( 20 ng/μl ) , odr-1::dsRed ( 60 ng/μl ) , pBluescript ( 30 ng/μl ) ] injected into wyEx4287 strain with overexpression of dma-1 in PDE ( Liu and Shen , 2012 ) ; wyEx8063[Pdat-1::GFP ( 20 ng/μl ) , Pdat-1::hsp-4 ( 40 ng/μl ) , Pdat-1::hpo-30 ( 30 ng/μl ) , odr-1::dsRed ( 90 ng/μl ) ]; For PDE UPR reporter experiments , wyEx8049[Phsp-4::HIS-24::GFPnovo2 ( 20 ng/μl ) , Pdat-1::mCherry ( 2 ng/μl ) , pBluescript ( 60 ng/μl ) , odr-1::dsRed ( 70 ng/μl ) ]; then use this line to ectopic express dma-1 in PDE , wyEx8065 [Pdat-1::dma-1::BFP ( 50 ng/μl ) , Pdat-1::mCherry ( 20 ng/μl ) , pBluescript ( 30 ng/μl ) , odr-1::GFP ( 20 ng/μl ) ] . wyEx4280 was used for FLP labeling ( Liu and Shen , 2012 ) . Following the protocol in ( Shen et al . , 2014 ) with some modifications , we first synchronized the culture by allowing 100–150 adult worms containing transgenic arrays ( raised at 20°C ) to lay eggs for 3 hr on seeded NGM plates . The eggs were heat-shocked at 33°C for 2 hr and then shifted to 20°C . After 60 hr , the PVD morphology was checked at the young adult stage . Images of fluorescently tagged fusion proteins were captured in live C . elegans using Plan-Apochromat 40×/1 . 3NA objective for whole PVD morphology and 63×/1 . 4NA for subcellular localization of fluorescent proteins on a Zeiss LSM710 confocal microscope ( Carl Zeiss , Germany ) . Animals were immobilized on 2% agarose pad using 10 mM levamisole ( Sigma–Aldrich , St . Louis , MO ) and oriented anterior to the left and dorsal up . Z-stacks were collected and the maximum intensity projection was used for additional analysis . For analyzing DMA-1::GFP intensity on tertiary dendrites ( middle and bottom panels in Figure 3H–J ) , XBP-1::GFP intensity during development ( Figure 5B–J ) and HIS-24::GFP UPR activity reporter ( Figure 5—figure supplement 1 , Figure 6—figure supplement 2 and Figure 7—figure supplement 2 ) , images were acquired using a Zeiss Axio Observer Z1 microscope equipped with a Plan-Apochromat 63×/1 . 4NA objective , Yokogawa spinning disk head ( Japan ) , 488 nm and 561 nm diode lasers ( Coherent , Santa Clara , CA ) , and a Hamamatsu ImagEm EMCCD camera ( Japan ) driven by MetaMorph ( Molecular Devices , Sunnyvale , CA ) . For 4° dendrite number counting , two PVD images ( labeled by wyIs581 ) from late L4 or young adults were stitched together in Adobe Photoshop ( San Jose , CA ) . The general shape and location of the primary dendrite ( the ‘backbone’ ) was recognized by a model-based neurite fiber tracing method ( Peng et al . , 2008 ) . Then the length of primary dendrite was determined by tracing the backbone and calculating the distance between adjacent identified pixels . Finally , the anterior part from cell body was divided into 8 equal segments while the posterior part was divided into 4 equal segments ( written in custom Matlab scripts ( Mathworks , Natick , MA ) ) . It should be noted that the length of each anterior segment is not equal to each posterior segment . The numbers of 4° dendrites whose secondary dendrites grew in each segment were counted manually . For 2° dendrite number counting , two PVD images ( labeled by wyIs581 or wyIs592 ) from late L4 or young adults were stitched together in Photoshop . Then the length of primary dendrite was determined manually by tracing the backbone and calculating the distance between adjacent identified pixels . The 2° dendrite number in each animal was counted manually , and this number was divided by the length of PVD primary dendrite ( per 100 μm ) . For measuring DMA-1::GFP intensity on 3° dendrites , we chose menorahs around the vulva region as ‘Proximal’ to avoid numerous puncta in dendrites close to cell body and chose menorahs around the middle point of anterior primary dendrite as ‘Distal’ to make sure we could get T-like branches in this region in ire-1 mutants . Two channel images were combined together by ImageJ ( Wayne Rasband ) , and a 2-pixel wide line was drawn along the tertiary branches ( avoiding obvious puncta ) and then the mean intensity values of two separated channels along this line were measured . After subtracting the background signal , the DMA-1::GFP signal was normalized to cytoplasmic mCherry . 3–5 tertiary branches were chosen for each spinning-disk image . To quantify the UPR activity , for different genotypes , the XBP-1::GFP or HIS-24:GFP mean intensity in the nucleus ( after background subtraction ) measured with ImageJ was normalized to the mean intensity of cytoplasmic mCherry in the same region using custom written Python ( Python Software Foundation , Beaverton , OR ) scripts . HIS-24::GFP intensity ( Figure 5—figure supplement 1 and Figure 7—figure supplement 2 ) was measured and quantified without normalization to cytoplasmic mCherry . All custom matlab and Python codes are provided in the Source code 1 . In comparisons of measurements such as fluorescence intensity or length of ectopic branches , we first tested for normality using a D'Agostino-Pearson test ( alpha = 0 . 05 ) . For data sets with normal distribution , we applied a two-tailed Student's t test for comparisons of two groups ( Figure 6—figure supplement 3G and Figure 4—figure supplement 1G ) . Comparisons involving multiple groups with multiple factors used two-way ANOVA and post hoc Sidak's multiple comparisons test ( Figure 3K ) . For data sets without normal distribution , we applied a two-tailed Mann–Whitney U-test for comparisons of two groups ( Figure 7H and Figure 7—figure supplement 2G ) . Comparisons involving multiple groups used Kruskal–Wallis one-way test and post hoc Dunn's test ( Figures 5K , 6K , Figure 5—figure supplement 1J and Figure 6—figure supplement 2J ) . To compare variables such as proportions we used χ2 test with Sidak correction for multiple comparisons ( Figure 4D and Figure 7F ) . All statistical tests were performed in Graphpad Prism ( San Diego , CA ) or in R ( R Development Core Team ) .	The brain consists of billions of cells called neurons that can rapidly send and receive information . At one end of the neuron , branched structures called dendrites receive signals from other cells . The number of dendrites and the amount of branching vary in different types of neurons . These patterns are crucial for each neuron to receive the information it needs . Abnormalities in dendrites affect brain activity and are associated with several diseases in humans . To make dendrites , the neuron needs to increase the amount of protein and other cell materials it produces . New proteins are made in a compartment called the endoplasmic reticulum and are folded into particular three-dimensional shapes with the help of chaperone proteins . These chaperones may be overwhelmed if protein production increases , leading to some proteins being folded incorrectly . This can activate a system called the unfolded protein response , which increases the number of chaperone proteins so that the proteins can be refolded correctly . However , it was not clear if neurons rely on the unfolded protein response , or another system , to cope with the increased levels of protein production needed to form complicated dendrite structures . Wei et al . studied a type of neuron called PVD—which has an elaborate network of dendrites—in nematode worms . The experiments show that the unfolded protein response is activated in these neurons as the dendrites form . Mutant worms that were missing a protein called IRE1 , which can activate the unfolded protein response , had dendrites with fewer branches than normal worms . The experiments also show that a protein called DMA-1—which is required for dendrites to form—was not able to fold correctly in the mutant worms . As a result , this protein remained in the endoplasmic reticulum instead of moving to the surface of the cell where it is usually found . Wei et al . 's findings reveal that the unfolded protein response plays a major role in allowing cells to increase protein production as the dendrites form . The next challenge is to understand how neurons coordinate transcription and activation of the unfolded protein response .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "neuroscience" ]	2015	The unfolded protein response is required for dendrite morphogenesis
Pests are a global threat to biodiversity , ecosystem function , and human health . Pest control approaches are thus numerous , but their implementation costly , damaging to non-target species , and ineffective at low population densities . The Trojan Female Technique ( TFT ) is a prospective self-perpetuating control technique that is species-specific and predicted to be effective at low densities . The goal of the TFT is to harness naturally occurring mutations in the mitochondrial genome that impair male fertility while having no effect on females . Here , we provide proof-of-concept for the TFT , by showing that introduction of a male fertility-impairing mtDNA haplotype into replicated populations of Drosophila melanogaster causes numerical population suppression , with the magnitude of effect positively correlated with its frequency at trial inception . Further development of the TFT could lead to establishing a control strategy that overcomes limitations of conventional approaches , with broad applicability to invertebrate and vertebrate species , to control environmental and economic pests . Pest species pose some of the greatest present-day challenges to native biota , global economies , and human health ( Naranjo et al . , 2015; Simberloff et al . , 2013 ) . With their emergence and spread linked to human trade , transport and the agricultural revolution , vertebrate , invertebrate and plant pest impacts have followed human global movement over past millennia like a footprint ( Hulme , 2009 ) . The environmental and agricultural losses caused by pests have been estimated at US$120 billion annually in the US alone ( Pimentel et al . , 2005 ) , while many also vector diseases of concern . For example , there are around 500 million human cases of the mosquito-borne diseases malaria and dengue fever annually ( World Health Organization , 2015; Bhatt et al . , 2013 ) , with other agents such as Zika virus rapidly spreading ( World Health Organization , 2016 ) . Considering the humanitarian , economic , and environmental values at stake , pest management requires large and recurrent government expenditure worldwide ( Stenseth et al . , 2003 ) . Conventional pest management programs typically rely on some form of lethal control , such as vertebrate shooting , trapping , and poisoning , or the area-wide application of pesticides to target weeds and invertebrates . The effects of these approaches are often temporary in nature and thus require regular re-application , and may have unwanted side effects on non-target species and the environment ( Bergstrom et al . , 2009; Tompkins and Veltman , 2006; Innes and Barker , 1999 ) . Furthermore , vertebrate culling is generally cost- and labor-intensive and can prove ineffective at low population densities ( Clout and Russell , 2006 ) , while blanket pesticide application is often hampered by the evolution of resistance in target species ( Tabashnik et al . , 2008; Innes and Barker , 1999 ) . Research efforts have thus focused on the development of novel control or eradication techniques that are target-specific , cost-effective even when applied at low population densities , and long-lasting in effect . One promising avenue lies in the development of techniques that impair the reproductive capacity of target pest species ( Cowan et al . , 2002; Courchamp and Cornell , 2000 ) . The most successful of such techniques employed to date is the Sterile Insect Technique ( SIT ) , whereby large quantities of sterilized males are introduced into target populations , reducing the reproductive success of the females with whom they mate ( Alphey et al . , 2010 ) . Although the SIT offers species-specificity , a major constraint lies in the need to continuously produce and release large numbers of sterile males for sustained population suppression , rendering eradication efforts time- and cost-intensive ( Dyck et al . , 2005; Alphey et al . , 2010 ) . Resource requirements could be greatly reduced if impairments to male fertility in a target population were heritable in nature . Emerging theory and experimental work suggests this is achievable via a prospective approach called the Trojan Female Technique ( TFT; Gemmell et al . , 2013 ) . The goal of the TFT is to use naturally occurring mutations in the mitochondrial DNA ( mtDNA ) , which impair male fertility but have no effects on females , to achieve multi-generational pest population suppression . Because mtDNA is typically maternally inherited ( White et al . , 2008; Birky , 1978 ) , such male-specific deleterious mtDNA mutations will to a large degree escape selection in the female germ line despite their associated high fitness cost to males , enabling their persistent inheritance across generations ( Frank and Hurst , 1996; Beekman et al . , 2014 ) . In theory , ‘Trojan Females’ carrying such mutations , and their female descendants , could continuously produce males with impaired fertility across generations , achieving perpetual numerical suppression of target populations ( Gemmell et al . , 2013 ) . Unlike other genetically based pest control approaches that involve transgenics , such as the Release of Insects carrying a Dominant Lethal ( RIDL ) ( Thomas et al . , 2000 ) , and the theorized use of gene-drives ( Webber et al . , 2015; Taylor and Gemmell , 2016 ) , the use of naturally occurring mutations means that TFT pest control would not necessarily require genome editing to progress . The TFT may thus offer a valuable alternative to emerging transgenic control techniques , whose potential use is currently subject to debates concerning safety and regulatory concerns ( Oye et al . , 2014; Lunshof , 2015 ) . The conceptual framework underpinning the TFT is based on a population genetic model that shows the maternal inheritance of mitochondria will facilitate the accumulation of deleterious mtDNA mutations that are male-biased in their effects ( Gemmell et al . , 2004; Frank and Hurst , 1996; Beekman et al . , 2014 ) . Recent empirical work in Drosophila melanogaster has substantiated this model , showing that the expression of fertility , longevity , and levels of nuclear gene expression are more sensitive to genetic variation in the mtDNA sequence in males than in females ( Camus et al . , 2012; Yee et al . , 2013; Innocenti et al . , 2011; Camus et al . , 2015 ) . Furthermore , particular mtDNA haplotypes have now been associated with sub- or complete-infertility in males , but with no apparent effects on female fertility , in Drosophila ( Dowling et al . , 2015; Yee et al . , 2013; Patel et al . , 2016; Wolff et al . , 2016b ) , seed beetles ( Dowling et al . , 2007 ) , hares ( Smith et al . , 2010 ) , and humans ( Ruiz-Pesini et al . , 2000 ) . Based on this theory and empiricism , Gemmell et al . , 2013 explored the conditions under which TFT mutations ( male-fertility impairing but female-benign ) could lead to population suppression . The results were encouraging , indicating that TFT haplotypes are predicted to cause suppression across a wide range of life-histories , with such effects not only persisting across generations , but also accumulating across successive introductions ( Gemmell et al . , 2013 ) . Building on those initial models ( Gemmell et al . , 2013 ) , empirical attention has focused on a particular mitochondrial haplotype sourced from a population of D . melanogaster in Brownsville ( USA ) , which has been associated with perturbed spermatogenesis and sperm maturation ( Clancy et al . , 2011 ) and is known to confer complete male sterility when placed alongside one isogenic nuclear background ( i . e . a nuclear background devoid of any allelic variation; Clancy et al . , 2011 ) , and consistent reductions in male fertility against a range of other nuclear backgrounds ( Dowling et al . , 2015; Yee et al . , 2013; Wolff et al . , 2016b ) . Sequence analyses revealed 10 single nucleotide polymorphisms ( SNPs ) that are unique to this Brownsville haplotype: one mutation located in the cytochrome b gene and nine others that reside within the A/T-rich control region ( mt:Cyt-b; Wolff et al . , 2016a ) . The SNP located in the mt:Cyt-b gene is non-synonymous , causing an amino acid change ( Ala278→Thr ) in complex III of the mitochondrial electron transfer chain , and it is this SNP that has previously been implicated as the putative fertility-reducing mutation ( Camus et al . , 2015; Clancy et al . , 2011 ) . This mt:Cyt-b SNP as the cause of the mtDNA-mediated male infertility has yet to be unambiguously confirmed; such confirmation would require further work to disassociate this mutation from the other nine SNPs that delineate the Brownsville haplotype from its counterparts , which would most likely be tractably accomplished by gene editing – an approach that remains in its infancy for the mitochondrial genome ( Wisnovsky et al . , 2016 ) . A recent study has provided further support for a key role for this SNP in fertility suppression ( Camus et al . , 2015 ) . Camus et al . ( 2015 ) demonstrated that the gene in which the SNP lies ( mt:Cyt-b ) experiences a four-fold decrease in expression in flies carrying the Brownsville haplotype relative to flies with other haplotypes , while expression of other mtDNA protein-coding genes is unaffected . Intriguingly , this Ala278→Thr mutation in the mt:Cyt-b gene occurs naturally in a range of other species , both vertebrate and invertebrate ( Clancy et al . , 2011 ) . Although the phenotypic implications of this mutation have not yet been screened outside of its putative effect in D . melanogaster , this indicates that male-fertility-reducing mtDNA haplotypes may routinely segregate in natural populations of metazoans ( Frank and Hurst , 1996; Beekman et al . , 2014; Gemmell et al . , 2004 ) . However , the practical utility of harnessing male-fertility-reducing haplotypes for pest control remains unclear on two fronts: first , whether the reductions in male fertility that they cause will indeed result in the numerical suppression of populations and second , whether any demonstrable suppression effects will persist across generations . There are several mechanisms by which impaired fertility in individual males may be compensated for at both individual and population scales . First , even males with impaired fertility may provide sufficient viable sperm for the complete fertilization of the eggs of females with which they mate in a population context . Second , even if a female is viable-sperm limited when mated with an impaired male , she may still obtain sufficient viable sperm through mating with other males . Third , even if population suppression initially occurs , as yet undetected pleiotropic effects on females could select against the mutation across generations . Fourth , even if the mutation persisted , the selection pressure imposed on males could select for nuclear modifiers that compensate for the effect of the TFT mitochondrial haplotype . Evolutionary theory and empiricism suggest that fertility effects associated with male-harming mtDNA mutations will often be reduced in such a way ( Yee et al . , 2013; Frank and Hurst , 1996; Gemmell et al . , 2004 ) . Ultimately , when present in large and panmictic populations , TFT haplotypes may be expected to undergo changes in frequency through genetic drift and directional or balancing selection ( Wolff et al . , 2014; Gregorius and Ross , 1984; Clark , 1984 ) . Furthermore , stochastic contractions and expansions in the target population may exacerbate the effects of genetic drift and facilitate the purging of introduced TFT haplotypes , even when they are not selected against ( White et al . , 2008; Rand et al . , 2001 ) . Here , we experimentally test the capacity of the Brownsville haplotype ( our candidate TFT haplotype ) to suppress large and panmictic laboratory populations of D . melanogaster . Persistent numerical suppression would provide proof-of-concept for the TFT , showing that the maternal mode of mtDNA inheritance can potentially be harnessed for a eukaryotic pest control technique that overcomes several limitations of conventional approaches . Trial populations were initiated with the TFT haplotype at four different starting frequencies ( 0% [control] , 25% , 50% , and 75% ) , with the expectation that the numerical suppression observed would increase with increasing TFT frequency , and these populations were then maintained for 10 generations under two environmental regimes . Under the first regime , populations were maintained in the ecological and demographic conditions in which they are typically maintained in the laboratory , and in which egg numbers per generation are carefully regulated . Predictions from an existing simulation model of D . melanogaster populations under such a regime ( see Supplementary information in [Wolff et al . , 2016b] ) are that the TFT haplotype utilized ( documented to reduce male breeding success by 29–69%; Dowling et al . , 2015; Wolff et al . , 2016b ) will cause mean population suppression of between 6 . 7–16 . 8% , 13 . 8–37 . 7% and 21 . 4–63 . 7% for the 25% , 50% and 75% TFT haplotype frequency , respectively . Such a magnitude of suppression , modeled under conditions of multiple mating , is predicted to translate into up to three times greater suppression in natural populations in which females re-mate at a relatively low frequency ( Wolff et al . , 2016b ) . Under the second regime , populations were maintained in conditions that allow them to experience stochastic contractions and expansions in population size that are more reflective of natural population dynamics . The purpose of this regime was to explore how such dynamics could influence the frequency of the TFT haplotype across generations , with genetic drift potentially leading to either haplotype purging ( with associated loss of population suppressive effects ) or fixation ( with associated increase in population suppressive effects ) . Together , our experiments document the first experimental test of the ability of a candidate TFT haplotype to cause and maintain population suppression . We demonstrate that the TFT haplotype caused significant numerical suppression in the laboratory Drosophila melanogaster populations relative to controls , with the magnitude of effect positively correlated with its frequency at trial inception . Furthermore , the suppressive effect persisted over the full length of the trial ( 10 generations ) , with no reduction in haplotype frequency . Our results thus provide proof-of-concept for the TFT , showing that uniparental inheritance of mtDNA could potentially be harnessed in the development of a pest control technique that would be broadly relevant across eukaryotes . We found an interactive effect of TFT treatment and generation number on population sizes ( Tables 1 and 2A ) , and on the frequency of the TFT mutation ( Tables 1 and 2B ) , across the 10 generations of the experiment ( Figure 1A ) . Replicates initiated with TFT haplotype frequencies of 0% or 25% stabilized at average population sizes of 72 . 99 and 72 . 57 individuals across the 10 generations , respectively . However , replicates initiated with TFT haplotype frequencies of 50% or 75% declined over the first six generations to average population sizes of 67 . 24 and 66 . 75 respectively , with this magnitude of suppression ( 8% ) maintained for the remainder of the experiment . 10 . 7554/eLife . 23551 . 003Table 1 . Mean offspring numbers , TFT haplotype frequencies , and genotyping outcomes for the two experiments ( regulated conditions and fluctuating conditions ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23551 . 00310 . 7554/eLife . 23551 . 004Table 1—source data 1 . Raw data for offspring number and TFT frequency for Experiments 1 and 2 . Offspring number per population for each of 10 generations in Experiment 1 and 2 . TFT frequency for each population at generations 1 , 5 and 10 for Experiment 1 and at generation 10 for Experiment 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 23551 . 004TFT treatment ( starting frequency ) 0% TFT25% TFT50% TFT75% TFTPopulations [n]21212121ExperimentRegulatedF1 Frequency0 . 000 . 26 ± 0 . 030 . 44 ± 0 . 030 . 63 ± 0 . 03F5 Frequency0 . 000 . 15 ± 0 . 030 . 63 ± 0 . 030 . 71 ± 0 . 02F10 Frequency0 . 000 . 17 ± 0 . 030 . 67 ± 0 . 030 . 80 ± 0 . 03Loss-400Fixation-003Heteroplasmy-000Mean offspring number ( F10 ) 74 . 15 ± 1 . 0072 . 05 ± 0 . 8967 . 43 ± 0 . 9966 . 38 ± 0 . 70Population extinction1000FluctuatingF10frequency0 . 000 . 35 ± 0 . 060 . 59 ± 0 . 060 . 75 ± 0 . 05Loss-410Fixation-144Heteroplasmy-020Mean offspring number ( F10 ) 80 . 47 ± 5 . 1281 . 19 ± 4 . 5978 . 56 ± 5 . 6385 . 86 ± 3 . 73Population extinction202010 . 7554/eLife . 23551 . 005Table 2 . ( A ) Linear mixed model showing effects of TFT treatment and generation number on mean offspring number , and ( B ) generalized linear mixed model of effects of TFT treatment and generation on TFT haplotype frequency of populations with regulated population size ( Experiment 1 ) . There was no evidence of overdispersion in the model of TFT haplotype frequency ( dispersion parameter = 0 . 766 ) , and addition of an observation-level random effect to the final model did not change this parameter ( dispersion parameter = 0 . 767 ) , nor the parameter estimates of the model . DOI: http://dx . doi . org/10 . 7554/eLife . 23551 . 005 ( A ) Offspring number ( B ) TFT frequencyFixed effectsχ2Dfpχ2DfpTFT treatment2 . 2330 . 5338 . 142<0 . 001Generation14 . 7390 . 106 . 2120 . 045TFT treatment × generation51 . 09270 . 00323 . 524<0 . 001Random effectsSDSDBiological replicate0 . 81--0--Experimental population1 . 11--0--Residual5 . 48-----* Standard deviation10 . 7554/eLife . 23551 . 006Figure 1 . Mean offspring number and haplotype frequencies under density-controlled population conditions ( Experiment 1 ) . ( A ) Mean offspring number ( ±SEM ) , and B ) mean haplotype frequencies ( ±SEM ) of experimental populations in Experiment 1 over 10 generations , under density-controlled conditions . Founding populations ( F0 ) were established with varying proportions ( 0% , 25% , 50% , 75% ) of the TFT haplotype . Each generation was propagated with 80 eggs . Genotyping was conducted at generations 1 , 5 , and 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 23551 . 00610 . 7554/eLife . 23551 . 007Figure 1—source data 1 . Raw data for offspring number and TFT frequency for Experiment 1 . Offspring numbers per population for each of 10 generations and TFT frequencies for each population at generations 1 , 5 and 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 23551 . 007 We genotyped females from each experimental population at generations 1 , 5 and 10 , and found that the population suppression effect across generations was associated with the frequency of the TFT haplotype ( χ2=21 . 37 , p=<0 . 003; Table 3 , Figure 2A–C ) . There was no evidence of consistent TFT haplotype purging; while the 25% TFT treatment ended ( at generation 10 ) at a mean haplotype frequency of 0 . 17 , ending frequencies were 0 . 67 and 0 . 80 for the 50% and 75% TFT treatments respectively ( Figure 1B; Table 1 ) . Over the course of the trial , the TFT haplotype was purged from four replicates and went to fixation in three ( out of total of 63 populations ) . 10 . 7554/eLife . 23551 . 008Table 3 . Linear mixed model of association of TFT haplotype frequency on mean offspring number in populations with regulated population size ( Experiment 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23551 . 008Fixed effectsχ2DfPTFT frequency21 . 3770 . 003Random effectsSDBiological replicate0 . 31--Experimental population0--Generation2 . 77--Residual6 . 40-- Standard deviation10 . 7554/eLife . 23551 . 009Figure 2 . Mean offspring number in relation to TFT haplotype frequency under density-controlled ( Experiment 1 ) and fluctuating population conditions ( Experiment 2 ) . Mean offspring number ( ±SEM ) across experimental populations ( A–C ) with regulated population size at generations 1 , 5 , and 10 in Experiment 1; and ( D ) with fluctuating population size at generation 10 in Experiment 2 . Haplotype frequencies were determined by genotyping seven females for each experimental population at each of three generations ( 1 , 5 , and 10; n = 1323 ) in Experiment 1; and by genotyping nine females for each experimental population at generation 10 ( n = 567 ) in Experiment 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 23551 . 00910 . 7554/eLife . 23551 . 010Figure 2—source data 1 . Raw data for offspring number and TFT frequency for Experiments 1 and 2 . Offspring numbers per population for each of 10 generations in Experiment 1 and 2 . TFT frequencies for each population at generations 1 , 5 and 10 for Experiment 1 and at generation 10 for Experiment 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 23551 . 010 The less-constraining rearing conditions of this experiment led to mean changes in population size between any two generations of approximately 30% across all experimental populations ( χ2=46 . 72 , p=<0 . 001; Figure 3; Table 4 ) , with four experimental populations ( two in each of the 0% and 50% TFT treatments ) going extinct . Under these conditions , there was no detectable effect of the TFT treatment on population size across the experiment ( Table 4A ) . Nor did we detect an association between the TFT haplotype frequency of each experimental population at generation 10 and the final population size ( Table 5 , Figure 2D ) . However , the TFT haplotype was still stably maintained in most cases; ending frequencies were 0 . 35 , 0 . 59 and 0 . 75 for the 25% , 50% and 75% TFT treatments , respectively ( Table 1 , Table 4B , Figure 3 ) . Over the course of the trial , the TFT haplotype was lost from five replicates and went to fixation in nine ( out of a total of 63 populations ) . In two replicates of the 50% TFT treatment , we detected 11 cases where flies carried both the TFT and wildtype haplotype ( five individuals in one replicate and six in the other ) , indicating at least two cases of biparental mtDNA inheritance in the experiment . 10 . 7554/eLife . 23551 . 011Figure 3 . Mean offspring number and haplotype frequencies under fluctuating population conditions ( Experiment 2 ) . Mean offspring number ( ±SEM , on vertical axis on left-hand side ) , and mean haplotype frequencies ( ±SEM , right-hand side ) at generation 10 , of experimental populations . Founding populations ( F0 ) were established with varying proportions ( 0% , 25% , 50% , 75% ) of fruit fly pairs harboring the mt:Cyt-b TFT mutation . Each generation , each experimental population was propagated with all offspring of the previous generation . DOI: http://dx . doi . org/10 . 7554/eLife . 23551 . 01110 . 7554/eLife . 23551 . 012Figure 3—source data 1 . Raw data for offspring number and TFT frequency for Experiment 2 . Offspring numbers per population for each of 10 generations and TFT frequency for each population at generation 10 . DOI: http://dx . doi . org/10 . 7554/eLife . 23551 . 01210 . 7554/eLife . 23551 . 013Table 4 . ( A ) Linear mixed model showing effects of TFT treatment and generation on mean offspring number; and ( B ) generalized linear mixed model of effects of TFT treatment on TFT haplotype frequency of populations with fluctuating population size ( Experiment 2 ) . When reanalyzing mean offspring number ( A ) , having excluded the 15 zero values in the dataset resulting from three vial extinctions ( two from the TFT 0% treatment , and one from the TFT 50% treatment ) , the effect of TFT treatment on offspring number remained statistically non-significant ( χ2=3 . 2 , p=0 . 36 ) . The binomial model of TFT frequency ( b ) indicated overdispersion ( overdispersion parameter = 1 . 85 ) , and thus an observation-level random effect was added ( experimental population ) to the model ( overdispersion parameter of final model = 1 . 07 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23551 . 013 ( A ) Offspring number ( B ) TFT frequencyFixed effectsχ2DfP2DfPTFT treatment4 . 9830 . 1720 . 332<0 . 001Generation46 . 729<0 . 001---Random effectsSDBiological replicate3 . 1--0--Experimental population11 . 2--1 . 32--Residual20 . 85----- Standard deviation10 . 7554/eLife . 23551 . 014Table 5 . Linear model of association of TFT haplotype frequency on mean offspring number in populations with fluctuating population size ( Experiment 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23551 . 014Fixed effectsχ2DfPTFT Frequency4 . 9590 . 839Random effectsSDBiological replicate0--Residual24 . 93-- Standard deviation Working with laboratory populations of D . melanogaster , we have demonstrated that the compromised male fertility caused by our candidate TFT haplotype can suppress population sizes across generations . The magnitude of suppressive effect was dependent on the frequency of the TFT haplotype and the conditions under which populations were propagated . When the experiment was conducted under regulated population sizes , persistent numerical suppression was observed ( Figure 1A ) . While all treatments reduced in population size over the first two generations , as the trial moved toward equilibrium dynamics , those seeded with at least 50% TFT haplotypes continued to decline to generation six , and then remained suppressed at sizes averaging 8% below control populations . No such effect was apparent for populations seeded with only 25% TFT haplotypes . These results raise two questions . First , why was the suppressive effect only 8% at the population scale , when a-priori modelling predicted suppression relative to controls of at least 21 . 4% in the 75% TFT haplotype frequency treatment ( see Supplementary information in Wolff et al . , 2016b ) . Second , why was there no apparent effect of the 25% TFT treatment ? These two issues are likely linked , and may be due to compensation at the scale of the individual , through females obtaining more fertile sperm than would be expected on an additive basis , either wholly from the sub-fertile TFT males with whom they mate and/or through multiple mating ( i . e . the TFT haplotype used may have caused less reduction in male breeding success than modelled , or females may have mated with more males than included in our modelling ( see Supplementary information in Wolff et al . , 2016b; Gemmell et al . , 2013 ) ) . Such compensation would explain why population size was not affected in the 25% TFT treatment ( complete compensation ) and was less than predicted in the 50% and 75% TFT treatments ( partial compensation ) . The reduction in haplotype frequency that occurred in the 25% TFT treatment ( Figure 1B; Table 1 ) may also have contributed . However , irrespective of these or other mechanisms that may be responsible , a significant population suppression effect of the TFT haplotype , which persisted to the end of the trial , was observed for populations of the 50% and 75% TFT treatments . When the experiment was conducted under conditions in which populations experienced large stochastic contractions and expansions in size , no suppression effects were detected . This was not driven by overall changes in TFT haplotype frequency over the course of the trial . While , as was expected , there were more cases of haplotype loss and fixation under this regime than under the more stable regime ( totals of 14 versus 7 such events ) , the haplotype went to fixation more often than it was purged , and no overall decline in TFT frequency from starting conditions was observed ( Figure 3 ) . It is thus most likely that the suppression effect of around 8% observed under the more stable experimental regime , was masked by the underlying population dynamics in the more stochastic regime . Although mean TFT haplotype frequency declined slightly ( from 25% to 17% ) across the 10 generations of Experiment 1 in the 25% TFT treatment , the frequency of the TFT haplotype actually increased in the 50% and 75% treatments ( to an average of 67% and 80% , respectively; Figure 1A–B ) of this experiment . Furthermore , in Experiment 2 , haplotype frequencies increased in the 25% and 50% treatments and were stably maintained in the 75% treatment at generation 10 . Thus , across six treatments and two experiments , the frequency of the TFT haplotype decreased in just one treatment , and this decrease was modest . These data suggest there was no strong selective pressure against the TFT haplotype , even though it was causing population suppression . This result further supports both the theory underpinning the TFT , that male-infertility caused by a mutation in the mtDNA will generally escape selection due to its maternal inheritance ( Frank and Hurst , 1996; Gemmell et al . , 2004; Beekman et al . , 2014 ) , and the previous anecdotal observations of no negative pleiotropic effects on female fertility being linked to the TFT haplotype ( which would also incur selection against it; Clancy et al . , 2011; Dowling et al . , 2015; Yee et al . , 2013 ) . Intriguingly , the frequency of the TFT haplotype increased in four of the six treatments ( Table 1 ) . While this contention requires further experimental testing , if haplotype frequency increases are occurring one potential explanation may lie in an observation from a previous study that the TFT haplotype is linked to increased pupal viability ( Wolff et al . , 2016b ) , suggesting an antagonistic pleiotropic effect ( low male fertility , but high pupal viability ) that is under positive selection due to its benefits to females . As noted in the introduction , persistent maintenance of TFT haplotypes within a population is expected to result in selection on males for nuclear modifiers that compensate for the negative TFT effect , and restore male fertility ( Frank and Hurst , 1996; Yee et al . , 2013; Wolff et al . , 2016b; Dowling et al . , 2015 ) . The capacity for an effective compensatory response will depend largely on levels of standing nuclear allelic variance already present within populations . Our previous work indicated that although the effects of the TFT haplotype on male fertility consistently conferred lower fertility in males relative to other haplotypes , it was indeed modulated by the nuclear background of different populations ( Wolff et al . , 2016b; Dowling et al . , 2015 ) . However , even though the experimental populations utilized in the current study were large and expected to maintain high levels of segregating nuclear allelic variance ( Gardner et al . , 2005; Griffin et al . , 2016 ) , there was no apparent rapid selection for fertility-restoring nuclear components over the course of our trials ( which would have been expressed as restoration in population sizes over time ) . Thus , although nuclear mutations could arise over time in a population that compensate for the negative effects of the TFT mutation , in the population of Drosophila that we used there were no apparent segregating nuclear modifiers that had the capacity to completely restore male fertility and be rapidly selected . Extending our experiments by placing the TFT haplotype alongside additional outbred nuclear backgrounds , including the nuclear background from the population the TFT haplotype was originally sourced from , could potentially inform at what frequency ( if at all ) such nuclear modifier alleles may occur . In this regard , it would also be interesting to evaluate whether the use of fertility-reducing mutations that have evolved naturally are likely to be more successful in the long-term ( in terms of heritability and sustained effect ) to suppress population size than the use of artificial gene-drive constructs , whose introduction are predicted to almost inevitably lead to the emergence of drive-resistant alleles in most natural populations ( Unckless et al . , 2017; Noble et al . , 2016 ) . Notably , if the suppression effect of these mutations ( be they natural or gene-drive constructs ) when placed into new target pest populations is large , this will presumably act to reduce both the efficacy by which selection can target standing nuclear variation , and the likelihood of spontaneous compensatory mutations , that restore fertility in the target population . Interestingly , we identified cases in which offspring were heteroplasmic for both the TFT and Dahomey mtDNA haplotypes in two of the replicate populations . Heteroplasmy has previously been found in Drosophila sourced from Brownsville ( Kann et al . , 1998 ) . Whether the Brownsville population is predisposed to sporadic episodes of biparental inheritance is unclear , but the repeated observation of low rates of biparental inheritance of mtDNA in populations across the globe suggests that paternal leakage may be common in Drosophila ( Wolff et al . , 2013; Nunes et al . , 2013; Dokianakis and Ladoukakis , 2014 ) . The intra-individual co-occurrence of both TFT and wildtype haplotypes enables the possibility for recombination between divergent mtDNA molecules to create novel mitochondrial haplotypes carrying the TFT mutation ( Ma and O'Farrell , 2015 ) . Whether the fertility-suppressing effect of the candidate TFT mutation ( s ) would be moderated by its placement alongside a different mitochondrial genetic background , within a recombinant haplotype , is unknown and would depend on the capacity for epistasis within the mitochondrial genome to affect fitness outcomes . Furthermore , to come into effect , such a novel recombinant haplotype must then be at a selective advantage if it is to become rapidly fixed within populations . However , a more likely scenario is that rare recombinant haplotypes will be purged under drift while segregating at low population frequencies , or under purifying selection if harmful to females ( Wolff et al . , 2011; Bergstrom and Pritchard , 1998; Ma and O'Farrell , 2016 ) . While we have provided proof-of-concept for the TFT , demonstrating experimentally that its male fertility effects can achieve persistent population suppression , the question remains of its utility for field application to pest populations . Critically , while the conceptual foundation for the work was based on TFT haplotypes conferring complete male sterility ( Clancy et al . , 2011 ) , subsequent work has demonstrated that reduced fertility is the more likely scenario ( Dowling et al . , 2015; Wolff et al . , 2016b ) , and our current laboratory trials indicate that compensation due to females remating with wild-type males , or other processes such as mitonuclear , or gene-by-environment interactions , could result in reduced levels of population suppression than would otherwise be predicted by the underpinning theory ( Gemmell et al . , 2013 ) . However , modeling predicts the magnitude of effect to be much greater in natural populations in which females are expected to re-mate at a relatively low frequency ( Wolff et al . , 2016b ) , due to lower encounter rates between individuals of each sex and lower densities of cohabitation ( Gemmell et al . , 2004 ) . Previous experiments have further revealed that individuals harboring the Brownsville haplotype exhibited increased pupal viability , which is likely to aid the introgression of the TFT haplotype into target populations ( Wolff et al . , 2016b ) . The utility of the TFT in the field will also depend on the efficacy of specific TFT haplotypes to decrease male fertility ( Gemmell et al . , 2013 ) . Multiple-release strategies can be employed in order to reach TFT haplotype frequencies required to achieve population suppression of natural populations . This way , eradication strategies may still be achievable even if the desired effect in population suppression is reliant on TFT haplotype frequencies that are high . In addition , there is potential to augment the sterilizing effects of the TFT haplotype through linking it with further candidate TFT mutations . Evolutionary theory and empiricism both suggest that plant and animal mitochondrial genomes should be naturally enriched for male-harming mtDNA mutations ( Innocenti et al . , 2011; Gemmell et al . , 2004; Frank and Hurst , 1996; Camus et al . , 2012; Beekman et al . , 2014; Dobler et al . , 2014 ) . Indeed , a male-sterilizing but female-benign mutation has recently been discovered in the gene encoding the cytochrome c oxidase subunit 2 ( mt:COII ) in D . melanogaster ( Patel et al . , 2016 ) . Linking multiple TFT mutations within a single TFT haplotype , or the release of multiple TFT strains each bearing a distinct set of TFT mutations holds great promise to further the capacity of the TFT to efficiently suppress population size . The pairing of multiple TFT mutations within the one mtDNA sequence should soon be quickly achievable , given the rapid advances in genome editing technologies ( Reddy et al . , 2015 ) . Thus , although the TFT does not necessarily require transgenics to progress , genome editing could enable the time- and cost-efficient placement of TFT mutations into target populations ( without the need for long-running mutagenic and breeding approaches to first generate and then implant the candidate TFT mutations ) . Placement of single mutations into test populations would also allow the unambiguous identification of the fertility-reducing mutation ( s ) harbored by the Brownsville haplotype , and to confirm whether it is indeed the mt:Cyt-b mutation that causes the decrease in male fertility . If confirmed , the utility of the mt:Cyt-b mutation holds particular promise for pest control given that this candidate mutation has already been identified in a broad range of invertebrate and vertebrate species ( Clancy et al . , 2011 ) . Combined with previous studies , and a solid theoretical conceptual basis , our results lend credence to the utility of the TFT as a novel approach to pest control , deserving of continued development . Underpinning work needs to continue in the fruit fly model system to both explore the effects of linking multiple candidate TFT mutations within single mtDNA sequences ( and quantifying their effects ) , and whether female-beneficial ( but male fertility-impairing ) haplotypes can be harnessed to drive the spread of TFT haplotypes through pest populations to effectively suppress population size . However , it would now also be timely to explore the capacity of TFT candidate mutations to decrease male fertility in other species , particularly real-world pest species that could be suitable targets for TFT control . If applicability and consistency of effect can be confirmed , the potential use of the TFT to suppress populations holds promise for a broad range of metazoan pests . The experiment harnessed a laboratory population of fruit flies that was originally collected in Dahomey ( Benin , West Africa ) in 1970 ( Partridge and Andrews , 1985 ) , and which has been kept at large effective population sizes since ( at 25°C on a 12:12 light: dark cycle ) . To maintain the high levels of nuclear allelic variation segregating within the Dahomey population ( Gardner et al . , 2005; Griffin et al . , 2016 ) , populations have been kept in large replicate populations on a discrete-generations cycle since these were obtained from Prof Linda Partridge in 2010 . This is achieved by propagating each generation with around 900 adult flies of 4 days of adult age , dispersed across three 250 ml bottles , each containing 75 ml of a potato-dextrose-agar food substrate . The flies are provided with a one to two hour ovipositioning period , after which the number of eggs per bottle is manually reduced ( trimmed ) to 300–350 . Adult flies are then removed from the bottles and then , for the subsequent generation , emerging adult offspring that eclose from each bottle are admixed prior to their re-sorting into three separate bottles to start the propagation procedure for the following generation . We initiated six replicates of the Dahomey population , and introgressed the TFT mtDNA haplotype ( sourced from Brownsville [BRO] Texas , USA; Rand et al . , 1994 ) into three of these replicates . The other three replicates were designated to the control , and they hosted their own coevolved mtDNA haplotype sourced from Dahomey [DAH] ( Partridge and Andrews , 1985 ) . While the BRO haplotype confers low male fertility , with no recorded negative effects on female fertility , the DAH haplotype is putatively healthy and confers normal fertility in both sexes ( Camus et al . , 2012; Yee et al . , 2013; Wolff et al . , 2016b; Dowling et al . , 2015; Camus et al . , 2015 ) . All six replicates went through the same handling procedures leading into the experiment , over successive generations , which ensured effective population sizes across all replicates were carefully controlled with the expectation that levels of segregating nuclear variance were highly similar across replicates . Specifically , to initiate replicates harboring the TFT haplotype , 45 virgin females were collected from a genetic strain in which the BRO haplotype is placed alongside an isogenic nuclear background , called w1118 ( Bloomington #5905; Ryder et al . , 2004 ) . These females were then crossed to 50 males from the Dahomey lab population . In the next generation , 45 virgin daughters were collected from each strain replicate , and again backcrossed to 50 males from the Dahomey lab population . This backcrossing procedure progressed for 12 generations . To initiate replicates harbouring the DAH haplotype , the same procedure was followed , but the haplotypes were sourced directly from the Dahomey lab population , via an initial mating of 45 virgin females to 50 males collected from Dahomey . Then in the next generation , the virgin daughters of this cross were backcrossed to males of the stock Dahomey population , and this procedure repeated each generation ( Figure 4 ) . 10 . 7554/eLife . 23551 . 015Figure 4 . Experimental breeding scheme . The TFT-bearing BRO haplotype and the putatively healthy DAH haplotype were introgressed into the outbred nuclear background Dahomey in three independent replicates ( i . e . BRO:Dahomey1-3; DAH:Dahomey1-3 ) . Experimental populations were established by supplementing DAH:Dahomey test populations ( DAH:Dahomey1-3 ) with varying contributions of fly pairs bearing the TFT haplotype ( 0% , 25% , 50% , 75% ) sourced from the corresponding BRO:Dahomey replicate population ( i . e . DAH:Dahomey1/BRO:Dahomey1; DAH:Dahomey2/BRO:Dahomey2; DAH:Dahomey3/BRO:Dahomey3 ) . For each of the three biological replicates and for each treatment class ( 0% , 25% , 50% , 75% ) , we established seven technical replicate populations . These experimental populations were further duplicated , one cohort providing populations for the regulated population size approach ( Experiment 1 ) , and one cohort for the fluctuating population size approach ( Experiment 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23551 . 015 Theoretically , each generation of introgression of the TFT haplotype into the Dahomey nuclear background increases the contribution of the Dahomey nuclear background by 50% , and thus , after 12 generations of introgression the contribution of Dahomey nuclear alleles to each TFT population replicate should have exceeded 99 . 98% . Thus , following the introgression procedure , we had six replicate strains , each of which contained a large representative sample of the nuclear alleles segregating within the Dahomey laboratory population; three of which however harbored the BRO haplotype harboring the candidate TFT mutations ( denoted BRO/Dahomey ) , and the other three the DAH haplotype ( denoted DAH/Dahomey ) . All experimental crosses involved 4-day-old flies in 40 ml vials containing 6 ml of potato-dextrose-agar medium ( at 25 . 0°C and a 12 hr: 12 hr light:dark cycle ) . Populations were maintained at a density of 80 flies per vial because at this density fly populations are sufficiently large to be stably maintained while limiting nutritional stress ( e . g . food scarcity ) which otherwise may impact development , fitness and behavior of affected fly populations ( Santos et al . , 1994 ) . Further , sensitivity analyses within our a-priori demographic modeling , which informed our TFT treatments , showed that predictions were robust with respect to a modelled lab population size of 80 individuals ( see Supplementary information in Wolff et al . , 2016b ) . To test the capacity of the candidate TFT haplotype to effect population suppression within a multi-generational framework , in each experiment we established experimental target populations ( DAH/Dahomey ) that were seeded with varying contributions of the TFT haplotype-bearing BRO/Dahomey individuals ( a TFT treatment with four levels ) . All populations were established with 80 adult fruit flies at 1:1 sex ratio , and with the TFT haplotype contributing to 0% ( control ) , 25% , 50% , and 75% of the starting population . All four levels of the TFT treatment were established for each of the three biological BRO/Dahomey replicates , with each BRO/Dahomey replicate matched to a corresponding DAH/Dahomey replicate . Within each strain replicate , each level of the TFT treatment was itself replicated seven times ( i . e . three biological replicate BRO:Dahomey populations × four treatment levels × seven technical replicates = 84 experimental populations; see Figure 4 ) . We conducted two separate experiments , which ran concurrently . In the first experiment , we matched the conditions ( in terms of density ) under which fly populations are typically maintained in our laboratory , in which egg numbers per generation are carefully regulated , and thus , the population size is maintained around a constant density of 80 individuals at both juvenile and adult life stages ( hereafter referred to as: Experiment 1; Regulated population size ) . Each experimental population was initiated using virgin flies at three days of age . The flies of each experimental population were then allowed 24 hr to mate , after which flies were transferred into vials with fresh food substrate for 4–6 hr for ovipositioning on day 4 , until each vial contained in excess of 80 eggs . Immediately after ovipositioning , flies were collected and stored at −20°C , and the number of eggs per population reduced to 80 eggs by manually removing surplus eggs in each vial . To select eggs for retaining , the circular-shaped food source was divided into stripes . Eggs within each stripe were then counted starting from one side of the circle moving toward the opposite side of the circle until 80 eggs had been counted . This way , eggs were selected from the periphery through to the center of the food source regardless of egg density . Surplus eggs were discarded . We also aimed to minimize sampling effects by regulating egg density prior to the manual cull of eggs , by having females lay eggs over a short time period of only 4 to 6 hr . This ensured a sufficient number of eggs per vial , and also ensured that the majority of eggs contained in any one vial was used to give rise to the next generation . Each clutch of eggs was counted twice to minimize error in egg counts . Despite this precaution , eggs can be covered by food and escape detection , thus for 11 out of 840 vials ( 1 . 3% ) offspring counts >80 were observed . The 80 eggs of each experimental population were then allowed to develop until eclosion to give rise to the next generation . Once eclosed into adults , flies of each population were transferred onto fresh food daily , until 4 days of age , when the next round of ovipositioning and egg-trimming occurred . We continued this procedure for 10 consecutive generations in total . In each generation , the total number of flies eclosing per population , from the initial pool of 80 eggs , was counted . The experiment was conducted blind to the identity of the experimental vials . In the second experiment , all procedures were identical with the exception that the 84 experimental populations were not subjected to egg-trimming each generation ( hereafter referred to as: Experiment 2; Fluctuating population size ) . Instead , once established , each ovipositioning period per experimental population was stopped when around 50% of the population vials were estimated to contain in excess of 80 eggs . Once this threshold was reached , flies were collected and stored at −20°C , eggs were left to develop to eclosion , and the number of eclosed flies per population in each generation was counted . This protocol diverges from the rearing conditions under which our laboratory populations are typically maintained , with the populations experiencing high levels of competition at both larval and adult stages for the 6 ml of available food , which routinely led to generations of population contractions , interspersed by generations of population expansions . This experimental design was used as an approximation for the demographic conditions natural populations may be exposed to , where single populations potentially transition through severe population bottlenecks , in which genetic drift would be expected to play a larger role in shaping frequencies of co-occurring mtDNA haplotypes relative to the regulated conditions of Experiment 1 . DNA from single females was extracted in 96-well format using Wizard Genomic DNA Purification Kit ( Promega , Madison , WI 53711 , USA ) following the manufacturer’s instructions for single sample extractions , and using a third of recommended volumes to adjust for 96-well plate well volume . We invested most genotyping resources on Experiment 1 , extracting DNA from seven females in each of three generations ( 1 , 5 , and 10 ) from each of the 63 populations that had been seeded with the TFT haplotype BRO ( 25% , 50% , and 75% BRO:Dahomey contribution; sample size: 63 populations × seven females × three generations = 1323 females ) . For Experiment 2 , we extracted DNA from nine females , all from generation 10 only , from each of the 63 populations that had been seeded with the TFT haplotype ( 25% , 50% , and 75% BRO:Dahomey contribution; sample size: 63 populations × nine females × one generation = 567 females ) . Populations established with DAH:Dahomey only ( control populations ) were not genotyped for either experimental cohort ( their genotypes were fixed at 0% TFT haplotype ) . All DNA extracts were adjusted to DNA concentrations of 5 ng*µl−1 and a final volume of 50 µl per sample . Genotyping was conducted using a custom iPLEX Gold genotyping assay on the Sequenom MassARRAY Analyzer four system at Geneworks Pty Ltd , Thebarton , Australia . The structure of the data is outlined in Figure 4 . We fitted linear mixed models to phenotype data ( offspring number ) , and generalized linear mixed models to genotype data ( frequency of the TFT haplotype ) , using the lme4 package 1 . 1 . 12 ( Bates et al . , 2015 ) in R 3 . 0 . 3 ( R Development Core Team , 2013 ) . In the phenotypic data analyses , the response variable ( number of offspring produced ) was modeled using a Gaussian distribution . Although this data is strictly count data , it conformed to a normal rather than Poisson distribution as expected of large sample sizes under the Central Limit Theorem . For example , the analysis of Experiment 1 contained only 1 zero value in the dataset . Although the analysis of Experiment 2 contained 15 zero values , removal of these values did not change the statistical inferences; and importantly , the residuals of these models were normally distributed . For the linear mixed models , fixed effects parameters were estimated using maximum likelihood estimation , and random effects were estimated using restricted maximum likelihood estimation . In the genotype data analyses , the frequency of the TFT haplotype was modeled as a binomial vector comprising the number of TFT haplotypes genotypes per vial and the number of wild-type haplotypes , using a binomial distribution and logit link . For both analyses , we treated both the identity of the experimental population ( i . e . individual vials ) nested within Biological Replicate , and Biological Replicate , as random effects , and both TFT treatment ( control , 25% TFT , 50% TFT , 75% TFT ) , and generation ( F1–F10 ) , as fixed effects . The TFT treatment control level was removed from analyses modeling the TFT haplotype frequency across the TFT treatments , since this level was invariably zero and its inclusion violated the model assumption of homogeneity of variance across classes . Generation was not added as a factor to the model of TFT haplotype frequency for Experiment 2 , since genotyping was only conducted at generation 10 in this experiment . Consequently , each data-point represented that of a specific experimental population ( i . e . 63 data points in the dataset , equaling 63 experimental populations ) , and thus , experimental population represented an observation-level effect in the models of Experiment 2 . For the binomial models of TFT haplotype frequency , the blmeco package ( Korner-Nievergelt et al . , 2015 ) indicated overdispersion; ‘experimental population’ ( an observation-level random effect ) was thus added to the models to account for this . We also examined the correlation between the number of offspring produced per experimental population and the TFT haplotype frequency , for both experiments . We fitted linear mixed models , with the number of offspring produced per experimental population as the response variable , and the TFT frequency of the same population as a fixed factor ( of eight levels in Experiment 1 , and 10 levels in Experiment 2 ) , with the identity of the experimental population nested within the Biological Replicate , Biological Replicate , and generation number added as random effects to the model of Experiment 1 , and with Biological Replicate added as a random effect for the model of Experiment 2 . Significance of fixed effects in each model was assessed using Type III sums-of-squares , χ2 tests in the car package of R ( Fox and Weisberg , 2011 ) .	Insect and other animal pests pose some of the greatest challenges to biodiversity , global economies and human health . The environmental and agricultural losses caused by pests have been estimated at 120 billion US dollars a year in the US alone . Many pests also spread diseases , such as dengue fever and malaria . A variety of different strategies are used to control pests , but their effects are generally short-lived and they are often ineffective when pest numbers are low . Furthermore , many of these strategies are harmful to other wildlife , such as bees . Most of the DNA within an animal cell is contained within a structure called the nucleus . However , some DNA is also found within other compartments called mitochondria . The Trojan Female technique has been proposed as a new strategy to control insect pests that harnesses naturally-occurring changes ( known as mutations ) in this mitochondrial DNA ( or mtDNA for short ) . Introducing mutations that lower the fertility of males , but have no effects on females , into a pest population should , in theory , lead to a long-lasting decline in the size of the population , even if it is relatively small to begin with . Wolff et al . tested this theory in fruit flies , which are often used as models of insects and other animals in research projects . Adding female fruit flies carrying a mutation in mtDNA that lowers male fertility ( known as “Trojan Females” ) into populations of fruit flies reduced the size of the population over several generations . The mutation was maintained in the population for at least ten generations , and no “rescue” mutations evolved in the nuclear DNA to compensate for the mtDNA mutation . This indicates that the Trojan Female technique could be effective at controlling pests , without the need for Trojan Females to be repeatedly released into the populations . The next steps following on from this work are to test this approach in economically important pest species , and to find out whether the approach is effective in various environments outside the laboratory . If these findings do indeed translate into these pests , then the Trojan Female technique may have the potential to be used to control a wide variety of different pest species from mosquitos through to rats .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "genetics", "and", "genomics" ]	2017	Introduction of a male-harming mitochondrial haplotype via ‘Trojan Females’ achieves population suppression in fruit flies
Histone tail modifications can greatly influence chromatin-associated processes . Asymmetrically modified nucleosomes exist in multiple cell types , but whether modifications on both sister histones contribute equally to chromatin dynamics remains elusive . Here , we devised a bivalent nucleosome system that allowed for the constitutive assembly of asymmetrically modified sister histone H3s in nucleosomes in Saccharomyces cerevisiae . The sister H3K36 methylations independently affected cryptic transcription in gene coding regions , whereas sister H3K79 methylation had cooperative effects on gene silencing near telomeres . H3K4 methylation on sister histones played an independent role in suppressing the recruitment of Gal4 activator to the GAL1 promoter and in inhibiting GAL1 transcription . Under starvation stress , sister H3K4 methylations acted cooperatively , independently or redundantly to regulate transcription . Thus , we provide a unique tool for comparing symmetrical and asymmetrical modifications of sister histone H3s in vivo . In eukaryotes , chromatin carries both genetic and epigenetic information that controls multiple cellular processes , such as DNA replication , transcription and genome organization ( Berger , 2007; Lawrence et al . , 2016; Papamichos-Chronakis and Peterson , 2013 ) . The basic unit of chromatin is the nucleosome , which comprises ~147 bp of DNA and a histone octamer formed by two copies of histone H2A-H2B and H3-H4 heterodimers ( Bentley et al . , 1984; Kornberg and Thomas , 1974; Luger et al . , 1997; Oudet et al . , 1975 ) . The packaging of DNA into nucleosomes affects sequence accessibility , and nucleosomes therefore regulate the activity of DNA-binding proteins ( Lee et al . , 1993; Wasylyk and Chambon , 1979 ) . Histones also appear to protect DNA from breaking and maintain the fidelity of both replication and transcription ( Carrozza et al . , 2005; Govind et al . , 2007; Joshi and Struhl , 2005; Keogh et al . , 2005; Pinskaya et al . , 2009 ) . The regulation of nucleic acid metabolism by nucleosomes is mediated through multiple post-translational modifications ( PTMs ) , such as methylation , acetylation , phosphorylation , and sumoylation ( Lawrence et al . , 2016 ) . Histone lysine methylation , especially on histone H3 , regulates chromatin structure and transcription ( Ng et al . , 2002; Vermeulen and Timmers , 2010; Wagner and Carpenter , 2012 ) . In budding yeast , the best-studied methylations on histone H3 are methylation of lysine at amino acid positions 4 , 36 , and 79 ( H3K4 , H3K36 and H3K79 , respectively ) . H3K4 di- and tri-methylation ( H3K4me2/3 ) is catalyzed by the Set1 complex ( also called the COMPASS complex ) and is associated with steady-state gene transcription; thus , H3K4me2/3 is considered to be an ‘activating’ mark in mammals . Conversely , in budding yeast , most of the evidence indicates that H3K4 methylation is a repressive mark ( Shilatifard , 2006; Weiner et al . , 2012 ) . H3K36 tri-methylation ( H3K36me3 ) by Set2 directs the deacetylation of histones , predominantly at the 3’ portion of gene open reading frames ( ORFs ) , to suppress spurious intragenic transcription initiation ( Carrozza et al . , 2005 ) . Methylation of H3K79 ( H3K79me ) affects telomeric heterochromatin structure because mutations at H3K79 as well as inactivation of its methyltransferase , Dot1 , lead to loss of telomere silencing ( Jones et al . , 2008; Ng et al . , 2002 ) . The functions of each modification are largely dissected by using histone mutations in combination with the inactivation of corresponding methyltransferases , under which circumstances the modifications on both sister histones are simultaneously removed , making it difficult to study the crosstalk between modifications on sister histones . Although two copies of each histone in a nucleosome possess identical protein sequences , histone-modification enzymes do not always modify sister histones simultaneously ( van Rossum et al . , 2012; Voigt et al . , 2013 ) . For example , symmetrical modification of histone lysines within a nucleosome is not globally required in HeLa cells ( Chen et al . , 2011 ) . In addition , in different cell types , a significant number of nucleosomes contain asymmetrically modified sister histones ( Fisher and Fisher , 2011; Mikkelsen et al . , 2007; Voigt et al . , 2012 ) . Furthermore , asymmetrically modified nucleosomes are present in embryonic stem cells but are symmetrically modified upon differentiation ( Voigt et al . , 2012 ) . Each of these studies suggests that sister histones within a single nucleosome may function independently in gene regulation . A synthetic system for the generation of asymmetrically modified nucleosomes has been used to study histone PTM crosstalk in vitro ( Lechner et al . , 2016 ) , but the lack of a genetic model system for studying asymmetric histone modifications in vivo has prevented exploration of the biological significance of this previously documented phenomenon . To investigate the individual contributions of sister histones and their modifications to chromatin structure and function , we employed a protein engineering strategy to mutate both copies of histone H3 in their interaction interface . After screening for mutants that were able to form histone H3 heterodimers but not H3 homodimers , we successfully set up a bivalent nucleosome system in the budding yeast Saccharomyces cerevisiae . By using this unique system , we validated the establishment of asymmetrically methylated H3K4 , H3K36 or K3K79 in chromatin in yeast in vivo . Furthermore , we examined the functions of asymmetrically modified sister histones in the regulation of chromatin structure and gene transcription . Our results revealed that modifications such as H3K4me , H3K36me or K3K79me on sister histone H3s could be independent of each other . In addition , the same modifications on both sister H3 histones can affect transcription in a cooperative , independent or redundant manner . Our study provides the first picture of the individual contributions of sister histones to chromatin dynamics in vivo . In S . cerevisiae , each canonical histone is encoded by two genes . H3 is encoded by HHT1 and HHT2 , and H4 is encoded by HHF1 and HHF2 . The histone genes are organized into a pair of divergently transcribed loci with HHT1-HHF1 and HHT2-HHF2 linked together . Owing to redundancy , deletion of either locus does not cause lethality ( Dollard et al . , 1994 ) . As asymmetrical modifications have previously been reported on histone H3 in vivo ( Voigt et al . , 2012 ) , we began by examining H3 . Previous structural work revealed that two molecules of histone H3 interact through their carboxy-terminal four-helix bundle to form a homodimer ( Luger et al . , 1997; Ramachandran et al . , 2011; White et al . , 2001 ) ( Figure 1A ) . We performed site-directed mutagenesis on the Ala110 , Ala114 and Leu130 residues of the HHT1 gene . These residues were chosen because they were spatially close and within the bundle region that interacts to form the H3 homodimer ( Luger et al . , 1997; Ramachandran et al . , 2011; White et al . , 2001 ) . These neutral amino acids were mutated to acidic or basic residues to make them electronegative or electropositive under physiological conditions . We reasoned that if we created an H3 allele with an electronegative ( or electropositive ) interface , it would not form homodimers , but it would interact with a different H3 allele with an electropositive ( or electronegative ) interface , thereby creating a heterodimer ( Figure 1A ) . Yeast cells lacking chromosomal HHT1 and HHT2 genes but containing the HHT1 gene on a counter-selectable URA3 plasmid were transformed with plasmids carrying the mutated histone H3 genes . We then screened for histone H3 mutants that did not support cell viability when loss of the wild-type ( WT ) HHT1 gene was counter-selected using 5-fluoroorotic acid ( 5-FOA ) ( Figure 1B ) . Only the H3 mutant bearing the A110E mutation survived ( Figure 1C ) , suggesting that the other 14 histone H3 mutants could not form a homodimer . Next , yeast cells were co-transformed with the pairwise plasmids carrying these 14 mutated histone H3 genes ( Figure 1D ) . Notably , only the H3A110D and H3L130H pairing was able to support cell viability on a 5-FOA plate ( Figure 1E; Figure 1—figure supplement 1 ) , allowing us to infer that the H3A110D and H3L130H mutants form a heterodimer that could be assembled into functional nucleosomes in vivo . Considering that histidine's positive charge is weakened under physiological pH conditions and may increase the risk for H3L130H self-interaction , we used the relatively weaker ADE3 promoter to reduce the expression of H3L130H ( Agez et al . , 2007; Antczak et al . , 2006 ) . This strain will be hereafter referred to as the H3D/H3H strain . To confirm that mutant histones H3A110D and H3L130H equally assembled into nucleosomes , we epitope-tagged one copy of H3 in H3D/H3H cells with Myc . After preparing mono-nucleosomes ( Figure 2—figure supplement 1 ) , we performed immunoprecipitations with an anti-Myc antibody and examined both Myc-tagged and untagged histone H3 . In the control , the chromatin from both the myc-H3 strain and the untagged H3 strain was mixed , and the immunoprecipitation of mononucleosomes using the anti-Myc antibody did not pull down untagged H3 ( Figure 2A , second lane ) . As the anti-H3 N-terminal antibody could not recognize Myc-tagged histone H3 ( Figure 2A ) , we normalized immunoprecipitated myc-H3L130H and myc-H3A110D to the same level . The amounts of the co-immunoprecipitated complementary H3A110D and H3L130H histones were identical ( Figure 2A ) , reflecting an equal incorporation of H3A110D and H3L130H into mononucleosomes in H3D/H3H cells . Next , we examined the ratio of H3A110D to H3L130H and the nucleosome positioning at the GAL1-10 gene locus in the H3D/H3H cells . GAL1-10 intergenic chromatin consists of a non-nucleosomal , UAS-containing hypersensitive region ( Lohr , 1984; Lohr and Hopper , 1985 ) surrounded by positioned nucleosomes ( Lohr and Lopez , 1995; Lohr et al . , 1987 ) . A chromatin immunoprecipitation ( ChIP ) assay showed almost the same enrichment of H3A110D and H3L130H at the GAL1 gene promoter ( Figure 2B ) , supporting our conclusion that mutant histones H3A110D and H3L130H were assembled into nucleosomes at a ratio of 1:1 in vivo . MNase digestion of the GAL1-10 promoter revealed that the nucleosome array on the GAL10 side of the UAS region displayed a similar digestion pattern in H3D/H3H and WT cells , but the nucleosome array on the GAL1 side showed a more evenly digested pattern in WT cells than in H3D/H3H cells ( Figure 2C ) , suggesting altered nucleosome stability in the GAL1 region in H3D/H3H cells . We next determined the functional viability of the H3D/H3H mutant using the histone shuffle strain ( LHT001 ) as a WT control . H3D/H3H mutant and WT cells exhibited identical growth rates in yeast extract peptone dextrose ( YPD ) medium at 23°C , 30°C and 37°C . In addition , when H3D/H3H cells were challenged by rapamycin ( data not shown ) or DNA-damage reagents , such as phleomycin or methyl methanesulfonate ( MMS ) , they showed nearly the same sensitivity as WT cells ( Figure 2D ) . Interestingly , compared with WT cells , H3D/H3H cells showed a reduced growth rate when cultured in raffinose or glycerol medium ( Figure 2E ) . We then checked the levels of multiple histone PTMs in WT and H3D/H3H strains by western blot and found no significant differences ( Figure 2F ) . Further , we performed a genome-wide RNA-Seq assay to examine the gene expression profiles in WT and H3D/H3H strains . Statistical analysis confirmed the reproducibility of the RNA-Seq results in each strain ( Figure 2—figure supplement 2 ) . The global gene expression profile of the H3D/H3H strain was found to be very similar to that of the WT strain ( Figure 2G ) , but we did see some genes with expression levels that varied between the H3D/H3Hand WT strains . Through Gene Ontology analysis ( see the Materials and methods for details ) , we found that most of the outliers were downregulated by histone H3 mutations . Interestingly , the genes encoding cytochrome-c reductase activity and ATPase activity were among the outliers ( Tzagoloff et al . , 1975 ) ( Figure 2—source data 2 ) . This finding might provide an explanation for the reduced growth of the H3D/H3H strain when glycerol was used as the carbon source ( Figure 2E ) . Taken together , the observations presented in Figure 2 indicated that the H3D/H3H strain behaved similar to the WT strain under most , but not all of the tested circumstances; thus , this strain provides a unique and valid system for analyzing asymmetrically modified sister histones . To address whether there is crosstalk between the amino-terminal tails of sister histone H3s in one nucleosome , we constructed strains that lacked the N-terminal 4–15 amino acids on one or both sister H3 histones ( Mann and Grunstein , 1992 ) . The H3DΔ4–15/H3H and H3D/H3HΔ4–15 strains contained one copy of N-terminal-deleted H3 , resulting in asymmetrically deleted histone H3 ( Figure 3A ) . The H3DΔ4–15/H3HΔ4–15 strain containing two copies of N-terminal-deleted H3 was also constructed and used as a negative control . The nucleosomes of the H3D/H3H ( treated as WT hereafter ) and mutant strains were precipitated , and western blots were performed to examine the levels of histone H3 N-terminal and K4 tri-methylation . Both histone H3 N-terminal and H3K4me3 signals in H3DΔ4–15/H3H and H3D/H3HΔ4–15 cells were reduced to approximately half of those observed in H3D/H3H cells ( Figure 3B and C ) . These results indicated that H3 N-terminal deletion on one sister H3 did not influence H3K4 methylation on the other . As the genes for H3A110D and H3L130H encoded compatible and functional histone H3 proteins , we anticipated that the substitution of K with R on one sister H3 would largely mimic unmethylated K . Thus , asymmetrically modified nucleosomes could be assembled in chromatin in vivo . To test this idea , we first introduced the K4R mutation into H3A110D ( H3DK4R ) or H3L130H ( H3HK4R ) in the H3D/H3H strain ( Figure 4A ) . Western blotting showed that H3K4me3 in H3DK4R/H3H or H3D/H3HK4R cells was approximately 50% lower than that in H3D/H3H cells , whereas little difference in H3K36me3 was detected among the tested strains ( Figure 4B and C ) . Therefore , these results suggest that the hybrid strains contain only mimics of asymmetrically deposited K4me3 . For sister H3 histones in a nucleosome , a lack of K4me3 in one tail did not influence K4me3 in the other tail , consistent with the observation in Figure 3B . In addition , H3K4me3 and H3K36me3 were independent of each other because the decrease in H3K4me3 did not alter the level of H3K36me3 . Cells that lack histone H3K4 methylation have an increased GAL1 induction level ( Pinskaya et al . , 2009 ) . To assess the effect of asymmetrical H3K4me3 on transcription , we assessed GAL1 mRNA levels in K4R mutant cells . Compared with H3D/H3H cells , H3DK4R/H3H and H3D/H3HK4R single-tail mutant cells showed a two-fold increase in GAL1 mRNA levels . Compared with single-tail mutant cells , H3DK4R/H3HK4R double-tail mutant cells showed a further two-fold increase in GAL1 mRNA levels ( Figure 4D and E ) . GAL1 mRNA levels were inversely proportional to H3K4me3 levels at the GAL1 promoter ( Figure 4F ) , suggesting a tight correlation between induction levels and H3K4me3 abundance . We also examined the enrichment of Gal4 binding to the GAL1 promoter using a ChIP assay . Gal4 is the primary activator of GAL1 transcription ( Johnston , 1987 ) . A moderate level of Gal4 recruitment to the GAL1 promoter was observed in the asymmetrical H3DK4R/H3H and H3D/H3HK4R mutant strains compared with that in their symmetrical H3D/H3H and H3DK4R/H3HK4R counterparts ( Figure 4G ) . Therefore , each K4me3-modified sister histone H3 contributed independently to GAL1 gene transcription , which is probably recognized and read by the GAL1 activator Gal4 . Set1C in yeast contains eight subunits , including Set1 , Spp1 and Sdc1 , and is responsible for methylating histone H3K4 ( Dehé and Géli , 2006; Roguev et al . , 2001 ) . Deletion of SET1 eliminates H3K4 mono- , di- and tri-methylation; deletion of SPP1 affects only H3K4 tri-methylation; and deletion of SDC1 affects di- and tri-methylation of H3K4 ( Pinskaya et al . , 2009 ) . To address which type of asymmetrical H3K4 methylation affects GAL1 transcription , and to confirm that the changes in gene expression were due to asymmetrical H3K4 methylation instead of the K4R mutation , we knocked out SET1 , SPP1 and SDC1 in the H3D/H3H , H3DK4R/H3H , H3D/H3HK4R and H3DK4R/H3HK4R strains and examined GAL1 levels in galactose medium . As the data show , loss of SPP1 , SDC1 and SET1 led to the upregulation of GAL1 transcription , which is consistent with previous findings ( Pinskaya et al . , 2009 ) . Meanwhile , an intermediate level of GAL1 expression was seen in spp1∆ H3DK4R/H3H and spp1∆ H3D/H3HK4R cells , whereas no significant difference was found in either sdc1∆ or set1∆ mutants ( Figure 4H ) . As distinguishing between the effects of H3K4me2 and H3K4me3 is difficult , we concluded that H3K4me2/3 but not mono-methylation of H3K4 on sister H3s contributed the most to GAL1 regulation . As both asymmetrical H3 N-terminal deletion and asymmetrical H3K4me3 were successfully assembled in chromatin , we constructed mutants that mimicked asymmetrical H3K36me . A K36R mutation was introduced into H3A110D ( H3DK36R ) or H3L130H ( H3HK36R ) in the H3D/H3H strain ( Figure 5A ) . The level of H3K36me3 and H3K4me3 on chromatin was examined by western blotting . When compared with H3D/H3H cells , H3DK36R/H3H or H3D/H3HK36R cells showed an approximately 50% decrease in H3K36me3 , whereas little difference in H3K4me3 was detected among the tested strains ( Figure 5B and C ) . These data indicated that the H3DK36R/H3H or H3D/H3HK36R mutants contained only mimics of asymmetrically deposited K36me3 , and loss of K36me3 on one tail did not affect K36me3 on the other tail . In addition , in agreement with the data shown in Figure 4B , H3K36me3 and H3K4me were independently regulated chromatin modifications . H3K36me3 directs deacetylation of histone H4 in gene-coding regions to suppress spurious intragenic transcription ( Carrozza et al . , 2005 ) . To address whether H3K36me3 on both sister histone H3s contributed to the regulation of cryptic transcription , we tested the level of intragenic initiation in the H3K36R mutants within the FLO8 , PCA1 and STE11 genes . Each of these genes is regulated by K36 methylation . Northern blot analysis showed that the loss of K36 methylation on H3 tails resulted in short transcripts of the tested genes , consistent with previous findings ( Carrozza et al . , 2005; Li et al . , 2007 ) . Compared with H3D/H3H cells and symmetrically mutated H3K36 cells , asymmetrical H3DK36R/H3H and H3D/H3HK36R cells exhibited an intermediate level of short transcripts ( Figure 5D ) . We next used anti-acetylated histone H4 antibodies to perform a ChIP assay on the 3’ ORF of the FLO8 , PCA1 and STE11 genes . H4 acetylation ( H4ac ) levels in H3DK36R/H3H and H3D/H3HK36R cells were intermediate relative to those in H3D/H3H cells and symmetrically mutated H3K36 cells . Moreover , H4ac levels were inversely correlated with H3K36me3 levels in the same region ( Figure 5E and F ) . In the absence of Set2 , the level of H4ac in the tested genes showed no significant differences in H3D/H3H , H3DK36R/H3H , H3D/H3HK36R and H3DK36R/H3HK36R cells ( Figure 5G ) . These observations indicated that the regulation of accurate transcription initiation was sensitive to the magnitude of H3K36me3 . Accordingly , H4ac levels were regulated by H3K36me3 on both sister histones . In light of these data , we concluded that H3K36me3 on either sister histone played an independent regulatory role in suppressing spurious intragenic transcription . H3K79 methylation regulates gene silencing in some telomere-proximal regions ( Takahashi et al . , 2011 ) . To address whether the H3K79 methylation of both sister histones is required to maintain silent chromatin near telomeres , we used strains in which H3K79 could be methylated at either one ( H3DK79R/H3H and H3D/H3HK79R ) or none ( H3DK79R/H3HK79R ) of the H3 sister histones ( Figure 6A ) . Western blot analysis revealed that H3K79me2/3 levels in H3DK79R/H3H and H3D/H3HK79R cells were approximately 50% lower than those in H3D/H3H cells ( Figure 6B and C ) , suggesting the incorporation of asymmetrical H3K79me into chromatin and that the methylation of K79 occurs independently on each sister H3 . We examined the transcription levels of COS12 , ERR1 and ERR3 , which are located proximal to the telomeric ends of chromosomes VIIL , XVR and XIIIR , respectively ( Takahashi et al . , 2011 ) . As expected , the K79R mutations on both sister H3s resulted in decreased silencing of those genes . Surprisingly , H3DK79R/H3H and H3D/H3HK79R cells containing asymmetrical H3K79me exhibited the same level of silencing loss as that of H3DK79R/H3HK79R or sir2∆ cells ( Figure 6D ) . A ChIP experiment confirmed that K79me levels at the promoters of the genes tested in H3DK79R/H3H and H3D/H3HK79R cells decreased to approximately half of those in H3D/H3H cells ( Figure 6E ) . Accordingly , the H4ac level in the ORF region was upregulated in K79R mutated cells ( Figure 6F ) . Collectively , these data reveal that K79me marks on both sister H3s act cooperatively to maintain gene silencing near telomeres . H3K4 , H3K36 and H3K79 methylation affects DNA double-strand break ( DSB ) repair ( Faucher and Wellinger , 2010; Jha and Strahl , 2014; Pai et al . , 2014 ) . Therefore , we examined the regulatory role of asymmetric H3K4 , H3K36 or H3K79 methylation in DSB repair . Mutant cells bearing asymmetrical methylated or non-methylated H3K4 , H3K36 or H3K79 were serially diluted and spotted onto plates containing various genotoxic chemicals , including phleomycin , hydroxyurea ( HU ) or MMS . H3K4R or H3K79R mutations on either one or two sister histones reduced cell growth in the presence of the tested genotoxins . Notably , the H3DK36R/H3HK36R mutant was hypersensitive to phleomycin and mildly sensitive to MMS . Compared with the wild type ( H3D/H3H ) and corresponding double-tail mutant , single-tail H3K36R or H3K79R mutants displayed an intermediate level of sensitivity to the genotoxic agents . The H3K4 mutants showed a similar level of sensitivity to HU and MMS , but single-tail H3K4R mutants displayed less growth in response to phleomycin treatment than did double-tail H3K4R mutants ( Figure 7 ) . Together , these observations suggest that , in response to DNA damage , H3K36me and H3K79me marks on sister histones functioned independently , whereas H3K4me marks on sister histones functioned cooperatively . Because the type of DNA damage triggered by the different genotoxic agents and the mechanisms of repair differ , we propose that the combination of sister histone modifications may influence DNA repair in different ways . Chromatin regulators do not appear to affect steady-state transcription , but instead are required for transcriptional reprogramming induced by environmental cues ( Weiner et al . , 2012 ) . For example , the genome-wide gene transcription profile of H3K4A cells was nearly the same as that of WT cells when the cells were grown under normal conditions , whereas differences were observed when the cells were challenged by multiple stress conditions ( Weiner et al . , 2012 ) . To further unravel the genome-wide function of sister H3K4me on transcription , we shifted the cultures of H3K4R mutants and H3D/H3H strains from 2% to 0 . 05% glucose in the medium , which mimics calorie restriction . After the cells were grown in 0 . 05% glucose medium for an hour , we performed RNA-Seq to examine the genome-wide gene-induction profiles , which are presented as fold-change ( level of induction ) . The fold-change value refers to the level of transcription in the induced strains divided by that in the uninduced strains . Of the 6000 genes in the yeast genome , approximately 2500 were altered by the H3K4 to R mutation in response to glucose starvation . Over 1500 genes’ fold-change ( MID , as defined in Materials and methods ) in both asymmetrical K4R mutants ( H3DK4R/H3H and H3D/H3HK4R ) fell between those of H3D/H3H cells and double K4R mutants ( Figure 8—figure supplement 1A ) . Statistical analysis by t-test and a gene skewness score ( GSS ) model described in the Materials and methods revealed that 22 genes’ fold-changes in asymmetrical K4R mutant ( H3DK4R/H3H and H3D/H3HK4R ) cells were nearly the same as those in symmetrical K4R mutant ( H3DK4R/H3HK4R ) cells ( Figure 8A , B; Figure 8—figure supplement 1B , Cluster I ) , indicating cooperativity of sister K4me at these loci . The fold-changes of 191 genes in asymmetrical K4R mutant ( H3DK4R/H3H and H3D/H3HK4R ) cells exhibited an intermediate state between those in symmetrical K4R mutant ( H3DK4R/H3HK4R ) and WT ( H3D/H3H ) cells ( p<0 . 05 ) ( Figure 8A , B; Figure 8—figure supplement 1B , Cluster II ) , suggesting that K4me on sister H3s independently regulates the expression of these genes . The fold-changes of 158 genes in asymmetrical K4R mutant ( H3DK4R/H3H and H3D/H3HK4R ) cells were nearly the same as those in WT H3D/H3H cells ( Figure 8A , B; Figure 8—figure supplement 1B , Cluster III ) , indicating redundancy of sister K4me at these loci . An approximately 50% decrease in H3K4me3 in asymmetrical K4R mutants was confirmed in the 5' ORFs of the YOR008C , YMR315W and YLR359W genes , which belong to the three clusters ( Figure 8—figure supplement 1C–E ) . These assessments suggest that , in response to glucose starvation stress , H3K4me on two sister histones in different gene loci impose their effects on transcription in a cooperative ( e . g . , Cluster I ) , independent ( e . g . , Cluster II ) or redundant ( e . g . , Cluster III ) manner . Interestingly , the genes in Cluster I and II were mostly upregulated ( log2fold change >0 ) , while the genes in Cluster III were mostly downregulated ( log2fold change <0 ) ( Figure 8B ) , suggesting that under glucose starvation stress , the transcription of upregulated genes may require more subtle regulation mechanisms , such as asymmetrical modification of sister histones . Although the H3K4R mutation in chromatin largely mimics K4me0 , it is not the same as K4me0 , and the phenotypes seen in H3DK4R/H3H and H3D/H3HK4R cells might not result from loss of K4me . To address this issue , we examined the genome-wide gene expression profile of set1∆ cells under glucose starvation , and compared its fold-change with that of K4R mutants ( Figure 8—figure supplement 1F ) . Many of the genes in Clusters I , II and III overlapped with genes that are regulated by SET1 deletion ( Figure 8C , I∩set1∆ , II∩set1∆ and III∩set1∆ , respectively ) , indicating that these overlapping genes are most probably regulated by K4me rather than the K4R mutation on sister H3s . To determine which pathways were regulated by asymmetrical K4me on sister H3s in response to glucose starvation , we carried out KEGG pathway analysis ( Huang et al . , 2009a , 2009b ) . Nine of the genes in the I∩set1∆ group and 45 of the genes in the III∩set1∆ group could not be mapped to any specific pathways in the KEGG database . Remarkably , 54 of the genes in the II∩set1∆ group , which is regulated by sister H3K4me , were enriched in the pathways involved in glycometabolism , such as carbon metabolism , TCA cycle , and fructose and mannose metabolism ( Figure 8D ) . KEGG pathway analysis was performed on genes that were regulated by both the H3DK4R/H3HK4R mutation and SET1 deletion under glucose starvation . Interestingly , the three pathways involved in glycometabolism in the II∩set1∆ were also found in the list ( Figure 8E ) . Therefore , under glucose starvation stress , a significant proportion of H3K4me-responsive genes are regulated by the fluctuation of H3K4me levels on sister H3s . Further analysis of the fold-changes of the genes in the glycometabolism-associated pathways revealed a pattern similar to that of Cluster II ( Figure 8—figure supplement 1G ) , suggesting that the independent regulatory mode of sister H3K4me is an important player in response to glucose starvation stress . Collectively , these results support the notions that the on-off regulatory mode for H3K4me is more likely to be applicable to the transcription of genes that do not specifically respond to external stimuli ( e . g . , genes in the I∩set1∆ and III∩set1∆ groups ) , whereas the fine-tuning mode evolved to regulate the transcription of genes involved in stress-responsive pathways ( e . g . , genes in the II∩set1∆ group ) . In a nucleosome , two canonical sister histones display identical sequences , suggesting that they have evolved to play similar roles in the regulation of chromatin structure and function . However , similar to post-translational chemical modifications on any protein , the modifications of histones provide an additional layer of chromatin regulation . Paradoxically , sister histones with either asymmetrical modification or coexistence of activation and repression marks have been found in different cell types ( Fisher and Fisher , 2011; Mikkelsen et al . , 2007; Voigt et al . , 2012 ) , raising the possibility that sister histones in a single nucleosome may function independently . In this study , we took advantage of a yeast system that allows for facile genetic manipulation of histones . We identified H3 mutations that prevented homodimer formation and allowed heterodimer formation . After a series of intentional and systematic screenings , we established a bivalent nucleosome system that enabled us to express and monitor sister histone H3s independently in vivo . Owing to the nature of the nucleosome , which is the basic unit structure of chromatin , any possible indirect effect ( s ) caused by knocking-in a histone mutation cannot be fully excluded from our analysis . Indeed compared with the parental strain , the H3D/H3H strain did display some minor differences in carbon source preference and nucleosome positioning ( e . g . , the GAL1 locus ) ( Figure 2C , E ) . In addition , when a mutation was introduced into one of the sister histone H3s , only half of the H3 could be modified . Therefore , the global level of modified H3 had a maximum value of 50% of the maximum in a normal cell . Discriminating between the biological consequences caused by forced asymmetrical modification and a forced decrease in total modification is difficult . Nevertheless , we have provided biochemical and functional evidence indicating that this unique genetic system is useful for studying asymmetrical modifications on sister histones . Using this system , we found that histone H3K4 methylation on one tail is independent of the other tail on the sister H3 histone ( Figure 3B ) , suggesting that Set1C binds and modifies one tail in a cis fashion . Consistently , mutation of K4R in one of the sister H3s did not affect the methylation of K4 on the other tail ( Figure 4B ) . Interestingly , the methylation of K36 or K79 on two sister H3s was also independent ( Figures 5B and 6B ) . The results of our genetic models are consistent with a previous observation that sister histones are not always modified in the same manner simultaneously ( Chen et al . , 2011 ) . Asymmetrically modified nucleosomes exist on chromatin ( Fisher and Fisher , 2011; Mikkelsen et al . , 2007; Voigt et al . , 2012 ) , but whether these asymmetrical modifications on sister histones function in manner similar to or different from that of symmetrical modifications remains largely unknown . In our study , we observed that K79me on both sister H3 histones was required for silencing telomere-proximal genes through regulation of the acetylation level of histone H4 ( Figure 6D ) , establishing a cooperative role for both sister histones in vivo . This mode of regulation was also seen for the genes in the I∩set1∆ group when the cells were challenged with glucose starvation ( Figure 8C ) . Different from K79me , H3K36me3 on two sister histone H3s did not appear to have a synergistic effect but rather had an additive effect on suppressing spurious transcription ( Figure 5D ) , indicating that two K36me3 marks on sister histone H3s altered chromatin structure independently . The same additive effect was observed in the genes grouped in II∩set1∆ ( Figure 8C ) , as well as in GAL1 transcription levels ( Figure 4E ) . In addition , K4me marks on sister histone H3s redundantly affected the transcription of the genes grouped in III∩set1∆ ( Figure 8C ) . Consistent with our observations in transcription , sister H3K4me exhibited different regulatory modes in response to various DNA-damage reagents ( Figure 7 ) . Thus , our data indicate that modifications on sister histones could employ a cooperative , independent , or redundant mode of regulation of chromatin-associated processes . However , why the genes in different loci are subjected to different regulatory mechanisms remains unclear . One possibility is that different gene loci are targeted by different readers , such as activators and repressors that sense the magnitude of H3K4me differently during transcription . This hypothesis is supported by the data in Figure 4F and G showing that differential marks of K4me3 on two sister histone H3s affected the enrichment of Gal4-activator binding to the GAL1 gene promoter , thereby fine-tuning the transcription of GAL1 . The chromatin readers for different genomic loci have not yet been well characterized , so providing a mechanistic explanation for the different performances of sister histone modifications in every case is difficult . Histones and their modifications are unique to eukaryotes , and they are important in the packaging of DNA into chromatin . From an evolutionary point of view , it may be that the possesion of two identical copies of each histone in the chromatin in eukaryotes rather than one copy is a sporadic outcome of natural selection . Previous high-throughput analysis showed that epigenetic regulation in the form of histone modification plays a far more pronounced role during gene induction/repression than during steady-state expression ( Weiner et al . , 2012 ) , suggesting the involvement of histone modifications in regulation of gene expression in response to changing environmental cues . In this study , we imposed glucose starvation on yeast cells to mimic an environmental cue . In response , the additive effect of H3K4me on gene transcription was recapitulated in the groups of genes that are enriched in pathways related to glycometabolism , such as carbon metabolism , TCA cycle , and fructose and mannose metabolism ( Figure 8D ) . These observations support the notion that in order to adapt to environmental stress , sister histones execute their fine-tuning regulation by differential modifications . In conclusion , this study provides new insights into how sister histones regulate the plasticity of chromatin structure , as well as gene transcription , and how epigenetic regulation evolves to address variable environmental cues . Given that combinatorial manipulations of sister histone H3 tails have encountered technical challenges in other model systems , the bivalent nucleosome system that we created in this study will be instrumental in further uncovering the role that combinatorial histone H3 modification crosstalk plays in regulating gene expression . In addition , our system for the genetic manipulation of sister histone H3s could be extended to an asymmetry study of sister histone H4s , which have N-terminal tail acetylations representing important epigenetic marks in various biological processes . Moreover , the genetic system that we created will be useful in examining the role that sister histones play in other biological processes , such as DNA repair and recombination , chromatin replication and heterochromatin assembly . Finally , since the protein sequences of histone H3s are highly conserved during evolution , it will be appealing to apply the same scheme to construct a bivalent nucleosome system in other model systems . However , the challenge might be much greater in higher eukaryotes because the copy numbers of histone genes in these organisms are much higher than those in yeast . All yeast strains used in this study were derived from yeast strain YPH500 ( Sikorski and Hieter , 1989 ) . The genotypes of the yeast strains are listed in Supplementary file 1 . The native promoter of HHT1 ( L130H ) in the strains derived from H3D/H3H was replaced with the ADE3 promoter . The histone shuffle strain ( LHT001 ) was constructed previously in our lab . Antibodies used in this study are listed in Supplementary file 2 . For galactose induction assays , cells were grown in YPD ( 10 g/L yeast extract , 20 g/L peptone , 2% dextrose ) to mid-log phase ( OD600 = 0 . 4–0 . 6 ) before being shifted to medium containing raffinose ( 10 g/L yeast extract , 20 g/L peptone , 2% raffinose ) overnight . Each sample was induced by 2% galactose for 10–30 min . Remaining samples in raffinose medium were taken as having an induction time of 0 min . For glucose starvation assays , samples were grown in YPD ( 2% glucose ) to mid-log phase and then shifted to medium containing 0 . 05% glucose for one hour . Yeast cells were cross-linked with 1% formaldehyde for 15 min at room temperature and then resuspended in lysis buffer ( 50 mM HEPES [pH 7 . 5] , 35 mM NaCl , 0 . 5% Na-Deoxycholate [wt/vol] , 5 mM EDTA , 1% Triton X-100 , 1 mM phenylmethylsulfonyl fluoride [PMSF] , protease inhibitor cocktail ) . Cells were lysed using glass beads and sonicated to shear the chromatin to fragment sizes of 200–400 bp . After centrifugation at 10 , 000 g for 10 min , the supernatant fraction was subjected to further fractionation with a 24 ml Superdex-200 column ( GE ) in IP buffer ( 10 mM Tris-HCl [pH 8 . 0] , 100 mM NaCl , 0 . 5 mM EDTA , 1 mM DTT ) . Fractions containing mononucleosomes were pooled for subsequent incubation with anti-Myc antibody and protein G sepharose beads ( GE ) overnight at 4°C . The beads were washed with wash buffer ( 50 mM HEPES [pH 7 . 5] , 150 mM NaCl , 0 . 5% Na-Deoxycholate [wt/vol] , 5 mM EDTA , 1% Triton X-100 ) and TE ( 10 mM Tris-HCl [pH 8 . 0] , 1 mM EDTA ) . Finally , the immunoprecipitated mononucleosomes were eluted from beads with elution buffer ( 10 mM Tris-HCl [pH 8 . 0] , 1 mM EDTA , 1% SDS [wt/vol] ) . Total RNA was isolated from yeast cells with an RNeasy mini kit ( Qiagen ) . cDNA was synthesized using the Fastquant RT kit ( Tiangen ) . 1 µl of the RT reaction was used in the subsequent real-time fluorescence quantitative PCR ( ABI ) . Primer pairs used in qRT-PCR were listed in Supplementary file 3 . The expression of GAL1 was normalized to the RNA levels of ACT1 , and the fold-changes were calculated by defining the relative mRNA level at 0 min as 1 . Preparation and digestion of yeast nuclei were performed as described previously ( Kent and Mellor , 1995; Wang et al . , 2011a ) . Yeast genomic DNAs were prepared with phenol-chloroform extraction followed by ethanol precipitation . The DNA was then digested by EcoRI and separated on a 1 . 6% agarose gel . Digestion patterns were analyzed by indirect-end-labeling . The [32P]dCTP incoroporated probe whose sequence was listed in Supplementary file 3 was used for hybridization . Yeast chromatin was prepared as described previously ( Peng and Zhou , 2012 ) . Specifically , mononucleosomes were purified as described previously for detecting the level of H3N , H3K4me3 , H3K36me3 and H3K79me2/3 . Chromatin was boiled for 10 min in SDS-PAGE loading buffer and separated in 15% SDS-PAGE , and then subjected to western blotting . The chromatin immunoprecipitation ( ChIP ) assay was performed as described previously ( Wang et al . , 2011b ) . We detected the linear range of all the antibodies . Then we loaded our samples in the linear range and performed a western blot . Quantification of the western blot signals was carried out using ImageJ software ( RRID:SCR_003070 ) . Total RNA was extracted using the Yeast RNA extraction kit ( Qiagen ) , resolved on agarose-formaldehyde gels and transferred to Hybond-N+ membrane ( GE ) . RNA was crosslinked to the membrane by UV irradiation . Hybridization was carried out in 7% SDS , 1 mM sodium pyrophosphate , 1 M Na2HPO4 , 150 mM NaH2PO4 , and 1 mM EDTA . Probes were generated by PCR . The method for constructing RNA-Seq libraries was modified from the TruSeq DNA sample preparation kit protocol ( Illumina ) . Briefly , total RNA was isolated using the RNeasy midi kit ( Qiagen ) . The mRNA was purified from total RNA by Dynaloligo ( dT ) beads ( Invitrogen , CA , USA ) . The first and second strand cDNAs were synthesized using the SuperScript III CellsDirect cDNA Synthesis Kit ( Invitrogen ) and the SuperScript Double-Stranded cDNA Synthesis Kit ( Invitrogen ) , respectively . The resulting double-stranded DNA was subjected to DNA repair and end-polishing ( blunt-end ) using the End-It DNA End-Repair Kit ( Epicentre ) . The DNA was then purified with the QIAquick PCR Purification Kit ( Qiagen ) and a dA-tail was added using the 3'−5' exo-Klenow Fragment ( NEB ) . The resulting purified fragments were ligated to adaptor oligo mix ( Illumina ) using Quick T4 DNA ligase ( NEB ) . The 200–500 bp ligation products were recovered from a 2% ( w/v ) agarose gel using the Qiagen gel extraction kit and were PCR amplified with Illumina primers using the KAPA HiFi HotStart kit . The 250–400 bp amplified products were purified again from a 2% agarose gel and used directly for high-throughput sequencing . The raw paired-end reads contained the adapter sequences: the P7 adapter ( read1 ) is 'AGATCGGAAGAGCACACGTCTGAACTCCAGTCAC' , the P5 adapter ( read2 ) is 'AGATCGGAAGAGCGTCGTGTAGGGAAAGAGTGT' . We used the FASTX Toolkit ( RRID:SCR_005534 ) to remove the adapter sequences . We trimmed the reads using TopHat ( RRID:SCR_013035 ) , only mapping the reads to the transcriptome of sacCer3 ( Apr . 2011 ) with the default parameter . For the mapped reads , we then extracted the reads that have the ‘NH:i:1’ field . In order to reduce the PCR duplicates' bias , we kept the maximal three records at the same position . To compare the gene expression profiles between WT ( LHT001 ) and H3D/H3H strains , the aligned reads were analyzed using Cuffdiff2 ( RRID:SCR_001647 ) ( Trapnell et al . , 2012 ) to determine the RPKM ( Reads Per Kilobase per Million mapped reads ) value for each sample . Genes with a change greater than or equal to two folds and p-value ≤ 0 . 001 were regarded as differentially expressed genes and listed in Figure 2—source data 2 . We identified 406 genes that were downregulated in the H3D/H3H sample and 243 genes that were upregulated in the H3D/H3H sample compared with the WT sample . We used FunSpec ( RRID:SCR_006952 , http://funspec . med . utoronto . ca/ ) to annotate the differentially expressed genes to get the GO enrichment results ( Robinson et al . , 2002 ) , which were presented in Figure 2—source data 2 . For RNA-Seq analysis in glucose starvation experiments , we quantified the number of genes for which at least one read was mapped ( RPKM≠0 ) . Fold changes in the transcription of genes under glucose starvation , for genes listed in Figure 2—source data 2 , were quantified as FCi , j = log2 ( ( RPKM_1i , j/RPKM_1act1 , j ) / ( RPKM_0i , j/RPKM_0act1 , j ) ) , where RPKM_1i , j and RPKM_0i , j refer to RPKM for gene i in sample j after glucose starvation for 1 and 0 hr , respectively . We excluded the gene i when p<0 . 05 ( t-test ) by comparing FCi , j in two independent experiments . For the remaining genes , we calculated the average FCi , j ( defined as FCai , j ) of gene i in sample j using two replicates . To evaluate whether the gene was potentially regulated by H3K4 methylation , we screened gene i of which FCai , H3D/H3H is significantly different ( p<0 . 05 ) from FCai , H3DK4R/H3HK4R and grouped it to set DH_4 R4R . Genes in set DH_4 R4R were listed in Figure 8—source data 2 . The skewness of gene transcript fold-change was defined by using the following model . If asymmetrically modified nucleosomes were involved in gene regulation , FCai , H3DK4R/H3H or FCai , H3D/H3HK4R should prerequisitely fall between FCai , H3D/H3H and FCai , H3DK4R/H3HK4R . We therefore pooled gene i of set DH_4 R4R into subset MID when Midi , j = 1 . The value of Midi , j was calculated using the following equation:Midi , j=FCai , j-FCai , H3D/H3H+FCai , j-FCai , H3DK4R/H3HK4RFCai , H3D/H3H-FCai , H3DK4R/H3HK4R The skewness score ( GSS ) of gene i in sample j was calculated by the equation:GSSi , j=log2FCai , j-FCai , H3DK4R/H3HK4RFCai , j-FCai , H3D/H3H , if Midi , j=1 Greater skewness of sample j to H3D/H3H leads to larger GSSi , j . Conversely , greater skewness of sample j to H3DK4R/H3HK4R leads to smaller GSSi , j . The results of Midi , j and GSSi , j were listed in Figure 8—source data 3 . The gene i in set MID was classified into three subsets: ( 1 ) II , FCai , H3DK4R/H3H and FCai , H3D/H3HK4R showing significant difference from both FCai , H3D/H3H and FCai , H3DK4R/H3HK4R . The maximum and minimum values of GSS in this subset were defined as GSSmax and GSSmin , respectively; ( 2 ) III , FCai , H3DK4R/H3H and FCai , H3D/H3HK4R exhibiting no difference from FCai , H3D/H3H but significant difference from FCai , H3DK4R/H3HK4R in the condition of each GSSi , H3DK4R/H3H and GSSi , H3D/H3HK4R larger than GSSmax; and ( 3 ) I , FCai , H3DK4R/H3H and FCai , H3D/H3HK4R significantly differing from FCai , H3D/H3H rather than FCai , H3DK4R/H3HK4R in the condition of each GSSi , H3DK4R/H3H and GSSi , H3D/H3HK4R smaller than GSSmin . They were listed in Figure 8—source data 4 . Data were analyzed by Pearson's product-moment test and Student t-test as indicated .	Inside each human cell , about two meters of DNA is wrapped around millions of proteins called histones , forming structures known as nucleosomes . Each nucleosome contains 147 letters of DNA code and two copies of four different histones – H2A , H2B , H3 and H4 – meaning eight proteins in total . The two copies of each histone protein found in a nucleosome are referred to as “sister” histones and are identical . Histone proteins have long tails that the cell can edit by adding chemical groups at specific positions . This changes the way the cell copies , uses and repairs its DNA . Previous studies show that identical sister histones can end up with different modifications . But , it was not clear what effect this had . To adress this issue , there are two questions to answer . What do asymmetric sister histones do in living cells ? And , does a modification to one histone affect its sister ? Gene editing could help scientists to understand the effect of asymmetrical tail modification by forcing cells to make non-identical sister histones . However , this is challenging because most animals studied in the laboratory have many copies of the genes for histones . Fruit flies , for example , have 23 copies of their histone genes . The single-celled yeast Saccharomyces cerevisiae has only two copies of its histone genes . Yet , even if one of these genes was replaced with a mutant gene and the other left unedited or “wild-type” , there would be nothing to stop the cell from forming nucleosomes in which both sister histones were still identical – that is to say , mutant with mutant or wild-type with wild-type . Now , Zhou , Liu et al . report a new method that allowed them to edit the tail sequence of one H3 histone but not its sister . First , they searched for , and found , a pair of mutant H3 genes , which encode two extremely similar but different H3 proteins that could bind to each other but not to themselves . As a result , yeast cells with the genes for these proteins could only form nucleosomes in which the sister H3 histones were non-identical . Next , Zhou et al . made a small change to the tail of one of the H3 sisters which meant it could not be modified . The resulting nucleosomes contain one H3 histone with a wild-type tail and one with a mutant tail . The cell could only modify one of them , mimicking natural asymmetrical modifications . The new technique revealed that modification of one sister does not affect the the other . It also revealed that modifications to sister histones can work both alone and together . In some cases , the cell needs only edit one tail to affect the use of a gene . Other times , it must edit both tails for greatest effect . This new tool is the first step in understanding the contribution of the tails of sister histones in living cells . In future , it should help to uncover the effect of different combinations of modifications . This could shed light on how cells control the use of different genes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2017	Independent manipulation of histone H3 modifications in individual nucleosomes reveals the contributions of sister histones to transcription
Drosophila larvae and adults possess a potent innate immune response , but the response of Drosophila eggs is poor . In contrast to Drosophila , eggs of the beetle Tribolium are protected by a serosa , an extraembryonic epithelium that is present in all insects except higher flies . In this study , we test a possible immune function of this frontier epithelium using Tc-zen1 RNAi-mediated deletion . First , we show that bacteria propagate twice as fast in serosa-less eggs . Then , we compare the complete transcriptomes of wild-type , control RNAi , and Tc-zen1 RNAi eggs before and after sterile or septic injury . Infection induces genes involved in Toll and IMD-signaling , melanisation , production of reactive oxygen species and antimicrobial peptides in wild-type eggs but not in serosa-less eggs . Finally , we demonstrate constitutive and induced immune gene expression in the serosal epithelium using in situ hybridization . We conclude that the serosa provides insect eggs with a full-range innate immune response . To combat infection , insects rely on humoral and local immune responses . The humoral immune response is characterized by the massive secretion of antimicrobial peptides into the hemolymph and is mainly exerted by the fat body . Epithelia and hemocytes play the main role in local immune defenses that comprise melanisation , local AMP production , phagocytosis , and encapsulation ( Lemaitre and Hoffmann , 2007; Ganesan et al . , 2011; Davis and Engstrom , 2012; Ferrandon , 2013; Ligoxygakis , 2013; Wang et al . , 2014 ) . The mechanisms regulating these innate immune responses have largely been uncovered with the aid of genetic and molecular studies in the fruit fly Drosophila melanogaster . When microbes invade the fly , their released peptidoglycans are sensed by peptidoglycan recognition proteins ( PGRPs ) and Gram-negative binding proteins ( GNBPs ) leading to the activation of the main immune signaling pathways . The meso-diaminopimelic acid-type ( DAP-type ) peptidoglycans of Gram-negative bacteria activate the IMD pathway , whereas the Lys-type peptidoglycans of Gram-positive bacteria activate the Toll pathway . The activation of the Toll pathway is mediated by a proteolytic cascade of serine proteases leading to the cleavage of the cytokine Spätzle , the ligand of the transmembrane receptor Toll . Activation of the immune signaling pathways leads to nuclear localization of the NF-kappaB factors Dorsal , Dif , or Relish that induce antimicrobial peptides ( AMPs ) . Other upregulated genes are prophenoloxidases ( proPOs which mediate melanisation ) and dual oxidase ( DUOX which produces reactive oxygen species ) . Drosophila has been extremely helpful uncovering those mechanisms , but research in other insects , such as the mealworm beetle Tenebrio molitor , has also generated insightful results . The biochemical details of pathway activation , for instance , have mainly been unraveled using this beetle ( See Park et al . , 2010 for review ) . With the availability of tools such as RNAseq and RNAi , more insect species are being established as model organism for innate immunity research ( Altincicek and Vilcinskas , 2007; Waterhouse et al . , 2007; Gerardo et al . , 2010; Johnston and Rolff , 2013; Johnston et al . , 2013; Zhu et al . , 2013 ) . In particular the red flour beetle ( Tribolium castaneum ) has received much attention in innate immune studies ( Zou et al . , 2007; Altincicek et al . , 2008 , 2013; Roth et al . , 2010; Contreras et al . , 2013; Milutinović et al . , 2013; Zhong et al . , 2013; Behrens et al . , 2014 ) . Comparative genome analysis has revealed that components of intracellular immune signaling pathways ( Toll , IMD , and JAK/STAT ) in Drosophila are 1:1 conserved in Tribolium ( Zou et al . , 2007 ) . The RNAi knockdown technology has shown that the IMD and Toll pathway are largely functionally conserved ( Shrestha and Kim , 2010; Yokoi et al . , 2012a , 2012b ) . Their activity does , however , not strictly depend on either Gram-negative or Gram-positive bacteria ( Yokoi et al . , 2012a , 2012b ) , but this distinction is also not completely black and white in Drosophila ( Leulier et al . , 2003; Leone et al . , 2008 ) . Nevertheless , species-specific family expansion and sequence divergence in the PGRP and AMP families indicate species-specific differences , possibly required for effective recognition and elimination of evolving pathogens ( Christophides et al . , 2002; Zou et al . , 2007; Altincicek et al . , 2008; Park et al . , 2010 ) . Not only larvae and adults but also insect eggs are constantly threatened by pathogens ( See Blum and Hilker , 2008; Kellner , 2008 for review ) . Serratia bacteria , for instance , have been found inside eggs of corn earworms and corn borers ( Bell , 1969; Lynch et al . , 1976 ) and can infect eggs in the laboratory ( Sikorowski et al . , 2001 ) . We have also shown that Serratia infection leads to reduced egg survival in the burying beetle Nicrophorus vespilloides ( Jacobs et al . , 2014 ) . Maternal investments have been proposed to counter microbial infections . Female medflies , for example , cover their eggs with antimicrobial secretions ( Marchini et al . , 1997 ) and in the absence of maternal care , eggs of earwigs die of fungal infection ( Boos et al . , 2014 ) . Two studies focusing on transgenerational immune priming , however , have shown that the antimicrobial activity of eggs is of internal origin ( Sadd and Schmid-Hempel , 2007; Zanchi et al . , 2012 ) . This is often implicitly interpreted as maternal loading of antimicrobials into the egg ( Moreau et al . , 2012 ) , but maternal transfer of bacteria to the eggs also leaves zygotic investment as possibility ( Trauer and Hilker , 2013; Freitak et al . , 2014 ) . Overall , it is ecologically relevant to gain a better understanding of the immune system in insect eggs . The zygotic response in Drosophila eggs , however , seems poor . It is not until late stage 15 , ( one of the latest stages in development when ectoderm and trachea have differentiated ) , that eggs show up to 25-fold upregulation of antimicrobial peptides ( Tan et al . , 2014 ) . This is incomparable to the upregulation in adult flies that is at least an order of magnitude larger . Except for Cecropin ( Tingvall et al . , 2001 ) , stage 11 embryos do not show any induction of antimicrobial peptides and cannot contain an infection of non-pathogenic bacteria , leading to reduced survival ( Tan et al . , 2014 ) . In strong contrast , we have shown that the eggs of Tribolium which were not even half way during development could upregulate several AMPs to levels comparable to the adult ( Jacobs and van der Zee , 2013 ) . This upregulation depends on the serosa , an extraembryonic epithelium that envelopes yolk and embryo ( Jacobs and van der Zee , 2013 ) . This membrane is present in all insects but was lost in a small group of higher Diptera ( the Schizophora ) to which Drosophila belongs ( Schmidt-Ott , 2000; Rafiqi et al . , 2008 ) . Although two maternal extracellular coverings , the chorion and the vitelline membrane , envelop the insect egg , the serosa is the first cellular epithelium surrounding the egg at the interface between the microbe rich external milieu on the one side and the yolk and embryo at the other side . Thus , the serosa could function as an immune competent barrier epithelium . This has been suggested before , as the NF-kappaB factor Dorsal is highly expressed in the presumptive serosa ( Chen et al . , 2000 ) . The absence of the serosa might account for the poor immune response in Drosophila eggs . To gain deeper insights into the role of the serosa , we chose Tribolium castaneum , a beetle that possesses a serosa like all non-Schizophoran insects . In this beetle , we can prevent the development of the serosa by parental RNA interference with Tc-zerknüllt1 ( Tc-zen1 ) . This technique generates Tribolium eggs with an amnion at the dorsal side , but without a serosa ( van der Zee et al . , 2005 ) . At the relative humidity of the air of the laboratory , normal larvae hatch from these eggs ( Jacobs et al . , 2013 ) . As Tc-zen1 is only expressed in the early serosa ( van der Zee et al . , 2005 ) and is not expressed anymore by the time the experiments are performed ( See discussion ) , we expect only to find effects that are a consequence of the absence of the serosa . We investigated the growth of bacteria in serosa-less and wild-type eggs , sequenced the whole transcriptome of naive and immune-challenged eggs with and without a serosal epithelium and confirmed constitutive and induced gene expression in the serosa by in situ hybridization . We conclude that the serosa is a frontier epithelium that provides immune competence to the insect egg . To examine the influence of the serosa on bacterial growth in infected eggs , and to standardize our infection method , we counted colony forming units ( cfu's ) directly after infection ( t = 0 ) and 6 hr later ( t = 6 ) ( Figure 1 ) . We pricked 24–40hr old eggs ( i . e . up to half-way during development ) with a tungsten needle dipped in a concentrated mix of Escherichia coli and Micrococcus luteus cultures ( see ‘Materials and methods’ ) . To determine cfu's , we shortly treated eggs with 0 . 5% hypochlorite to sterilize the outside . Untreated eggs did hardly contain bacteria that grow on LB agar plates ( on an average three cfu's were found ) . Sterile injury did not increase this number ( Figure 1 , lower lines ) . In contrast , septic injury introduced on average 53 bacteria into wild-type eggs and 49 into serosa-less eggs . These numbers increased on average to 747 cfu's in wild-type eggs and to 7260 cfu's in serosa-less eggs . When we use the formula N ( t ) = N ( 0 ) ekt , the specific bacterial growth rate k in wild-type eggs is 0 . 44 hr−1 , whereas k = 0 . 83 hr−1 in serosa-less eggs . This means that bacteria grow twice as fast in serosa-less eggs and suggests that the serosa exerts an immune function . 10 . 7554/eLife . 04111 . 003Figure 1 . Counts of colony forming units ( cfu's ) after sterile and septic injury . Green lines represent bacterial growth in wild-type eggs . Red lines represent bacterial growth in Tc-zen1 RNAi ( serosa-less ) eggs . Sterile injury did not introduce bacteria ( lower lines: average of 2 cfu's found at t = 0 and an average of 5 cfu's found at t = 6 ) . Septic injury introduced on average 53 bacteria into wild-type eggs and 49 into serosa-less eggs . These numbers increased to 747 ± 106 cfu's in wild-type eggs ( green upper line ) and to 7260 ± 1698 cfu's in serosa-less eggs ( red upper line ) at t = 6 . This means that bacteria propagate twice as fast in serosa-less eggs ( p < 0 . 01 , as determined by a Pearson's chi-square test ) . Suspensions of 10 eggs were used per LB agar plate ( see ‘Materials and methods’ ) , and 10 plates were analyzed per treatment and time point , giving rise to the error bars presented in the graph ( standard error ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04111 . 003 To characterize this immune function , we sequenced the whole transcriptome of wild-type eggs , Tc-zen1 RNAi ( serosa-less ) eggs , and control RNAi eggs without injury , after sterile injury , and after septic injury ( Figure 2 ) . The control RNAi consists of an injection of a 500 bp dsRNA derived from a vector sequence without target in the Tribolium castaneum genome . For these nine different treatments , three biological replicates were carried out ( independent RNAi , independent injury ) giving a total of 27 samples ( Figure 2 ) . Illumina next generation sequencing resulted in over 970 million cDNA reads with over 49 billion bp sequence information . Approximately , 72% of the reads could be mapped to Tribolium gene models built on the 3 . 0 genome assembly ( Richards et al . , 2008 ) ( Supplementary file 1 ) . We found expression of 14 , 903 of the total of 16 , 541 predicted genes , of which 13 , 464 genes were expressed in wild-type , control , and Tc-zen1 RNAi eggs and 1440 genes were expressed in a subset of these treatments . These numbers confirm the quality of the deep sequencing data . 10 . 7554/eLife . 04111 . 004Figure 2 . Experimental setup . ( A ) We collected eggs from wild-type , control RNAi , and Tc-zen1 RNAi beetles overnight . These eggs were incubated for 24 hr at 30°C to ensure development of the serosa . Eggs are then maximally 40 hr old , while total developmental time is close to 85 hr at 30°C . Eggs were pricked with a sterile needle ( sterile injury ) , pricked with a mix of E . coli and M . luteus ( septic injury ) , or remained untreated ( naive ) . They were incubated for another 6 hr at 30°C before total RNA was extracted for RNAseq . To analyze the immune response , the transcriptomes of sterilely injured eggs and of septically injured eggs were compared to naive eggs . This was done for wild-type , control , and Tc-zen1 RNAi eggs . ( B ) We collected three biological samples for each combination of egg-type ( wild-type , control RNAi , or Tc-zen1 RNAi ) and treatment ( naive , sterile injury , or septic injury ) giving a total of 27 biological samples . DOI: http://dx . doi . org/10 . 7554/eLife . 04111 . 004 First , we identified the immune-responsive genes by determining differential expression of genes between naive eggs on the one hand and sterilely injured eggs or septically injured eggs on the other hand . We only considered genes with at least a twofold change in expression and an adjusted p-value smaller than 0 . 01 . This gave a total of 415 differentially expressed genes in the sterilely injured eggs compared to the naive eggs , and a total of 538 differentially regulated genes in septically injured eggs compared to naive eggs . This shows that Tribolium eggs possess an extensive transcriptional response upon infection . To obtain a global impression of the kind of genes differentially regulated upon infection in wild-type and control RNAi eggs , we assigned gene ontology terms ( GO-terms ) to all Tribolium genes . As no GO-term annotation is available for Tribolium , we blasted Tribolium genes against Drosophila and used the Drosophila GO-terms of the best hit . Using the Wallenius approximation ( Young et al . , 2010 ) , we found several highly over-represented GO-term categories with a p-value below 0 . 001 in both wild-type eggs ( Figure 3A ) and control RNAi eggs ( Figure 3B ) . The over-represented categories are mostly immune related . This indicates that our approach does not depend on artefacts generated by pricking eggs ( e . g . delayed development ) but mainly identifies genes involved in the innate immune response . 10 . 7554/eLife . 04111 . 005Figure 3 . Types of genes that are differentially regulated . ( A ) Significantly over-represented GO-terms among the genes induced in wild-type eggs after septic injury ( p < 0 . 001 ) . ( B ) Significantly over-represented GO-terms among the genes induced in control RNAi eggs after septic injury ( p < 0 . 001 ) . These categories indicate that the detected differential regulation does not result from artefacts induced by treatments ( such as death or delayed development ) and show that Tribolium eggs display an elaborate immune response . DOI: http://dx . doi . org/10 . 7554/eLife . 04111 . 005 To obtain a more detailed analysis of the immune response in wild-type and control eggs , we focused on 368 genes that have been annotated as immune genes ( Zou et al . , 2007; Altincicek et al . , 2013 ) ( Supplementary files 4–9 ) . Of these genes , 78 were differentially regulated in wild-type eggs upon septic injury ( Table 1 and Supplementary file 2 and 5 ) , while 95 immune genes were differentially regulated in control RNAi eggs ( Table 1 and Supplementary file 2 and 7 ) . This indicates that RNAi itself leads to an increased number of differentially regulated genes upon bacterial challenge but , more importantly , shows that Tribolium eggs possess an elaborate immune response . In the following sections , we take a closer look at the exact genes involved in this extensive immune response . 10 . 7554/eLife . 04111 . 006Table 1 . Number of differentially expressed immune genes in Tribolium castaneum eggsDOI: http://dx . doi . org/10 . 7554/eLife . 04111 . 006Wild-type sterile injuryWild-type septic injuryControl sterile injuryControl septic injuryTc-zen1 sterile injuryTc-zen1 septic injuryMicrobial recognition417260831000Extracellular signal transduction and modulation276321033434104520Intracellular transduction pathways ( Toll/IMD/JNK/JAK-STAT ) 213222633221Execution/stress1202021642475231Total45862165710722313972updownupdownupdownupdownupdownupdownBlue = induction , red = repression . To investigate the role of the serosa in the immune response , we compared the transcriptional response of wild-type and control eggs to the response of serosa-less eggs . Of all 538 genes differentially regulated upon bacterial challenge , 481 genes are only responsive in eggs with a serosa . The vast majority , 276 genes , are differentially regulated in both wild-type and control eggs but not in serosa-less eggs ( Figure 4B ) . In the serosa-less Tc-zen1 RNAi eggs , merely 57 genes are differentially regulated upon microbial challenge , despite our finding that RNAi rather increases the number of immune responsive genes . Of all 368 Tribolium genes that are annotated as immune genes ( Zou et al . , 2007; Altincicek et al . , 2013 ) , only nine were differentially regulated upon infection in serosa-less eggs ( Table 1 and Supplementary file 2 ) . Except for serpin24 , all of the other eight genes were also differentially regulated in response to sterile injury , indicating that they do not respond to infection but to wounding . Notably , none of the AMPs is induced upon infection in serosa-less eggs , neither proPO1 nor the DUOX ortholog Hpx11 ( Supplementary file 2 ) . Thus , the serosa is essential for the early immune response of the Tribolium egg . These data corroborate our previous qPCR study showing that AMP and PGRP upregulation upon infection is abolished in serosa-less eggs ( Jacobs and van der Zee , 2013 ) . To see if we could also independently confirm serosa-dependent induction of some of our newly identified candidates , we performed qPCR on the transmembrane recognition protein of the IMD pathway PGRP-LC , the serine proteases cS-P8 , SPH-H57 , SPH-H70 , the serine protease inhibitors serpin24 and serpin26 , the Toll receptor toll3 and the novel potential AMPs TC004646 , TC007763 , TC007857 , TC008806 , and TC015479 ( Figure 5 ) . The fold-changes detected by qPCR after sterile and septic injury of wild-type eggs match the values found in the RNAseq data . The largest deviation was found for the potential AMP TC007858 that is upregulated 156 times upon septic injury in our qPCRs but 484 times according to the RNAseq data ( Figure 5J ) . Most importantly , all qPCRs convincingly showed the absence of induction in Tc-zen1 RNAi eggs , thus providing independent support for our conclusion that the serosa is required for the immune response in Tribolium eggs . 10 . 7554/eLife . 04111 . 009Figure 5 . RT-qPCR verification of immune gene expression . The expression levels of several immune genes was verified by RT-qPCR . Expression shown relative to the expression in naive eggs , the mean fold change of the biological replicates ( based on two technical replicates ) is plotted and error bars show the standard error . Black bars represent expression after sterile injury , white bars represent expression after septic injury . Expression levels measured by RT-qPCR show very similar results as the expression levels measured by RNAseq ( See Supplementary file 2 ) . ( A ) PGRP-LC , ( B ) SPH-H57 , ( C ) SPH-H70 , ( D ) cSP-P8 , ( E ) serpin24 , ( F ) serpin26 , ( G ) toll3 , ( H ) TC004646 , ( I ) TC007763 , ( J ) TC007858 , ( K ) TC008806 , ( L ) TC015479 . See ‘Materials and methods’ for experimental details . DOI: http://dx . doi . org/10 . 7554/eLife . 04111 . 009 To investigate if it is the serosal epithelium itself that expresses the identified immune genes and to exclude indirect effects , we performed in situ hybridization on two AMPs ( thaumatin1 and attacin1 ) of which mRNA length permitted in situ detection . In naive eggs , we could not detect thaumatin1 or attacin1 expression . In contrast , expression was obvious in challenged eggs ( Figure 6 ) . In these eggs , brown melanisation was found at the site of injury ( asterix in Figure 6A and A′ and arrowhead in Figure 6G ) and the individual nuclei of the serosa can be distinguished from the oversaturated DAPI signal marking the germ-band ( Figure 6B , E , H ) ( Handel et al . , 2000 ) . The thaumatin1 expression clearly associates with the large polyploid serosal nuclei and not with the dense cells of the germ-band ( overlay in Figure 6C and C′ ) . A deeper focal plane of a different egg demonstrates exclusive expression in the overlying serosa on the outer surface ( Figure 6D , D′ ) and not in the underlying embryo proper ( Figure 6E , F ) . Also attacin1 expression consistently associated with the large polyploid serosal nuclei ( Figure 6G–I′ ) . 10 . 7554/eLife . 04111 . 010Figure 6 . In situ hybridization showing expression of AMP genes in the serosa upon septic injury . ( A–F ) Thaumatin1 in situ hybridization . ( A ) Superficial view . Thaumatin1 is expressed around the site of injury ( asterix ) . Brown melanisation is observed around the site of injury . ( A′ ) Magnification of the expression area shown in ( A ) . Asterix marks the site of injury . ( B ) DAPI counterstaining of the same egg as in ( A ) . The large polyploid serosal nuclei can be distinguished from the oversaturated DAPI signal from the germ-band . Head lobes to the left . ( B′ ) magnification of ( B ) . ( C ) Overlay of the in situ hybridization shown in A and the DAPI staining shown in ( B ) . The thaumatin1 expression associates with the large polyploid serosal nuclei and is not found in the embryo proper . ( C′ ) Magnification of the expression area shown in ( C ) . ( D ) Focal plane through the egg . Thaumatin1 is expressed in a thin outer layer at the surface of the egg . ( D′ ) Magnification of the expression area shown in ( D ) . ( E ) DAPI staining of the same egg shown in ( D ) . The embryo is brightly visible . Head to the left . ( E′ ) Magnification of E . ( F ) Overlay of the in situ hybridization shown in D and the DAPI staining shown in ( E ) . ( F′ ) Magnification of the expression area . ( G ) Attacin1 in situ hybridization . Brown melanisation is visible around the site of injury ( arrowhead ) . ( G′ ) Magnification of the anterior region of the egg shown in ( G ) . ( H ) DAPI staining of the same egg shown in G . The germ-band is brightly stained ( head to the left ) and the separate large serosal nuclei are visible . ( H′ ) Magnification of the anterior of the egg shown in ( H ) . ( I ) Overlay of the in situ hybridization shown in G and the DAPI staining shown in ( H ) . Attacin1 is expressed in the large serosal cells covering the germ-band and is not expressed in the dense cells of the germ band . ( I′ ) Magnification of the anterior of the egg shown in ( I ) . The attacin1 staining associates with the large serosal nuclei . DOI: http://dx . doi . org/10 . 7554/eLife . 04111 . 010 Thus , it is the serosal epithelium itself that expresses these AMPs upon infection . Although we cannot exclude an indirect role of the serosa in the expression of the other identified immune genes , we propose that the serosa itself expresses these genes and thus regulates the described immune response involving melanisation , the generation of reactive oxygen species , and the massive production of AMPs . To discover immune genes that are constitutively expressed in the serosa , we compared the transcriptomes of naive Tc-zen1 RNAi eggs to naive wild-type eggs . We found 44 immune genes that have serosa-dependent expression ( Table 3 ) . Of these genes , more than 75% is involved in the recognition of microbes and extracellular signal transduction such as PGRP-LA , many serine proteases and Spz4 and Spz5 ( Table 3 ) . In contrast , most of the genes of the intracellular signal transduction were present in Tc-zen1 RNAi eggs at similar levels as in wild-type eggs . Notably , the transmembrane receptor toll3 exhibits higher expression in unchallenged eggs with a serosa than in eggs without a serosa . These data indicate that the serosa is an immune competent epithelium that expresses many genes involved in bacterial recognition and transduction of this recognition to receptor activation . 10 . 7554/eLife . 04111 . 011Table 3 . Differentially regulated immune genes in naive wild-type eggs compared to naive Tc-zen1 RNAi eggsDOI: http://dx . doi . org/10 . 7554/eLife . 04111 . 011Gene IDDescriptionFold changeFDR adjusted p-valueGene IDDescriptionFold changeFDR adjusted p-valueExtracellular signal transduction and modulationTC000247cSPH-H22 . 70<0 . 01TC005754serpin225 . 26<0 . 01TC000248cSPH-H34 . 11<0 . 01TC006255serpin240 . 690 . 03TC000249cSPH-H45 . 16<0 . 01TC011718serpin271 . 62<0 . 01TC000740SPH-H179 . 28<0 . 01TC006726Spz43 . 00<0 . 01TC000829SPH-H188 . 26<0 . 01TC013304Spz5122 . 56<0 . 01TC007026cSPH-H7829 . 79<0 . 01Microbial recognitionTC012390SPH-H1291 . 60<0 . 01TC002789PGRP-LA3 . 950 . 02TC000495cSP-P86 . 57<0 . 01TC014664TEP-B2 . 900 . 02TC000497cSP-P104 . 50<0 . 01TC005976PSH3 . 43<0 . 01TC000547SP-P132 . 41<0 . 01TC006978C-type lectin114 . 52<0 . 01TC000635SP-P162 . 54<0 . 01TC013911C-type lectin 1318 . 21<0 . 01TC004160cSP-P449 . 79<0 . 01Toll-signalling pathwayTC004624cSP-P520 . 52<0 . 01TC004438Toll32 . 28<0 . 01TC004635cSP-P5351 . 56<0 . 01IMD-signalling pathwayTC005230cSP-P61250 . 00<0 . 01TC014708NFAT2 . 01<0 . 01TC006033SP-P681 . 54<0 . 01Execution mechanismsTC009090cSP-P912 . 80<0 . 01TC005375hexamerin20 . 38<0 . 01TC009092cSP-P933 . 00<0 . 01TC005493Heme peroxidase 13 . 84<0 . 01TC009093cSP-P9427 . 76<0 . 01TC015234Heme peroxidase 26 . 30<0 . 01TC013277cSP-P1363 . 04<0 . 01TC010356Scavenger receptor-B130 . 600 . 03TC013415SP-P14111 . 85<0 . 01TC015854Scavenger receptor-B21 . 91<0 . 01TC000760serpin15 . 29<0 . 01TC014946Scavenger receptor-B529 . 29<0 . 01TC005750serpin181 . 92<0 . 01TC000948Scavenger receptor-B6163 . 90<0 . 01TC005752serpin202 . 30<0 . 01TC014954Scavenger receptor-B91 . 96<0 . 01SP = serine protease; SPH = non-catalytic serine protease; cSP = clip-domain serine protease . To confirm constitutive expression of these identified genes , we performed in situ hybridization on naive eggs . We chose the receptor toll3 that shows two times higher expression in eggs with a serosa and the scavenger receptor B5 that shows 30 times higher expression in eggs with a serosa ( Table 3 ) . We found ubiquitous expression of toll3 in the egg ( Figure 7A ) . Although toll3 was clearly expressed in the serosa ( partly detached from the egg Figure 7A′ ) , we also detected expression in the embryo . As in situ hybridization is not a quantitative technique , and because the serosal cells are flat and thin , it is possible that we could not detect the twofold higher expression in the serosa . For scavenger receptor B5 that has a 30-fold higher expression in eggs with a serosa , we did find clear expression in the serosal epithelium ( Figure 7D ) , whereas the underlying germ-band was not stained ( Figure 7F and F′ ) . We propose that all genes listed in Table 3 are constitutively expressed in the serosa and thus make the serosa an immune-competent frontier epithelium . 10 . 7554/eLife . 04111 . 012Figure 7 . Constitutive expression of immune genes in the serosa . ( A–C ) Toll3 in situ hybridization . ( A ) Toll3 is expressed in the flat and thin serosal cells ( partly detached from the egg ) but also in the germ rudiment ( head lobes to the right ) . ( A′ ) Magnification of the area indicated with an arrow in ( A ) . ( B ) DAPI staining of the same egg shown in ( A ) . The bright staining of the germ-band can be distinguished from the large nuclei of the serosa . ( C ) Overlay of the in situ hybridization shown in A and the DAPI staining shown in ( B ) . ( C′ ) Magnification of ( C ) . Toll3 is expressed in cells of the serosa . ( D–F ) Scavenger receptor B5 in situ hybridization . ( D ) Scavenger receptor B5 shows expression in every serosal cell at the surface . ( D′ ) Magnification of ( D ) . ( E ) DAPI staining of the same egg shown in ( D ) . The germ-band is brightly stained ( head to the left ) and the staining of the serosal nuclei is clearly visible when not overwhelmed by staining of the dense nuclei of the germ-band . ( E′ ) Magnification of ( E ) . The serosal nuclei are visible . Bright staining of the germ-band to the right . ( F ) Overlay of the in situ hybridization shown in D and the DAPI staining shown in ( E ) . Scavenger receptor B5 expression follows the serosal nuclei and is not detected in the germ-band . ( F′ ) Magnification of ( F ) . Scavenger receptor B5 mRNA is detected around the large polyploid serosal nuclei and not around the dense nuclei of the germ rudiment . DOI: http://dx . doi . org/10 . 7554/eLife . 04111 . 012 Taken together , we have shown that the eggs of the beetle Tribolium castaneum display an extensive transcriptional immune response . This response is entirely dependent on the serosa , an extraembryonic epithelium that envelops yolk and embryo . This immune competent frontier epithelium constitutively expresses some immune genes and can induce massive amounts of AMPs . Tribolium castaneum eggs can mount a full-range innate immune response involving antimicrobial peptides , melanisation , and the production of reactive oxygen species . This response depends entirely on the extraembryonic serosa , an immune competent frontier epithelium that is absent in Drosophila . The Tribolium stock used for this study was the T . castaneum wild-type strain , San Bernardino . Stock keeping and Tc-zen1 RNAi were performed as described in van der Zee et al . ( 2005 ) . The control dsRNA was synthesized from a 500-bp vector sequence cloned from the pCRII vector ( Invitrogen , Waltham , MA , USA ) using the primers 5′-TGCCGGATCAAGAGCTACCAA-3′ and 5′-TGTGAGCAAAAGGCCAGCAA-3′ and has no targets in the Tribolium castaneum genome ( See also Jacobs et al . , 2013; Jacobs and van der Zee , 2013 ) . Infection experiments were performed as described in Jacobs and van der Zee ( 2013 ) . 24- to 40-hr old eggs ( total developmental time is close to 85 hr ) were pricked with a sterile tungsten needle or with a tungsten needle dipped in a concentrated mix of E . coli and M . luteus cultures ( bacteria provided by D Ferrandon , Strasbourg ) or were not pricked at all . To allow comparison to the extensive body of work in Drosophila , we have used the same strains of E . coli and M . luteus as are traditionally used in Drosophila ( Ferrandon et al . , 2007 ) . 6 hr later , eggs were used for RNA isolation or in situ hybridization . Cfu's were determined directly after infection ( t = 0 ) or 6 hr after infection ( t = 6 ) . Eggs were shortly washed for 15 s in a 0 . 5% hypochlorite solution to sterilize the outside and rinsed with water . 10 eggs were pooled and homogenized in 100 µl water with a sterile pestle . For t = 0 , 25 µl of this suspension was directly plated on LB agar plates; for t = 6 these 100 microliters were either diluted 50 times in 50 µl water ( for wild-type eggs ) or 500 times in 50 µl water ( for Tc-zen1 RNAi eggs ) . Of these dilutions , 25 µl was plated on LB agar plates . Colonies were counted after an overnight incubation at 37°C , and average numbers of cfu's were calculated per egg . For each combination of time and treatment , the cfu's were measured 10 times . Statistical significance was determined by performing a Pearson's chi-square test . Bacterial load of wild-type eggs increased to on average 32 , 975 cfu's after 24 hr , but at this time point comparisons to Tc-zen1 RNAi eggs were unreliable as bacteria might have reached a maximum . At t = 6 , bacteria were still in their exponential growth phase and the formula N ( t ) = N ( 0 ) ekt could be used to calculate the specific growth rate . For RNAseq and qPCR , total RNA of approximately 300 eggs was extracted using TRIzol extraction ( Invitrogen ) after which the RNA was purified and DNA digested on column with the RNeasy kit ( Qiagen , Venlo , Netherlands ) . We collected three biological samples for each of the 9 treatments , giving a total of 27 biological samples ( Figure 2 ) . cDNA library synthesis and sequencing was performed by the ZF-screens ( Leiden , the Netherlands ) sequencing company on an Illumina HiSeq2500 sequencer . Sequencing reads were mapped with CLC genomics workbench 6 using the first 51 bp with the highest sequencing quality and score values over 20 , allowing 2 mismatches to the reference sequence of the Tribolium genome 3 . 0 which was obtained from Ensemble ( Flicek et al . , 2013 ) . The mismatch cost was set to 2 , the insertion cost to 3 , the deletion cost to 3 , the length fraction to 0 . 5 , and the similarity fraction was set at 0 . 8 . To calculate statistical differences of the expression levels of genes between treatments , we utilized the DESeq package ( Anders and Huber , 2010 ) in Bioconductor ( Gentleman et al . , 2004 ) in R ( R Development Core Team , 2009 ) . The p values were adjusted for multiple testing with the Benjamini–Hochberg procedure , which determines the false discovery rate ( FDR ) . We trimmed the data to only contain genes that are induced more than twofold or repressed more than twofold . To minimize false discovery rate , we set the cut-off value for significant genes to an FDR of <0 . 01 . DESeq was used to normalize the count data , calculate mean values , fold changes , size factors , variance and p values ( raw and adjusted ) of a test for differential gene expression based on generalized linear models using negative binomial distribution errors . Sequence homology searches of predicted reference gene sequences and subsequent functional annotation by gene ontology terms ( GO ) and InterPro terms ( InterProScan , EBI ) were determined using the BLAST2GO software suite v2 . 6 . 6 ( Conesa et al . , 2005 ) . First , homology searches were performed through BLASTX against sequences of the Drosophila protein database with a cut-off value of 1 . 0E-10 . Subsequently , GO classification annotations were created after which InterPro searches on the InterProEBI web server were performed remotely by utilizing BLAST2GO . RNA was collected as described under ‘Sample collection for transcriptional analysis’ . The quality of RNA preparation was confirmed spectro-photometrically and on gel . One microgram of total RNA was used for cDNA synthesis . First strand cDNA was made using the Cloned AMV First Strand Synthesis kit ( Invitrogen ) . Each qRT-PCR mixture ( 25 µl ) contained 2 . 5 ng of cDNA , and the real-time detection and analyses were done based on SYBR green dye chemistry using the qPCR kit for SYBR Green I ( Eurogentec , Seraing , Belgium ) and a CFX96 thermocycler ( Bio-rad , Hercules , CA , USA ) . Thermal cycling conditions used were 50°C for 2 min , 95°C for 10 min , then 50 cycles of 95°C for 15 s , 60°C for 30 s , 72°C for 30 s; this was followed by dissociation analysis of a ramp from 65°C to 95°C with a read every 0 . 5°C . Relative quantification for each mRNA was done using the Livak method ( Livak and Schmittgen , 2001 ) . The values obtained for each mRNA were normalized by RPL13a mRNA amount for Tribolium ( Primers as in Lord et al . , 2010 ) . Total RNA for each treatment was isolated two times ( biological replication ) and each sample was measured by qRT-PCR twice ( technical replication ) . The primers used for qPCR are in Supplementary file 3 . In situ hybridizations involving alkaline phosphatase-based visualization of DIG-labelled probes were essentially performed as described in Tautz and Pfeifle ( 1989 ) , but without the proteinase K step . Eggs were fixed for 20 min in a 1:1 mix of heptane and 3 . 7% formaldehyde in PBST . As the serosa tightly associates with the vitelline membrane , we used Tc-CHS1 RNAi eggs ( Jacobs et al . , 2013 ) , making it possible to manually dissect eggs containing the serosa from the vitelline membrane . The following primers were used to amplify 500-bp fragments of thaumatin1 , attacin1 , toll3 , and scavenger receptor B5 . Thaumatin1 FW 5′-CTAAGCGAAGGGGGTTTCGT-3′ RV 5′-TTTGTGGTCATCGTAGGCGT-3′ Attacin1 FW 5′-ATCGTCCAAGACCAGCAAGG-3′ RV 5′-GAAGCGGTGGCTAAACTGGA-3′ Toll3 FW 5′-AACTGGGAGGTTTTGCACAC-3′ RV 5′-AACTCCATTTTCCCCCAAAC-3′ SR-B5 FW 5′-AGCCAGGGAGTTCATGTTCG-3′ RV 5′-TGATTTGGTAACGGACGGCA-3′ PCR fragments were cloned into the TOPO II vector ( Invitrogen ) , according to the manufacturer's protocol . From these plasmids , templates for probe synthesis were amplified using M13 primers . DIG-labelled probes were synthesized using the MEGAscript kit ( Ambion , Austin , Texas , USA ) , according to the manufacturer's protocol , but with Roche RNA-labelling mix ( Roche , Basel , Switzerland ) . The data discussed in this publication have been deposited in NCBI's Gene Expression Omnibus ( Barrett et al . , 2013 ) and are accessible through GEO Series accession number GSE54018 ( http://www . ncbi . nlm . nih . gov/geo/query/acc . cgi ? acc=GSE54018 ) .	Insects are among the most numerous and diverse creatures on Earth , and over a million different species of insects have been described . Insects have a hard exoskeleton that protects their segmented bodies , and adult insects and their young are also well protected from pathogens . To fight off infection by bacteria or viruses , these creatures release antimicrobial molecules in the fluid that bathes their internal organs . Insects can also mount a localized immune response that kills off invading microbes . Most of what scientists have learned about the insect immune system has come from studying fruit flies . While much of the knowledge gained has been applicable to other insects , there is an important exception—fruit fly eggs are incredibly vulnerable to infection . Eggs from other insects are far better protected . In some species , the mother insect protects her eggs either through scrupulous care or by coating them with her own antimicrobial fluids . However , it was unclear if insect eggs could also defend themselves and counter an infection with a strong immune response . To better understand the immune response in insect eggs , Jacobs et al . studied the eggs of red flour beetles . These beetles are common agricultural pests that eat stored grains and are often studied by scientists in the laboratory . The beetle eggs share a trait with all other insect eggs that is missing from fruit flies and some other flies; the beetle eggs have an extra layer—called the serosa—that envelops the yolk and the developing embryo . To test whether this extra layer provides immune protection for the egg , Jacobs et al . used a technique called RNA interference to prevent the formation of the serosa . Beetle eggs either with or without a serosa were then pricked with a bacteria-covered object , and Jacobs et al . observed that the bacteria grew twice as fast in the eggs lacking a serosa compared with the eggs that had a serosa . Next , Jacobs et al . examined gene expression in response to the infection in the eggs . Over 500 genes that are expressed after an infection were identified , and of these genes , 481 were only expressed in eggs with a serosa . Three of these genes , including two that encode antimicrobial molecules , were looked at in more detail , and found to be only expressed within the serosa , indicating that the serosa is the most likely source of the egg's immune response . Importantly , Jacobs et al . found that eggs with a serosa produce the same immune system response as adult insects and concluded that most insect eggs are far from defenseless and are capable of fending off infection .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "immunology", "and", "inflammation" ]	2014	The extraembryonic serosa is a frontier epithelium providing the insect egg with a full-range innate immune response
The endosomal sorting complexes required for transport ( ESCRTs ) constitute hetero-oligomeric machines that catalyze multiple topologically similar membrane-remodeling processes . Although ESCRT-III subunits polymerize into spirals , how individual ESCRT-III subunits are activated and assembled together into a membrane-deforming filament remains unknown . Here , we determine X-ray crystal structures of the most abundant ESCRT-III subunit Snf7 in its active conformation . Using pulsed dipolar electron spin resonance spectroscopy ( PDS ) , we show that Snf7 activation requires a prominent conformational rearrangement to expose protein-membrane and protein-protein interfaces . This promotes the assembly of Snf7 arrays with ~30 Å periodicity into a membrane-sculpting filament . Using a combination of biochemical and genetic approaches , both in vitro and in vivo , we demonstrate that mutations on these protein interfaces halt Snf7 assembly and block ESCRT function . The architecture of the activated and membrane-bound Snf7 polymer provides crucial insights into the spatially unique ESCRT-III-mediated membrane remodeling . The endosomal sorting complexes required for transport ( ESCRTs ) are membrane remodeling machinery that mediate diverse fundamental cellular processes , including the biogenesis of multivesicular body ( MVB ) during receptor down-regulation ( Katzmann et al . , 2001 ) , enveloped virus budding ( Garrus et al . , 2001 ) , cytokinesis ( Carlton and Martin-Serrano , 2007 ) , plasma membrane repair ( Jimenez et al . , 2014 ) , nuclear pore complex assembly ( Webster et al . , 2014 ) , and nuclear envelope reformation ( Olmos et al . , 2015; Vietri et al . , 2015 ) . Originally identified using yeast genetics , ESCRTs package ubiquitinated transmembrane proteins into intraluminal vesicles ( ILVs ) that bud into the interior of the late endosome , creating a MVB that ultimately delivers cargos into the yeast lysosome ( vacuole ) . The ESCRT pathway achieves receptor sorting through an elaborate division of labor . Upstream ESCRT components , ESCRTs-0 , I , and II , assemble into stable hetero-multimers to sort ubiquitinated cargo on the endosomal surface by binding ubiquitin and endosomal lipid , phosphatidylinositol 3-phosphate ( PI ( 3 ) P ) . In addition , ESCRT-II sets the architecture and initiates the assembly of the ESCRT-III complex , which together with Vps4 is responsible for remodeling endosomal membranes ( Henne et al . , 2011; Hurley and Hanson , 2010 ) . ESCRT-III is a unique protein complex in that it is metastable and conformationally dynamic , forming hetero-oligomeric filaments of multiple subunits on membranes ( Saksena et al . , 2009; Teis et al . , 2008 ) . Its subunits are inactive monomers in the cytoplasm , which activate and assemble into spiraling polymers on endosomes to drive cargo sequestration , membrane invagination and constriction ( Buchkovich et al . , 2013; Hanson et al . , 2008; Henne et al . , 2012; Wollert and Hurley , 2010 ) . ESCRT-III is a hetero-polymer of four “core' subunits of Vps20 , Snf7/Vps32 , Vps24 and Vps2 ( Babst et al . , 2002 ) , and 'accessory' subunits of Ist1 , Did2/Vps46 , Vps60 ( Rue et al . , 2008 ) and Chm7 ( Horii et al . , 2006 ) . Although all ESCRT-III subunits share a common domain organization , each subunit appears to contribute a specific function . ESCRT-II directly engages Vps20 to trigger a sequential activation and ordered assembly of ESCRT-III subunits at endosomes ( Teis et al . , 2010 ) . Vps20 nucleates the homo-oligomerization of the most abundant ESCRT-III subunit , Snf7 , which then recruits Vps24 and Vps2 ( Teis et al . , 2008 ) . Vps2 finally engages the Vps4 complex for ESCRT-III disassembly ( Lata et al . , 2008b; Obita et al . , 2007 ) , making individual subunits available for subsequent rounds of vesicle formation . ESCRT-mediated membrane remodeling produces membrane curvature that pushes away from the cytoplasm , which is topologically opposite to that of the 'classical' clathrin and COP-I/II vesicle budding reactions . This unique membrane bending topology highlights an ancient and central role of the ESCRT machinery in cellular remodeling events . However , due to the relative instability and heterogeneity of ESCRT-III polymers , high-resolution structural studies have generally been problematic . Structural work on Snf7 in particular has been difficult , due to its ability to assemble readily into polymers that interfere with crystallization . Ultimately , atomic-resolution structural information is necessary to understand how ESCRT-III achieves ordered assembly and membrane remodeling in diverse cellular pathways . Even with limited structural information , previous studies have revealed distinct regions of Snf7 critical to ESCRT function . Snf7 contains a highly structured 'core' domain of four α-helices ( Muziol et al . , 2006 ) . The C-terminus , in contrast , is less structured , including an α-helix ( α5 ) that folds back against the core domain in cis to mediate autoinhibition ( Lata et al . , 2008a ) , a microtubule interacting and transport ( MIT ) -interacting motif ( MIM ) for Vps4 recognition ( Obita et al . , 2007 ) , and an α-helix ( α6 ) for Bro1/Alix interaction ( McCullough et al . , 2008 ) ( Figure 1A ) . 10 . 7554/eLife . 12548 . 003Figure 1 . X-ray Crystal Structure of Snf7core ( A ) The domain organization of Snf7 . The core domain used for X-ray crystallography is shown in blue . ( B ) Overlay of ribbon and space-filling models of the X-ray crystal structure of Snf7core . ( C ) Electrostatic surface potential of Snf7core with positively charged regions in blue ( +10kcal/e- ) and negatively charged regions in red ( -10kcal/e- ) . See also Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 00310 . 7554/eLife . 12548 . 004Figure 1—figure supplement 1 . Protein purification of Snf7core ( A ) A superdex-200 gel filtration size exclusion chromatogram of Snf7core . ( B ) A SDS-PAGE Coomassie brilliant blue staining of the gel filtration fractions corresponding to Snf7core . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 00410 . 7554/eLife . 12548 . 005Figure 1—figure supplement 2 . 2Fc-Fo simulated-annealing composite-omit electron density maps contoured at 1 . 0σ of Snf7core open conformations ( A ) A and ( B ) B . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 00510 . 7554/eLife . 12548 . 006Figure 1—figure supplement 3 . Superimposing Snf7core ( blue ) with ( A ) CHMP4Bα1-α2 ( cyan ) ( PDB: 4ABM ) , with ( B ) CHMP3α1-α4 ( purple ) ( PDB: 3FRT ) , with ( C ) CHMP6α1 ( red ) ( PDB: 3HTU ) Snf7core , and with ( D ) IST1α1-α6 ( grey ) ( PDB: 3FRR ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 006 How is Snf7 activated to promote ESCRT-III assembly ? Numerous studies indicate that ESCRT-III subunits are activated by intramolecular conformational changes that promote protein-protein interactions ( Henne et al . , 2012; Lata et al . , 2008a; Saksena et al . , 2009; Schuh et al . , 2015; Shen et al . , 2014 ) , but the structural basis for this is obscure . Available X-ray crystal structures of the autoinhibited Vps24 ( Muziol et al . , 2006 ) and Ist1 ( Bajorek et al . , 2009; Xiao et al . , 2009 ) suggest that these conformational changes involve the disruption of intramolecular interactions between the basic N-terminus and the acidic C-terminus . Upon releasing this autoinhibition , Snf7 subunits assemble into higher-order protofilaments or spirals ( Cashikar et al . , 2014; Hanson et al . , 2008; Henne et al . , 2012; Shen et al . , 2014 ) with a range of different morphologies and dimensions . Here , we present two X-ray crystal structures that unravel the molecular mechanism governing Snf7 conformational activation and polymer assembly . By selectively removing its autoinhibitory C-terminus , we determine the first crystal structure of the Snf7 core domain in the active conformation at 1 . 6 Å resolution . Surprisingly , rather than adopting a rigid four-helix coiled-coil , the core domain undergoes a large-scale conformational rearrangement to extend into a highly elongated structure . This conformational change not only extends a cationic membrane-binding surface , but also exposes hydrophobic and electrostatic protein interacting surfaces for polymerization . In vitro reconstitution and pulsed dipolar electron spin resonance spectroscopy ( PDS ) demonstrate that full-length Snf7 adopts the same active conformation and assembles into ~30 Å periodic protofilaments on a near-native lipid environment . Using negative stain transmission electron microscopy ( TEM ) and quantitative flow cytometry , we further demonstrate that mutations on key protein interfaces halt Snf7 assembly and block ESCRT function in vivo . Collectively , the molecular architecture of the activated and polymeric ESCRT-III Snf7 filament provides a detailed structural explanation for the mechanism underlying ESCRT-III-mediated membrane remodeling . Despite reconstituting and visualizing ESCRT-III assembly with the resolution of TEM , it was unclear how Snf7 is conformationally activated , and how this activation coordinates the assembly of Snf7 polymers on membranes . To answer these questions , we sought to determine the structure of Snf7 at atomic resolution . Because Snf7 intermolecular interactions rely primarily on core-core and core-membrane interactions ( Figure 1A ) ( Buchkovich et al . , 2013; Henne et al . , 2012 ) , we purified Snf7core to homogeneity ( Figure 1—figure supplement 1 ) . We then crystallized and solved X-ray crystal structures of Snf7core in two conformations at 1 . 6 Å and 2 . 4 Å resolutions , respectively . The structures were determined by molecular replacement using CHMP4Bα1-α2 ( PDB: 4ABM ) ( Table 1 , Figure 1—figure supplement 2 ) . Although two conformations were determined , they share a similar overall tertiary structure with one notable exception discussed further below . 10 . 7554/eLife . 12548 . 007Table 1 . Crystallographic Data Collection and Refinement StatisticsDOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 007Snf7coreConformation AConformation BWavelength ( Å ) 0 . 9780 . 978Resolution range ( Å ) 50 - 2 . 4 ( 2 . 49 - 2 . 40 ) 50 - 1 . 6 ( 1 . 6 - 1 . 55 ) Space groupP21P21Unit cella = 29 . 5Å b = 52 . 2Å c = 54 . 5Å α = 90°β = 97 . 5°γ = 90°a = 29 . 9Å b = 46 . 2Å c = 44 . 6Å α = 90°β = 98 . 5°γ = 90°Total reflections23263 ( 1946 ) 73723 ( 6034 ) Unique reflections6376 ( 612 ) 16849 ( 1581 ) Multiplicity3 . 6 ( 3 . 2 ) 4 . 4 ( 3 . 8 ) Completeness ( % ) 97 . 99 ( 93 . 72 ) 95 . 77 ( 90 . 65 ) Mean I/sigma ( I ) 8 . 04 ( 2 . 91 ) 8 . 85 ( 1 . 35 ) Wilson B-factor54 . 0325 . 39Rmerge0 . 0884 ( 0 . 249 ) 0 . 0782 ( 0 . 997 ) CC1/20 . 988 ( 0 . 968 ) 0 . 995 ( 0 . 590 ) CC*0 . 997 ( 0 . 992 ) 0 . 999 ( 0 . 861 ) Rwork0 . 259 ( 0 . 398 ) 0 . 210 ( 0 . 330 ) Rfree0 . 262 ( 0 . 533 ) 0 . 225 ( 0 . 356 ) Number of non-hydrogen atoms9821097macromolecules975992water7105Protein residues123125RMS ( bonds ) ( Å ) 0 . 0150 . 006RMS ( angles ) ( o ) 1 . 240 . 81Ramachandran favored ( % ) 9599Ramachandran outliers ( % ) 1 . 70Clashscore21 . 569 . 9Average B-factor91 . 139 . 7macromolecules91 . 238 . 9solvent69 . 247 . 2 All previous ESCRT-III X-ray crystal structures adopt a canonical four α-helical core domain fold ( Bajorek et al . , 2009; Muziol et al . , 2006; Xiao et al . , 2009 ) . When we superimposed our Snf7 structure with available ESCRT-III structures ( Figure 1—figure supplement 3 ) , we were surprised to note that Snf7core does not fold into four α-helices , but instead , it contains only three α-helices that pack into a highly elongated structure ( Figures 1B–C ) . Although the α1/2 hairpin is relatively unchanged , α3 and α4 undergo large-scale structural rearrangements from the proposed autoinhibited ESCRT-III fold . α2 extends into a ~90 Å long α-helix combining the α2 and α3 segments that were distinct α-helices in previously defined ESCRT-III structures ( Figures 2A–B ) . The angle of the flexible loop between α3 and α4 also changes , which enables α4 to position in different orientations relative to the α1-3 hairpin . Despite the conformational change , we designated this elongated α-helix as α2/3 to maintain a consistent numbering scheme for conserved ESCRT-III helices . 10 . 7554/eLife . 12548 . 008Figure 2 . Conformational Rearrangement of Snf7 ( A–B ) Ribbon diagrams of ( A ) a homology model of closed Snf7core ( Henne et al . , 2012 ) and ( B ) the X-ray crystal structure of open Snf7core . ( C ) A close-up view of the side chain interaction between Gln90 and Met130 . ( D ) Western blotting and subcellular fractionation of snf7Δ yeast exogenously expressing SNF7 or snf7Q90C M130Cwith and without copper ( II ) 1 , 10-phenanthroline . ( E ) Schematic showing closed and open Snf7core with cysteines ( red dots ) before and after SDS-denaturing . ( F ) Snf7 site-directed spin-labeling with MTSL ( red ) . ( G–H ) Distance between Glu88 and His118 of ( G ) closed and ( H ) open Snf7 shown in ribbon . ( I and K ) Time domain signals and distance distributions from DEER spectroscopy of ( I ) Snf7R52E E88C H118C in solution , and simulated closed and open Snf7core E88C H118C using MMM , and ( K ) Snf7R52E E88C H118C: Snf7R52E ( 1:0 , 1:1 , 1:2 , and 1:8 ) with liposomes . ( J ) Schematic showing liposome sedimentation for DEER . MTSL-labeled Snf7 proteins ( blue oval ) and liposomes ( grey circle ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 00810 . 7554/eLife . 12548 . 009Figure 2—figure supplement 1 . Conceptual model for the Mup1-pHluorin MVB sorting assay . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 00910 . 7554/eLife . 12548 . 010Figure 2—figure supplement 2 . Sequence alignments of Snf7 α2 and α4 , with conserved Gln90 and Met130 shown in red , and quantitative MVB sorting data for snf7Δ yeast exogenously expressing SNF7 , snf7Q90C , snf7M130C , and snf7Q90C M130C . Error bars represent standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 01010 . 7554/eLife . 12548 . 011Figure 2—figure supplement 3 . Time domain signals and distance distributions from DEER spectroscopy of full-length Snf7R52E E88C H118C , Snf7R52E H118C G140C and Snf7R52E E88C G140C . ( A ) Ribbon models of closed and open Snf7core showing inter-residue distances between E88 , H118 and G140 . ( B–D ) Time domain signals and distance distributions from DEER spectroscopy of ( B ) full-length Snf7R52E E88C H118C , and full-length Snf7R52E E88C H118C:Snf7R52E ( 1:1 ) in solution , ( C ) full-length Snf7R52E H118C G140Cin solution and simulated closed and open Snf7core H118C G140C using MMM , and ( D ) full-length Snf7R52E E88C G140C in solution and simulated closed and open Snf7core E88C G140C using MMM . Blue shaded portions of the distributions indicate distance ranges that can be attributed to open and closed conformations . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 011 Previous studies suggested that a 'closed' Snf7 becomes activated by the displacement of α5 away from the core domain ( Henne et al . , 2012; Lata et al . , 2008a ) . Using a homology model of closed Snf7 ( Henne et al . , 2012 ) ( Figure 2A ) , we identified close proximity between conserved residues Gln90 ( α2 ) and Met130 ( α4 ) in the four-helix coiled-coil ( Figure 2C ) . We applied a cysteine-based crosslinking strategy to directly monitor the conformational states of Snf7 in vivo . We mutated both Gln90 and Met130 to cysteines , and expressed this mutant in snf7Δ yeast ( Figure 2—figure supplements 1 and 2 ) . Since conformationally active Snf7 resides on endosomal membranes , we performed subcellular fractionation and collected the supernatant ( S13 ) and the membrane-enriched pellet ( P13 ) fractions . Western blotting analysis showed that Snf7Q90C M130C migrated to ~37 kDa , comparable to cysteineless Snf7 . We then oxidized both fractions using copper ( II ) 1 , 10-phenanthroline . Strikingly , in the S13 fraction , ~50% of Snf7Q90C M130C migrated faster , indicating a conformationally closed Snf7 species . Notably , in the P13 fraction , the migration shift did not occur ( Figures 2D–E ) . This is indicative of distinct conformations between the cytoplasmic and the endosome-bound states , and suggests that Snf7 on endosomal membranes adopts an open conformation in which α4 is displaced away from α2 . To investigate Snf7 activation at a structural level , we applied the PDS technique of double electron-electron resonance ( DEER ) and monitored full-length Snf7 in solution and bound to liposomes . As an approach to characterize protein conformations ( Borbat and Freed , 2007; Borbat and Freed , 2014; Jeschke , 2012 ) , PDS can provide distance constraints with a range of ~10–90 Å by measuring the magnitude of the dipolar coupling between spins of unpaired electrons in nitroxide spin labels ( Hubbell et al . , 2000 ) . Snf7 assembles into spiraling protofilaments on membranes , presenting two challenges: ( 1 ) to characterize the conformational state of Snf7 building blocks; and ( 2 ) to determine the protofilament assembly from these structural elements . To determine whether Snf7 activation induces the 'open' conformation we observed by X-ray crystallography , we selected two solvent-accessible residues , Glu88 ( α2 ) and His118 ( α3 ) , predicted to be separated by a short distance of 20 Å in the closed state ( Figure 2G ) , and an expected longer distance of 45 Å in the open state ( Figure 2H ) . We labeled these two sites with a nitroxide spin label , MTSL ( Figure 2F ) , and then obtained the distance distribution for the full length Snf7 in solution . The result showed a wide distance spread of ~15–50 Å ( Figure 2I ) , corresponding to large amplitude motions of the spin labeled positions , but not a distinct closed or open state . Thus , soluble Snf7 is structurally heterogeneous , suggesting that it is conformationally dynamic ( Figure 2—figure supplement 3 ) . To map the active conformation of Snf7 , we reconstituted spin-labeled full-length Snf7R52E polymers on lipid membranes , where R52E is a previously characterized activation mutant that induces Snf7 polymerization ( Henne et al . , 2012 ) . We mixed the double-labeled Snf7R52E E88C H118C proteins with liposomes and collected the membrane-bound Snf7 polymers by ultracentrifugation ( Figure 2J ) . Intriguingly , membrane-bound Snf7R52E E88C H118C produced a strong ~30 Å peak . We also observed a significant population of distances at 40–50 Å , but diminished signal at ~20 Å ( Figure 2K ) . We postulated that both the inter- and intra-subunit interspin distances contribute these signals . To isolate the intra-subunit interspin distance , we next produced magnetically diluted samples ( Borbat and Freed , 2007; Dzikovski et al . , 2011 ) by mixing double-labeled Snf7R52E E88C H118C with unlabeled Snf7R52E in ratios ranging from 1:1 to 1:8 . We observed that the signal changed significantly up to 1:2 dilution , then less for the maximal 1:8 dilution ( Figure 2K and Figure 3—figure supplement 2 ) , showing approach to the infinite dilution limit . The data for the 1:8 dilution is characteristic of a single long distance of 45 Å with a moderate distance distribution , as expected for spin labels on an α-helix separated by 29 residues . In summary , the reconstructed distance distributions are consistent with structural rearrangements that transform α2 and α3 into one continuous α-helix in the membrane-bound active conformation . As we did not observe short distances corresponding to the closed conformation , we conclude that only the open conformation is present in Snf7 polymers assembled on membranes . Therefore , the large-scale conformational rearrangement observed in the crystal structures is fully consistent with the PDS data of the full-length Snf7 conformations on the membranes . While examining the arrangement of Snf7 molecules in the crystal lattice , we noted that multiple Snf7 protomers are arrayed into polymeric lattices , reminiscent of the protofilaments previously observed by TEM ( Henne et al . , 2012 ) . Each of the ~100 Å long α1–3 hairpin tilts by ~27° and polymerizes into a ~45 Å diameter single protofilament , with each protomer exhibiting a repeat distance of ~30 Å ( Figure 3A ) . 10 . 7554/eLife . 12548 . 012Figure 3 . Membrane-bound Snf7 Protofilament with ~30 Å Periodicity ( A ) Overlay of ribbon and space-filling models of a 7-mer Snf7 protofilament with measured dimensions . ( B and D ) Time domain signals and distance distributions from DEER spectroscopy of ( B ) full-length Snf7R52E T20C , Snf7R52E K35C , and Snf7R52E E88C with liposomes , ( D ) full-length Snf7R52E K60C , Snf7R52E H118C , and Snf7R52E G140C with liposomes . ( C and E ) Schematic showing the spin label positions in a Snf7 protofilament . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 01210 . 7554/eLife . 12548 . 013Figure 3—figure supplement 1 . Time domain signals and distance distributions from DEER spectroscopy of full-length Snf7R52E K60C A66C in solution and full-length Snf7R52E K60C A66C: Snf7R52E ( 1:0 , 1:2 ) with liposomes , and schematic showing the locations of the spin label positions in a Snf7 protofilament . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 01310 . 7554/eLife . 12548 . 014Figure 3—figure supplement 2 . Time domain signals and distance distributions from DEER spectroscopy of full-length Snf7R52E E88C H118C: Snf7R52E ( 1:0 , 1:2 . 5 , 1:4 , 1:8 ) with liposomes and simulated Snf7core E88C H118C: Snf7core ( 1:0 , 1: ∞ ) polymers using MMM , and schematic showing the locations of the spin label positions in a Snf7 protofilament . The full-length Snf7R52E E88C H118C: Snf7R52E ( 1:0 and 1:8 ) with liposomes datasets are re-plotted from Figure 2K as shown in fine lines . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 01410 . 7554/eLife . 12548 . 015Figure 3—figure supplement 3 . Quantitative MVB sorting data for snf7Δ yeast exogenously expressing SNF7 , snf7T20C , snf7K35C , snf7K60C , snf7E88C , snf7H118C , snf7G140C , snf7K60C A66C and snf7E88C H118C . Error bars represent standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 01510 . 7554/eLife . 12548 . 016Figure 3—figure supplement 4 . Representative TEM images of recombinant full-length Snf7R52E K35C , Snf7R52E E88C , Snf7R52E K60C A66C , and Snf7R52E E88C H118C labeled with MTSL . Scale bars , 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 016 The spacing of protomers in the crystal is also in agreement with our DEER results of the full-length Snf7 protofilaments assembled on liposomes . We performed a series of DEER measurements on the single-labeled protein at several key positions . Specifically , we selected the middle of the α1-3 hairpin ( Thr20 , Lys35 and Glu88 ) , and at both ends of α2/3 ( Lys60 and His118 ) , and the end of α4 ( Gly140 ) to spin label Snf7 . This allowed us to probe the interface between adjacent Snf7 protomers to establish their mutual orientation . Importantly , these cysteine-substituted Snf7 mutants are capable of assembling into protofilaments in vitro and do not impair MVB sorting in vivo ( Figure 3—figure supplement 3 and 4 ) . Consistent with the extensive amount of inter-molecular contacts revealed in the Snf7 crystal , we observed moderately broad distance distributions , specifically at 28–32 Å for T20C , K35C and E88C ( Figures 3B–C ) , and at 32–36 Å for K60C , H118C and G140C ( Figures 3D–E ) . The modulation depths of the time-domain echo signals indicate a ~3-spin system , in agreement with the crystalline arrangement of Snf7 , where each protomer has two neighboring protomers . The magnetic dilution ( Figure 3—figure supplements 1 and 2 ) readily removed the intersubunit couplings , indicating that protofilaments do not make extensive contacts homogenous with each other . Based on this series of single-cysteine DEER scanning and the double-cysteine magnetic dilution experiments , we conclude that Snf7 packing adopts a parallel arrangement in a single-layer array with a period of ~30 Å , and the reconstituted full-length Snf7 spirals on liposomes adopt a packing pattern similar to the Snf7core crystals . Thus , our X-ray crystal structures provide a foundation for in-depth study of the membrane-bound Snf7 polymer . In the Snf7 protofilament , the protomer ( i ) interacts with the next protomer ( i+1 ) through both hydrophobic and electrostatic interactions ( Figures 4A–B ) , burying ~1060 Å2 of solvent-accessible surface area per protomer . The assembly of the extended α2/3 helix exposes a hydrophobic surface on α3 , which was buried in the closed state . This enables the α2/3 helix of protomer ( i ) to interact with α2/3 of its neighboring protomer ( i+1 ) ( Figure 4C ) . Notably , Gln90 , which interacts with Met130 in cis in the closed state , interacts with Met107 in trans in the open state . 10 . 7554/eLife . 12548 . 017Figure 4 . Hydrophobic and Electrostatic Interactions in a Snf7 Filament ( A–B ) Ribbon models of a Snf7 protofilament . The hydrophobic protein interface is shown in black dash-line and the electrostatic interface in grey dash-dot line . ( C–D ) Close-up views of the hydrophobic interface between α2/3i and α3i+1 and the electrostatic interface between α1i and α2/3i+1 . Protomer ( i ) shown in yellow and protomer ( i+1 ) in red . ( E ) Conceptual model for the Mup1-pHluorin MVB sorting assay . Vacuole ( v ) . ( F ) Quantitative MVB sorting data for snf7Δ yeast exogenously expressing empty vector , SNF7 , snf7L121D , snf7I117E , snf7M114E , snf7M107E , snf7T103E , snf7L99K , snf7M104E , snf7L101E , snf7A97K , snf7I94E , snf7Q90K , snf7M87E , and snf7T83E . Error bars represent standard deviations . ( G ) Quantitative MVB sorting data for snf7Δ yeast exogenously expressing empty vectors , SNF7 , snf7R25E H29E K36Eand empty vector , empty vector and snf7E95K E102K E109K , and snf7R25E H29E K36E and snf7E95K E102K E109K . Error bars represent standard deviations . ( H ) Representative TEM images of recombinant full-length Snf7R52E , Snf7R52E Q90K , Snf7R52E M107E , Snf7R52E R25E H29E K36E , Snf7R52E E95K E102K E109K , and Snf7R52E R25E H29E K36E and Snf7R52E E95K E102K E109K ( 1:1 ) . Scale bars , 200 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 01710 . 7554/eLife . 12548 . 018Figure 4—figure supplement 1 . Hydrophobic Interface Mutant Analysis . ( A ) Representative TEM images of recombinant full-length Snf7R52EI94E and Snf7R52EM114E . Scale bars , 200nm . ( B ) Superdex-200 gel filtration size exclusion chromatograms of Snf7R52E , Snf7R52EI94E and Snf7R52E M107E . Related to Figure 4H . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 01810 . 7554/eLife . 12548 . 019Figure 4—figure supplement 2 . Western blotting analyses of snf7Δ yeast expressing SNF7 , snf7L121D , snf7I117E , snf7M114E , snf7M107E , snf7T103E , and snf7L99K , and SNF7 , snf7M104E , snf7L101E , snf7A97K , snf7I94E , snf7Q90K , snf7M87E , and snf7T83E . G6PDH used as loading controls . Sequence analyses of Snf7 α2/3 with conserved residues shown in gold and dark red . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 01910 . 7554/eLife . 12548 . 020Figure 4—figure supplement 3 . Quantitative MVB sorting data for snf7Δ yeast exogenously expressing empty vector , SNF7 , snf7R25E , snf7H29E , snf7K36E , snf7E95K , snf7E102K , and snf7E109K , and empty vector , SNF7 , snf7R25E K36Eand vector , vector and snf7E95K E109K , snf7R25E K36Eand snf7E95K E109K . Error bars represent standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 02010 . 7554/eLife . 12548 . 021Figure 4—figure supplement 4 . Western blotting analyses of snf7Δ yeast expressing SNF7 , snf7R25E H29E K36E , and snf7E95K E102K E109K . Sequence analyses of Snf7 α2/3 with conserved residues shown in gold and dark red . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 02110 . 7554/eLife . 12548 . 022Figure 4—figure supplement 5 . Western blotting analyses of ex vivo P13 fractions BMOE crosslinking by Snf7K35C with Snf7K60C , Snf7A63C , Snf7K69C , Snf7Q75C , Snf7E81C , Snf7E88C , Snf7E95C , and Snf7E102C . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 022 To validate the interactions present in this hydrophobic interface , we performed site-directed mutagenesis and tested each mutant in vivo by an established quantitative MVB sorting assay ( Buchkovich et al . , 2013; Henne et al . , 2012 ) . This assay monitors the efficiency of fluorescence quenching after internalization and MVB sorting of Mup1-pH ( the plasma membrane methionine transporter , Mup1 , fused to the pH-sensitive GFP-derivative , pHluorin ) ( Figures 4E and Figure 2—figure supplement 1 ) . As a result , mutants M104E , L101E , A97K , I94E , Q90K , M87E and T83E showed severe sorting defects , with sorting efficiencies from 12% to 76% , and mutants L121D , I117E , M114E , M107E , T103E and L99K from 7% to 34% ( Figure 4F and Figure 4—figure supplement 2 ) . Correspondingly , we previously demonstrated that the L121D mutant blocks Snf7 polymerization in vivo and in vitro , and missorts the MVB cargo carboxypeptidase S , Cps1 ( Saksena et al . , 2009 ) . Furthermore , recombinant Snf7R52E Q90K , Snf7R52E I94E , Snf7R52E M107E , and Snf7R52E M114E proteins were able to be purified to homogeneity , but unable to generate protofilaments visible by TEM ( Figure 4H and Figure 4— figure supplement 1 ) . We also observed electrostatic interactions between α1 of protomer ( i ) and α2/3 of protomer ( i+1 ) ( Figure 4D ) . This interaction is also dependent upon the extension of α2/3 , and appears to be important for the positioning of α1 in the protofilament . To validate whether these inter-protomer electrostatic interactions occur in vivo , we generated and tested charge-inversion mutations , snf7R25E H29E K36E and snf7E95K E102K E109K , which resulted in severe sorting defects of 16% and 44% , respectively . Strikingly , when co-expressing both mutants in trans , MVB sorting was restored to 91% ( Figure 4G and Figure 4—figure supplements 3 and 4 ) . Consistently , Glu95 has been previously indicated to be involved in Snf7 inter-protomer contacts ( Shen et al . , 2014 ) . These results are further supported by ex vivo crosslinking experiments . In the Snf7 polymer-enriched P13 fraction , cysteine-substituted Lys35 ( α1 ) can be specifically crosslinked to cysteine-substituted Glu95 ( α2 ) or Glu102 ( α3 ) in trans ( Figure 4—figure supplement 5 ) . Furthermore , co-incubating recombinant Snf7R52E R25E H29E K36E and Snf7R52E E95K E102K E109K proteins resulted in protofilament formation , but no protofilaments were detected when each mutant was tested individually ( Figure 4H ) . Altogether , these in vivo and in vitro data provide strong evidence that the observed hydrophobic and electrostatic interfaces are required for Snf7 polymerization in vivo , and that the Snf7 protofilament observed in the crystal lattice is physiologically relevant . We next mapped the previously determined Snf7 membrane-interacting region ( Buchkovich et al . , 2013 ) onto the Snf7 polymer structure ( Figure 5A ) . Strikingly , several key conserved lysine residues , K60 K64 K68 K71 K79 ( α2 ) , and K112 K115 ( α3 ) , which were in distinct α-helices in the closed state , are arranged on an elongated and solvent-exposed surface ideal for interacting with the acidic endosomal membrane . The electrostatic membrane-binding regions of all Snf7 protomers face the same direction in the polymer , allowing for a continuous membrane-binding interface ( Figure 5B ) . Thus , the crystal structure of Snf7 polymers reveals a mechanism for coupling polymerization to stable membrane association . 10 . 7554/eLife . 12548 . 023Figure 5 . Electrostatic Protein-membrane Interactions in a Snf7 Filament ( A ) A Snf7 protofilament in ribbons placed on a lipid membrane in spheres ( grey ) ( Heller et al . , 1993 ) . ( B ) Electrostatic surface potential showing the membrane interacting surface of a Snf7protofilament with positively charged regions in blue ( +10kcal/e- ) and negatively charged regions in red ( -10kcal/e- ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 023 Notably , ESCRT-III subunits utilize multiple hydrophobic and electrostatic interfaces to interact with endosomal membranes ( Buchkovich et al . , 2013 ) . Consistent with this , we observed that α1 of Snf7 is moderately positively charged ( Figure 1C ) , and cannot rule out that at some stage of vesicle biogenesis it also comes in contact with the membrane . Comparison of our Snf7core crystal structures we determined revealed two distinct conformations . Although both structures exhibit an open conformation , we noted two different orientations of α4 with respect to the α1/2 hairpin . In open conformation A , α4 extends in the protofilament plane , whereas in open conformation B , α4 is positioned perpendicular to the protofilament plane . Superimposing the two conformations revealed that α4 can rotate by at least ~90° along the axis of the α2/3 helix ( Figures 6A–B ) . Despite the large differences in α4 positioning , α4 makes a similar interaction with the α1/2 hairpin of a Snf7 protomer in a neighboring protofilament in both structures ( Figure 6C and Figure 6—figure supplement 4 ) . This supports a model in which the assembly of the extended α2/3 helix upon Snf7 activation results in two key events: ( 1 ) α4 can no longer bind in cis to its own protomer; ( 2 ) α5 is displaced from the α1/2 hairpin . Together , this enables α4 to contact the α1/2 hairpin of another protomer in trans on a neighboring protofilament . 10 . 7554/eLife . 12548 . 024Figure 6 . Snf7 α4 in Inter-Filament Interactions ( A–B ) Snf7core conformations A ( green ) and B ( blue ) superimposed . ( B ) 90° rotation and superimposing with a closed CHMP3 ( purple ) using its α3 as a reference . ( C ) Overlay of ribbon and space-filling models of the Snf7core crystal packing of the open conformation A . The dash-line box represents the interfilament contacts . Arrows represent inter-protofilament orientations . ( D–E ) Close-up views of the hydrophobic interface between α1/2i ( blue ) and α4j ( yellow ) of open conformations ( D ) A and ( E ) B . ( F ) Quantitative MVB sorting data for snf7Δ yeast exogenously expressing empty vector , SNF7 , snf7V126E , snf7M130E , snf7I133E , snf7A51E , snf7L55E , and snf7L67E . Error bars represent standard deviations . See also Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 02410 . 7554/eLife . 12548 . 025Figure 6—figure supplement 1 . Representative TEM images of recombinant full-length Snf7R52E V126E and Snf7R52E I133E . Scale bars , 200nm . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 02510 . 7554/eLife . 12548 . 026Figure 6—figure supplement 2 . Western blotting analyses of snf7Δ yeast expressing SNF7 , snf7A51E , snf7L55E , snf7L67E , snf7V126E , snf7M130E , and snf7I133E . G6PDH as a loading control . Sequence analyses of Snf7 α1/2 and α4 with conserved residues shown in blue or gold . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 02610 . 7554/eLife . 12548 . 027Figure 6—figure supplement 3 . Quantitative MVB sorting data for snf7Δ yeast exogenously expressing SNF7 , snf7E102P , snf7N59P , and snf7L121P . Error bars represent standard deviations . Overlay ribbon models of ( upper right ) closed ( purple ) and open ( blue ) Snf7core with Glu102 shown in sticks , and ( lower right ) open conformation A ( green ) and B ( blue ) with Asn59 and Leu121 shown in sticks . Arrows represent conformational rearrangements . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 02710 . 7554/eLife . 12548 . 028Figure 6—figure supplement 4 . An overlay of ribbon and space-filling models of the Snf7core crystal packing of the open conformation B . The dash-line box represent the interfilament contacts shown in Figure 6E . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 02810 . 7554/eLife . 12548 . 029Figure 6—figure supplement 5 . Superimposing of Snf7core subunit ( i ) ( blue ) , ( j ) ( yellow ) and CHMP3α1-α5 ( purple ) of open conformations A ( upper ) and B ( lower ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 029 Consistent with the DEER data that two Snf7 protofilaments do not make extensive contacts with each other and do not assemble into homogeneous arrangements , this interfilamental interface only buries 474 Å2 of solvent-accessible surface area per protomer . To test whether these observed interfilamental interactions were functionally important , we mutated residues at their α1/2i-α4j interface ( Figures 6D–E ) . Notably , Met130 , which interacts with Gln90 in the closed state , is involved in this interface in the open state . Snf7 mutants of A51E , L55E , L67E , V126E , M130E and I133E led to drastic loss-of-function , with sorting efficiencies from 9% to 55% in vivo , and were unable to assemble into protofilaments in vitro ( Figure 6F and Figure 6—figure supplements 1 and 2 ) . To gain insights into the importance of the local rearrangement of the Snf7 α3/4 loop in vivo , we mutated the conserved α3/4 loop residue Leu121 to Pro to constrain the rotational angle between α3 and α4 . The L121P mutant exhibited a MVB sorting efficiency of 32% , compared to that of the α1/2 loop residue Asn59 mutant N59P of 75% ( Figure 6— figure supplement 3 ) , suggesting that the α3/4 loop functions as an important flexible 'hinge' that may facilitate different architectural stages of Snf7 polymers ( Figure 6C and Figure 6—figure supplement 4 ) . Interestingly , studies have previously shown that the tip of the α1/2 hairpin is important for intra- and inter-molecular contacts of ESCRT-III subunits . For example , X-ray crystal structures of CHMP3 and IST1 are autoinhibited through an intramolecular contact between the α1/2 hairpin and α5 ( Figure 6—figure supplement 5 ) ( Bajorek et al . , 2009; Muziol et al . , 2006 ) ; and the Ist1-Did2 co-crystal structure revealed that the MIM1 of CHMP1B forms an intermolecular contact with the α1/2 hairpin of Ist1 ( Xiao et al . , 2009 ) . To gain insights into any functionally important surfaces on the Snf7 structure , we performed CONSURF analysis ( Celniker et al . , 2013 ) . As a result , we identified seven highly conserved regions in the Snf7core domain ( Figure 7A and Figure 7— figure supplement 1 ) . Strikingly , all of them map to regions of Snf7 assigned specific functions in either polymer assembly or membrane interaction: regions ( 1 ) and ( 2 ) are located on opposite sides of the extended α2/3 helix and stabilize intrafilamental protein-protein interactions; region ( 3 ) is located towards the N-terminus of α1 and region ( 4 ) towards the middle of the α2/3 helix , forming the intrafilamental electrostatic interacting surfaces; region ( 5 ) corresponds to the beginning of α2 , which we previously identified as a cationic membrane-binding surface; regions ( 6 ) and ( 7 ) are the tip of the α1/2 hairpin and the hydrophobic side of α4 , which together stabilize interfilamental interactions . Thus , the Snf7 protein-protein interactions revealed from our X-ray crystal structures and the protein-membrane interactions previously identified ( Buchkovich et al . , 2013 ) are evolutionally conserved . 10 . 7554/eLife . 12548 . 030Figure 7 . Models of Snf7 activation , polymer assembly and membrane remodeling ( A ) Space-filling CONSURF models with high conservation ( purple ) and low conservation ( cyan ) . Interacting protomers shown in ribbon ( blue ) . Seven conserved regions with assigned functions labeled . Gray arrows indicate the flexibility of α4 . ( B ) Speculative cartoons illustrating four stages in ESCRT-mediated vesicle budding . ( C ) Space-filling models and schematic cartoons of Snf7core in closed and open states with membrane ( grey ) . ( D ) Space-filling and close-up ribbon models of a 25-mer Snf7 single filament with membrane . ( E ) Space-filling and close-up ribbon models of a 23-mer Snf7 normal mode analysis filament with membrane ( grey ) . ( F ) Schematic of a Snf7 homo-polymer in the neck of a nascent ILV with positive and negative membrane curvatures . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 03010 . 7554/eLife . 12548 . 031Figure 7—figure supplement 1 . Alignment of Snf7core protein sequences from Saccharomyces cerevisiae ( Sc ) , Homo sapiens ( Hs ) , Mus musculus ( Mm ) , Xenopus laevis ( Xl ) , Drosophila melanogaster ( Dm ) , Caenorhabditis elegans ( Ce ) , Schizosaccharomyces pombe ( Sp ) and Lokiarchea ( Spang et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 03110 . 7554/eLife . 12548 . 032Figure 7—figure supplement 2 . A ribbon model of a supercomplex of Vps25-Vps20-Snf7 . The first Snf7’s α1 was used for superimposing with the Vps20 α1 ( Im et al . , 2009 ) ( PDB: 3HTU ) for molecular docking . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 03210 . 7554/eLife . 12548 . 033Figure 7—figure supplement 3 . Architectures of Snf7 protofilaments ( A ) A representative TEM image of recombinant Snf7R52E ( left ) and a space-filling model of a 61-mer Snf7α1-3 straight filament shown in the same scale ( right ) . ( B ) A representative TEM image of recombinant full-length Snf7R52E , Vps24 and Vps2 ( 2:1:1 ) ( left ) , and space-filling and close-up view of ribbon models of a 97-mer Snf7α1-3 superhelix generated by normal mode analysis with measured dimensions ( right ) . TEM scale bars , 50nm . DOI: http://dx . doi . org/10 . 7554/eLife . 12548 . 033 A linear Snf7 filament has a simple two-dimensional geometry , and thus is incapable of mediating the drastic three-dimensional membrane remodeling required for membrane deformation and vesicle formation at the endosome ( Figure 7B ) . Since we know that Snf7 can form spirals on a membrane surface , we asked how a linear Snf7 filament ( Figure 7D and Figure 7—figure supplement 3 ) could be transformed into a circular array . Due to the heterogeneity of a Snf7 double filament , we utilized a single linear Snf7 filament to determine a plausible curved Snf7 filament using normal mode analysis ( NMA ) , a simple non-detailed simulation technique used to probe large-scale macromolecular motions by assessing flexibility intrinsic to the structure of a protein ( Suhre and Sanejouand , 2004 ) . Remarkably , without any dramatic structural rearrangement within each protomer or alterations of the protein-protein interface , a linear two-dimensional Snf7 filament can bend into a ~70 nm diameter three-dimensional superhelix with turn length of ~62 nm , reminiscent of the structure of the Snf7/Vps24/Vps2 co-assembly previously observed ( Henne et al . , 2012 ) ( Figure 7—figure supplement 3 ) . Importantly , this 3D helical array aligns the cationic membrane-binding surfaces on the outside of the superhelix , ideal for a Snf7 polymer to stabilize a negatively curved membrane surface ( Figure 7E ) . A classic model of ESCRT-III activation involves the disruption of intramolecular interactions between α5 and the core domain ( Henne et al . , 2012; Lata et al . , 2008a ) . In the present study , we provide surprising structural evidence that this activation requires further rearrangement within the core domain itself . Consistent with the available CHMP4Bα1/2 structure ( Martinelli et al . , 2012 ) , α1/2 folds into a rigid ~70 Å α-helical hairpin , which forms intramolecular contacts with at least three short α-helices , α3 , α4 , and α5 in the closed state . Notably , upon activation , all of these interactions are reorganized to extend the hairpin to a ~90 Å structure available for intermolecular contacts . Comparison of the closed and open states reveals that α4 is displaced by ~60 Å as the molecule opens . Intriguingly , a recent small-angle X-ray scattering ( SAXS ) study showed that Vps20 exists as a 94 Å extended 'open' conformation but it is incapable of homo-polymerization ( Schuh et al . , 2015 ) . Based on these structural insights , we propose a detailed 'lifecycle' of Snf7 activation and polymerization: 1 ) In the cytoplasm , Snf7 exists in a dynamic equilibrium of mixed intermediates between the open and closed states; 2 ) on endosomes , Vps20 α1 directly associates with the ESCRT-II subunit Vps25 ( Im et al . , 2009 ) , allowing Vps20 to function as a nucleator to engage an open Snf7 from the cytoplasm ( Figure 7—figure supplement 2 ) ; 3 ) the open conformation of Snf7 with an extended α2/3 helix presents a cationic membrane-interacting surface to orient itself on endosomes; 4 ) the N-terminal membrane ANCHR motif further stabilizes Snf7 on the endosomal surface ( Figure 7C ) ; 5 ) the endosomal recruitment shifts the conformational equilibrium and thus triggers a 'domino effect' of Snf7 opening and promotes Snf7 polymerization into a ~30 Å periodic array of ordered inter-protomer contacts . In agreement with this , an averaged 32 . 5 Å inter-subunit distance was observed in C . elegans Vps32 spirals ( Shen et al . , 2014 ) . Because X-ray crystal structures of both Vps24 ( Muziol et al . , 2006 ) and Ist1 ( Bajorek et al . , 2009 ) were determined in their autoinhibitory conformations with an unresolved 'linker' between the core and α5 , the four-helix core domain has been treated as a rigid body that remains unaltered between the open and closed states . However , a previous SAXS study suggested that Vps24 can adopt both a 75 Å globular and a 105 Å extended conformation ( Lata et al . , 2008a ) , implying that the core domain extension may be a common theme of ESCRT-III activation . Due to this unexpected conformational change , previous ESCRT-III polymer studies using the 'closed' conformation as a building unit may need careful reevaluation . The ESCRT-III Snf7 filament-mediated membrane remodeling is conceptually reminiscent of other membrane remodeling machinery , including bacterial FtsZ ( Osawa et al . , 2008 ) . Interestingly , both Snf7 and FtsZ/FtsA can drive cytokinetic abscission , and they share at least three distinct structural characteristics: electrostatic protein-membrane interactions , membrane insertion of an amphipathic helix , and oligomeric protein scaffolding . Despite these similarities , the major difference between FtsZ and Snf7 is that FtsZ requires nucleotide hydrolysis to drive its conformational dynamics . The propagation of conformational changes in the FtsZ polymer is thus coupled with the architectural changes that promote membrane fission . In contrast , ESCRT-III does not bind nor hydrolyze nucleotides to regulate its conformation , but it recruits the AAA-ATPase Vps4 for its disassembly . Although Snf7 can be activated by specific point mutations in vitro , the conformational switching in vivo appears to be tightly regulated by other ESCRT components to prevent pre-mature polymer assembly . During MVB biogenesis , ESCRT-II binds two copies of Vps20 , which then nucleates the homo-oligomerization of Snf7 . However , in enveloped viral budding and cytokinetic abscission , Bro1/Alix directly bridges ESCRT-I to ESCRT-III , by binding to the C-terminal α6 of Snf7 ( McCullough et al . , 2008 ) . We speculate that this interaction may directly dissociate the C-terminal autoinhibitory region to trigger Snf7 polymer assembly . Furthermore , CHMP7 was recently shown to trigger Snf7 assembly during nuclear envelope reformation ( Vietri et al . , 2015 ) , highlighting the distinct spatial and temporal regulation of Snf7 activation between different ESCRT-dependent processes . ILVs that bud into the endosomal lumen contain no outer vesicle coat , yet show consistent diameters , suggesting ESCRTs regulate vesicle size . Somewhat paradoxically , ESCRT-III cannot shape the vesicle exterior because it is segregated in the cytoplasm by the limiting membrane of the endosome . Instead , the ESCRT-III filament appears to predominantly drive membrane deformation by sculpting the neck interior of a growing vesicle . This membrane sculpting requires an intricate balance of competing curvatures . Snf7 has been shown to localize to both the curved necks of invaginations and along highly curved edges of membranes ( Buchkovich et al . , 2013; Fyfe et al . , 2011; Wollert and Hurley , 2010 ) ( Figure 7F ) . Our collective studies on Snf7 address the balance of membrane curvatures associated with ILV formation . The ANCHR motif of Snf7 acts to sense and stabilize the positively curved rim of the invagination ( Figure 7C ) . Coinciding with this positive curvature stabilization , the helical Snf7 polymer acts as a circular scaffold that triggers and stabilizes the negatively curved circumference of the neck of the invagination ( Figure 7E ) . We propose that as a two-dimensional ESCRT-III spiral elongates into a three-dimensional superhelix , the tight membrane binding of the 'corkscrew' concentrates transmembrane cargoes ahead of the leading edge of the forming and narrowing filament , packaging them into the nascent ILV ( Figure 7B ) . Despite the reconstitution and high-resolution analysis of Snf7 polymers , key questions remain . The most pressing are the mechanisms governing inter-ESCRT-III subunit interactions , particularly , Vps24 and Vps2 , required for the ESCRT-III architectural changes , and a precise mechanochemical role of the AAA-ATPase Vps4 complex during the final membrane constriction and scission coupled with ESCRT-III disassembly . Additional structural studies together with new assays are necessary for further addressing these challenging but exciting questions . The DNA sequence encoding Saccharomyces cerevisiae Snf7core ( residues 12–150 ) was subcloned into a pET28a vector with an N-terminal His6-Sumo tag . Recombinant proteins were overexpressed in Escherichia coli Rosetta cells and purified by TALON metal affinity resin . The His6-Sumo tag was removed by Ulp1 protease at 4°C overnight . The mixture was further purified by Superdex-200 gel filtration . The peak corresponding to Snf7core was pooled and concentrated in a buffer of 300 mM NaCl , 20 mM HEPES pH7 . 4 . Snf7core ( conformation A ) was crystallized in a hanging-drop vapor diffusion system at 4°C by mixing protein ( 5 . 7 mg/mL ) with reservoir solution containing 100 mM NaCl , 100 mM MES:NaOH pH5 . 5 , 3% PEG20 , 000 in 1:1 ratio ( v/v ) . Crystals were transferred into the same solution supplemented with 30% glycerol before cooling to liquid nitrogen temperature under atmosphere pressure . Snf7core ( conformation B ) crystals were grown in 110 mM NaCl , 70 mM MES:NaOH pH5 . 5 , 6% PEG20 , 000 and subject to high-pressure cryo-cooling ( Kim et al . , 2005 ) . The crystals were mounted in oil on a pin with a piece of steel piano wire attached to the base , pressurized to 200MPa and cooled to liquid nitrogen temperature . The pressure was then released while keeping the temperature unaltered . X-ray diffraction data was collected on Snf7 crystal 'A' to 2 . 4 Å at MacCHESS beam line F1 of Cornell High Energy Synchrotron Source . The crystal belonged to space group P21 with unit cell dimensions a=29 . 5 Å b=52 . 2 Å c=54 . 5 Å α=90° β=97 . 5° γ=90° . X-ray diffraction data was collected on crystal 'B' to 1 . 6 Å . It belonged to space group P21 with unit cell dimensions a=29 . 9 Å b=46 . 2 Å c=44 . 6 Å α=90° β=98 . 5° γ=90° . Diffraction data were processed using HKL-2000 ( Otwinowski and Minor , 1997 ) . There is one Snf7 molecule in the asymmetric unit of both crystals . The structures were solved using Phaser in Phenix ( Adams et al . , 2010 ) by molecular replacement with CHMP4Bα1-α2 ( PDB: 4ABM ) as a search model . Refinement and density modification were performed in Phenix . Model building was performed using Coot ( Emsley and Cowtan , 2004 ) . Throughout this study , structural images were generated with PyMOL using the 1 . 6 Å structure unless otherwise noted . All Snf7 protein purification for PDS and TEM analyses were performed as previously described ( Henne et al . , 2012 ) . Briefly , Snf7 constructs were subcloned into a pET23d bacterial expression vector ( Novagen ) with an N-terminal His6-tag . Recombinant proteins were overexpressed by Escherichia coli BL21 or C41 cells , purified by TALON metal affinity resin and eluted in 150 mM NaCl , 20 mM HEPES pH7 . 4 and 400 mM Imidazole . The elution fractions were pooled and further purified by Superdex-200 gel filtration in a buffer of 150 mM NaCl , 20 mM HEPES pH7 . 4 . Recombinant Snf7 cysteine-substituted proteins were purified and enriched on TALON resin , and spin-labeled with 1 μg/mL S- ( 1-oxyl-2 , 2 , 5 , 5-tetramethyl-2 , 5-dihydro-1H-pyrrol-3-yl ) methyl methanesulfonothioate , MTSL ( Santa Cruz Biotech ) dissolved in acetonitrile at 4°C overnight . The spin-labeled proteins were eluted in 150 mM NaCl , 20 mM HEPES pH7 . 4 , 400 mM Imidazole and further purified by Superdex-200 gel filtration in a buffer of 150 mM NaCl , 20 mM HEPES pH7 . 4 to remove unreacted spin labels . For soluble protein samples , spin-labeled proteins were buffer exchanged in a 10 kDa molecular weight cutoff protein concentrator ( Millipore ) to ~80% deuterium buffer of 150 mM NaCl , 20 mM HEPES pD7 . 4 supplemented with 30% ( v/v ) glycerol-d8 . For liposome-reconstituted protein samples , 1 mg/mL of 800 nm diameter 60% 1 , 2-dioleoyl-sn-glycero-3-phosphocholine ( DOPC ) , 30% 1 , 2-dioleoyl-sn-glycero-3-phospho-L-serine ( DOPS ) , 10% phosphatidylinositol 3-phosphate ( PI ( 3 ) P ) liposomes were generated as previously described ( Buchkovich et al . , 2013 ) . 25 μL of 10-30 μM proteins and 25 μL 1 mg/mL liposomes were coincubated at room temperature for 15 min and ultracentrifuged in a TLA-100 rotor ( Beckman Coulter ) for 10 min at 70 , 000 rpm at 20°C . A total of 6 liposome pellets were combined and resuspended in 20 μL deuterium buffer of 150 mM NaCl , 20 mM HEPES pD7 . 4 supplemented with 15% ( v/v ) glycerol-d8 , resulting in a sample of ~10-30 μM protein: ~3 mg/mL lipid for DEER measurements . 20 μL spin-labeled samples were loaded into 1 . 8 mm inner diameter Pyrex sample tubes ( Wilmad-LabGlass ) , shock frozen in liquid nitrogen prior to DEER measurements . DEER measurements were performed at 60 K using a home-built Ku band 17 . 3 GHz pulse electron spin resonance spectrometer ( Borbat et al . , 1997; Borbat et al . , 2013 ) . A four-pulse DEER sequence ( Jeschke and Polyhach , 2007 ) was used routinely with the detection π/2- and π-pulses having widths of 16 and 32 ns and pump π-pulse of 16 ns . The detection pulse sequence was applied at the low-field spectral position , while pumping was performed at a lower by 70 MHz frequency positioned at the central maximum . A 32-step phase cycle ( Gamliel and Freed , 1990 ) was applied to suppress unwanted contributions to the signal . Nuclear modulation effects caused by surrounding protons were suppressed by averaging the data from 4 measurements with slightly different separations of the first two pulses , i . e . advanced by 9 . 5 ns for subsequent measurement . Depending on spin-labeled protein concentration , distance , and phase relaxation time , DEER data were usually acquired in less than 12 hr . Time-domain DEER data , V ( t ) , were reconstructed into distance distributions using standard approaches ( Borbat and Freed , 2007; Borbat and Freed , 2014; Jeschke , 2012; Jeschke and Polyhach , 2007 ) . First , the signal decay due to intermolecular spin interactions was removed from V ( t ) by approximating the latter points ( about a half of the record ) of lnV ( t ) with a low-order polynomial , usually nearly a straight line , and subtracting it out from lnV ( t ) so that the antilog yields u ( t ) . Once normalized as V ( t ) =u ( t ) u ( 0 ) , it serves as a typical form of DEER data presentation , while u ( t ) -1 gives background free data , which was subsequently converted to a distance distribution between spin pairs with L-curve Tikhonov regularization ( Chiang et al . , 2005a ) followed , when needed , by maximum entropy method refinement ( Chiang et al . , 2005b ) . The modulation depth , defined as 1-V ( ∞ ) , where V ( ∞ ) is the asymptotic value of V ( t ) , was used to report on the presence and extent of multispin effects ( Bode et al . , 2007 ) . For mapping Snf7 conformation , we employed double spin-labeled Snf7 and magnetic dilution ( Borbat and Freed , 2007; Dzikovski et al . , 2011; Meyer et al . , 2014; Pornsuwan et al . , 2013 ) . Figure 3—figure supplement 1 demonstrates a benchmark magnetic dilution study of double-labeled Snf7R52E K60C A66C with unlabeled Snf7R52E . This spin pair at the tip of the α1/2 hairpin was selected as a reference for inspecting bound protein conformational variability and the conditions for isolation of intramolecular distances . The distance of this construct in solution is ~20 Å , in agreement with spin-label modeling into a homology structure ( Figure 2A ) using MMM ( molecular multiscale modeling ) software package ( Polyhach et al . , 2011 ) . Generic MTSL rotamer library for 298 K was used to determine conformations of attached spin labeled cysteine side chains and produce distance distributions between pairs of labeled sites . Distance distributions FWHMs were in the range of 0 . 4-1 . 2 nm . Respective background free time-domain data were generated with the help of the same package . Consistently , Snf7 was found to be structurally more heterogeneous in solution , producing broad distributions based on DEER data for a set of double-labeled Snf7 full-length constructs ( Figure 2— figure supplement 3 ) . Intriguingly , this study revealed a distinct Snf7 conformation in the protofilaments , which manifests itself as a very narrow distance distribution already at mild magnetic dilution ( 1:2 ) , thus pointing to a low extent of intermolecular contacts . In the absence of unlabeled proteins ( 1:0 magnetic dilution ) , Snf7R52E K60C A66C in liposome samples produced broad distributions , which showed a range of distances to neighbors with ~30 Å being dominant ( Figure 3—figure supplement 1 ) . In addition , the large modulation depth indicated coupling to at least two neighbors . This indicated that for isolating longer distances considerably higher dilution ratios would be desirable . Figures 2I , K and Figure 3—figure supplement 2 illustrate subsequent application of this method to the membrane-bound Snf7R52E E88C H118C . Note that in Figure 2I , the reconstructed distance distribution of soluble Snf7R52E E88C H118C is normalized at a 4x scale than the MMM simulation data to illustrate the structural heterogeneity . In Figure 2K , the reconstructed distance distributions have a large component of ~30 Å originating from distances to immediate neighbors similar to the benchmark case , the magnetically diluted samples have a single peak at ~45 Å that is dominant with only a small fraction at 30 Å that could still be noticed at 1:8 dilution . A dilution factor in excess of 15 would be necessary to fully reveal the expected signal shape , however the 1:8 dilution sample already has ~5 μM protein concentration , making larger ratios problematic to study . Snf7 polymeric packing was assayed by inspecting intermolecular dipolar couplings for various single-labeled constructs assembled in protofilaments on liposome membranes . Whereas the most pronounced distance is expected to be determined by the proximal neighbors , the widths of distance distributions ( Figures 3B–E ) obtained in these scans are likely to have contributions from the couplings to more distant neighbors and in addition by the complex nature of the Snf7 polymer in a liposome-reconstituted system where Snf7 spiraling double- protofilaments are observed , and the orientation relative to each other is heterogeneous ( Cashikar et al . , 2014; Henne et al . , 2012 ) . Notably , while searching for 'tip-to-tip' contacts possible in double protofilaments , we did not identify any spin labeled position with a distinct short proximity that is expected to occur at the contacting edge of the single filament , thus ruling out this scenario . We also did not discern any significant distance variation as the spin labeled position is moved from one end of the α1-3 hairpin to the other , thus ruling out any alternating protomer packing in the protofilaments . So far , only parallel protomer packing in a single-layer filament is consistent with the data ( see also Result ) . Subcellular fractionation experiment was performed as previously described ( Buchkovich et al . , 2013 ) . Briefly , 30 OD600nmV of mid-log yeast cultures were spheroplasted in Zymolyase and lyzed in 1 mL of 50 mM Tris pH7 . 4 , 1 mM EDTA , 200 mM sorbitol with protease inhibitors ( Roche ) . Lysates were cleared at 500 xg for 5 min at 4°C , and then fractionated by centrifugation at 13 , 000 xg for 10 min at 4°C . The supernatant ( S13 ) fraction was collected . The pellet ( P13 ) fraction was resuspended in 1 mL lysis buffer . Both fractions were then precipitated by 10% trichloroacetic acid for at least 30 min and washed by acetone twice . The oxidizing chemical copper ( II ) 1 , 10-phenanthroline was prepared freshly . 9 mg copper ( II ) sulfate was dissolved in 250 μL ionic buffer of 150 mM postasium acetate , 5 mM magnesium acetate , 250 mM sorbitol , 20 mM HEPES pH7 . 0 . 20 mg 1 , 10-phenanthroline was dissolved in 500 μL ethanol . Both solutions were mixed creating a brilliant aqua-colored solution with white precipitate . 7 μL copper ( II ) 1 , 10-phenanthroline solution was added into 450 μL of S13 or P13 fractions , and incubated at 4°C for 15 min . Samples were then precipitated by 10% trichloroacetic acid , washed twice by acetone and subjected for western blotting analysis . 30 OD600nmV of mid-log yeast cultures were spheroplasted , lyzed and fractionated . The 1 mL P13 fractions were equally divided into two subfractions . Subfraction 1 was treated with 20 μL DMSO and subfraction 2 with 20 μL 20 mM bismaleimidoethane ( BMOE ) ( Life Technologies ) in DMSO for 2hours at 4°C . Excessive BMOE was quenched by adding 0 . 2 μL 1 M dithiothreitol . Samples were then precipitated by 10% trichloroacetic acid , washed twice by acetone and subjected for western blotting analysis . The quantitative Mup1-pHluorin ESCRT cargo-sorting flow cytometry assay , negative stain TEM , and western blotting were performed as previously described ( Buchkovich et al . , 2013; Henne et al . , 2012 ) . See Supplementary file 1 for a list of plasmids and yeast strains used . Saccharomyces cerevisiae Snf7 protein sequence was input as a query sequence for a protein BLAST analysis using the algorithm of blastp ( protein-protein BLAST ) . The top 100 sequences from the result were subjected for ClustalW sequence alignment . The multiple sequence alignment and the Snf7 conformation B structure were then used as input for conservational analysis using the CONSURF server ( Ashkenazy et al . , 2010; Berezin et al . , 2004; Celniker et al . , 2013 ) . The overall conservation scores calculated using the Bayesian method were color-coordinately mapped onto the Snf7 structure shown in Figure 7A . Calculation of the normal modes of the Snf7 polymer was preformed on the elNémo server ( Suhre and Sanejouand , 2004 ) , by using a 25-mer of Snf7 of conformation B as an input structure . To model a circular structure with a diameter of ~65–70 nm , perturbation parameters of DQMIN of -10000 , DQMAX of 10000 , and DQSTEP of 2000 were applied . This yielded 3 nontrivial normal modes numbered 7 , 8 and 9 . The lowest frequency nontrivial normal mode , mode 7 , was used . Using Coot , the middle 12 protomers of the No . 7 normal mode were selected and then superimposed in a head-to-tail fashion to manually generate a 23-mer and 94-mer shown in Figures 7E and Figure 7—figure supplement 3 . Coordinates and structure factors for Snf7core have been deposited in the RCSB Protein Data Bank ( http://www . rcsb . org ) under accession PDB ID 5FD7 ( open conformation A ) and 5FD9 ( open conformation B ) .	A cell constantly remodels its surface to adapt to its environment , as well as to replace old or damaged proteins . To achieve this , cell-surface receptors are taken inside the cell and delivered to organelles called endosomes , where a molecular machine called ESCRT governs the receptors’ fate . Distinct ESCRT complexes remodel the endosomal membrane to form vesicle packages that encapsulate the receptor proteins . These vesicles bud off into the endosome , which is then targeted to another organelle called the lysosome where the receptor proteins are degraded . If the vesicles are unable to make their deliveries , the resulting sustained receptor activity can lead to numerous developmental and neurodegenerative diseases , as well as cancer . Remarkably , the ESCRT machinery also plays critical roles during cell division and the release of the human immunodeficiency virus ( HIV ) from host cells . Previous studies demonstrated that a particular ESCRT complex , called ESCRT-III , forms spiraling filaments on membranes to generate vesicles . However , how the individual components of ESCRT-III assemble into such filaments was a mystery . Now , Tang et al . have determined the first X-ray crystal structures of the main component of ESCRT-III , a polymer of the protein called Snf7 , and thus uncovered how these membrane-bound Snf7 spirals assemble . Using a combination of cell biology , genetics and biochemistry techniques , Tang et al . also demonstrated that the Snf7 structures are necessary for ESCRT-III to work correctly inside living cells . Despite this achievement , key questions remain . The main one is how the other subunits of ESCRT-III interact and work together to remodel the membrane to form the vesicle packages at the endosomes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2015	Structural basis for activation, assembly and membrane binding of ESCRT-III Snf7 filaments
Contractile actomyosin networks have been shown to power tissue morphogenesis . Although the basic cellular machinery generating mechanical tension appears largely conserved , tensions propagate in unique ways within each tissue . Here we use the vertebrate eye as a paradigm to investigate how tensions are generated and transmitted during the folding of a neuroepithelial layer . We record membrane pulsatile behavior and actomyosin dynamics during zebrafish optic cup morphogenesis by live imaging . We show that retinal neuroblasts undergo fast oscillations and that myosin condensation correlates with episodic contractions that progressively reduce basal feet area . Interference with lamc1 function impairs basal contractility and optic cup folding . Mapping of tensile forces by laser cutting uncover a developmental window in which local ablations trigger the displacement of the entire tissue . Our work shows that optic cup morphogenesis is driven by a constriction mechanism and indicates that supra-cellular transmission of mechanical tension depends on ECM attachment . The shape of animal organs evolved by natural selection under constrains imposed both by organ physiology in the adult and tissue mechanics during embryogenesis . Throughout development , genetic programs coordinate the behavior of single cells allowing the self-assembly of coherent tissues and tridimensional organs . Regardless of the nature of the process ( i . e . either cell migration , epithelial bending or cell intercalation ) , mechanical tensions need to be transmitted at a supra-cellular scale for organ morphogenesis to occur . Mechanical forces , however , are generated by the contractile cytoskeleton of the constituent cells of a tissue ( Mammoto et al . , 2013; Heisenberg and Bellaiche , 2013 ) . The main force generator during morphogenesis results from the molecular interaction between myosin II motors and the actin filaments at the cellular cortex ( Salbreux et al . , 2012 ) . This actomyosin contractile apparatus sustains cortical tension , pulling cells into shape during development and tissue homeostasis . Contractile forces are then transmitted to neighboring cells and to the extracellular matrix ( ECM ) through cadherin and integrin receptors , allowing individual cell contributions to be integrated into tensions at the tissue/organ level ( Papusheva and Heisenberg , 2010; Lecuit et al . , 2011 ) . Regardless the morphogenetic context , actomyosin contractile forces are resisted both by cellular adhesions and by the compression of the internal cytoskeleton itself . This results in a balance of forces that stabilizes transiently cell and tissue shapes for each stage of the developmental program that builds up a given organ . Live-imaging studies have examined actomyosin architecture and dynamics in different morphogenetic models . The emerging picture reveals a wide variety of cortical actomyosin behaviors and localizations depending on the tissue context . Initial reports , focused in epithelial constriction processes , revealed pulsatile myosin flows preceding the periodic contraction of the cellular cortex . This has been reported in Drosophila epithelia either at the apical cortex , during mesoderm invagination or germ-band extension ( Martin et al . , 2009; Gorfinkiel and Blanchard , 2011; Roh-johnson et al . , 2012; Rauzi et al . , 2010 ) , or at the basal surface during egg chamber elongation ( He et al . , 2010 ) . Oscillatory actomyosin flows can be coupled to the stabilization of the cells in a 'constricted' state after each pulse , thus resulting in a progressive ( i . e . ratcheted ) reduction of the cellular apex ( Martin et al . , 2009; Rauzi et al . , 2010 ) . Alternatively , the cell cortex may oscillate , contracting and relaxing , without a net reduction of the area over time ( He et al . , 2010; Solon et al . , 2009 ) . Furthermore , actomyosin flows may direct epithelial morphogenesis operating in a continuous non-pulsatile manner , as described during zebrafish epiboly ( Behrndt et al . , 2012 ) . Notably , the actomyosin network localizes in circumferential ( i . e . junctional ) belts in the vertebrate neural tube ( Nishimura et al . , 2012 ) , instead of medio-apically as observed in several Drosophila epithelia ( Gorfinkiel and Blanchard , 2011; Martin et al . , 2009 ) and in gastrulating cells in Xenopus ( Kim and Davidson , 2011 ) . In the context of the current study , although actomyosin distribution has been analyzed during optic cup morphogenesis in vertebrates ( Chauhan et al . , 2009; Martinez-morales et al . , 2009 ) , its dynamics has not been examined in vivo . Vertebrate eye development has been a common subject of interest for classical embryologists as well as modern developmental geneticists ( Spemann , 1901; Fuhrmann , 2010; Sinn and Wittbrodt , 2013 ) . The process entails first the protrusion of the eye progenitors to form the lateral optic vesicles , and subsequently the infolding of this tissue into bi-layered optic cups ( Li et al . , 2000; Schmitt and Dowling , 1994; Hilfer , 1983; Schook , 1980 ) . Live imaging followed by cell tracking of retinal progenitors in zebrafish revealed that optic vesicle bulging is driven by the rearrangement and epithelialization of individual cells ( Brown et al . , 2010; Rembold et al . , 2006; England et al . , 2006; Ivanovitch et al . , 2013 ) . In contrast to teleosts , in amniotes and cartilaginous fishes optic vesicles develop by epithelial folding from an already hollow neural tube ( Lowery and Sive , 2004 ) . The morphogenesis of the vertebrate optic cup has also been examined in live imaging studies , both in teleost models ( Kwan et al . , 2012; Martinez-morales et al . , 2009; Picker et al . , 2009; Heermann et al . , 2015 ) , as well as in self-organized organs from ES-cultured cells in mammals ( Nakano et al . , 2012; Eiraku et al . , 2011 ) . Although optic cup formation seems less divergent among vertebrates than vesicles’ evagination , some particularities in cell behavior have been observed and different mechanisms proposed . In mouse embryos , contractile filopodia connecting neural retina and lens epithelia have been shown to adjust the final curvature of both epithelia ( Chauhan et al . , 2009 ) . However , optic cup development can be recapitulated in vitro in ES cells aggregates suggesting that the morphogenetic program is to a large extent intrinsic . Using this in vitro model , it has been hypothesized that optic cup invagination is driven by the apical constriction of the neuroepithelial cells located at the rim between the presumptive retina and RPE domains ( Eiraku et al . , 2011 , 2012 ) . Tracking of individual cells in zebrafish has shown that epithelial flow through this rim contributes to neural retina expansion ( i . e . at the expenses of the RPE ) and optic cup folding ( Heermann et al . , 2015; Kwan et al . , 2012; Picker et al . , 2009 ) . Whether cell involution and apical constriction at the rim are species-specific mechanisms or operate coordinately in the same organism is still an open question . Finally , we previously postulated the basal constriction of the neuroblasts as an active mechanism contributing to optic cup morphogenesis ( Martinez-Morales et al . , 2009; Martinez-Morales and Wittbrodt , 2009 ) . The polarized trafficking of integrin receptors toward the basal surface of the epithelial cells plays an essential role during retinal morphogenesis in teleosts . We showed that this process is controlled by the molecular antagonism between the trans-membrane protein opo and the clathrin adaptors numb and numb-like ( Bogdanovic et al . , 2012 ) . In opo medaka mutants , basal feet appear wider and disorganized in the retina ( Martinez-morales et al . , 2009 ) . Although this observation suggests a progressive reduction of the neuroblasts feet , the constriction process has not been formally examined in vivo . Through quantitative imaging , here we characterize the pulsed contractile behavior of the retinal neuroblasts during optic cup folding in zebrafish . We explore actomyosin dynamics and show that accumulation of myosin foci in scattered cells is associated with contraction of the cellular feet . We show that interference with myosin II function or laminin-mediated basal attachment impairs cell contractility and affect retina folding . To further characterize this morphogenetic process at tissue level , we locally ablate the neuroepithelium to map mechanical tensions through development . This approach identified a narrow developmental window in which local ablation of the retina at its basal surface triggers the global displacement of the retinal epithelium . Our work shows that the myosin-dependent generation of constrictions forces in individual neuroblast and their transmission at a supra-cellular scale play an essential role during optic cup folding in zebrafish . To formally show that basal constriction is taking place as the optic cup forms , we investigated the behavior of retinal precursors by live-imaging analysis . Retinae from the zebrafish line tg ( vsx2 . 2:GFP-caax ) , in which precursors’ plasma membrane is uniformly labeled , were imaged through morphogenesis starting at 17 hpf ( Figure 1A–H; Video 1 ) . Tissue recordings evidenced a complete epithelial organization shortly after 17 hpf , with mitotic rounding happening apically throughout the entire folding process . In agreement with previous reports , cell involution was also observed at the rim between the RPE and neural retina , particularly from 20 hpf on and at the posterior ( i . e . temporal ) border of the cup ( Heermann et al . , 2015; Kwan et al . , 2012; Picker et al . , 2009 ) . As morphogenesis proceeds , GFP-caax signal become brighter at the basal side in the central retina , suggesting an increased membrane density in this region . Moreover , whereas the length of the apical edge of the retina increased significantly , the basal length remained invariant ( Figure 1I ) . This observation , in conjunction with the previously reported increase ( 1 . 5x ) in retinal cells number within this developmental window ( Kwan et al . , 2012 ) , suggests a progressive narrowing of the basal feet between 17 and 24 hpf . Cell elongation , a common phenomenon in many constricting epithelia ( Sawyer et al . , 2010 ) , does not occur during retinal folding , as the width of the tissue remained constant ( ≈50 µm ) through the process ( Figure 1J–L ) . 10 . 7554/eLife . 15797 . 003Figure 1 . Folding of the retinal epithelium in zebrafish . ( A–H ) Time series of optical sections show the progression of retinal morphogenesis starting at 17 hpf ( dorsal view ) in a tg ( vsx2 . 2:GFP-caax ) embryo . Arrowheads point to mitotic divisions at the apical surface . Apical and basal edges are indicated at 60 ( purple ) and 420 ( orange ) min . See also Video 1 . ( I ) Quantification of the perimeter of the apical and basal edges between 18 and 24 hpf . ( J–L ) Retinal width remains constant throughout retinal folding as revealed in tg ( vsx2 . 2:GFP-caax ) embryos . Error bars indicate s . d . of the mean . ( n = 3; T-test ) . Antero-posterior and medio-lateral axes are indicated . Scale bars = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 00310 . 7554/eLife . 15797 . 004Figure 1—figure supplement 1 . Imaging setup and segmentation . ( A , A’ ) Schematic representation of the imaging setup . Confocal planes for panels B–D are indicated in A’ . ( B–D ) Optical sections through a 20 hpf tg ( vsx2 . 2:GFP-caax ) retina showing basal ( orange in C ) and apical ( purple in D ) planes . Mitotic figures ( m ) and antero-posterior axis ( a–p ) are indicated . fb = forebrain . ( E–G ) Automatic cell segmentation ( E–E’ ) and manual tracking of the segmented cells through time ( F , G ) are shown . Scale bars = 50 µm in B–D and 5 µm in E–G . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 00410 . 7554/eLife . 15797 . 005Figure 1—figure supplement 2 . Neuroblasts’ area quantification during eye morphogenesis . ( A ) Quantification of average cell areas at the apical and basal sides at 19 , 20 and 21 hpf . A total of 24 cells from three different embryos were recorded either at the apical or at the basal side . Error bars indicate SE of the mean ( n = 24 ) . Statistical significance was determined after T-test . ( B–C ) The percentage of cells showing a contraction , or relaxation larger than 20% over a 25 min period is indicated for the three different stages . A total of 24 cells were monitored at both basal ( B ) and apical ( C ) surfaces . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 00510 . 7554/eLife . 15797 . 006Video 1 . Time lapse of zebrafish optic cup folding . Optical section from a tg ( vsx2 . 2:GFP-caax ) embryo showing the folding of the retinal tissue . Imaging starts at 17 hpf . Antero-posterior and medio-lateral axes are indicated . See also Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 006 To investigate directly the constriction process , we examined the dynamics of both apical and basal neuroblasts’ surfaces within the most critical morphogenetic window , between 19 and 21 hpf , using again the tg ( vsx2 . 2:GFP-caax ) line . Processed images were segmented and individual cell areas tracked through time ( Figure 1—figure supplement 1 ) . During this developmental window , basal areas shrank significantly ( 40% ) and progressively from 25 . 4 ± 1 . 7 to 15 . 3 ± 1 . 5 µm2 ( n = 24 ) . Maximum basal constriction was observed between 19 and 20 hpf when most of the cells significantly reduced their area in a 30 min period ( 74 . 2% and 66 . 7% respectively; Figure 1—figure supplement 2 ) . Interestingly , this developmental window coincides with the acute bending of the retinal epithelium ( Figure 1 ) . In contrast , apical areas remained constant between 19 and 20 hpf and even expanded ( 28% ) at later stages , between 20 and 21 hpf ( Figure 1—figure supplement 2 ) . Live imaging analyses revealed periodic contractions occurring at apical and basal cell surfaces ( Video 2 , Figure 2 ) , which may resemble the pulsatile behavior observed in constricting epithelia in both vertebrate and invertebrate tissues ( Martin et al . , 2009; Solon et al . , 2009; Rauzi et al . , 2010; He et al . , 2010; Kim et al . , 2011 ) . As previously reported for Drosophila epithelia ( Martin et al . , 2009 ) , analysis of pulsed contractions in adjacent retinal cells revealed that these are mostly asynchronous ( Figure 2—figure supplement 1 ) . The analysis of individual cells from three independent retinas showed that 76% of the apical ( n = 43 ) and 90% of the basal ( n = 46 ) oscillations presented no major correlation with those of their neighbors ( Pearson correlation coefficient R < \|0 . 5\| ) . Comparison of the pulsatile behavior at both epithelial planes revealed significant differences . Although both surfaces oscillate with a similar frequency of 50 ± 12 . 5 mHz ( ≈20 ± 5 s; n = 26 cells ) , the peak-to-peak amplitude is considerably larger at the basal 11 . 1 ± 1 . 3 µm2/min than at the apical surface 4 . 1 ± 0 . 57 µm2/min ( Figure 2—figure supplement 1 ) . Of note , whereas a progressive reduction of cell area was apparent at the basal side , cells did not display a net constriction at the apical side over a 25-min period ( Figure 2 ) . This observation confirms the basal constriction of the retinal neuroepithelium during optic cup morphogenesis . 10 . 7554/eLife . 15797 . 007Figure 2 . Quantitative analysis of membrane oscillations in tg ( vsx2 . 2:GFP-caax ) embryos . Cell area dynamics at the basal ( A–-D ) and apical ( E–H ) surfaces is shown for three individual cells ( color coded ) . Absolute basal ( D ) and apical ( H ) areas in µm2 are represented versus time for the individual cells . The mean area indicates a progressive constriction of the basal , but not apical surfaces over time ( D , H ) . Scale bars = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 00710 . 7554/eLife . 15797 . 008Figure 2—figure supplement 1 . Quantitative analysis of cell pulses . ( A–-B ) Single cell recordings of area variations in µm2 ( purple ) and constriction changes in µm2/min ( orange ) at the basal ( A ) and apical ( B ) surfaces are represented over time . ( C—D ) The evaluation of constriction rates in adjacent cells shows asynchronous pulsing . ( E–G ) Distribution of correlation coefficients between neighboring cell pairs is represented as bins for basal ( E ) and apical ( F ) oscillations , as well as in a box plot ( G ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 00810 . 7554/eLife . 15797 . 009Video 2 . Membrane oscillations at the basal and apical surfaces . Maximum projection of 3 z-stacks ( over a total of 1 µm ) at the basal and apical surfaces in a tg ( vsx2 . 2:GFP-caax ) retina show the oscillatory behavior of the cell membranes over a period of 35 min . Images were acquired every 5 s . Scale bars = 10 µm . See also Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 009 As periodic contractions occur at both neuroblasts’ ends with a similar frequency , we next ask whether apical and basal surfaces oscillate synchronically . To answer this issue , we generated retinal clones by blastomere transplantation from tg ( vsx2 . 2:GFP-caax ) donor embryos into wild-type late-blastula hosts . Live-imaging analysis of singularized tg ( vsx2 . 2:GFP-caax ) neuroblasts along the apico-basal axis allowed the simultaneous recording of variations in the length of the apical and basal edges at 20 hpf ( Video 3 ) . Quantitative analysis of 10 individual cells revealed a poor correlation between the pulses at apical and basal ends ( R < \|0 . 5\| in all cells examined ) , thus indicating that these surfaces oscillate largely in an independent manner ( Figure 3 ) . A second emerging question was whether apical cell rounding during mitosis may affect either basal constriction or apical expansion . To address this issue , the distance between the two cells flanking mitotically active neuroblasts was measured through time . Whereas quantitative analysis of distance variation showed a transient expansion of the apical domain as the cells divide , this was recovered once mitoses were resolved ( Figure 3—figure supplement 1 ) . Thus , both the apical and basal net distances at the beginning and end of the process did not change significantly ( T-test; n = 10 ) . This observation is in agreement with previous data showing that cell mitoses did not play a major role for optic cup formation ( Kwan et al . , 2012 ) . 10 . 7554/eLife . 15797 . 010Figure 3 . Analysis of tg ( vsx2 . 2:GFP-caax ) clones show uncoupled oscillations at apical and basal surfaces . ( A ) Scheme of transplantation experiment at sphere stage . ( B ) Confocal microscopy image showing transmitted light and GFP expression for transplanted clones ( white arrows ) at 20 hpf . Antero-posterior ( a–p ) axis is indicated . ( C–E ) Confocal microscopy time-lapse images show length variation of basal ( orange ) and apical ( purple ) edges through time in a transplanted clone . The orientation of the apico-basal ( a–b ) axis is indicated . Scale bars = 50 µm in B and 10 µm in C–E . ( F ) Quantification of the basal ( orange ) and apical ( purple ) length variation for an individual clone showing no correlation between the oscillations ( R = 0064 ) . ( G ) Box plot showing the distribution of apical vs basal oscillations correlation coefficients for 10 transplanted neuroblasts from five different retinas . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 01010 . 7554/eLife . 15797 . 011Figure 3—figure supplement 1 . Mitotic rounding impact on basal constriction and apical expansion . ( A–D ) Confocal microscopy time-lapse images showing a mitosis in a tg ( vsx2 . 2:GFP-caax ) retina at 20 hpf . Dashed white lines highlight flanking cells . Arrows indicate apical ( purple ) and basal ( orange ) distance variation . The orientation of the apico-basal ( a–b ) axis is indicated . ( E–F ) The graphs show the quantification of distance variation ( % ) for the apical ( E ) and basal ( F ) sides . The mitotic event ( red arrow ) results only in a transient expansion of the apical domain . Error bars indicate standard error of the mean ( n = 10 , from three different retinas ) . ( G–I ) Confocal microscopy time-lapse images showing a mitosis occurring in the apical plane in a tg ( vsx2 . 2:GFP-caax ) retina at 20 hpf . Dashed white arrows indicate apical distance variation along the mitotic axis . Neighboring cells are indicated with colored dots . ( J ) Quantification of apical distance variation ( % ) along the mitotic axis for five different cells confirms a transient expansion of the apical domain . Scale bars = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 01110 . 7554/eLife . 15797 . 012Video 3 . Analysis of tg ( vsx2 . 2:GFP-caax ) clones show uncoupled oscillations at apical and basal surfaces . Maximum projection of 3 z-stacks ( over a total of 1 µm ) along the apico-basal axis shows the oscillatory behavior of apical and basal edges simultaneously in tg ( vsx2 . 2:GFP-caax ) clones . Images were acquired every 8 s . See also Figure 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 012 Oscillatory cell contractions and epithelial bending have been associated to the periodic accumulation of the cortical actomyosin network . To investigate this phenomenon in constricting retinal cells , we first examined actin dynamics during optic cup morphogenesis . To follow dynamic changes in cell area and F-actin simultaneously , we injected utrophin-GFP RNA in one-cell stage embryos of the transgenic line tg ( vsx2 . 2:lyn-tdTomato ) and then performed live imaging analyses at 20 hpf focusing on the basal neuroblasts surface ( Video 4 ) . As previously reported for vertebrate neuroepithelial cells ( Nishimura et al . , 2012 ) , actin accumulated circumferentially ( i . e . junctional ) rather than medially as observed in constricting Drosophila epithelia ( He et al . , 2010; Martin et al . , 2009 ) ( Figure 4A–F ) . In addition , we observed that actin accumulated at the basal surface and oscillated with a frequency similar to membrane pulses ( Video 4 ) . To detect whether there is a relationship between cortical actin accumulation and basal area changes , both parameters were quantified after segmentation and a cross-correlation analysis was performed . This analysis showed a positive association between actin accumulation and basal area expansion , with a cross-correlation coefficient of 0 . 40 ± 0 . 16 ( median 0 . 35 ) , as calculated for 26 cells from three different experiments ( Figure 4G , H ) . In order to evaluate the significance of our results , we compared our experimental data with simulated random and sinusoidal signals of similar statistical properties . Coefficients of simulated random data were significantly lower than our observations in vivo , indicating that the cells display a significant positive correlation between actin accumulation and basal area changes ( Figure 4I ) . Hence , cell area expansion and actin accumulation occur simultaneously or with time lags shorter than 5 s ( i . e . our sampling rate limitation ) . Furthermore , when we plotted cross-correlation coefficients as a function of the actin intensity , we observed higher coefficients corresponding to cells with higher actin intensity rates ( Figure 4J ) . Taken together , these results indicate that the molecular mechanism responsible for the fast oscillations in the vertebrate retina differs in important aspects from that controlling the pulsatile behavior in constricting epithelia in Drosophila . In retinal neuroblasts , peripheral actin accumulation is associated with basal ends’ expansion , whereas in Drosophila cell contraction is linked to medial condensation of actin . 10 . 7554/eLife . 15797 . 013Figure 4 . Basal actin dynamics in constricting retinal cells . ( A–F ) Actin dynamics , as revealed by utrophin-gfp , and membrane oscillations were simultaneously examined by time lapse in the line tg ( vsx2 . 2:lyn-tdTomato ) at 20 hpf ( see Video 4 ) . Note that F-actin localizes mainly at the cellular cortex . Scale bars = 10 µm . ( G ) Normalized basal area rate ( orange ) and normalized utrophin-gfp rate ( green ) are shown over time for a cell displaying a high correlation between actin oscillations and membrane expansion . Area rate and Utrophin-gfp rate were normalized dividing by the mean of their absolute values . ( H ) Normalized auto-correlation ( grey line ) and cross-correlation ( orange ) are shown for cell represented in G . Maximum cross-correlation ( 0 . 8 ) is indicated . ( I ) Box plot comparison of cross-correlation results between actin vs . membrane oscillations , simulated random and simulated sinusoidal signals shows a significant ( p<0 . 001; T-test; n = 26 ) positive correlation between actin accumulation and basal area expansion . ( J ) Scattered plot showing the dependency of cross-correlation coefficients ( n = 26 ) on mean actin intensity rates . Linear regression line ( orange ) and linear correlation coefficient ( 0 . 38 ) are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 01310 . 7554/eLife . 15797 . 014Video 4 . Actin dynamics in constricting retinal cells . ( Upper panel ) Maximum projection of 3 z-stacks ( over a total of 1 µm ) along the apico-basal axis shows actin oscillatory activity in tg ( vsx2 . 2:lyn-tdTomato ) embryos at 20 hpf . Retinal basal surface ( region within the square ) is magnified in lower panels . ( Lower panels ) Time lapse shows the simultaneous recording of membrane behavior , as revealed by lyn-tdTomato ( left panel ) , and actin dynamics , as revealed by Utrophin-GFP ( right panel ) . Images were acquired every 5 s . Scale bars = 10 µm . See also Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 014 To investigate myosin dynamics , we then carried out time-lapse studies through optic cup folding in tg ( actb1:myl12 . 1-eGFP ) embryos . At the organ level , myosin accumulations were detected both at the apical lens and basal retina epithelia ( Figure 5A–H; Video 5 ) . This is in agreement with the bending of these tissues toward their apical and basal surfaces , respectively . When examined in relation to basal membrane oscillations , as revealed by lyn-tdTomato , myosin dynamics showed a behavior different from that of actin . Basal myosin accumulates in scattered cortical foci , which have an average stability in the range of minutes , 4 ± 0 . 5 min ( Figure 5I–Q ) . Treatment of embryos for 1 hr with blebbistatin , a specific inhibitor that blocks myosin in an actin-detached state ( Kovacs et al . , 2004 ) , severely interfered with myosin dynamics in the retina , increasing significantly the stability of the foci to 21 . 5 ± 2 . 4 min ( Figure 5Q , Video 6 ) . 10 . 7554/eLife . 15797 . 015Figure 5 . Myosin accumulates in basal foci during optic cup morphogenesis . ( A–D ) Live-imaging analysis of tg ( actb1:myl12 . 1-eGFP ) embryos reveals myosin accumulation at the apical lens ( purple arrowheads ) and basal retina ( orange arrowheads ) between 19 and 20 . 5 hpf . Antero-posterior ( a–p ) and medio-lateral ( m-l ) axes are indicated . ( E–H ) Myosin accumulates in transient foci ( orange arrows ) at the basal cortex . ( I–P ) Time-lapse analysis of myosin foci at the basal surface plane in embryos injected with lyn-tdTomato RNA reveals that the protein accumulates at the peripheral cortex in scattered cells . ( Q ) The box plot shows a significant difference in foci stability between control and blebbistatin ( 150 µM ) treated embryos ( T-test , n = 21 ) . fb = forebrain; nr = neural retina; lv = lens vesicle . Scale bars = 50 µm in A–D , 20 µm in E–H , and 10 µm in I–P . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 01510 . 7554/eLife . 15797 . 016Video 5 . Myosin dynamics during optic cup morphogenesis . Live imaging analysis of tg ( actb1:myl12 . 1-eGFP ) embryos reveal myosin accumulation at apical lens and basal retina epithelia . Movie starts at 19 hpf . Antero-posterior ( a-p ) axis is indicated . Images were acquired every 20 s . Scale bar 50 µm . See also Figure 5DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 01610 . 7554/eLife . 15797 . 017Video 6 . Myosin foci dynamics at the basal surface . Live-imaging analysis of myosin distribution at the basal surface in 20 hpf tg ( actb1:myl12 . 1-eGFP ) embryos shows cortical localization of myosin foci in scattered cells ( left and middle panels ) . Membrane oscillations were simultaneously examined by injection of lyn-tdTomato RNA ( merged in left panel with myl12gfp ) . Treatment of tg ( actb1:myl12 . 1-eGFP ) embryos with blebbistatin ( 150 µM ) severely blocks myosin dynamics at the basal surface ( right panel ) . Images were acquired every 5 s . Scale bar 10 µm . See also Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 017 Live-imaging analysis along the apico-basal retinal axis showed that myosin foci correlate with basal membrane indentations ( i . e . transient shortenings of the apico-basal axis ) , suggesting active pulling of the basal lamina ( Figure 6A–F; Video 7 ) . To quantitatively analyze this phenomenon , we measured simultaneously myosin intensity and apico-basal axis shortening ( Figure 6G–H ) . The analysis of 25 individual foci revealed a significant shortening of the apico-basal axis upon myosin accumulation for most of the events examined , with an average shortening of 2 . 3 ± 1 . 4 ( SD ) µm ( Figure 6I ) . Correlative analysis of basal membrane dynamics and myosin accumulation in tg ( actb1:myl12 . 1-eGFP ) embryos injected with lyn-tdTomato RNA revealed that a large proportion of the cells containing myosin foci contract significantly their basal surface ( Figure 6J–P; Figure 6—figure supplement 1 ) . In contrast , the oscillatory behavior and average area of the cells neighboring those with myosin foci was not affected upon myosin accumulation ( Figure 6P; Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 15797 . 018Figure 6 . Myosin accumulation correlates with basal membrane displacement . ( A–E ) Time series of optical sections from tg ( actb1:myl12 . 1-eGFP ) embryos show discrete myosin foci ( labeled 1 , 2 , 3 ) and basal surface displacement . ( F ) Basal edges were color-coded for each time point and overlapped to illustrate the transient indentations of the basal surface associated to myosin foci . ( G–H ) Quantitative recording over time of myosin intensity and apico-basal axis shortening for a couple of representative foci . The focus in G is #2 in A–F . ( I ) Box plot showing the maximum shortening of the a-b axis for 25 foci from 12 different retinas . ( J–O ) Correlative analysis of basal area ( revealed by lyn-tdTomato ) and myosin dynamics is shown for three neighbor cells ( color-coded ) . ( P ) Quantitative analysis of cell area changes and myosin intensity for the three neighboring cells . Note that only the cell accumulating myosin contracts . Scale bars =10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 01810 . 7554/eLife . 15797 . 019Figure 6—figure supplement 1 . Myosin accumulation correlates with basal contraction . ( A–E ) Quantitative analysis of myosin intensity ( green lines ) and cell area changes for five cells containing myosin foci ( purple lines ) and their neighboring cells ( orange dashed lines ) . Note the contraction of the cells upon myosin accumulation . ( F ) Box plot showing average cell contraction ( µm2 ) at the peak of myosin accumulation for 22 different cells containing myosin foci ( purple ) and 35 neighboring cells . Myosin accumulating cells undergo a significant contraction of their basal area , as determined by T-test . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 01910 . 7554/eLife . 15797 . 020Figure 6—figure supplement 2 . Myosin inhibition impairs basal constriction . ( A–L ) Live-imaging analysis of cell area dynamics in control ( A–F ) and blebbistatin-treated ( G–L ) tg ( vsx2 . 2:GFP-caax ) embryos . Progressive constriction is observed in individual cells ( asterisk ) in control , but not in blebbistatin-treated tissue . ( M ) Basal area variation rate is shown for representative control and blebbistatin-treated cells . ( N ) Average peak amplitude of the cell area rate is considerably reduced in treated cells ( T-test , n = 12 ) . ( O ) Blebbistatin treatment significantly inhibited basal constriction over a considered period of 25 min , blocking the cells in a relaxed state ( T-test ) . Scale bars = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 02010 . 7554/eLife . 15797 . 021Figure 6—figure supplement 3 . Myosin inhibition interferes with optic cup folding . ( A , B ) Optic cup folding is also impaired in blebbistatin-treated embryos as assessed by the retinal opening angle ( indicated with green dashed lines ) . ( C ) Quantitative analysis of retinal opening angles show a significant delay in optic cup folding in embryos treated with 50 and 200 µM blebbistatin ( one-way ANOVA followed by Tukey test , n = 15 ) . fb = forebrain; nr = neural retina . Scale bars = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 02110 . 7554/eLife . 15797 . 022Video 7 . Myosin foci dynamics and basal membrane indentations upon blebbistatin treatment . Live-imaging analysis of myosin dynamics at the basal surface both in control ( upper panel ) and blebbistatin treated ( 150 µM; lower panel ) 20 hpf embryos from the line tg ( actb1:myl12 . 1-eGFP ) . Note the increased stability of the myosin foci and the reduced contractility of the basal surface in the retina of the blebbistatin-treated embryos . Images were acquired every 10 s Scale bar = 10 µm . See also Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 022 As we mentioned , myosin inhibition stabilized cortical myosin foci . Blebbistatin treatment also impaired contractility at the basal surface of the retina . Thus , basal membrane indentations associated to myosin foci appeared largely attenuated ( Video 7 ) , suggesting an inefficient mechanical coupling . In addition , when basal membrane oscillations were examined in the tg ( vsx2 . 2:GFP-caax ) line , treatment for one hour with blebbistatin abolished the pulsatile behavior and impaired basal constriction by blocking the cells in a relaxed state ( Figure 6—figure supplement 2 , Video 8 ) . This result indicates that although myosin levels do not oscillate with basal area changes , its activity is required to maintain the pulsatile dynamics . Finally , sustained treatment with blebbistatin for 3 hr significantly delays the folding of the optic cup ( Figure 6—figure supplement 3 ) . This finding , however , needs to be interpreted cautiously , as myosin inhibition may interfere with optic cup folding either by blocking basal constriction or through any other acto-myosin-dependent morphogenetic mechanism . 10 . 7554/eLife . 15797 . 023Video 8 . Membrane oscillations at the basal surface in control and blebbistatin-treated embryos . Maximum projection of 3 z-stacks ( over a total of 1 µm ) at the basal surface in tg ( vsx2 . 2:GFP-caax ) retinae show cell membranes oscillatory behavior over a period of 25 min in control ( left panel ) and blebbistatin treated ( 150 µM; right panel ) embryos . Note that blebbistatin treatment abolishes the oscillatory behavior and blocks the cells in a relaxed state . Images were acquired every 5 s . Scale bars = 10 µm . See also Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 023 We have previously shown that integrin-mediated adhesion to the ECM plays a fundamental role during optic cup folding in medaka ( Martinez-morales et al . , 2009; Bogdanovic et al . , 2012 ) . To specifically interfere with this process in zebrafish , we knocked down lamc1 , a core component of laminin trimer , the mutation of which results in ocular malformations ( Domogatskaya et al . , 2012; Lee and Gross , 2007 ) . To this end we employed morpholinos previously reported to phenocopy the zebrafish lamc1 mutation sleepy ( sly ) ( Parsons et al . , 2002; Ivanovitch et al . , 2013 ) . Comparative examination of sly mutants and lamc1 morphants revealed a similar optic cup phenotype ( Figure 7A–C ) , both interfering with the folding of the epithelium , as indicated by measurement of retinal opening angles at 24 hpf ( Figure 7D–F ) . Live-imaging analysis of tg ( vsx2 . 2:GFP-caax ) morphant retinas revealed that basal oscillations are not reduced upon lamc1 knockdown; on the contrary , their average peak amplitude was significantly increased by 45% ( n = 22 ) . Interestingly , the progressive reduction of the cellular feet observed in control retinas ( Figure 2 ) was severally impaired in embryos injected with lamc1 morpholinos ( lamc1Mo ) , and basal cell areas appeared significantly larger when compared to the control situation ( Figure 7—figure supplement 1 ) . This observation indicates that laminin-dependent adhesion to the ECM is required for effective basal constriction . 10 . 7554/eLife . 15797 . 024Figure 7 . Optic cup folding , basal contractility and myosin dynamics depend on lamc1 function . ( A–C ) General embryo morphology for wild type , lamc1 morphants and sly ( lamc1-/- ) mutants at 24 hpf . Retinal opening is indicated with a dashed line . ( D–E ) Retinal morphology in tg ( vsx2 . 2:GFP-caax ) both wild type and lamc1Mo-injected embryos , at 24 hpf . Ventral opening angle ( white ) and retinal contour ( orange ) are indicated with dashed lines . ( F ) Frequency distribution of retinal opening angles is shown for controls ( either wild type or p53Mo-injected ) , lamc1Mo injected , or sly mutants . ( G–N ) Time-lapse analysis of tg ( actb1:myl12 . 1-eGFP ) wild type and lamc1Mo-injected embryos show dynamic accumulation of myosin foci ( green arrows ) at the basal surface . ( O ) Analysis of myosin foci reveals that they are significantly more stable in lamc1Mo-injected embryos ( T-test ) . ( P ) The box plot shows that transient indentations of the basal surface are significantly diminished in lamc1Mo-injected embryos ( T-test ) . h = heart; nr = neural retina; lv = lens vesicle . Scale bars = 200 µm in A–C , 50 µm in D–E , and 10 µm in G–N . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 02410 . 7554/eLife . 15797 . 025Figure 7—figure supplement 1 . Analysis of membrane oscillations reveal impaired basal constriction in lamc1 morphant embryos . ( A–B ) Cell area dynamics in control ( A ) and lamc1Mo ( B ) tg ( vsx2 . 2:GFP-caax ) embryos is shown for three representative cells . The mean area of the three cells is shown as red dotted lines . ( C ) Average peak amplitude of the cell area rate is significantly increased in lamc1 morphant cells ( T-test , n = 22 ) . ( D ) Basal feet area is larger and basal constriction , over the recorded period of 25 min , appears significantly inhibited in lamc1Mo retinas ( T-test , n = 22 ) . Mean ± SEM is represented . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 025 To investigate myosin dynamics in the folding retina of lamc1-deficient embryos , morpholinos were injected in the tg ( actb1:myl12 . 1-eGFP ) line . Live-imaging analysis of lamc1Mo and control sibling embryos revealed that myosin foci are still observed in the morphant retinae ( Figure 7G–N ) . However , in the morphant tissue , foci were significantly more stable than in the wild-type siblings , and more importantly , basal membrane indentations associated to them appeared attenuated ( Figure 7O , P; Video 9 ) . This result suggests that deficient adhesion to the ECM also results in a less efficient transmission of mechanical tensions and hence reduced contractility at the basal feet . 10 . 7554/eLife . 15797 . 026Video 9 . Myosin foci dynamics and basal membrane indentations in wild-type and lamc1 morphants . Live-imaging analysis of myosin dynamics at the basal surface both in control ( upper panel ) and lamc1Mo-injected ( lower panel ) 20 hpf embryos from the line tg ( actb1:myl12 . 1-eGFP ) . Note the increased stability of the myosin foci and the reduced contractility of the basal surface in lamc1 morphants . Images were acquired every 10 s Scale bar = 10 µm . See also Figure 7 . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 026 To examine how mechanical tensions are distributed in the folding epithelium , we performed laser ablations experiments at different stages of optic cup morphogenesis . In order to visualize membranes displacement during tension release , local ablations were carried out in tg ( vsx2 . 2:GFP-caax ) embryos , either at the apical or at the basal surfaces of the tissue . Laser-induced cuts trigger a limited expansion of the wounded area and a local relaxation of the tissue , as determined by optical flow analysis ( Figure 8—figure supplement 1 ) . For most of the stages analyzed , tissue relaxation affected only neuroblasts immediately adjacent to the wounded area . However , laser ablations within a developmental window corresponding to a 125°–140° opening of the optic cup resulted in a global tissue relaxation that affected bending of the entire epithelium ( Video 10 ) . At this specific stage , tension release triggered a noticeable folding of the retinal tissue toward its basal surface . To quantitatively investigate membrane displacement after laser ablation in retinal tissues , we carried out an optical flow analysis of the movies ( Figure 8A–C; Video 12 ) , which allow determining retraction speeds at different stages and locations within the tissue ( Figure 8D , E; Figure 8—figure supplement 1 ) . Statistical analysis of optical flow data confirmed that maximum retraction speeds are significantly higher only for retinas displaying a 125º–140º bending ( Figure 8F , G ) . This observation indicates that the balance between tensile forces and tissue resistance that maintains organ shape is particularly unstable within a narrow developmental window that coincides with the acute constriction of the basal feet at 19 hpf ( Figure 1 ) . In contrast to the global reaction observed upon basal ablation , which triggers the displacement of the peripheral retina , apical ablation only affected the morphology of the central retina but no peripheral retraction was observed ( Movie 11; Figure 8—figure supplement 2 ) . The differential tissue response upon ablation at the apical and basal surfaces , together with our previous observations on lamc1 requirement for basal contractility ( Figure 7 ) prompted us to investigate tissue behavior in lamc1 morphants . The analysis of retraction speeds in laser-ablated tissues at the critical 125°–140° stage showed that global relaxation of the optic cup is attenuated in lamc1 knockdown retinas ( Figure 8—figure supplement 3 ) . This data indicates that the laminin-mediated attachment to the ECM is essential for the transmission of mechanical tensions throughout the folding tissue . 10 . 7554/eLife . 15797 . 027Figure 8 . Optical flow analysis of tissue displacement upon laser ablation at different stages of folding . ( A–C ) Analysis of pixel displacement after laser ablation at the basal surface is shown for retinas at 170° , 130° , and 80° of bending . Red arrowheads indicate the ablation point . Particles’ motion vectors are indicated with a color code: Colors correspond to the direction of the displacement and color intensity to its magnitude . Note maximum displacement 20 s after ablation in 130°-stage retina . Scale bar = 50 µm . See Video 12 . ( D–E ) Average tissue retraction speed profiles over time are shown for different stages of optic cup folding ( represented as angle bins ) , both at the central ( D ) or distal ( E ) positions in the retina . ( F–G ) Box plot representation of maximal retraction speeds at the different stages , represented as angle bins . For each stage , median values ( red bars ) and sample sizes are indicated . One-way ANOVA analysis followed by Dunnett’s multiple comparison tests show significant differences ( p<0 . 01** ) only at 125–140º-stage . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 02710 . 7554/eLife . 15797 . 028Figure 8—figure supplement 1 . Tissue local relaxation upon laser ablation: Optical flow analysis of tissue displacement . ( A–B ) Laser ablation experiments at the basal surface of the retina imaged along the apico-basal axis ( A ) and basal plane ( B ) Red arrowheads indicate the ablation point . Time 0 corresponds to the first frame after the ablation . Tissue reaction through time is shown at higher magnification ( A–B ) and particles’ motion is indicated with a color code . ( C ) Different colors correspond to the direction of the displacement and color intensity to its magnitude . Regions selected for optical flow quantification in Figure 8 are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 02810 . 7554/eLife . 15797 . 029Figure 8—figure supplement 2 . Optical flow analysis of retinal tissue displacement upon apical vs basal laser ablation . ( A–B ) Laser ablation experiments at the apical ( A ) or basal ( B ) surfaces of the retina in wild-type embryos . Red arrowheads indicate the ablation point . Scale bar = 50 µm . ( C–D ) Tissue retraction speed profiles at different retinal positions ( color-coded ) are represented for apical ( C ) vs basal ( D ) ablations . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 02910 . 7554/eLife . 15797 . 030Figure 8—figure supplement 3 . Optical flow analysis of tissue displacement upon laser ablation in wild type vs . lamc1_Mo tissues . ( A–B ) Laser ablation experiments at the basal surface of the retina in wild type ( A ) and lamc1Mo ( B ) tissues . Red arrowheads indicate the ablation point . Scale bar = 50 µm . ( C ) Box plot representation of maximal retraction speeds for control and morphant tissues both at the central and distal ( peripheral ) retina . For each stage , median values ( red bars ) and sample sizes are indicated . Two-way ANOVA analysis shows that retraction speeds are significantly reduced in lamc1 morphants ( p<0 . 05* ) . ( D–E ) Tissue retraction speed profiles at different retinal positions ( color-coded ) are represented over time for wild type ( D ) and lamc1Mo ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 03010 . 7554/eLife . 15797 . 031Video 10 . Laser ablation experiments at the basal surface of the retina through optic cup folding . Local cell ablations were carried out in tg ( vsx2 . 2:GFP-caax ) retinae at different stages . Ablation points are indicated with green arrowheads . Retinal folding angles are indicated . Note the global tissue relaxation upon ablation at 130º . Images were acquired every seconds . Scale bar = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 03110 . 7554/eLife . 15797 . 032Video 11 . Comparative analysis of focal ablations at the apical or basal surface of the retina . Ablations were carried out in tg ( vsx2 . 2:GFP-caax ) retinas with a 130° opening . Ablation points are indicated with green arrowheads . Peripheral tissue displacement is indicated with white arrowheads . Scale bar = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 03210 . 7554/eLife . 15797 . 033Video 12 . Optical flow analysis of tissue displacement upon laser ablation at different stages of optic cup folding . Ablation points are indicated with white arrowheads . Particles’ motion vectors are indicated with a color code . Images were acquired every seconds . Scale bar = 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 033 In the current study , we have characterized the morphogenetic behavior of retinal precursors during zebrafish optic cup folding by live imaging . Our quantitative analysis demonstrates that retinal neuroblasts undergo a progressive constriction of their basal surface . Previous reports have described the involution of outer layer progenitors into the presumptive neural retina domain as a mechanism driving the formation of the eye chamber ( Picker et al . , 2009; Kwan et al . , 2012; Heermann et al . , 2015 ) . Our observations are also consistent with these reports , thus suggesting that basal constriction and cell involution cooperate during eye morphogenesis in zebrafish . Comparative analysis of our data and previous studies ( Heermann et al . , 2015 ) indicate that , although both mechanisms overlap substantially , they are staggered events . Whereas basal constriction occurs mainly during the primary folding of the retinal epithelium between 18 and 20 hpf , cell involution through the rim is limited during this period and becomes more prominent at later stages between 20 and 24 hpf . It is tempting to speculate that these mechanisms might be coupled . Thus , basal constriction may generate centripetal tensions facilitating cell involution and , conversely , cell involution may relieve tissue resistance supporting a constriction-dependent optic cup folding . However , the precise cellular mechanisms driving cell involution are currently unknown , and hence exploring this possibility will require further investigation . Here , we have described that retinal precursors undergo fast pulsations both at their apical and basal surfaces . We then examined both membrane and actomyosin dynamics at the basal surface , where the progressive constriction takes place . Although , in principle , the neuroblasts’ periodic pulsations share some features with the oscillations observed in other constricting epithelia ( Kim and Davidson , 2011; Martin et al . , 2009; Roh-johnson et al . , 2012; Solon et al . , 2009 ) , there are fundamental differences . In most epithelial cells , pulsations are more regular in frequency and amplitude than in retinal precursors , and their average oscillation frequency range between 1 and 5 min ( Gorfinkiel and Blanchard , 2011 ) . This is in contrast to irregular fast oscillations ( ≈20 s ) here described in the zebrafish retina . A second fundamental difference concerns the organization of the actomyosin fibers in the shrinking surface of the tissue . In most of the constricting epithelia so far examined , contractile actomyosin fibers accumulate in a medioapical domain . From this domain , centripetal tension responsible for cell contraction is generated and transmitted to surface junctions ( Martin et al . , 2009; Roh-johnson et al . , 2012; He et al . , 2010 ) . Interestingly , F-actin turnover is required for this medioapical localization of the actomyosin meshwork , its efficient attachment to cellular junctions , and the generation of centripetal tension ( Jodoin et al . , 2015 ) . In contrast , our data show that both actin and myosin fibers accumulate at the cellular cortex in the zebrafish retina . Cortical distribution of actomyosin fibers has also been described in the folding neural tube ( Nishimura et al . , 2012 ) , thus suggesting that it may be a common feature in elongated neuroepithelial cells regardless the tissue is bending toward its apical o basal surface . It has been shown that medial and cortical actomyosin pools have different mechanical properties in epithelial cells ( Rauzi et al . , 2010 ) . In the light of this finding , our observation that the molecular mechanism driving fast oscillations in retinal neuroblasts differs substantially from that previously reported in constricting epithelia is not surprising . Whereas medioapical accumulations of actomyosin precede periodic cellular contractions in most epithelia analyzed , we observed that cortical actin accumulation correlates positively with basal membrane expansion in retinal precursors . Local actin assembly at the leading edge has been described as a positive force driving membrane extension in lamellipodia and axonal growth cones ( Pollard and Borisy , 2003; Levayer and Lecuit , 2012; Medeiros et al . , 2006 ) . Our data may suggest a similar mechanism as responsible for the pulsatile behavior of the retinal precursors , but confirming this hypothesis will require further analysis . Here , we show that cortical myosin accumulation does not correlate in time with the fast oscillations of the membrane . In spite of this , our data does not allow to rule out a myosin role in the maintenance of the pulsatile state . On the contrary , blebbistatin treatment severely impaired membrane pulses , suggesting that myosin basal activity is necessary to maintain the fast oscillatory behavior . Our data also show that myosin accumulates at the basal cortex in discrete foci , which have an average stability of approximately 4 min and are distributed in scattered cells across the epithelial field . Remarkably , a large proportion of the retinal cells accumulating basal myosin foci are contracting both along the apico-basal and basal plane axes . These episodic contractions at the basal surface can be inhibited either by blocking myosin activity or by interfering with the adhesive properties of the extracellular matrix . Taken together , our observations suggest a working model for the ratcheted constriction of the epithelium ( Figure 9 ) . According to this hypothetical model , retinal precursors would experience non-ratcheted fast membrane oscillations . Pulsatile behavior without a net reduction of cell area has also been reported in several epithelial contexts ( He et al . , 2010; Solon et al . , 2009; Roh-johnson et al . , 2012 ) . Superimposed to these fast oscillations , the episodic accumulation of myosin at the basal surface in scattered cells would mediate their progressive ( i . e . ratcheted ) constriction . Then , individual contributions would add up over time to cause the constriction of the entire neuroepithelium . At a tissue level , our laser ablation experiments indicate that the global balance between mechanical tensions and tissue resistance becomes transiently unstable within a limited developmental window ( 19–20 hpf ) . This critical period , in which local ablations at the basal surface trigger global tissue rearrangement , coincides with the acute bending of the optic cup epithelium and the active constriction of the neuroblasts’ feet . Upon lamc1 knockdown both basal contractility and global tissue response to laser ablation are attenuated . This suggests that the ECM plays a fundamental role in the transmission of mechanical tensions generated by individual cells at the tissue level . In agreement with this concept , previous reports have shown that optic cup morphogenesis largely depends on integrin function ( Martinez-morales et al . , 2009; Bogdanovic et al . , 2012; Nakano et al . , 2012 ) . 10 . 7554/eLife . 15797 . 034Figure 9 . A working model for the basal constriction of the retinal epithelium . ( A ) Representation of the retinal epithelium during eye morphogenesis showing the distribution of cortical actomyosin , integrins and ECM at the basal surface of the tissue . Apical junctions and focal adhesion components have been included as a reference for apico-basal polarity . ( B ) Schematic diagram representing the condensation of nonmuscle myosin II foci at the basal surface in wild type and lamc1Mo retinas . Both fast pulsating cells ( orange ) and myosin-enriched constricted cells ( green ) are depicted . Weakly constricting neuroblast feet are represented in pale green . The final form of the organ is also shown for wild type and lamc1 deficient embryos . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 034 The formation of the eye chamber offers an excellent model to understand basal constriction in epithelia . This study has revealed significant differences in cell and actomyosin dynamics between retinal folding and previously characterized apical constriction processes . To what extent these different features can be attributed to the neuroepithelial character of the retina or are a common theme in epithelial layers undergoing basal constriction remains an open question . Adult AB/Tübingen ( AB/Tu; RRID:ZIRC_ZL1/RRID:ZIRC_ZL57 ) wild-type zebrafish strain , transgenic lines tg ( vsx2 . 2:GFP-caax ) ( Gago-rodrigues et al . , 2015 , tg ( actb1:myl12 . 1-eGFP ) ( Behrndt et al . , 2012 ) , and the mutant strain sleepy ( slym86; RRID:ZFIN_ZDB-GENO-090402-2; Parsons et al . , 2002 ) were maintained and bred under standard conditions ( Westerfield , 2000 ) . The line tg ( vsx2 . 2:lyn-tdTomato ) was generated by recombining the medaka vsx2 . 2 promoter ( Martinez-morales et al . , 2009 ) with the membrane reporter Lyn-tdTomato in the backbone of the destination vector pDestTol2CG ( Kwan et al . , 2007 ) . All embryos were staged in hours post-fertilization ( hpf ) as described ( Kimmel et al . , 1995 ) . All experiments conform national and European Community standards for the use of animals in experimentation . Transgenic embryos were anesthetized using 0 . 04% MS-222 ( Sigma ) , embedded in 0 . 8% low-melting agarose in E3 medium , and mounted on 35 mm glass-bottom dishes ( WPI-Fluorodish ) . Time-lapse analyses were performed on a Leica SP5 confocal microscope with a 20x/0 . 75 IMM multi-immersion objective . Optical sections containing either apical or basal surfaces were identified by z-stacks in resonant mode throughout the entire retinal epithelium ( Figure 1—figure supplement 1 ) . To determine the orientation of the neuroepithelium along the apico-basal axis and the position of apical and basal surfaces , a z-stack ( with 1µm spatial resolution ) was taken across the entire retina at the beginning and at the end of each time series . We used this information to establish confocal planes for live imaging 1–3 µm below the surfaces . Then small z-stacks ( 3 planes over a total of 1 µm ) were recorded every 5 or 8 s at the selected planes , 1 µm below the apical or basal surfaces . Long-term recordings along the apico-basal axis were performed using the galvano scanner . Time-lapse images were processed using Fiji ( RRID:SCR_002285; Schindelin et al . , 2012 ) . Different plugins were used for maximum intensity projection of z-stacks , signal intensity quantification in selected regions of interest ( ROIs ) , and measurement of angles and distances . To measure the length of the apical and basal edges of the retina ( Figure 1 ) , we selected a single stack at the central retina and outlined tissue borders using the Fiji tool freehand . For automatic detection of cell edges and tracking of individual cells through time we used Packing Analyzer v2 . 0 , which is based on a watershed algorithm for cell identification ( Aigouy et al . , 2010 ) . Unique RGB codes were assigned to each cell by Packing Analyzer V2 . 0 in tracked images . Individual images were examined manually to correct for automatic segmentation mistakes . Only those cells that could be tracked unambiguously through time were considered for quantification ( Figure 1—figure supplement 1; Video 13 ) . Once cell areas were quantified , the constriction rates were calculated as the first derivative of time and represented with Excel ( Microsoft ) ( Figure 2—figure supplement 1 ) . For automatic actin intensity measurements ( Figure 4 ) , individual cell profiles ( as revealed by lyn-tdTomato ) were segmented and tracked using Packing Analyzer V2 . 0 . This software generates unique RGB codes and masks for every tracked cell . Then , a MATLAB ( Mathworks ) script was used to overlap cell masks with images showing F-actin ( Utrophin-GFP ) and to quantify average intensity per cell area . 10 . 7554/eLife . 15797 . 035Video 13 . Membrane oscillations in an optical section from a tg ( vsx2 . 2:GFP-caax ) embryo and Packing Analyzer v2 . 0 automatic cell edge detection ( represented by unique RGB codes ) are shown in parallel movies . Scale bar = 10 µm . See also Figure 1_figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 15797 . 035 For cross-correlation analyses of oscillatory signals we use the following equation: ( ƒ ⋆ g ) [n] = F−1 {F {ƒ*} · F {g}}; where F−1 denotes the inverse Fourier transform . We use the autocorrelation , the cross-correlation of a signal with itself , to normalize the cross-correlation and obtain a cross-correlation coefficient ranging from −1 ( maximum inverse correlation ) to +1 ( maximum correlation ) . Fertilized Tg ( vsx2 . 2:GFP-caax ) and wild-type eggs were incubated at low density ( 50 eggs per dish ) at 28°C until 4hpf . Then embryos were dechorionated by pronase treatment ( 375 µg/ml ) and gently washed with E3 medium . Cells from the blastula cap of donor embryos were collected with a glass needle ( Borosilicate Glass Capillaries GC100-10; 1 . 0 mm × 58 mm , 6´´ . Harvard Apparatus ) and implanted into the caps of host embryos . After cell transfers were completed , host and donor embryos were incubated at 28ºC . Once the desired developmental stage is reached ( 20 hpf ) , GFP-positive embryos were selected and prepared for in vivo live imaging . Apical and basal oscillations were simultaneously recorded for 10 transplanted neuroblasts from five different retinas . To visualize actin dynamic , we used utrophin-GFP as a reporter . The plasmid pCS2:Utrophin-GFP ( Burkel et al . , 2007 ) was used to synthesize the corresponding RNA . The construct was first linearized with NotI ( Takara ) , and RNA was synthesized using the mMESSAGE mMACHINE SP6 kit ( Ambion ) . Capped utrophin-GFP RNA was then precipitated with 4M LiCl , quantified , and injected into Tg ( vsx2 . 2:lyn-tdtomato ) embryos at one-cell stage ( 200 pg per embryo ) . Antisense lamc1morpholino oligonucleotides ( MO ) were purchased from Gene Tools , LLC . Lamc1Mo 5’-TGTGCCTTTTGCTATTGCGACCTC-3’ blocks translation , is complementary to the 5’ sequence of lamc1 and has been shown to phenocopy ocular malformations observed for the lamc1 mutation sly ( Ivanovitch et al . , 2013; Parsons et al . , 2002 ) . The lamc1Mo was injected into tg ( vsx2 . 2:GFP-caax ) and tg ( actb1:myl12 . 1-eGFP ) embryos at one-cell stage at a concentration of 1 pmol per embryo . To prevent potential apoptotic effects , a p53MO ( p53MO: 5’-GCGCCATTGCTTTGCAAGAATTG-3’ ) , was co-injected with Lamc1Mo at a concentration of 0 . 5 pmoles per embryo . Control embryos were injected in parallel with p53MO alone . Transgenic embryos were selected at the appropriate developmental stages , dechorionated with forceps , embedded in 0 . 8% low melting point agarose , and mounted onto 35 mm petri dishes as described above . Embryos were carefully oriented with the dorsal head surface contacting the coverslip and were imaged using a 40x objective . In order to be able to record time-lapse movies with sufficient time resolution ( ms ) for an optical flow analysis , we used a spinning disk confocal microscope ( RoperScientific ) , achieving a time resolution of 0 . 5 s for all experiments in this work . Laser ablations were performed by applying a short wavelength laser ( 405 nm ) at single cell membranes for 450 ms , either at the basal or apical surfaces of the neuroretinal tissue . Laser pulses were controlled using iLas software ( Roper Scientific ) . For the statistical analysis of maximal ablation speeds , ablated retinas were sorted in 15° bins . In order to assess the retraction speed of the neuroretinal tissue after laser ablation , we measured optical flow between consecutive frames . To compare pixel intensity between frames , we employed the Lucas-Kanade method , which groups neighboring pixels together assuming similar motion for them ( Barron et al . , 1994 ) . The algorithm Good Features to Track was used for the pixel-wise detection of features to track ( Shi and Tomasi , 1994 ) . Both methods are available as programming functions at the computer vision open source library , OpenCV ( Bradski and Kaehler , 2008 ) . Different positions at the central and distal retina and the apical and basal surfaces of the neuro-epithelium were considered for optical flow measurements . For each region , 11 points were tracked and their speed values median-averaged . Retraction speed graphs have been Gaussian smoothed . In order to allow direct comparison between different experiments , speed profiles for each retina analyzed were normalized to their median values .	Tissues and organs form into their final shapes because the cells in a developing embryo generate forces that alter their shape and position . Networks of fibres made from actin and myosin proteins generate these forces , and because the fibres can assemble in many different ways inside cells , they allow the cells to move and change shape in many different ways . Forces in some tissues can cause flat sheets of cells to bend . These sheets of cells are attached on one side ( their “basal” surface ) to a collection of membranes and molecules that are known as the extracellular matrix . When the cells in the sheet progressively shrink at their basal surface , causing the sheet to bend towards the extracellular matrix , this is known as basal constriction . Nicolás-Pérez et al . have used high-resolution imaging to record how basal constriction helps the optic cup – the main chamber of the eye – to form in zebrafish embryos . This imaging confirmed that a sheet of precursor cells progressively bends towards its basal surface to form the curved shape of the eyeball . Further analysis revealed that this basal constriction happens when myosin fibres accumulate in clusters along the basal surface of some of the precursor cells . The resulting contraction of the basal surface of the cells relies both on the tension generated by myosin inside the cell and on the cells being attached properly to the extracellular matrix . Using a laser beam , Nicolás-Pérez et al . also destroyed small parts of the basal surface of the retina . This procedure allows the mechanical tension distribution throughout the developing eye to be mapped . Laser ablations revealed a narrow time window during development when destroying small parts of the basal surface can cause the entire sheet of cells to relax , preventing it from curving to form the shape of the eye . Sheets of precursor cells are important building blocks of the nervous system , yet researchers only have limited knowledge of the processes that enable them to fold or bend into a final shape . As such , the findings of Nicolás-Pérez et al . will contribute to a wider understanding of how cells and tissues behave while the brain is forming .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2016	Analysis of cellular behavior and cytoskeletal dynamics reveal a constriction mechanism driving optic cup morphogenesis
Grid cells in the medial entorhinal cortex ( MEC ) respond when an animal occupies a periodic lattice of ‘grid fields’ in the environment . The grids are organized in modules with spatial periods , or scales , clustered around discrete values separated on average by ratios in the range 1 . 4–1 . 7 . We propose a mechanism that produces this modular structure through dynamical self-organization in the MEC . In attractor network models of grid formation , the grid scale of a single module is set by the distance of recurrent inhibition between neurons . We show that the MEC forms a hierarchy of discrete modules if a smooth increase in inhibition distance along its dorso-ventral axis is accompanied by excitatory interactions along this axis . Moreover , constant scale ratios between successive modules arise through geometric relationships between triangular grids and have values that fall within the observed range . We discuss how interactions required by our model might be tested experimentally . A grid cell has a spatially modulated firing rate that peaks when an animal reaches certain locations in its environment ( Hafting et al . , 2005 ) . These locations of high activity form a regular triangular grid with a particular length scale and orientation in space . Every animal has many grid cells that collectively span a wide range of scales , with smaller scales enriched dorsally and larger scales ventrally along the longitudinal axis of the MEC ( Stensola et al . , 2012 ) . Instead of being smoothly distributed , grid scales cluster around particular values and thus grid cells are partitioned into modules ( Stensola et al . , 2012 ) . Consecutive pairs of modules have scale ratios in the range 1 . 2–2 . 0 ( Stensola et al . , 2012; Barry et al . , 2007; Krupic et al . , 2015 ) . The scale ratio averaged across animals is constant from one pair of modules to the next and lies in the interval 1 . 4 ( Stensola et al . , 2012 ) to 1 . 7 ( Barry et al . , 2007; Krupic et al . , 2015 ) , suggesting that the grid system favors a universal scale ratio in this range . Encoding spatial information through grid cells with constant scale ratios is thought to provide animals with an efficient way of representing their position within an environment ( Moser et al . , 2008; Fiete et al . , 2008; Mathis et al . , 2012; Wei et al . , 2015; Stemmler et al . , 2015; Sanzeni et al . , 2016; Mosheiff et al . , 2017 ) . Moreover , periodic representations of space permit a novel mechanism for precise error correction against neural noise ( Sreenivasan and Fiete , 2011 ) and are learned by machines seeking to navigate open environments ( Cueva and Wei , 2018; Banino et al . , 2018 ) . These findings provide motivation for forming a modular grid system with a constant scale ratio , but a mechanism for doing so is unknown . Continuous attractor networks ( Fuhs and Touretzky , 2006; Burak and Fiete , 2009 ) , a leading model for producing grid cells , would currently require discrete changes in scales to be directly imposed as sharp changes in parameters , as would the oscillatory interference model ( Burgess et al . , 2007; Hasselmo et al . , 2007 ) or hybrid models ( Bush and Burgess , 2014 ) . In contrast , many sensory and behavioral systems have smooth tuning distributions , such as preferred orientation in visual cortex ( Issa et al . , 2008 ) and preferred head direction in the MEC ( Taube et al . , 1990 ) . A self-organizing map model with stripe cell inputs ( Grossberg and Pilly , 2012 ) and a firing rate adaptation model with place cell inputs ( Urdapilleta et al . , 2017 ) can generate discrete grid scales , but their ratios are not constant or constant-on-average unless explicitly tuned . Here , we present a simple extension of the continuous attractor model that adds excitatory connections between a series of attractor networks along the dorso-ventral axis of the MEC , accompanied by an increase in the distance of inhibition . The inhibition gradient drives an increase in grid scale along the MEC axis . Meanwhile , the excitatory coupling discourages changes in grid scale and orientation unless they occur through geometric relationships with defined scale ratios and orientation differences . Competition between the effects of longitudinal excitation and lateral inhibition self-organizes the complete network into a discrete hierarchy of modules . Certain grid relationships are geometrically stable , which makes them , and their associated scale ratios , insensitive to perturbations . The precise ratios that appear depend on the balance between excitation and inhibition and how it varies along the MEC axis . We show that sampling across a range of these parameters leads to a distribution of scale ratios that matches experiment and is , on average , constant from the smallest to the largest pair of modules . Continuous attractors are a powerful general method for self-organizing neural dynamics . To our knowledge , our results are the first demonstration of a mechanism for producing a discrete hierarchy of modules in a continuous attractor system . We assemble a series of networks along the longitudinal MEC axis , numbering them z = 1 , 2 , . . . , 12 from dorsal to ventral ( Figure 1A ) . Each network contains the standard 2D continuous attractor architecture of the Burak-Fiete model ( Burak and Fiete , 2009 ) . Namely , neurons are arranged in a 2D sheet with positions ( x , y ) , receive broad excitatory drive ( Bonnevie et al . , 2013 and Figure 1B ) , and inhibit one another at a characteristic separation on the neural sheet ( Figure 1C; see Materials and methods for a complete description ) . In our model , this inhibition distance l is constant within each network but increases from one network to the next along the longitudinal axis of the MEC . With these features alone , the population activity in each network self-organizes into a triangular grid whose lattice points correspond to peaks in neural activity ( Figure 2A ) . Importantly , the scale of each network’s grid , which we call λ ( z ) , is proportional to that network’s inhibition distance l ( z ) ( ‘uncoupled’ simulations in Figure 3A ) . Also , network grid orientations θ show no consistent pattern across scales and among replicate simulations with different random initial firing rates . Following the standard attractor model ( Burak and Fiete , 2009 ) , the inhibitory connections in each network are slightly modulated by the animal’s velocity such that the population activity pattern of each network translates proportionally to animal motion at all times ( Materials and methods ) . This modulation allows each network to encode the animal’s displacement through a process known as path-integration , and projects the network grid pattern onto spatial rate maps of single neurons . That is , a recording of a single neuron over the course of an animal trajectory would show high activity in spatial locations that form a triangular grid with scale Λ ( Figure 2C ) . Moreover , Λ ( z ) for a neuron from network z is proportional to that network’s population grid scale λ ( z ) , and thus also proportional to its inhibition distance l ( z ) ( uncoupled simulations in Figure 3B ) . To be clear , we call Λ the ‘spatial scale’; it corresponds to a single neuron’s activity over the course of a simulation and has units of physical distance in space . By contrast , λ , the ‘network scale’ described above , corresponds to the population activity at a single time and has units of separation on the neural sheet . Similarly , Θ ( z ) describes the orientation of the spatial grid of a single neuron in the network z; we call Θ the ‘spatial orientation . ’ Like the network orientations θ discussed above , spatial orientations of grids show no clustering ( uncoupled simulations in Figure 3B ) . With an inhibition distance l ( z ) that increases gradually from one network to the next ( Figure 1C ) , proportional changes in network and spatial scales λ ( z ) and Λ ( z ) lead to a smooth distribution of grid scales ( uncoupled simulations in Figure 3A , B ) . To reproduce the experimentally observed jumps in grid scale between modules , the inhibition distance would also have to undergo discrete , sharp jumps between certain adjacent networks . In summary , a grid system created by disjoint attractor networks will not self-organize into modules . Module self-organization can be achieved with one addition to the established features listed above: we introduce excitatory connections from each neuron to those in the preceding network with approximately corresponding neural sheet positions ( Figure 1D; see Materials and methods for a complete description ) . That is , a neuron in network z ( more ventral ) with position ( x , y ) will excite neurons in network z – 1 ( more dorsal ) with positions that are within a distance d of position ( x , y ) . In other words , the distance d is the ‘spread’ of excitatory connections , and we choose a constant value across all networks comparable to the inhibition distance l ( z ) . The self-organization of triangular grids in the neural sheet and the faithful path-integration that projects these grids onto single neuron spatial rate maps persist after introduction of inter-network coupling ( Figure 2G ) . Network and spatial scales λ ( z ) and Λ ( z ) still increase from network z = 1 ( dorsal ) to network z = 12 ( ventral ) . Yet , Figure 3A , B shows that for the coupled model , these scales exhibit plateaus that are interrupted by large jumps , disrupting their proportionality to inhibition distance l ( z ) , which is kept identical to that of the uncoupled system ( Figure 1C ) . Collecting scales across all networks illustrates that they cluster around certain values in the coupled system while they are smoothly distributed in the uncoupled system . We identify these clusters with modules M1 , M2 , and M3 of increasing scale . Note that multiple networks at various depths z can belong to the same module . Moreover , coupling causes grid cells that cluster around a certain scale to also cluster around a certain orientation ( Figure 3A , B ) , as seen in experiment ( Stensola et al . , 2012 ) . The uncoupled system does not demonstrate co-modularity of orientation with scale , that is two networks with similar grid scales need not have similar orientations unless this is imposed by an external constraint . In summary , excitatory coupling between grid attractor networks dynamically induces discreteness in grid scales that is co-modular with grid orientation , as observed experimentally ( Stensola et al . , 2012 ) , and as needed for even coverage of space by the grid map ( Sanzeni et al . , 2016 ) . Not only does excitatory coupling produce modules , it can do so with consistent scale ratios and orientation differences . For the coupled system depicted in Figure 2 , scale ratios and orientation differences between pairs of adjacent modules consistently take values 1 . 74 ± 0 . 02 and 29 . 5 ± 0 . 4° , respectively ( mean ± s . d . ; Figure 3C ) . These values are robust to a variety of parameter perturbations , coupling architectures , and sources of noise . We can make the inhibition distance profile l ( z ) less or more concave ( Figure 4A , B ) , or we can implement excitatory connections with different properties by reversing their direction ( Figure 4C ) , including connections in both directions ( Figure 4D ) , or allowing the coupling spread to vary with network depth ( Figure 4E ) . In each case , the same scale ratio of ≈1 . 7 and orientation difference of ≈30° persist . We can also reduce the number of neurons by a factor of 9 without affecting the scale ratio and orientation difference ( Figure 4F ) . Similar results are obtained with neural inputs corrupted by independent Gaussian noise ( Figure 4G ) and with randomly shifted excitatory connections , which adds another form of coupling imprecision in addition to spread ( Figure 4H ) . Finally , simulations with spiking dynamics following Burak and Fiete ( 2009 ) also demonstrate a preference for scale ratios of ≈1 . 7 and orientation differences of ≈30° , albeit with greater variability ( Figure 4I ) . We can intuitively understand this robust modularity through the competition between lateral inhibition within networks and longitudinal excitation across networks . In the uncoupled system , grid scales decrease proportionally as the inhibition distance l ( z ) decreases from z = 12 to z = 1 . However , coupling causes areas of high activity in network z to preferentially excite corresponding areas in network z – 1 , which encourages adjacent networks to share the same grid pattern ( z = 10 & 11 in Figure 3D ) . Thus , coupling adds rigidity to the system and provides an opposing ‘force’ against the changing inhibition distance that attempts to drive changes in grid scale . This rigidity produces the plateaus in network and spatial scales λ ( z ) and Λ ( z ) that delineate modules across multiple networks . At interfaces between modules , coupling can no longer fully oppose the changing inhibition distance , and the grid pattern changes . However , the rigidity fixes a geometric relationship between the grid patterns of the two networks spanning the interface . In the coupled system of Figure 2 and Figure 3 , module interfaces occur between networks z = 4 and 5 and between z = 9 and 10 . The network population activity overlays of Figure 3D reveal overlap of many activity peaks at these interfaces . However , the more dorsal network ( with smaller z ) at each interface contains additional small peaks between the shared peaks . In this way , adjacent networks still share many corresponding areas of high activity , as favored by coupling , but the grid scale changes , as favored by a changing inhibition distance . Pairs of grids whose lattice points demonstrate regular registry are called commensurate lattices ( Chaikin and Lubensky , 1995 ) and have precise scale ratios and orientation differences , here respectively 3 ≈ 1 . 7 and 30° , which match the results in Figure 3C and Figure 4 . In summary , excitatory coupling can compete against a changing inhibition distance to produce a rigid grid system whose ‘fractures’ exhibit stereotyped commensurate lattice relationships . These robust geometric relationships lead to discrete modules with fixed scale ratios and orientation differences . In our model , commensurate lattice relationships naturally lead to field-to-field firing rate variability in single neuron spatial rate maps ( z = 8 in Figure 2G , for example ) , another experimentally observed feature of the grid system ( Ismakov et al . , 2017; Dunn et al . , 2017; Diehl et al . , 2017 ) . At interfaces between two commensurate lattices , only a subset of population activity peaks in the grid of smaller scale overlap with , and thus receive excitation from , those in the grid of larger scale . The network with smaller grid scale will contain activity peaks of different magnitudes; this heterogeneity is then projected onto the spatial rate maps of its neurons . Adjusting the balance between excitatory coupling and a changing inhibition distance produces other commensurate lattice relationships , each of which enforces a certain scale ratio and orientation difference . To explore this competition systematically , we use a smaller coupled model with just two networks , z = 1 and 2 , and vary three parameters: the coupling spread d , the coupling strength umag , and the ratio of inhibition distances between the two networks l ( 2 ) /l ( 1 ) ( Appendix 1 ) . For each set of parameters , we measure network scale ratios and orientation differences produced by multiple replicate simulations ( Figure 5—figure supplement 1 and Figure 5—figure supplement 2 ) . We find that as the excitation-inhibition balance is varied by changing umag and l ( 2 ) /l ( 1 ) , a number of discretely different relationships appear , which can be summarized in ‘phase diagrams’ ( Figure 5A , B ) . In many regions of the phase diagrams , these lattice relationships are commensurate , each with a characteristic scale ratio and orientation difference ( Figure 5C ) . When parameters are chosen near a boundary between two regions , replicate simulations may adopt either lattice relationship or occasionally be trapped in other metastable relationships due to variations in random initial conditions ( Figure 5—figure supplement 2 ) . At larger umag in both phase diagrams , there are fewer regions as l ( 2 ) /l ( 1 ) varies because a higher excitatory coupling strength provides more rigidity against gradients in inhibition distance ( Figure 5A , B ) . However , a larger coupling spread d would cause network z = 2 to excite a broader set of neurons in network z = 1 , softening the rigidity imposed by coupling and producing a wider variety of lattices in Figure 5B than Figure 5A . Also in Figure 5B , when excitation is weak and approaching the uncoupled limit , there is a noticeable region dominated by incommensurate lattices , in which the two grids lack consistent registry or relative orientation , and grid scale is largely determined by inhibition distance ( Figure 5—figure supplement 2 ) . Figure 5B also contains a larger region of discommensurate lattices ( although strictly speaking , in condensed matter physics , they would be termed commensurate lattices with discommensurations; Chaikin and Lubensky , 1995 ) . Discommensurate networks have closely overlapping activities in certain areas that are separated by a mesh of regions lacking overlap called discommensurations ( Figure 5D ) . They exhibit ranges of scale ratios 1 . 1–1 . 4 and orientation differences 0°–10° that ultimately arise from a single source: the density of discommensurations , whose properties can also be explained through excitation-inhibition competition . Stronger coupling drives more activity overlap , which favors sparser discommensurations and lowers the scale ratio and orientation difference . However , a larger inhibition distance ratio drives the two networks to differ more in grid scale , which favors denser discommensurations . To better accommodate the discommensurations , grids rotate slightly as observed previously in a crystal system ( Wilson , 1990 ) . Figure 5E confirms that scale ratios and orientation differences vary together as the discommensuration density changes . Thus , by changing the balance between excitation and inhibition , a two-network model yields geometric lattice relationships with various scale ratios and corresponding orientation differences . All the commensurate relationships ( Figure 5C ) and almost the entire range of discommensurate relationships ( Figure 5D ) have scale ratios that fall in the range of experimental measurements , which is roughly 1 . 2–2 . 0 ( Stensola et al . , 2012; Barry et al . , 2007; Krupic et al . , 2015 ) . The scale ratios and orientation differences in both the commensurate and discommensurate cases are robust against activity noise and coupling noise ( Figure 5—figure supplement 3 ) . As mentioned above , discommensurate lattices have a range of allowed geometries ( Figure 5D , E ) , but they can still produce modules in a full 12-network grid system with a preferred scale ratio and orientation difference . However , these values do not cluster as strongly as they do for a commensurate relationship , which is geometrically precise . The phase diagrams of Figure 5 provide guidance for modifying a 12-network system that exhibits a [3 , 30∘] relationship to produce discommensurate relationships instead . We make the inhibition distance profile l ( z ) shallower ( Figure 6A ) and increase the coupling spread d by 50% . Network activity overlays of these new simulations reveal grids obeying discommensurate relationships ( Figure 6B , C ) , which are projected onto single neuron spatial rate maps through faithful path-integration ( Figure 6—figure supplement 1A ) . Across replicate simulations with identical parameter values but different random initial firing rates , the discommensurate system demonstrates greater variation in scale and orientation ( Figure 6D ) than the commensurate system of Figure 3 does . Nevertheless , analysis of each replicate simulation reveals clustering with well-defined modules ( Figure 6E and Figure 6—figure supplement 1B ) . These modules have scale ratio 1 . 39 ± 0 . 10 and orientation difference 6 . 7 ± 3 . 5° ( mean ± s . d . ; Figure 6F ) . The preferred scale ratio agrees well with the mean value observed experimentally in Stensola et al . ( 2012 ) . Conceptually , we can interpret the greater spread of scales and orientations in terms of coupling rigidity . Excitatory coupling , especially when the spread is larger , provides enough rigidity in the discommensurate system to cluster scale ratios and orientation differences but not enough to prevent variations in these values . The degree of variability observed in Figure 6D , E appears consistent with experimental measurements , which also demonstrate spread ( Stensola et al . , 2012; Barry et al . , 2007 ) . A few module pairs in Figure 6F exhibit a large orientation difference >10° . This is not expected from a discommensurate relationship , and indeed , inspecting the network activities reveals adjacent networks trapped in a relationship with low activity overlap and large orientation difference ( Figure 6G ) . In the context of a grid system that otherwise obeys commensurate or discommensurate geometries containing more overlap , we call this less common relationship a ‘defect . ’ We distinguish between these relationships and the incommensurate lattices discussed above , which also have low activity overlap . Defects arise when the excitatory coupling is strong , and incommensurate lattices arise when this coupling is weak . Also , defects have smaller scale ratios <1 . 1 and larger orientation differences >10° , whereas incommensurate lattices have larger scale ratios >1 . 3 and any orientation difference ( Figure 5B and Figure 5—figure supplement 2 ) . Thus , networks governed by discommensurate relationships also cluster into modules with a preferred scale ratio and orientation difference within the experimental range ( Stensola et al . , 2012; Krupic et al . , 2015 ) . Due to lower coupling rigidity compared to commensurate grid systems , they exhibit increased variability and occasional defects across replicate simulations . As in the commensurate case , discommensurate lattice relationships also create field-to-field firing rate variability in single neuron spatial rate maps . At interfaces between two discommensurate lattices , population activity peaks lack overlap at discommensurations and exhibit overlap in between them . Thus , only a subset of peaks in the grid of smaller scale receive excitation from the grid of larger scale; those located at discommensurations do not . As activity patterns translate on the neural sheets during path-integration , a grid cell in the network with smaller scale will have lower firing rate when a discommensuration moves through it , leading to firing rate variability ( see Figure 6—figure supplement 2 for an example ) . So far , each set of 12-network simulations contained replicates with identical parameter values and exhibited a single dominant lattice relationship . We now present results with different parameter values to imitate biological network variability across animals . This procedure leads to modules with different commensurate and discommensurate relationships ( Figure 7A and Figure 7—figure supplement 1 ) . There is no longer a single preferred scale ratio or orientation difference ( Figure 7B ) , but patterns emerge due to the predominance of discommensurate and commensurate relationships . Recall from Figure 6F that discommensurate module pairs exhibit scale ratios ≈1 . 4 and orientation differences ≈7° . Combined with [3≈1 . 7 , 30∘] module pairs , we find a bimodal distribution of orientation differences around 7° and 30° , consistent with experimental data ( Krupic et al . , 2015 ) , and positive correlation between scale ratio and orientation difference . Modules with low scale ratio but high orientation difference decrease this correlation; they arise from defects ( Figure 6G ) . Figure 7—figure supplement 2 illustrates how modules observed experimentally may be governed by a variety of lattice relationships . Scale ratios across the assorted simulations span a range of values , but their averages are constant across module pairs . That is , the median scale ratio does not change between the pair of modules with smaller scales and the larger pair ( Figure 7C ) . Similarly , mean values are respectively 1 . 52 ± 0 . 05 and 1 . 53 ± 0 . 05 ( mean ± s . e . m . ) for module pairs M2 and M1 and M3 and M2 . Combining data from both module pairs gives scale ratio 1 . 52 ± 0 . 03 ( mean ± s . e . m . ) , which agrees well with the mean value of 1 . 56 from Krupic et al . ( 2015 ) . Stensola et al . ( 2012 ) reports a slightly smaller mean value of 1 . 42 ± 0 . 17 ( mean ± s . d . ; re-analyzed by Wei et al . , 2015 ) , but its broad distribution of scale ratios overlaps considerably with ours . Moreover , we find that the normalized scale difference does change its median across module pairs ( Figure 7D ) . This result that scale ratios are constant on average but scale differences are not matches experiment ( Stensola et al . , 2012 ) . Thus , although our model can produce modules with fixed scale ratios , allowing for a range of network parameters also produces modules with a range of scale ratios . Nevertheless , the scale ratio averaged over these parameters is still constant across module pairs , a key feature of the grid system that holds even if scales are not governed by a universal ratio ( Stensola et al . , 2012 ) . Excitatory coupling locks networks into scales and orientations imposed by more ventral networks . Disrupting the coupling frees networks from this rigidity , which can change scales and orientations far from the disruption . We demonstrate this effect by inactivating one network z = 7 midway through the simulation ( Figure 8A ) . This corresponds experimentally to disrupting excitatory connections at one location along the dorsoventral MEC axis . After the lesion , grid cells ventral to the lesion location ( z ≥ 8 ) are unaffected , but those dorsal to the lesion location ( z ≤ 6 ) change scale and orientation and form a single module ( Figure 8B–D ) . Network z = 6 is no longer constrained by larger grids of more ventral networks , so its scale decreases . The coupling that remains from z = 6 to 1 then rigidly propagates the new grid down to network z = 1 . This post-lesion module M1 has larger scale and 30º orientation difference compared to the pre-lesion M1; these changes also appear as corresponding changes in the scale ratio and orientation difference between modules M2 and M1 ( Figure 8E ) . Immediate changes in grid scale and/or orientation observed at one location along the longitudinal MEC axis due to a lesion at another location would strongly support the presence of the excitatory coupling predicted by our model . Moreover , the anatomical distribution of the changes would indicate the directionality of coupling; those in grid cells dorsal to the lesion would indicate ventral-to-dorsal coupling and those ventral to the lesion would indicate dorsal-to-ventral coupling . We have also considered the consequences of certain incomplete lesions . A regional lesion , in which a corner of the lesioned network z = 7 is preserved , causes each more dorsal network to contain regions with different scales ( Figure 8—figure supplement 1 and Figure 8—video 1 ) . These differences are not large enough to create a new module close to the lesioned network ( z = 5 and 6 ) , so scale ratios and orientations are not strongly affected . However , different regions of each network will independently transition to the smallest module farther away from the lesioned network ( z = 1 to 4 ) . Thus , one network corresponding to a single location along the dorso-ventral MEC axis can contain grid cells belonging to two modules . Experimentally , grid modules do overlap in their anatomic extent along the MEC axis ( Stensola et al . , 2012 ) ; our model predicts that this overlap may be enhanced by a regional lesion . Note that some neurons also appear to show band-like spatial rate maps ( z = 4 and 6 in Figure 8—figure supplement 1A ) , whose experimental observation has been reported ( Krupic et al . , 2012 ) but disputed ( Navratilova et al . , 2016 ) . We also performed a decimation-type lesion , in which one neuron of every 3 × 3 block is preserved in the lesioned network . This impedes the motion of the grid pattern on the neural sheet in more dorsal networks ( Figure 8—video 2 ) and thus destroys single neuron grid responses in those networks ( Figure 8—figure supplement 1D ) . We propose that the hierarchy of grid modules in the MEC is self-organized by competition in attractor networks between excitation along the longitudinal MEC axis and lateral inhibition . We showed that such an architecture , with an inhibition distance that increases smoothly along the MEC axis , reproduces a central experimental finding: grid cells form modules with scales clustered around discrete values ( Stensola et al . , 2012; Barry et al . , 2007; Krupic et al . , 2015 ) . The distribution of scales across modules in our model quantitatively matches experiments . Different groups have reported mean scale ratios of 1 . 64 ( 6 module pairs ) , 1 . 42 ( 24 module pairs ) , and 1 . 56 ( 11 module pairs ) ( Barry et al . , 2007; Stensola et al . , 2012; Krupic et al . , 2015 ) . These data could be interpreted as an indication that the grid system has a preferred scale ratio roughly in range of 1 . 4–1 . 7 . As we showed , our model naturally produces a hierarchy of modules with scale ratios in this range; its network parameters lead to both commensurate and discommensurate grids ( Figure 5 ) . On the other hand , the data on scale ratios between individual pairs of modules actually span a range of values in the different experiments: 1 . 6–1 . 9 , 1 . 1–1 . 8 , and 1 . 2–2 . 0 ( Barry et al . , 2007; Stensola et al . , 2012; Krupic et al . , 2015 ) . This suggests that the underlying mechanism that produces grid modules must be capable of producing different scale ratios as its parameters vary . This is indeed the case for our model , in which variation of network parameters produces a realistic range of scale ratios ( Figure 7 ) . Despite variability across individual scale ratios , experiments strikingly reveal that the average scale ratio is the same from the smallest pair of modules to the largest pair , whereas the average scale difference changes across the hierarchy ( Stensola et al . , 2012 ) . Our model robustly reproduces this observation ( Figure 7C , D ) because its fundamental mechanism of geometric coordination between grids enforces constant-on-average scale ratios even with variation in parameters among individual networks . Our model requires that grid orientation be co-modular with scale , as observed in experiment ( Stensola et al . , 2012 ) . Studies characterizing the statistics of orientation differences between modules are limited , but values seem to span the entire range 0°–30° , with some preference for values at the low and high ends of this range ( Krupic et al . , 2015 ) . Our model can capture the entire range of orientation differences with discommensurate relationships favoring small differences and commensurate relationships favoring large differences ( Figure 5 ) . Overall , our model predicts a positive correlation between scale ratio and orientation difference ( Figure 5E and Figure 7B ) , which can be tested experimentally . Existing datasets ( Stensola et al . , 2012; Krupic et al . , 2015 ) have a confound—animals are tested in square and rectangular enclosures which have distinguishable orientations marked by the corners . Grid orientations can anchor to such features ( Stensola et al . , 2015 ) , either through the integration of visual and external cues ( Raudies and Hasselmo , 2015; Savelli et al . , 2017 ) , or through interaction with boundaries ( Bush and Burgess , 2014; Krupic et al . , 2016; Giocomo , 2016; Evans et al . , 2016; Hardcastle et al . , 2017; Keinath et al . , 2018; Ocko et al . , 2018 ) . Experiments in circular or other non-rectangular environments may help disambiguate the effects of such anchoring . Our model also predicts that orientation differences between modules will be preserved between environments with different geometries since the differences are internally generated by the dynamics of the network . This effect has been observed ( Krupic et al . , 2015 ) . Our model produces consistent differences in firing rate from one grid field to another for some grid cells . This variability is structured because it arises at module interfaces from the selective excitation of some network activity peaks in the smaller-scale grid by the overlapping activity peaks of the larger-scale grid . Such an explanation for firing rate variability has been suggested by Ismakov et al . ( 2017 ) . Signatures of structured variability can be sought in experimental grid cell recordings ( see Figure 6—figure supplement 2 for an example ) . However , these signatures may be obscured by other sources of grid variability , such as proposed inputs from place cells ( Dunn et al . , 2017 ) and the observed modulation of grid fields by reward ( Butler et al . , 2019; Boccara et al . , 2019 ) , which may in turn be also related to hippocampal input . Our model requires excitatory coupling between grid cells at different locations along the longitudinal MEC axis , either through direct excitation or disinhibition ( Fuchs et al . , 2016 ) . Short-range excitatory connections between principal neurons in superficial MEC layers have been discovered recently through patch clamp experiments ( Fuchs et al . , 2016; Winterer et al . , 2017 ) . These neurons also make long-range projections to superficial layers of the contralateral MEC ( Varga et al . , 2010; Fuchs et al . , 2016 ) , where they connect to other principal cells ( Zutshi et al . , 2018 ) . The validity of our model would be bolstered if similar connections were found between locations along the MEC that correspond to different grid modules . The presence of excitatory coupling can also be tested indirectly . We predict that the destruction of grid cells , or inactivation of excitatory coupling ( Zutshi et al . , 2018 ) , at a given location along the axis will change grid scales and/or orientations at other locations ( Figure 8 ) . The presence of noise correlations across modules , as previously investigated but not fully characterized ( Mathis et al . , 2013; Tocker et al . , 2015 ) , would suggest connections between modules . Such correlations , and perhaps even lattice relationships , could be observed via calcium imaging of the MEC ( Heys et al . , 2014; Gu et al . , 2018 ) . The effect of environmental manipulations on grid relationships has been suggested to demonstrate both independence ( Stensola et al . , 2012 ) and dependence ( Krupic et al . , 2015 ) across modules . However , ( Keinath et al . , 2018 ) showed that apparent deformations of grids after changes in environmental shape may result in part from learned interactions with boundaries , perhaps mediated by border cells . Thus , environmental deformation paradigms may not be ideal tests of our model due to confounding boundary effects ( Keinath et al . , 2018; Ocko et al . , 2018 ) . Our predictions may be altered by synaptic plasticity , which we do not implement in our model . Spike-timing-dependent plasticity rules are capable of creating the recurrent inhibitory architecture required by continuous attractor models of a single grid module ( Widloski and Fiete , 2014 ) . As for our model with multiple modules , synaptic plasticity within the inhibitory connections may resolve the competition between excitation and inhibition by adjusting the inhibition distance in each network to the value favored by the rigidity of excitatory coupling . In that case , lesioning one network would not affect the grid scales of other networks , although changes in orientation differences may be observed over time due to attractor drift . Nevertheless , our proposed geometric mechanism could still govern the initial formation of modules with certain scale ratios before plasticity fully takes effect . Since spatial grid scales are both proportional to inhibition distance l and inversely proportional to velocity gain α ( Burak and Fiete , 2009 and Materials and methods ) , we also simulated excitatorily coupled networks with a depth-dependent velocity gain α ( z ) and a fixed inhibition distance l ( Appendix 2 ) . In contrast to simulations in one dimension ( J Widloski and I Fiete , personal communication , October 2017 ) , while we observed module self-organization , the system gave inconsistent results among replicate simulations and lacked fixed scale ratios . Moreover , recent calcium imaging experiments suggest that activity on the MEC is arranged a deformed triangular lattice ( Gu et al . , 2018 ) , as predicted by the continuous attractor model ( Burak and Fiete , 2009 ) , and that regions with activity separated by larger anatomic distances contain grid cells of larger spatial scale . These observations support a changing inhibition distance over a changing velocity gain as a mechanism for producing different grid scales , under the assumption that anatomic and network distances correspond to each other . Our results differ from previous work on mechanisms for forming grid modules . Grossberg and Pilly hypothesize that grid cells arise from stripe cells in parasubiculum , and that discreteness in the spatial period of stripe cells leads to modularity of grid cells ( Grossberg and Pilly , 2012 ) . However , stripe cells have only been observed once ( Krupic et al . , 2012; Navratilova et al . , 2016 ) , and the origin of discrete periods with constant-on-average ratios in stripe cells would then need to be addressed . Urdapilleta , Si , and Treves propose a model in which discrete modules self-organize from smooth gradients in parameters in a model where grid formation is driven by firing rate adaptation in single cells ( Urdapilleta et al . , 2017 ) . They also utilize excitatory coupling among grid cells along the longitudinal MEC axis . However , this model does not have a mechanism to dynamically enforce the average constancy of grid scale ratios , which is a feature of the grid system ( Stensola et al . , 2012 ) . Furthermore , it produces modules with orientation differences near zero and does not demonstrate values near 30° ( Krupic et al . , 2015 ) . Our model naturally produces constant-on-average scale ratios and allows for a wide range of orientation differences . Moreover , over the past few years , multiple reports have provided independent experimental support for the importance of recurrent connections among grid cells ( Couey et al . , 2013; Dunn et al . , 2015; Fuchs et al . , 2016; Zutshi et al . , 2018 ) and for the continuous attractor model in particular ( Yoon et al . , 2013; Heys et al . , 2014; Gu et al . , 2018 ) . Our work establishes that continuous attractor networks can produce a discrete hierarchy of modules with a constant-on-average scale ratio . The competition generated between excitatory and inhibitory connections bears a strong resemblance to the Frenkel-Kontorova model of condensed matter physics , in which a periodic potential of one scale acts on particles that prefer to form a lattice of a different , competing scale ( Kontorova and Frenkel , 1938 ) . This model has a rich literature with many deep theoretical results , including the calculation of complicated phase diagrams involving ‘devil’s staircases’ ( Bak , 1982; Chaikin and Lubensky , 1995 ) which mirror those of our model ( Figure 5 ) . Under certain conditions , our model produces networks with quasicrystalline approximant grids that are driven by networks with standard triangular grids at other scales ( Appendix 3 ) . Quasicrystalline order lacks periodicity , but contains more nuanced positional order ( Levine and Steinhardt , 1986 ) . This phenomenon wherein quasicrystalline structure is driven by crystalline order in a coupled system was recently observed for the first time in thin-film materials that contain Frenkel-Kontorova-like interactions ( Förster et al . , 2013; Förster et al . , 2016; Paßens et al . , 2017 ) . Commensurate and discommensurate lattice relationships are a robust and versatile mechanism for self-organizing a grid system whose scale ratios are constant or constant on average across a hierarchy of modules . We demonstrated this mechanism in a basic extension of the continuous attractor model with excitatory connections between networks . This model is amenable to extensions that capture other features of the grid system , such as fully spiking dynamics , learning of synaptic weights ( Widloski and Fiete , 2014 ) , the union of our separate networks into a single network spanning the entire MEC , and the addition of border cell inputs or recurrent coupling between modules to correct path-integration errors or react to environmental deformations ( Hardcastle et al . , 2015; Keinath et al . , 2018; Ocko et al . , 2018; Pollock et al . , 2017; Mosheiff and Burak , 2019 ) . We implemented the Burak-Fiete model ( Burak and Fiete , 2009 ) as follows ( Source code 1 ) . Networks z=1 , … , h each contain a 2D sheet of neurons with indices 𝐫= ( x , y ) , where x=1 , … , n and y=1 , … , n . Neurons receive broad excitatory input a⁢ ( 𝐫 ) from the hippocampus , and , to prevent edge effects , those toward the center of the networks receive more excitation than those toward the edges . Each neuron also inhibits others that lie around a length scale of l⁢ ( z ) neurons away in the same network z . Moreover , every neuron belongs to one of four subpopulations that evenly tile the neural sheet . Each subpopulation is associated with both a preferred direction 𝐞^ along one of the network axes ±𝐱^ or ±𝐲^ and a corresponding preferred direction 𝐄^ along an axis ±𝐗^ or ±𝐘^ in its spatial environment . A neuron at position 𝐫 in network z has its inhibitory outputs w⁢ ( 𝐫;z ) shifted slightly by ξ neurons in the 𝐞^⁢ ( 𝐫 ) direction and its broad excitation modulated by a small amount proportional to 𝐄^⁢ ( 𝐫 ) ⋅𝐕 , where 𝐕 is the spatial velocity of the animal . Note that lowercase letters refer to attractor networks at each depth z in which distances have units of neurons , and uppercase letters refer to the animal’s spatial environment in which distances have physical units , such as centimeters . In addition to these established features ( Burak and Fiete , 2009 ) , we introduce excitatory connections u⁢ ( 𝐫 ) from every neuron 𝐫 in network z to neurons located within a spread d of the same 𝐫 but in the preceding network with depth z−1 . u⁢ ( 𝐫 ) is constant for all networks . These components lead to the following dynamical equation for the dimensionless neural firing rates s⁢ ( 𝐫 , z , t ) : ( 1 ) τs ( r , z , t+Δt ) −s ( r , z , t ) Δt+s ( r , z , t ) ={∑r′w ( r−r′+ξe^ ( r′ ) ;z ) s ( r′ , z , t ) +∑r′u ( r−r′ ) s ( r′ , z+1 , t ) +a ( r ) [1+αE^ ( r ) ⋅V ( t ) ]}+ . Inputs to each neuron are rectified by {c}+=0 for c<0 , c for c≥0 . Δt is the simulation time increment , τ is the neural relaxation time , and α is the velocity gain that describes how much the animal’s velocity 𝐕 modulates the broad inputs a⁢ ( 𝐫 ) . Note that s can be treated as a dimensionless variable because Equation 1 is invariant to scaling of s and a by the same factor . We use velocities 𝐕⁢ ( t ) corresponding to a real rat trajectory ( Hafting et al . , 2005; Burak and Fiete , 2009 ) . Details are provided in Appendix 1 . The broad excitatory input is ( 2 ) a ( r ) ={amage−afallrscaled2rscaled<10rscaled≥1 , where rscaled= ( x−n+12 ) 2+ ( y−n+12 ) 2/n2 is a scaled radial distance for the neuron at 𝐫= ( x , y ) , amag is the magnitude of the input , and afall is a falloff parameter . The inhibition distance for network z is ( 3 ) l⁢ ( z ) =[lminlexp+ ( lmaxlexp-lminlexp ) ⁢z-1h-1]1/lexp , which ranges from lmin=l⁢ ( 1 ) to lmax=l⁢ ( h ) with concavity tuned by lexp . More negative values of lexp lead to greater concavity; for lexp=0 , we use the limiting expression l⁢ ( z ) =lmin ( h-z ) / ( h-1 ) ⁢lmax ( z-1 ) / ( h-1 ) . The recurrent inhibition profile for network z is ( 4 ) w ( r;z ) ={−wmagl ( z ) 21−cos⁡[πr/l ( z ) ]2r<2l ( z ) 0r≥2l ( z ) , where wmag is the magnitude of inhibition . We scale this magnitude by l⁢ ( z ) -2 to make the integrated inhibition constant across z . The excitatory coupling is ( 5 ) u ( r ) ={umagd21+cos⁡[πr/d]2r<d0r≥d , where umag and d are the magnitude and spread of coupling , respectively . In analogy to wmag , we scale umag by d-2 . To determine spatial grid scales , orientations , and gridness , we consider an annular region of the spatial autocorrelation map that contains the six peaks closest to the origin . Grid scale is the radius with highest value , averaging over angles . Grid orientation and gridness are determined by first averaging over radial distance and analyzing the sixth component of the Fourier series with respect to angle ( Weber and Sprekeler , 2019 ) . The power of this component divided by the total Fourier power measures ‘gridness’ and its complex phase measures the orientation . Grid cells are subject to a gridness cutoff of 0 . 6 . For each replicate simulation , we cluster its grid cells with respect to scale and orientation using a k-means procedure with k determined by kernel smoothed densities ( Stensola et al . , 2012 ) . See Appendix 1 for full details .	In a room , we have a sense of our location relative to the doors and to objects within the room . This is because the brain constructs a mental map of our current environment . As we move around the room , neurons called grid cells fire whenever we are in specific locations . But these locations are not random . They correspond to the corners of a grid of tessellating triangles on the floor , a little like the dots in a regular polka-dot pattern . Grid cells fire whenever we stand on one of the dots . This enables the brain to keep track of where we are and where we are heading . But the brain does not use just a single grid cell map to represent a room . Instead , it uses multiple maps with different spatial scales . These maps differ in the distance between the points at which each grid cell fires , that is , the distance between the polka dots . Some maps have many small triangles , providing high resolution spatial information . Others have fewer , larger triangles . This is similar to how we use maps with different spatial scales when driving between cities versus walking around a single neighborhood . A set of grid cell maps with the same spatial scale—and the same orientation—is known as a grid cell module . Animal experiments suggest that different individuals use a similar combination of grid cell modules that can efficiently map rooms . But how can the brain reliably produce this particular combination ? Using a computer model to simulate networks of grid cells , Kang and Balasubramanian identify a mechanism that enables the brain to spontaneously organize into the previously observed combination . The interactions between networks—in particular the balance of inhibitory and excitatory activity—determine the arrangement of grid cell modules . This process still works even with random fluctuations in network activity . Grid cells occupy a brain region that degenerates early in the course of Alzheimer's disease . This may explain why some patients experience difficulty finding their way around as one of their first symptoms . To develop effective treatments , scientists need to understand how neural circuits within this brain region work , and how the disease process disrupts them . The computer model of Kang and Balasubramanian brings the research community a step closer to achieving this . It also provides insights into how neuronal networks self-organize , which is relevant to other brain functions too .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "physics", "of", "living", "systems", "neuroscience" ]	2019	A geometric attractor mechanism for self-organization of entorhinal grid modules
Defective Ca2+ handling is a key mechanism underlying hepatic endoplasmic reticulum ( ER ) dysfunction in obesity . ER Ca2+ level is in part monitored by the store-operated Ca2+ entry ( SOCE ) system , an adaptive mechanism that senses ER luminal Ca2+ concentrations through the STIM proteins and facilitates import of the ion from the extracellular space . Here , we show that hepatocytes from obese mice displayed significantly diminished SOCE as a result of impaired STIM1 translocation , which was associated with aberrant STIM1 O-GlycNAcylation . Primary hepatocytes deficient in STIM1 exhibited elevated cellular stress as well as impaired insulin action , increased glucose production and lipid droplet accumulation . Additionally , mice with acute liver deletion of STIM1 displayed systemic glucose intolerance . Conversely , over-expression of STIM1 in obese mice led to increased SOCE , which was sufficient to improve systemic glucose tolerance . These findings demonstrate that SOCE is an important mechanism for healthy hepatic Ca2+ balance and systemic metabolic control . The endoplasmic reticulum ( ER ) is a key cellular organelle coordinating a variety of essential processes such as protein synthesis and secretion , lipid biosynthesis , glucose metabolism and redox reactions . Additionally , ER is the main site of Ca2+ storage in the cell ( Gardner et al . , 2013; Hotamisligil , 2010 ) . Structurally , ER comprises a complex network of membranes spread throughout the cytoplasm , which establishes physical and functional contacts with many other organelles including mitochondria , endosomes , lipid droplets and the plasma membrane ( Lynes and Simmen , 2011; Phillips and Voeltz , 2016 ) . Given its critical role in the cell , the functionality of the ER is tightly monitored by proteins that can communicate stress signals to other proteins or compartments of the cell in order to restore ER function . The most well-known adaptive pathway in the ER is the unfolded protein response ( UPR ) , which has the general goal of restoring ER function by enhancing ER folding capacity , decreasing protein translation and increasing protein degradation ( Gardner et al . , 2013; Hetz et al . , 2015; Hotamisligil , 2010; Wang and Kaufman , 2016 ) . Over the past decade it has been widely documented that chronic stress imposed by excess nutrients and energy leads to ER dysfunction and UPR activation in a variety of metabolic tissues in both mouse models and obese humans , which is associated with cellular stress , inflammation , and metabolic dysfunction ( Boden et al . , 2008; Gregor et al . , 2009; Hotamisligil , 2010; Nakatani et al . , 2005; Ozcan et al . , 2004; Sharma et al . , 2008; Wang and Kaufman , 2016 ) . Alleviation of ER stress by chemical or molecular chaperones dramatically improves metabolic control and insulin sensitivity and reduces inflammation in mouse models of obesity ( Fu et al . , 2015; Kammoun et al . , 2009; Ozawa et al . , 2005; Ozcan et al . , 2006 ) as well as in humans ( Kars et al . , 2010; Xiao et al . , 2011 ) . More recently , it has become clear that Ca2+ storage in the ER is compromised in the setting of obesity and other metabolic diseases ( Arruda and Hotamisligil , 2015; Ozcan and Tabas , 2016 ) . Ca2+ in the ER is essential for chaperone-mediated protein folding and secretion , as well as for the function of metabolic enzymes . Ca2+ concentration in the ER is tightly regulated by coordinated action between SERCA , which pumps Ca2+ from the cytosol into the ER lumen , and the IP3 receptor ( IP3R ) or Ryanodine receptor , which release Ca2+ from the ER into the cytosol ( Clapham , 2007 ) . In obesity , hepatic SERCA activity is compromised ( Fu et al . , 2011; Meikle and Summers , 2017; Rong et al . , 2013 ) while the activity of the IP3R1 Ca2+ channel is increased ( Arruda et al . , 2014; Feriod et al . , 2017; Wang et al . , 2012 ) . These alterations result in decreased ER Ca2+ content , loss of folding capacity , ER stress , inflammation , impaired insulin action and abnormal glucose metabolism ( Arruda and Hotamisligil , 2015 ) . Additionally , in this setting a ‘leaky’ ER contributes to both higher cytosolic Ca2+ and mitochondrial Ca2+ uptake , which has important implications for cytosolic Ca2+ signaling and mitochondrial dysfunction seen in metabolic diseases ( Arruda et al . , 2014; Feriod et al . , 2017; Ozcan et al . , 2013; Wang et al . , 2012; Xiao et al . , 2011 ) . Accordingly , strategies to restore hepatic ER Ca2+ levels , such as overexpression of SERCA or suppression of IP3R , improve ER function and promote metabolic homeostasis in mouse models of obesity ( Arruda et al . , 2014; Feriod et al . , 2017; Fu et al . , 2011; Wang et al . , 2012 ) . In a similar fashion , administration of compounds that increase ER Ca2+ content and improve ER function , e . g . SERCA agonists ( Kang et al . , 2016 ) , IP3R antagonists ( Ozcan et al . , 2012 ) , or azoromide ( Fu et al . , 2015 ) , improves metabolic health in obese mice . In addition to SERCA and IP3Rs , Ca2+ homeostasis in the ER is sensed and regulated by a third major adaptive/homeostatic system orchestrated by the STIM-Orai complex . STIM proteins ( STIM1 and STIM2 ) are found in the ER and sense luminal Ca2+ through N-terminal Ca2+ binding EF hand domain . In the resting state , STIM proteins are bound to Ca2+ and spread evenly throughout the ER membrane . Upon ER Ca2+ release , STIM forms oligomers ( seen in confocal microscopy as puncta ) , which translocate to junctions between ER and plasma membrane ( PM ) , where they couple with the PM channel protein Orai1 . This coupling results in the import of Ca2+ from the extracellular compartment to the cytosol , providing spatial Ca2+ signals that then influence cellular signaling and promote Ca2+ replenishment into the ER lumen through SERCA , in the process known as store operated Ca2+ entry or SOCE ( Feske , 2007; Prakriya and Lewis , 2015; Soboloff et al . , 2012; Wu et al . , 2007 ) . Given that obesity leads to impaired ER Ca2+ handling and this is a key mechanism associated with ER dysfunction in the obese condition , we examined whether alterations in SOCE system exist and contribute to the inability of ER to restore and or maintain Ca2+ levels . Here we show that STIM1 is modified and SOCE is defective in hepatocytes from obese mice due to the inability of STIM1 to translocate and couple with Orai1 at the PM . Hepatocytes lacking STIM1 displayed stress , increased glucose production and insulin resistance , whereas over-expression of STIM1 in the liver of obese mice significantly improved glucose intolerance . These findings support a critical role for STIM1-mediated SOCE in the maintenance of healthy Ca2+ balance in the ER , insulin action , and systemic glucose metabolism . In order to determine the impact of obesity on SOCE in the liver , we isolated primary hepatocytes from lean wild-type ( WT ) and genetically obese ( Lepob/ob ) mice and measured Ca2+ influx through STIM/Orai1 using the ratiometric calcium dye , Fura-2AM ( Poenie and Tsien , 1986 ) . First , the cells were incubated in Ca2+-free medium , and ER Ca2+ store depletion was induced by the addition of the SERCA inhibitor thapsigargin ( Tg ) . In agreement with earlier reports ( Arruda et al . , 2014 ) the initial rise in cytosolic Ca2+ induced by Tg , which reflects the ER Ca2+ content , was significantly lower in hepatocytes isolated from obese animals compared with WT cells ( Figure 1A ) . Next , we induced SOCE through STIM/Orai1 by substituting the Ca2+-free media with a media containing 2 mM Ca2+ . As depicted in Figure 1A , the Ca2+ entry in primary hepatocytes from obese mice was markedly reduced relative to control cells . To further confirm that the observed Ca2+ entry was mediated by SOCE , we added the SOCE inhibitor 2-apb , which completely blocked Ca2+ influx in both genotypes ( Figure 1A , dotted lines ) . This finding is consistent with a report of impaired SOCE in the steatotic hepatocytes from Zucker rats ( Wilson et al . , 2015 ) . As we observed impaired SOCE in hepatocytes from obese animals we next evaluated whether the expression levels of the main SOCE components were altered in obesity . As shown in Figure 1—figure supplement 1A , we did not detect differences in expression of Orai1 in the livers of mice with genetic or high-fat diet ( HFD ) -induced obesity . Despite a modest increase in the mRNA levels of stim1 and stim2 , protein levels of STIM1 , STIM2 and Orai1 remained unchanged in the livers of mice with both genetic and diet induced obesity ( Figure 1B and Figure 1—figure supplement 1B ) . However , phospho-JNK ( pJNK ) , a marker of inflammatory stress ( Hirosumi et al . , 2002 ) and phospho-Calmodulin kinase ( pCaMKII ) , a marker of elevated cytosolic Ca2+ ( Ozcan et al . , 2012 ) , were increased in the livers of obese animals ( Figure 1B and Figure 1—figure supplement 1B ) . Thus decreased Ca2+ import through SOCE in hepatocytes derived from obese mice appears to be independent of the expression levels of critical SOCE components . We therefore asked whether the decreased Ca2+ import through SOCE in hepatocytes from obese mice was a result of a functional alteration in STIM1 translocation in the ER membrane . To evaluate the activity of STIM1 proteins in obesity , we first validated antibodies for endogenous immunostaining ( Figure 1; Figure 1—figure supplement 1C ) . Next , we isolated primary hepatocytes from lean and obese mice , induced Ca2+ store depletion with Tg and determined STIM1 localization by immunostaining . As shown in Figure 1C and D and Figure 1—figure supplement 1D , while STIM1 was evenly distributed in the ER membrane of resting cells , Tg treatment led to a dramatic translocation to areas of the ER membrane in close proximity with the PM . This effect was observed within 5–10 min of Tg treatment and persisted for 30 min . However , in cells from obese ( Lepob/ob ) mice , STIM1 distribution at baseline was punctate throughout the entire ER ( Figure 1D and quantified in Figure 1E ) , and its translocation to areas of ER/PM junction in response to Tg was dramatically impaired . As shown in Figure 1F and Figure 1—figure supplement 1E , in WT cells , Tg treatment increased the number of high-intensity pixels , indicating the formation of the punctate structures , which represent oligomerization of STIM proteins . In contrast , hepatocytes derived from obese mice displayed higher intensity pixel counts at baseline with no further increase following Tg treatment . Defective STIM1 translocation in hepatocytes from obese animals is also illustrated in the line graphs showing the representative intensity profile of individual cells before and after Tg treatment ( Figure 1G and also see Figure 1—figure supplement 1F for image analysis details ) . In WT cells , Tg treatment increased the signal intensity at the edge of the cell and decreased it in the cytosol , quantified as the STIM1 ratio between these compartments ( Figure 1H ) . In hepatocytes from obese mice , the average signal intensity at baseline was higher than in WT cells , and did not change significantly in response to Tg ( Figure 1H ) , indicating that STIM1 translocation is markedly defective in these cells . To examine STIM1 translocation after Tg stimulation in more detail and using an alternative analytical tool , we performed total internal reflection fluorescence ( TIRF ) microscopy , in which fluorophores in the vicinity of the cell surface are selectively excited . As shown in Figure 1I , in WT cells Tg treatment induced STIM1 puncta formation and accumulation in areas close to the PM . This response was diminished in cells derived from obese animals ( Figure 1J ) . In these experiments , Na+K+ ATPase was used as a constitutive marker of PM and its staining pattern did not change in the presence of Tg . We also evaluated the time course of the defect in STIM1 trafficking during the development of obesity in mice . As shown in Figure 1—figure supplement 2A and Figure 1—figure supplement 2B , after 3 and 5 weeks of HFD , STIM1 showed some degree of translocation towards the plasma membrane in resting cells , indicating that short term HFD exposure is sufficient to trigger ER Ca2+ release , the driving force for STIM1 translocation . Additionally , at 3 weeks HFD , STIM1 translocation stimulated by Tg was not significantly impaired compared with cells derived from chow fed animals . However , the defect in STIM1 translocation induced by Tg was apparent in cells isolated from mice fed a HFD for 5 weeks and was more pronounced after 7 and 11 weeks of HFD ( Figure 1—figure supplement 2A and Figure 1—figure supplement 2B ) . These data support that defective STIM1 translocation may be a common feature of obesity in independent experimental models ( Lepob/ob and HFD ) and that this phenotype occurs independent of blood glucose levels and prior to the emergence of marked hyper-insulinemia and insulin resistance found in obesity ( Figure 1—figure supplement 2C ) . Next , we examined the activation and translocation of STIM2 in cells from lean ( WT ) and obese ( Lepob/ob ) mice . The role of STIM2 in the regulation of SOCE has been explored only recently , and its activation seems to result from smaller fluctuations in Ca2+ due to its lower affinity for the ion , while STIM1 is activated when the extent of ER depletion increases ( Berna-Erro et al . , 2017; Oh-Hora et al . , 2008; Prakriya and Lewis , 2015 ) . In hepatocytes , STIM2 is expressed at a lower level than STIM1 ( López et al . , 2012; Williams et al . , 2001 ) , however it was detectable both by staining and by western blot analysis . We did not observe aberrant STIM2 puncta formation in cells from Lepob/ob mice but a tendency for translocation at baseline . Upon Tg-induced store depletion , we observed a marked STIM2 translocation to areas close to the PM in both WT cells and Lepob/ob cells , however the degree of translocation in Lepob/ob cells was less pronounced than in WT cells ( Figure 1—figure supplement 3A ) . Thus , the obesity-related defect in STIM2 translocation was present but less dramatic than that observed for STIM1 . Overall , these findings suggest that defective SOCE in primary hepatocytes from obese mice is predominantly related to aberrant STIM1 puncta formation and inefficient STIM1 re-localization from the bulk ER membrane to ER/PM junction areas after ER Ca2+ store depletion . In order to gain insight into mechanisms underlying this phenomenon , we considered the possibility that altered STIM1 post-translational modification could be involved in its defective translocation capacity in the context of obesity . Two known post-translational modifications of STIM1 which influence its trafficking are phosphorylation and O-GlcNAcylation ( Pozo-Guisado et al . , 2013; Zhu-Mauldin et al . , 2012 ) . Phosphorylation of STIM1 at Ser621 and Ser575 regulate its interaction with the microtubule plus-end-tracking protein EB1 , enabling STIM1 to move in the ER membrane ( Pozo-Guisado et al . , 2013 ) . Therefore , we asked whether a reduction in STIM1 phosphorylation at these sites may explain its defective trafficking in obesity . Surprisingly however , we found that phosphorylation of STIM1 at Ser621 and Ser575 was actually increased in liver lysates from obese mice ( Figure 2A ) . This indicates that lack of phosphorylation at these residues does not underlie the defective translocation of STIM1 , and suggests that hepatic STIM1 may be released from EB1 in the basal state , potentially as a response to the decreased ER Ca2+ level observed in hepatocytes from obese mice . STIM1 can also be modified by O-linked N-acetyl glucosamine ( O-GlcNAc ) , a post-translational modification of serine or threonine amino acids . Previous work has shown that STIM1 O-GlcNAcylation impairs the ability of the protein to move in the ER membrane and to form punctate structures ( Zhu-Mauldin et al . , 2012 ) . Additionally , conditions of nutrient and substrate excess , including obesity , lead to increase in cellular O-GlcNAcylation levels ( Vosseller et al . , 2002; Yang and Qian , 2017; Dentin et al . , 2008; Yang et al . , 2008 ) . These studies indicate the possibility that metabolic stress may interfere with SOCE via O-GlcNAcylation of STIM1 , which disrupts its proper trafficking . To test this hypothesis , we first examined global O-GlcNAcylation in primary hepatocytes from WT and obese mice . As shown in Figure 2B , and in agreement with previous observations ( Baldini et al . , 2016 ) , global O-GlcNAcylation was higher in hepatocytes derived from obese animals compared with their controls . This effect was amplified by treatment with PugNac , a specific inhibitor of O-GlycNAcase ( OGA ) , the enzyme that catalyzes removal of OglcNAc sugars from proteins . Next , we performed immunoprecipitation of O-GlcNAc-modified proteins using an OglcNAc-specific antibody from hepatocyte lysates . We found that STIM1 was present among the OglcNAcylated proteins and it was more abundant in cells from obese mice ( Figure 2C ) . As a complementary approach , we utilized biotinylated-succinylated wheat germ agglutinin ( succinylated-WGA ) , a lectin that preferentially binds N-acetylglucosamine versus other sugars ( Baldini et al . , 2016; Hu et al . , 2010 ) . Following precipitation with streptavidin-conjugated magnetic beads , STIM1 was detected in the pool of proteins modified by O-GlcNac at higher levels in samples derived from obese animals relative to WT controls ( Figure 2D ) . Additionally , as shown in Figure 2E , OGT precipitates with STIM1 in hepatocytes derived from obese animals , indicating that OGT and STIM1 exist in a complex in obesity , consistent with higher levels of O-GlcNac-modified STIM1 in this condition . To explore whether STIM1 is also modulated by OglcNAcylation in the HFD context and study the time course of this modification , we isolated hepatocytes from animals fed a HFD for 3 , 5 and 11 wks and pull down STIM1 using succinylated-WGA . As shown in Figure 2—figure supplement 1A , STIM1 was progressively modified by O-GlcNac in this context starting at 3 weeks on HFD . In order to test if O-GlcNAc modification may alter STIM1 function in hepatocytes , we first overexpressed OGT in Hepa1-6 cells . As expected , overexpression of OGT strongly increased global protein modification with O-GlcNac ( Figure 2—figure supplement 1B ) . We then co-expressed OGT and STIM1 tagged with a Flag peptide ( Figure 2—figure supplement 1C ) to determine whether STIM1 can be directly O-GlcNacylated by OGT in this system . Immunoprecipitation of STIM1-Flag with a Flag specific antibody demonstrated its enhanced O-GlcNAc modification in OGT overexpressing cells ( Figure 2F ) . Additionally , STIM1 is able to directly bind to OGT in this cell model ( Figure 2F ) , similar to what we observed in hepatocytes derived from obese animals ( Figure 2E ) . Having established this cellular system with enhanced STIM1 O-GlcNacylation , we then assessed its translocation in cells transfected with OGT labeled with RFP , using un-transfected cells in the same culture plate as internal controls for quantification ( Figure 2G ) . These experiments showed that STIM1 puncta formation and translocation following Tg treatment was markedly impaired in cells overexpressing OGT compared with control cells ( Figure 2G , H and I ) . Additionally , overexpression of OGT led to a significant impairment in SOCE compared with the controls ( Figure 2J ) , with no significant differences in the response to Tg ( data not shown ) . Next , to determine whether inhibition of the O-GlcNac modification could rescue STIM1 translocation defect in Lepob/ob cells , we used adenoviral gene delivery to express an shRNA targeting OGT to down-regulate OGT and thus the O-GlcNacylation capacity of the cell . As can be seen in Figure 2K and L , the expression of OGT shRNA lead to a 60% decrease in its expression and activity . As shown in Figure 2M and N , down-regulation of OGT resulted in increased STIM1 translocation capacity compared with cells expressing a scrambled shRNA . Taken together these data indicate that increased modification of STIM1 by OglcNAc could , at least in part , underlie defective STIM1 translocation and reduced hepatic SOCE in the context of obesity . We then started to investigate the impact of defective STIM1-mediated SOCE on ER homeostasis , cellular stress responses and metabolic regulation in STIM1-deficient hepatocyte cell models . These included primary hepatocytes derived from mice with genetic STIM1 deficiency specifically in the liver ( stim1fl/fl Alb;Cre , identified here as stim1ΔLIVER ) and Hepa1-6 cells stably expressing shRNAs targeting stim1 ( Figure 3—figure supplement 1A ) or stim2 ( Figure 3—figure supplement 1B ) ( identified here as shSTIM1 and shSTIM2 respectively ) . Primary hepatocytes derived from stim1ΔLIVER mice displayed lower ER Ca2+ content and absence of Tg-triggered SOCE compared to controls ( stim1fl/fl ) ( Figure 3A ) . Similarly , Hepa 1–6 cells in which STIM1 expression was down regulated showed markedly blunted SOCE ( Figure 3—figure supplement 1C ) . Imbalances in ER Ca2+ content trigger cellular stress responses through various pathways ( Arruda and Hotamisligil , 2015; Fu et al . , 2012; Ozcan and Tabas , 2016 ) . In order to examine these responses in our cellular models of STIM deficiency , we measured phosphorylation of JNK as a benchmark measure of cellular stress and inflammatory activation , and phosphorylation of eIF2α as a marker of UPR activation . As shown in Figure 3B , in stim1fl/fl control cells , treatment with Tg induced phosphorylation of JNK and eIF2α at 30 min followed by a decrease in phosphorylation levels back to baseline after 90 min . However , in STIM1-deficient cells , the phosphorylation of JNK and eIF2α was enhanced at baseline and these cells displayed a stronger and more persistent response to Tg , with stress markers only returning to basal levels at 120 min . A similar profile was observed in Hepa1-6 cells with shRNA-mediated suppression of STIM1 ( Figure 3—figure supplement 1D ) . Interestingly , suppression of STIM2 alone did not alter the cellular response to Tg ( Figure 3—figure supplement 1E ) . Altogether , these data demonstrate that absence of core components of SOCE leads to elevated and prolonged stress responses in hepatocytes . Increased cellular stress and inflammation are associated with impaired insulin action and defective glucose and lipid metabolism ( Fu et al . , 2012; Hirosumi et al . , 2002; Hotamisligil , 2017 ) . Accordingly , we found that primary hepatocytes derived from stim1ΔLIVER mice displayed impaired phosphorylation of IRβ and AKT in response to insulin relative to cells from stim1fl/fl littermates ( Figure 3C ) . Likewise , gene silencing of stim1 in Hepa1-6 cells resulted in decreased insulin signaling ( Figure 3—figure supplement 1F ) . Additionally , primary hepatocytes isolated from stim1ΔLIVER mice maintained on HFD showed higher levels of glucose production stimulated by the gluconeogenesis substrates glycerol , pyruvate and glutamine compared to stim1fl/fl derived hepatocytes ( Figure 3E ) . In agreement with our finding that STIM2 suppression does not amplify stress responses , we found that STIM2-deficient cells retained normal insulin responsiveness ( Figure 3—figure supplement 1G ) . Recently , it has been shown that SOCE may regulate lipid metabolism ( Maus et al . , 2017; Wilson et al . , 2015 ) and animals with inducible STIM1/2 whole body deficiency exhibit increased lipid accumulation in multiple tissues such as skeletal muscle , heart and liver , as a consequence of impaired lipolysis and fatty acid oxidation ( Maus et al . , 2017 ) . We observed that primary hepatocytes isolated from stim1ΔLIVER mice showed modestly increased accumulation of lipid droplets at baseline and after incubation with 1 mM oleic acid and 40 µM of palmitic acid ( Figure 3D ) . These results suggest that in addition to higher stress levels , insulin resistance and increased glucose production , SOCE deficiency may result in abnormal lipid metabolism . We next assessed the systemic metabolic impact of STIM1 deficiency in hepatocytes in vivo in the stim1ΔLIVER mice described above . As shown in Figure 3—figure supplement 2A , liver lysates from stim1ΔLIVER mice displayed marked reduction in STIM1 and a mild increase in STIM2 protein . On a chow diet , stim1ΔLIVER mice gained weight equivalently to stim1fl/fl mice ( Figure 3—figure supplement 2B ) and did not exhibit alterations in glucose tolerance and triglyceride content ( Tg ) ( data not shown ) . In the HFD-fed mice , the weight gain was similar between genotypes ( Figure 3—figure supplement 2B ) . Interestingly , at 6 weeks on a HFD , we observed a mild , but significant , glucose intolerance in the stim1ΔLIVER animals compared with the stim1fl/fl controls ( Figure 3—figure supplement 2C ) . Additionally , at this time point , insulin sensitivity was impaired ( Figure 3—figure supplement 2D ) . No significant differences were observed in Tg content between the two genotypes , although a tendency to higher Tg levels was detected in stim1ΔLIVER mice ( Figure 3—figure supplement 2E ) . In long term HFD ( 20 weeks ) , the differences in glucose intolerance , insulin signaling and Tg content between stim1fl/fl and stim1ΔLIVER groups were very mild and did not reach statistical significance ( Figure 3—figure supplement 2F , G and H ) . Interestingly , although we haven’t observed overall changes in lipid accumulation in the liver specific STIM1 deficient animals , we have noticed that these animals showed higher content of microvesicular steatosis compared with controls where the majority of the lipid droplets are presented as large droplets ( Figure 3—figure supplement 2I ) . In light of the mild in vivo metabolic phenotype of constitutive hepatocyte-specific STIM1 deletion during development , we considered that STIM2 upregulation ( Figure 3—figure supplement 2 ) or alternative systems could compensate for the lack of STIM1 all throughout the embryonic life . Therefore , we examined the effect of acute deletion of STIM1 in hepatocytes , using adenovirus mediated gene delivery to express albumin-Cre recombinase ( Alb;Cre ) in adult stim1+/+ and stim1fl/fl mice . After weaning , littermate animals were subjected to HFD for 4 weeks ( for a short-term stress induction ) ( Figure 3F ) . Adenoviral delivery of Alb;Cre recombinase resulted in ~60% deletion of hepatic STIM1 ( Figure 3G ) . Importantly , there was no compensatory upregulation of STIM2 or Orai1 in this setting . Indeed , one week after adenovirus administration , the acute deletion of STIM1 resulted in significantly higher fasting glucose levels ( Figure 3H ) without any difference in insulin levels ( Figure 3I ) and body weight ( data not shown ) . Interestingly , following STIM1 deletion , mice also exhibited significantly impaired glucose tolerance ( Figure 3J ) . In summary , acute SOCE dysfunction in the liver results in amplified cellular stress and glucose intolerance in mice fed a HFD . Based on these observations , and the marked SOCE defect associated with obesity we hypothesized that recovery of SOCE function in the livers of obese mice could improve metabolic homeostasis in obesity . To test this hypothesis , we used a hepatocyte-specific adenovirus system to exogenously express either GFP ( control ) or STIM1-YFP in the Lepob/ob primary hepatocytes and mice ( Figure 4A ) . The functionality of the STIM1-YFP fusion protein has been verified previously ( Liou et al . , 2005 ) and confirmed by us ( Video 1 ) . Using this approach , we asked whether exogenous expression of STIM1-YFP would be sufficient to overcome the SOCE defects in primary hepatocytes from obese animals . As shown in Figure 4B , in control Lepob/ob cells expressing GFP , STIM1 formed puncta at baseline and Tg-induced translocation of the protein was impaired , as we previously observed ( Figure 1D ) . Delivery of STIM1-YFP lead to a significant increase in the protein level in liver cells . Interestingly , in non treated cells , we observed that some degree of the STIM1 protein was already localized in areas of close contact with the plasma membrane even prior to stimulation ( Figure 4C ) . Additionally , exogenously expressed STIM1 was able to translocate towards the plasma membrane after Tg treatment ( Figure 4C and D ) . To examine STIM1 translocation in Lepob/ob hepatocytes with higher resolution , we also performed TEM in cells expressing exogenous STIM1 fused with HRP . STIM1-HRP-expressing cells were fixed and treated with diaminobenzidine ( DAB ) in the presence of H2O2 . HRP catalyzes the polymerization and deposition of DAB , which recruits electron-dense osmium , providing contrast and revealing the localization of STIM1 . As shown in Figure 4E , exogenous delivery of STIM1-HRP resulted in rescue of translocation in cells from obese animals in a manner similar to that observed in lean , wild type cells . Taken together , these data indicate that exogenously increasing the amount of STIM1 protein is sufficient to partially overcome the STIM1 translocation defect in cells from obese mice . The ability of STIM1-YFP overexpression to overcome , at least in part , the translocation defect of endogenous STIM1 in Lepob/ob cells suggests that at least some proportion of STIM1 in this setting may escape the O-GlycNac modification . To examine the degree of O-GlcNacylation of overexpressed STIM1-YFP in Lepob/ob cells , we used an O-GlcNac specific antibody to immunoprecipitate all proteins modified by O-GlycNacylation in GFP or STIM1 YFP overexpressing cells and examined STIM1 protein by immunoblotting . As shown in Figure 4F , STIM1-YFP overexpression resulted in an ~20 fold increase in the expression of STIM1 protein . However , the amount of O-GlcNac modified STIM1 did not increase proportionally , indicating that in fact a significant amount of the overexpressed STIM1 escapes this post translation modification . Additionally , it has been previously shown that overexpression of STIM1 in HeLa cells leads to morphological changes in the ER with an increased cortical ER ( Orci et al . , 2009 ) . In agreement with this finding , electron micrographs of primary hepatocytes derived from Lepob/ob mice show that STIM1-YFP expression led to a remodeling of ER with abundant ER stacks apposed to the PM , although the degree of ER remodeling varied from cell to cell , likely due to variable STIM1 expression levels ( Figure 4—figure supplement 1A ) . This result suggests that , in addition to having more STIM1 that translocates to PM , overexpression of STIM1-YFP also remodels the ER , favoring the proximity of STIM1 to Orai at the plasma membrane . After verifying that overexpression of STIM1-YFP was able to rescue STIM1 translocation defects in obese cells , we examined whether this actually resulted in improved SOCE . As shown in Figure 4G overexpression of STIM1-YFP in Lepob/ob cells resulted in increased ER Ca2+ and SOCE compared to control cells . Thus , overexpression of STIM1 was a successful intervention to overcome the STIM translocation and SOCE defects , resulting in improved ER calcium handling . Next , we asked whether the effects of the overexpression of STIM1-YFP in Lepob/ob primary hepatocytes would impact overall metabolism in vivo , and evaluated the effect of liver-specific STIM1 expression on systemic glucose metabolism in the Lepob/ob mice following the protocol displayed in Figure 5A . Introduction of STIM1 by adenovirus gene delivery led to increased STIM1 protein levels in the liver . Interestingly , increased STIM1 expression was also accompanied by increased mRNA expression of Orai1 and with an increase in SERCA ( Atp2a2 ) mRNA and protein levels ( Figure 5B and C ) . This is in accordance with previous work showing that the modulation of expression STIM1 affects the expression of other Ca2+ channels or pumps , as SERCA , likely due to alterations in cytosolic Ca2+ levels ( Abell et al . , 2011 ) . Remarkably , liver-specific overexpression of STIM1 led to a significant improvement in glucose tolerance relative to control ( adGFP ) animals as evaluated by a glucose tolerance test ( Figure 5D ) . Improved glucose tolerance resulting from overexpression of STIM1 was associated with enhanced in vivo insulin signaling evaluated by direct insulin injection into the livers that resulted in increased AKT phosphorylation ( Figure 5E ) . Additionally , expression of genes involved in glycolysis were increased in livers overexpressing STIM1 , accompanied by decreased levels of PCK1 ( PEPCK ) , a rate limiting enzyme for gluconeogenesis ( Figure 5F and G ) . Given the reported role of STIM1 in lipid metabolism , we also evaluated if overexpression of STIM1 impacted the excessive lipid accumulation in the livers of Lepob/ob animals . As shown in Figure 5H , exogenous expression of STIM1 led to a reduction in total Tg content in the liver tissue of Lepob/ob mice compared to controls expressing GFP . Accordingly , STIM1 overexpression led to increase mRNA expression of genes involved in fatty acid oxidation , such as CPT1 , while no change was observed in genes involved in lipogenesis ( Figure 5—figure supplement 1A and B ) . Based on these results we conclude that SOCE is critical for glucose and lipid metabolism and that increasing hepatic SOCE through overexpression of STIM1 is able to revert , at least in part , the deleterious impact of obesity on systemic glucose homeostasis . In the last several years , a growing number of studies have laid the foundation for the concept that disrupted intracellular Ca2+ homeostasis in metabolic tissues is a key component of ER dysfunction and metabolic deterioration ( Arruda and Hotamisligil , 2015; Ozcan and Tabas , 2016 ) . The abnormal Ca2+ handling in the ER of hepatocytes in this setting is multi-faceted , involving both impairment of SERCA activity ( Fu et al . , 2011 ) as well as increased activity of IP3R ( Arruda et al . , 2014; Feriod et al . , 2017; Wang et al . , 2012 ) . This impacts not only Ca2+ levels in the ER and protein folding , but also mitochondrial oxidation and ROS production as well as the activation of cytosolic stress signaling pathways . Supporting the critical role of this biology , human genetic studies have shown that SNPs in SERCA ( Varadi et al . , 1999 ) and IP3R ( Shungin et al . , 2015 ) are associated with metabolic diseases such as obesity and diabetes . Here , our work reveals that another key mechanism supporting ER Ca2+ homeostasis , STIM-mediated SOCE , is also defective in obesity with important implications for the metabolic dysfunction in hepatocytes . Strikingly , we demonstrate that STIM1 gain-of-function improves cellular stress responses and glucose homeostasis in obese mice , underscoring the therapeutic potential of targeting Ca2+ homeostasis through STIM1 for metabolic disease treatment . Our interest in this biology was heightened by our initial observation that in hepatocytes isolated from obese ( Lepob/ob ) mice , STIM1 was present in an aberrant punctate pattern along the ER membrane . One possible explanation was that Lepob/ob hepatocytes display increased SOCE even in baseline conditions in order to respond to low ER Ca2+ levels induced by dysfunction of SERCA and IP3R1 . However , our detailed examination revealed that in cells of obese mice , STIM1 puncta were not localized in areas of close proximity of the PM , but rather were distributed all throughout the cell away from the PM . Additionally , when we depleted ER Ca2+ stores , STIM1 was not able to fully translocate to the PM , resulting in significantly reduced SOCE . Hence , metabolic stress impairs this critical component of ER Ca2+ homeostasis in obese liver tissue . One question arising from these observations is the nature of the mechanism underlying STIM1 dysfunction in obesity . STIM activation and translocation is modulated by a complex array of factors that include post-translational modification of the protein itself , protein-protein interactions ( for review Lopez et al . , 2016 ) , the interaction of STIM1 with lipids of the ER membrane and the ability of the ER itself to remodel and form ER/PM junctions ( Derler et al . , 2016; Lopez et al . , 2016; Prakriya and Lewis , 2015 ) . Here we found that in hepatocytes derived from Lepob/ob mice , STIM1 displays significantly increased levels of O-GlcNAcylation , a post-translation modification that causes impaired STIM1 trafficking and function ( Zhu-Mauldin et al . , 2012 ) . Indeed , overexpression of OGT in hepatocytes was sufficient to induce abnormal STIM1 puncta formation and impaired activation of SOCE while down-regulation of OGT in primary hepatocytes from obese mice partially rescued the STIM1 translocation defect . It will be interesting to explore the detailed location and impact of these modifications in future studies . In silico analysis , published literature and data in dbOGAP , suggest potential target sites coinciding with the domains involved in STIM1/Orai1 interactions with Orai1 in the CAD/SOAR domain or in the polybasic motif located at the C terminal of the protein that may be critical in oligomerization and puncta formation ( Liou et al . , 2007 ) . Regardless , protein modification with O-GlcNAc is a reflection of the nutritional status , as it integrates glucose , amino acid , fatty acid and nucleotide flux through the hexosamine biosynthetic pathway ( HBP ) to produce N-acetyl-glucosamine ( UDP-GlcNAc ) , the obligatory substrate for OGT . Hence , conditions of nutrient and substrate excess , including obesity , lead to increased cellular O-GlcNAcylation levels ( Vosseller et al . , 2002; Yang and Qian , 2017; Dentin et al . , 2008; Yang et al . , 2008 ) . Taken together , our findings demonstrate that O-GlcNAc modification of STIM1 alters organelle Ca2+ regulation with consequences for ER function and systemic metabolism and provide an important mechanistic insight into regulation of this critical metabolic integration point through organelle homeostasis in obesity . Another important finding reported here is that impaired SOCE in hepatocytes by STIM1 deficiency was sufficient to induce markers of ER stress and inflammation , to impair of insulin action , increase glucose production and induce excessive lipid droplet accumulation . Recently , it has been reported that cells from patients with loss-of-function mutations in STIM1 or ORAI accumulate lipid droplets as a consequence of their inability to increase cAMP , mobilize fatty acids from lipid droplets , activate lipolysis , and oxidize fatty acids ( Maus et al . , 2017 ) . Interestingly , tamoxifen-induced whole body STIM1/2 deficiency leads to higher lipid content in the heart , muscle and liver ( Maus et al . , 2017 ) . Here , although we detected increased accumulation of lipid droplets in hepatocytes lacking STIM1 in vitro , these phenomena were not clearly observed in vivo . Hepatocyte-specific deletion of STIM1 led to a mild metabolic phenotype in vivo suggesting that the activation of compensatory systems occurs when STIM1 is deleted during development . Indeed , STIM2 expression levels were slightly increased in STIM1 deficient liver . Furthermore , in lean animals , the rest of the ER calcium handling systems are intact , which may allow maintenance of the normal equilibrium . In contrast , acute down-regulation of STIM1 showed a more marked effect on glucose homeostasis . Finally , we show here that exogenous expression of STIM1 in primary hepatocytes from obese mice is sufficient to overcome SOCE defects and partially correct ER Ca2+ levels and STIM1 re-localization upon Ca2+ store depletion . Notably , we showed that a significant part of the exogenously expressed STIM1 is able to escape the O-GlcNAc modification , probably by a mass effect , and responds properly to ER Ca2+ depletion . Additionally , overexpression of STIM1 in hepatocytes from Lepob/ob animals led to remodeling of the ER with an increase in the amount of cortical ER possibly facilitating STIM1 translocation and Orai1 coupling . Indeed , it is known that STIM1 regulates the structure of the ER through its role in Tip Attachment Complex movement ( Grigoriev et al . , 2008; Westrate et al . , 2015 ) , coordinating the growth and shrinkage of the ER tubule . Importantly , we show that upregulation of SOCE through STIM1 replenishment in the liver of obese mice results in significant metabolic benefit including improved insulin signaling and glucose tolerance and decreased lipid content likely as a result of enhanced fatty acid oxidation . When STIM1 is exogenously supplied to the liver tissue in obese mice , we also observed an increase in glycolytic gene expression . This is in agreement with recent report showing that in T cells SOCE regulates the expression of glycolytic genes through the modulation of the transcription factor NFAT ( Vaeth et al . , 2017 ) . Overexpression of STIM1 also led to increased SERCA levels , which is necessary for the delivery of Ca2+ into the ER . In fact , it is proposed and validated experimentally and computationally that an adaptive feedback loop between components of Ca2+ machinery exist to keep cytosolic Ca2+ levels under control ( Abell et al . , 2011 ) . Increased SERCA expression in the context of STIM1 overexpression may indeed be a critical component of the metabolic benefit resulting from STIM1 expression in the liver in obese mice . It is remarkable that the individual manipulation of multiple proteins involved in the maintenance of ER Ca2+ levels- SERCA , IP3R , and STIM1- all result in a similar phenotype . These findings underscore the importance of ER Ca2+ homeostasis and ER function to metabolic health , and indicate the strong potential for targeting these pathways to combat metabolic disease . All in vivo studies are approved by the Harvard Medical Area Standing Committee on Animals . Unless stated otherwise , mice were maintained from 4 to 20 weeks on a 12-hour-light/12 hr-dark cycle in the Harvard T . H . Chan School of Public Health pathogen-free barrier facility with free access to water and to a standard laboratory chow diet ( PicoLab Mouse Diet 20 #5058 , LabDiet ) . No specific power analysis was used to estimate sample size . The sample size and number of replicates for this study were chosen based on previous experiments performed in our lab and others ( Arruda et al . , 2014; Fu et al . , 2011 ) . We used two mouse models of obesity , the leptin-deficient Lepob/ob mouse , and HFD-induced obesity . For the former , wild-type mice in the C57BL/6J genetic background ( Stock no . 000664 ) and Lepob/ob mice ( Stock no . 000632 ) were purchased from Jackson Laboratories at 6–7 weeks of age and used for experimentation between 8–12 weeks of age . For the latter , male C57BL/6J mice were purchased from Jackson Laboratories and placed on HFD ( D12492: 60% kcal% fat; Research Diets ) for up to 20 weeks . Control mice of the same age were fed with a low fat diet ( PicoLab Mouse Diet 20 #5053 , LabDiet ) . In animal experiments , all measurements were included in the analysis unless they fell more than two standard deviations from the mean . Mice carrying floxed alleles for stim1 ( referred here as stim1fl/fl ) on a C57BL/6J background were kindly provided by Dr . Anjana Rao , La Jolla Institute for Allergy and Immunology . To generate hepatocyte-specific STIM deficient mice , stim1fl/fl mice were bred to C57BL/6J mice expressing CRE recombinase under the control of the albumin promoter ( Alb;Cre ) . Alb;Cre -mediated recombination of floxed stim1 alleles was detected in genomic DNA by PCR . Age-matched littermates were used for the study of adult mice . For HFD studies , pups were placed on low-fat control diet ( 5053 ) for 1 week at weaning , followed by up to 20 weeks on HFD . The control chow group remained on low-fat diet . For the adenovirus-mediated stim1 down-regulation study , stim1fl/+ ( het ) animals were crossed to each other to generate a group containing stim1+/+ , stim1fl/+ and stim1fl/fl homozygous mice . Pups of the three genotypes were weaned and fed a low-fat diet ( 5053 ) for 1 week followed by 4 weeks of HFD . The animals were transferred to a BL2 facility where they received adenovirus-expressing Alb;Cre-recombinase ( 1 × 109 IFU/mouse ) intravenously . Metabolic studies were performed between day 7–12 after infection . The animals were sacrificed after 14 days of infection for ex vivo experiments . For exogenous STIM1 expression , purified , de-salted adenovirus ( serotype 5 , Ad5 ) expressing STIM1-YFP was purchased from Vector BioLabs with the agreement of Dr . Alexei Tepikin from University of Liverpool . The adenovirus was administered to 8–10 week-old Lepob/ob mice intravenously , at a titer of 1 × 109 IFU/mouse . Metabolic studies were performed between day 7–12 after infection . The animals were sacrificed after 14 days of infection for ex vivo experiments . Animals were subjected to an intraperitoneal ( i . p . ) glucose injection ( lean: 1 . 5 g kg−1 , obese: 0 . 5–1 . 0 g kg−1 ) after overnight fasting , and blood glucose levels were measured throughout the first 120 min of the metabolic response . Liver tissues ( approximately 100 mg ) were homogenized in 1 . 2 mL of water . Next , 100 µL of the homogenate was transferred to a 1 . 5 mL tube and 125 µL of chloroform and 250 µL of Methanol were added . Samples were vortexed briefly and incubated for 5 min . Additional 125 µL of chloroform was added . Next , 125 µL of water was added and samples were vortexed and centrifuged at 3000 rpm for 20 min at 4°C . Around 150 µL of the lower phase was collected in a 1 . 5 mL tube and dried ( evaporated ) in a heated vacuum oven . Thereafter , lipids were re-suspended in 300 uL of ethanol . Triglyceride in this solution was measured by a Randox Tg Kit ( cat number TR213 ) from Randox Laboratories . Animals were anesthetized using 2 mg/ml xylazine combined with 2 mg/ml ketamine in PBS and the livers were perfused with 50 mL of buffer I ( content described below ) through the portal vein with an osmotic pump set to the speed of ~4 mL/min until the liver turned pale . The speed was gradually increased until ~7 mL/min afterwards . When the entire buffer I had been infused , it was substituted for 50 mL of buffer II . The buffers should be kept at ~37°C during the entire procedure . After perfusion , the primary hepatocytes were carefully released and sedimented at 500 rpm for 2 min , washed two times and suspended with Williams E medium supplemented with 5% CCS and 1 mM glutamine ( Invitrogen , CA ) . To separate live from dead cells , the solution of hepatocytes was layered on a 30% Percoll gradient and centrifuged ~1500 rpm for 10–15 min . The healthy cells were recovered at the bottom of the tube and plated for experimentation . Buffer I contained: 11 mM Glucose; 200 µM EGTA; 1 . 17 mM MgSO4 heptahydrated; 1 . 19 mM KH2PO4; 118 mM NaCl; 4 . 7 mM KCl; 25 mM NaHCO3 , pH 7 . 32 . Buffer II contained: 11 mM Glucose; 2 . 55 mM CaCl2; 1 . 17 mM MgSO4 heptahydrated; 1 . 19 mM KH2PO4; 118 mM NaCl; 4 . 7 mM KCl; 25 mM NaHCO3; BSA ( fatty acid free ) 7 . 2 mg/mL; 0 . 18 mg/mL of Type IV Collagenase ( Worthington Biochem Catalog: LS004188 ) . BSA and collagenase were added immediately before use . Primary hepatocytes derived from stim1fl/fl and stim1ΔLIVER animals kept on high fat diet were isolated and incubated in fresh William’s E medium with 5% fetal bovine serum ( FBS ) for 4 hr and thereafter , incubated in 0 . 1% FBS overnight . Next day , cells were washed twice and incubated with DMEM without glucose ( Invitrogen ) supplemented with 10 mM HEPES in the presence of 2 mM pyruvate , 2 mM glutamate and 20 mM glycerol for 4 hr . The glucose concentrations in the media were determined by Amplex Red glucose/glucose oxidase assay system ( Invitrogen ) . Liver tissues were homogenized in cold lysis buffer containing 50 mM Tris-HCl ( pH 7 . 4 ) , 2 mM EGTA , 5 mM EDTA , 30 mM NaF , 10 mM Na3VO4 , 10 mM Na4P2O7 , 40 mM glycerophosphate , 1 % NP-40 , and 1% protease inhibitor cocktail . After ~20 min incubation under mild agitation in the cold , the homogenates were centrifuged for 15 min at 9000 rpm to pellet cell debris , and protein concentrations were determined by BCA . Finally , the samples were diluted in 5x Laemmli buffer and boiled for 5 min . The protein lysates were subjected to SDS-polyacrylamide gel electrophoresis , as previously described ( Arruda et al . , 2014; Fu et al . , 2012 ) . The antibodies used in this study are listed in Table 2 . All the immunoblots were incubated with primary antibody overnight at 4°C , followed by incubation with secondary antibody conjugated with horseradish peroxidase ( Amersham Biosciences ) for 1–3 hr at room temperature . Individual membranes were visualized using the enhanced chemiluminescence system ( Roche Diagnostics ) . Immunoprecipitation ( IP ) : For the immunoprecipitation studies , primary antibodies against STIM1 and O-GlcNAc were cross-linked on to protein-A magnetic beads with Dimethylpimelimidate ( DMP ) . Cells lysates were prepared in a IP lysis buffer containing 50 mM Tris-HCl ( pH 7 . 4 ) , 1 mM EDTA , 150 mM NaCl , 10 mM Na3VO4 , 1 % NP-40 , and 1% protease inhibitor cocktail and 10 µM of PugNac . Protein content was adjusted to be the same in all samples and 300–400 µg of protein were incubated with the beads overnight . The next day , samples were washed 3x with IP buffer and eluted in Laemmli buffer and boiled for 5 min . Immunoblotting was performed as described above . For the IP of Flag tagged STIM1 , the lysates were incubated with anti-flag M2 magnetic beads from Sigma and eluted by competition with 3x flag peptide . Tissues in TRIzol ( Invitrogen ) were disrupted using TissueLyser ( Qiagen ) . To obtain RNA , trizol homogenates were mixed with chloroform vortexed thoroughly and centrifuged 12000 g for 20 min at 4°C . The top layer was transferred to another tube and mixed with isopropanol and centrifuged again at 12000 g for 20 min at 4°C . The RNA found in the precipitate was washed twice with 70% Ethanol and diluted in RNAse free water . Complementary DNA was synthesized using iScript RT Supermix kit ( Biorad ) . Quantitative real-time PCR reactions were performed in duplicates or triplicates on a ViiA7 system ( Applied Biosystems ) using SYBR green and custom primer sets or primer sets based on Harvard primer bank . Gene of interest cycle thresholds ( Cts ) were normalized to 18S ribosomal RNA ( Rn18s ) house keeper levels by the ΔΔCt . The primers used for qPCR are listed in Table 1 . For thapsigargin ( Tg ) stimulation , primary hepatocytes or Hepa1-6 cells were seeded on 3 . 5 mm round glass dishes ( 1 . 5 mm ) in Williams Medium in the presence of 5% CCS overnight at 37°C , 5% CO2 . The following morning , cells were washed , and in the afternoon they were incubated for a few minutes in Ca2+-free medium ( described in the Ca2+ imaging section ) followed by 1 µM Tg treatment ( or vehicle , DMSO ) for the duration indicated in the figure legends . Cells were fixed with 4% paraformaldehyde for 10 min at room temperature ( RT ) and washed 3x in PBS , before a 20 min permeabilization using 0 . 2% Triton-X100 in 2% BSA at RT . Primary antibodies were diluted 1:200 for STIM1 ( Cell Signaling #4916 ) , 1:100 for STIM2 ( Cell Signaling #4917 ) , for Na+/K+ ATPase α−1 -Alexa Fluor 488 ( Sigma #16–243 ) and for Anti-KDEL antibody , Alexa fluor 488 ( Abcam , ab184819 ) in PBS and the cells were incubated in this solution over night at 4°C in the dark . The next day , cells were washed 3x , including one long wash for more than 10 min . Secondary antibody was diluted 1:2500 in PBS , and the cells were incubated with it at RT for 1 hr in the dark . The cells were washed 3x , including one long wash , and if needed , Hoechst was used as nuclear marker , diluted 1:1000 in PBS for 10 min at RT . For the experiments described in Figure 1—figure supplement 1D , Figure 1—figure supplement 2A , B cells were imaged at the Confocal and Light Microscopy Core Facility at the Dana Farber Cancer Institute with a Yokogawa CSU-X1 spinning disk confocal system ( Andor Technology , South Windsor , CT ) with a Nikon Ti-E inverted microscope ( Nikon Instruments , Melville , NY ) , using a 60x or 100x Plan Apo objective lens with a Hamamatsu OrcaER camera . Andor iQ software ( Andor Technology , South Windsor , CT ) was used for acquisition parameters , shutters , filter positions and focus control . For Figures 1C , D , 3G and 5D , cells were imaged at the Harvard Medical School Imaging Core with a Yokogawa CSU-X1 spinning disk confocal system ( Andor Technology , South Windsor , CT ) with an iXon EMCCD camera and Metamorph was the software used for acquisition parameters , shutters , filter positions and focus control . For Figure 2H cells were image with a Yokogawa CSU-X1 spinning disk confocal system ( Andor Technology , South Windsor , CT ) with a Nikon Ti-E inverted microscope ( Nikon Instruments , Melville , NY ) , using a 60x or 100x Plan Apo objective lens with Zyla cMOS camera and NIS elements software was used for acquisition parameters , shutters , filter positions and focus control . Image analysis was performed using Fiji software . All the images in the same experiment were analyzed using the same parameters . For Figure 1E , puncta number was counted using a custom made macro . After background subtraction of 15 , images were subject to ‘Filter Gaussian blur’ of 1 . 00 and a threshold was set to define puncta from background signal . In order to localize the individual puncta ‘Find Maxima’ tool was used and puncta were counted using ‘Analyze Particles’ tool . For Figure 2J , images were background subtracted and single cells or groups of cells either expressing OGT or not ( identified by RFP expression ) , were cropped from the original image . Puncta above a threshold of 50 were counted using ‘Analyze Particles’ , and normalized to number of cells in the field . Puncta pixel intensities: For Figure 1F and Figure 1—figure supplement 2B , a 125 × 125 pixels box was placed in a representative region of the cell and pixel intensities were recorded in the entire box using a custom made macro . Intensities were distributed into a histogram with R , and plotted in Prism . No further normalization was performed , as the same number of pixels was evaluated in all the images . Plot profile in Figures 1G , 2L and 5E: A thin box of 40 × 250 pixels was placed across the cell and the average intensity contained in the box was plotted as a profile with Fiji . Edge/cyto measurements: for Figures 1H , 2N , 5D , Figure 1—figure supplement 2B: A small box was hand-drawn at a region of the edge of the cell ( at the PM ) , and the same size box was placed right next to the edge , in the cytoplasmic region . The mean STIM signal across each box was recorded , and the ratio between the reading at the edge of the cell and in the cytoplasm was derived after background signal subtraction . Such pairs of boxes were drawn in four representative places in each cell , from 2 to 5 cells per field . Areas were chosen only on cell membranes containing neighbor cells . For the TIRF images , cells were stained as described above using anti-STIM1 antibody and anti- Na+/K+ ATPase α−1 , Alexa Fluor 488 antibody to stain the PM . In each experiment epifluorescence images were acquired and the PM marker was used to define the PM area for the TIRF imaging . The images were acquired a Zeiss Elyra PS . 1 microscope with a 60 or 100 × 1 . 46 N . A . oil immersion TIRF objective ( Carl Zeiss GmbH , Germany ) in TIRF mode with an Andor EM-CCD camera ( 512 × 512 pixels ) at the Harvard center for Biological Imaging , Cambridge , MA . Cells were loaded with 4 μM Fura-2AM and 1 µM Pluronic F-127 in HBSS for ~60 min at room temperature . Before imaging , the cells were washed and kept in a medium containing 10 mM Hepes , 150 mM NaCl , 4 mM KCl , 2 mM CaCl2 , 1 mM MgCl2 , 10 mM D-glucose , pH 7 . 4 . Ca2+-free medium was prepared similarly to the buffer described above , in the absence of CaCl2 and in the presence of 2 mM EGTA and 3 mM MgCl2 . Cells were stimulated with 1 µM Tg for indirect ER Ca2+ measurements , and in some cases supplemented with 100 µM 2-APB for SOCE inhibition . Ratiometric Fura-2AM imaging was performed by alternatively illuminated with 340 and 380 nm light for 250 ms ( Lambda DG-4; Sutter Instrument Co . ) , using an Olympus IX70 with 40x objective . Emission light >510 nm was captured using a CCD camera ( Orca-ER; Hamamatsu Photonics ) . Both channels were collected every 5 s , background corrected and analyzed with Slidebook and custom-made R-scripts . For the Fluo-4 experiments described in Figure 2J , the same procedure described above as used except that the cells were incubate with Fluo-4 for 30 min to 1 hr . The images were acquired on a Leica SP8X and LAS X software was used for acquisition parameters , shutters , filter positions and focus control . Loss of function experiments were performed in Hepa1-6 cells ( a mouse hepatocyte cell line from ATCC , derivative of the BW7756 mouse hepatoma that arose in C57/L mouse ) for STIM1 , STIM2 and ORAI1 . This cell line was negative for mycoplasma testing . Five different shRNA sequences targeting different regions of each gene were obtained from MISSION shRNA Library ( Sigma ) along with the Scramble shRNA control . Stable cell lines were generated by lentiviral infection of Hepa1-6 cells overnight in media ( DMEM and 10% CCS and polybrene ) . After infection cells were incubated in growth media ( DMEM and 10% CCS ) and selected with 3 µg/µl puromycine . The knockdown efficiency of each shRNA was evaluated by western blot analysis and the experiments were done with the sequence that generated the highest efficiency . Cells were frozen at passage two after selection and fresh cells were thawed for each experiment . The sequences of the shRNA used in the study are listed in Table 1 . For the overexpression of OGT , a hOGT/RFP co-expression plasmid was generated using a self-cleaving P2A peptide linker . RFP was first cloned into pcDNA6 Myc/His B ( Life Technologies V22120 ) at BamHI/XhoI sites . OGT-P2A was generated by PCR ( F: 5’-CCC GGT ACC GCC ACC ATG GCGTCTTCCGTGGGCAA; R: 5’- CCC GGA TCC AGG TCC AGG GTT CTC CTC CAC GTC TCC AGC CTG CTT CAG CAG GCT GAA GTT AGT AGC TCC GCT TCC TGCTGACTCAGTGACTTC ) and cloned into the above vector at KpnI/BamHI sites . The OGT-RFP plasmid was transfected into , cells using lipofectamine 2000 overnight in OptiMEM media and the experiment done after 36–48 hr after transfection . For the STIM1-flag pull downs , 3xFLAG-mSTIM1 was generated by PCR ( F: 5’- CCC GGATCC GCC ACC ATG GAC TAC AAA GAC CAT GAC GGT GAT TAT AAA GAT CAT GAC ATC GAT TAC AAG GAT GAC GAT GAC AAG GATGTGTGCGCCCGTCTTGCCCTGT; R: 5’- CCC GCGGCCGC TTA CTA CTTCTTAAGAGGCTTCTTAA ) and cloned into pENTR1A no CCDB vector ( Addgene #17398 ) at BamHI/NotI sites and then transferred to pLenti CMV Puro DEST ( w118-1 ) ( Addgene # 17452 ) using Gateway LR Clonase II Enzyme Mix ( Life Technologies 11791 ) . STIM1-flag was transfected in HEK293T cells and after 24–36 hours cells were trated with 3 ug/mL of puromycin for selection and a stable cell line was generated . For insulin signaling studies , cells were serum deprived for 4 hr before stimulus with 3 nM of insulin for 3 and 6 min . Cells were washed in ice-cold PBS and snap-frozen in liquid nitrogen . For the insulin infusions in vivo , following 6 hr of food withdrawal , mice were anaesthetized with an intraperitoneal injection of 2 mg/ml xylazine and 2 mg/ml ketamine , and insulin ( 0 . 75 IU kg−1 ) or phosphate buffered saline ( PBS ) in 200 µl volume was infused into the portal vein . Three minutes after infusion , tissues were removed and frozen in liquid nitrogen and kept at –80°C until processing . For lipid loading experiments primary hepatocytes were isolated in the morning and kept in Williams E medium . After the cells were settled , Williams medium containing 1 mM Oleic acid and 40 µM palmitic acid was added to the culture . Cells before and after 16 hr of lipid loading and stained with BODIPY and Hoechst to visualize lipid droplets and nucleus , respectively . For TEM , primary hepatocytes were fixed in a 1:1 ratio in a fixative buffer and Williams Medium . The fixative buffer contained: 4 parts of FP stock ( 2 . 5% PFA , 0 . 06% picric acid in 0 . 2M Sodium Cacodylate buffer pH 7 . 4 ) and 1 part of 25% glutaraldehyde for at least 2 hr . The samples were then placed in propyleneoxide for 1 hr and infiltrated in a 1:1 mixture of propyleneoxide and TAAB 812 Resin mixture ( Marivac Canada Inc . St . Laurent , Canada ) . Sectioning and imaging: ultrathin sections ( about 90 nm ) were generated using a Reichert Ultracut-S microtome , JEOL microscope . Images were recorded with an AMT 2 k CCD camera . For the STIM-HRP labeled cells , primary hepatocytes were infected with adenovirus expressing STIM1-HRP . The construct was kindly provided by Dr . Lewis from Stanford University and cloned in an adenovirus ( serotype 5 , Ad5 ) . Primary hepatocytes were fixed in a buffer containing 2% glutaraldehyde in 0 . 1 M sodium cacodylate at room temperature for 1 hr , then quickly moved to ice and washed with 0 . 1 M sodium cacodylate buffer for 10 min . HRP was visualized with 0 . 5 mg ml diaminobenzidine and 0 . 03% hydrogen peroxide in 0 . 1 M sodium cacodylate buffer for 5 min-20min . The reaction was stopped by five washes with cold 0 . 1 M sodium cacodylate buffer .	Obesity is a chronic metabolic disorder . Some people’s genetics make them more vulnerable to the condition , and it is generally caused by eating too much and moving too little . The resulting surplus of nutrients affects the cells and organs of the body in several adverse ways . For example , excessive nutrients impair a compartment within cells called the endoplasmic reticulum . This compartment is where many proteins and fats are made and transported . It is also the site for a lot of metabolic processes , and the main place in the cell where calcium ions are stored . Many proteins need calcium ions to work properly , including metabolic enzymes . In obesity , the endoplasmic reticulum becomes less able to store calcium ions . A protein called STIM1 senses and regulates the levels of calcium ions in the endoplasmic reticulum . When calcium levels drop , STIM1 moves along the endoplasmic reticulum membrane towards the part that is next to the cell surface . Here , STIM1 joins up with a calcium channel called Orai1 . The STIM1-Orai1 complex allows calcium ions to enter the cell and replenish its levels in the endoplasmic reticulum . Arruda , Pers et al . have now asked if STIM1 is altered in obesity and , if so , whether it contributes to the endoplasmic reticulum’s inability to maintain proper calcium levels . High-resolution microscopy and biochemical techniques confirmed that STIM1 is indeed compromised in liver cells from obese mice . In these cells , STIM1 was found in unusual small clusters . It also could not move along the endoplasmic reticulum membrane when calcium levels dropped . As a result of these navigational errors , STIM1 failed to couple with Orai1 , meaning less calcium could enter the cell . Further work identified that a small sugar molecule that is added onto STIM1 in obesity is behind its reduced ability to move accurately . Arruda , Pers et al . next created mice that lacked STIM1 in their liver . These mice showed signs of metabolic abnormalities . Notably , when STIM1 levels were experimentally increased in obese mice , it restored calcium levels in the endoplasmic reticulum closer to normal , and improved metabolism too . Thus , regulating calcium levels in the endoplasmic reticulum via proteins such as STIM1 is essential for maintaining a healthy metabolism . Interventions to correct calcium levels may have therapeutic promise to combat metabolic diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2017	Defective STIM-mediated store operated Ca2+ entry in hepatocytes leads to metabolic dysfunction in obesity
Red blood cell ( RBC ) invasion by malaria merozoites involves formation of a parasitophorous vacuole into which the parasite moves . The vacuole membrane seals and pinches off behind the parasite through an unknown mechanism , enclosing the parasite within the RBC . During invasion , several parasite surface proteins are shed by a membrane-bound protease called SUB2 . Here we show that genetic depletion of SUB2 abolishes shedding of a range of parasite proteins , identifying previously unrecognized SUB2 substrates . Interaction of SUB2-null merozoites with RBCs leads to either abortive invasion with rapid RBC lysis , or successful entry but developmental arrest . Selective failure to shed the most abundant SUB2 substrate , MSP1 , reduces intracellular replication , whilst conditional ablation of the substrate AMA1 produces host RBC lysis . We conclude that SUB2 activity is critical for host RBC membrane sealing following parasite internalisation and for correct functioning of merozoite surface proteins . The phylum Apicomplexa comprises a diverse group of protozoan organisms , many of which are obligate intracellular parasites of clinical or veterinary importance . A feature of these parasites is their possession of invasive forms that actively penetrate host cells . In the asexual blood stages of infection by malaria parasites ( Plasmodium species ) , merozoites invade red blood cells ( RBCs ) , replicating intracellularly to produce merozoites that egress to invade fresh RBCs . Invasion is a rapid , multi-step process that includes binding , merozoite reorientation , discharge of secretory organelles called rhoptries and micronemes , formation of an electron-dense ‘tight junction’ ( TJ ) between the merozoite apical end and the RBC membrane , and actinomyosin-powered entry into the RBC through this structure , with concurrent formation of a membrane-bound parasitophorous vacuole ( PV ) within which the parasite develops ( Aikawa et al . , 1978; Dvorak et al . , 1975; Weiss et al . , 2015; Bannister et al . , 1975 ) . Invasion ends with sealing of the RBC behind the intracellular parasite , concomitant with pinching off of the nascent PV membrane ( PVM ) in a membrane scission event such that the PVM is eventually non-contiguous with and internal to the RBC membrane . Invasion is typically followed by transformation of the host RBC into a shrunken ‘spiky’ state called echinocytosis , which typically resolves within minutes . The parasite rapidly transforms into an amoeboid ‘ring’ form . Whilst many advances have been made over recent decades in understanding invasion , particularly in the most lethal malaria species Plasmodium falciparum and Plasmodium knowlesi , as well as in the related apicomplexan parasite Toxoplasma gondii , the mechanisms underlying PV formation are obscure and nothing is known of the molecular mechanism ( s ) responsible for host cell membrane sealing . Early electron microscopic ( EM ) studies showed that invading merozoites shed a ‘fuzzy coat’ of ~20 nm-long bristle-like fibres ( Aikawa et al . , 1978; Bannister et al . , 1975 ) . These are likely composed predominantly of the major glycosyl phosphatidylinositol ( GPI ) -anchored surface protein MSP1 , which is synthesised as an abundant , ~200 kDa precursor at the plasma membrane of developing parasites ( Holder et al . , 1992 ) . Minutes before egress , MSP1 is proteolytically cleaved into several fragments by a parasite protease called SUB1 ( Koussis et al . , 2009; Yeoh et al . , 2007 ) . These fragments form a noncovalent complex on the merozoite surface ( McBride and Heidrich , 1987 ) where , together with several partner proteins , it facilitates membrane rupture at egress ( Das et al . , 2015 ) . During invasion the bulk of the MSP1 complex is shed from the merozoite ( Riglar et al . , 2011; Blackman et al . , 1996; Blackman et al . , 1990; Blackman et al . , 1991 ) as a result of a single further cleavage catalysed by a second , membrane-bound protease called SUB2 ( Harris et al . , 2005; Hackett et al . , 1999; Barale et al . , 1999 ) . SUB2 also mediates shedding of two merozoite surface integral membrane proteins called PTRAMP and AMA1 , which are released from micronemes at around egress ( Howell et al . , 2003; Green et al . , 2006; Siddiqui et al . , 2013; Thompson et al . , 2004 ) . PTRAMP acts as an RBC binding ligand ( Siddiqui et al . , 2013 ) whilst AMA1 plays a central role in TJ formation ( reviewed in Harvey et al . , 2014 ) . To access its substrates , SUB2 is also released from micronemes onto the merozoite surface , where it translocates to the posterior pole of the parasite just before or during invasion ( Riglar et al . , 2011; Harris et al . , 2005 ) . Despite evidence that SUB2 is essential ( Uzureau et al . , 2004; Zhang et al . , 2018; Bushell et al . , 2017 ) and that shedding of MSP1 and AMA1 is important and can aid evasion of invasion-inhibitory antibodies ( Blackman et al . , 1990; Olivieri et al . , 2011; Guevara Patiño et al . , 1997; Lazarou et al . , 2009 ) , the molecular function of SUB2-mediated shedding is unknown . Here , we used genetic modification of P . falciparum SUB2 and two of its substrates to examine the essentiality and function of SUB2 in the erythrocytic lifecycle . We show that SUB2 depletion results in defects in merozoite surface protein shedding and sealing of the host RBC upon invasion , leading to either abortive invasion with loss of host RBC haemoglobin , or developmental arrest of the intracellular parasite . Our findings highlight SUB2 as a key mediator of parasite viability and host RBC membrane integrity . The P . falciparum sub2 gene encodes a type I integral membrane protein with a large ectodomain incorporating a subtilisin-like protease module ( Hackett et al . , 1999; Barale et al . , 1999 ) . We designed a construct to integrate into the sub2 locus by homologous recombination , producing a modified locus in which the entire second exon encoding the crucial catalytic Ser961 , transmembrane domain ( TMD ) and cytoplasmic domain was flanked ( floxed ) by loxP sites ( Figure 1A ) . Integration also fused a triple hemagglutinin ( HA3 ) epitope tag to the SUB2 C terminus , as achieved previously ( Riglar et al . , 2011; Harris et al . , 2005 ) . Transfection of the construct into the DiCre-expressing P . falciparum 1G5DC clone ( Collins et al . , 2013a ) resulted in outgrowth of WR99210-resistant parasites expressing HA3-tagged SUB2 . Immunofluorescence analysis ( IFA ) of two integrant clones of the modified parasite line ( called SUB2HA3:loxP ) with anti-HA3 antibodies showed a signal consistent with the previously-determined location of SUB2 in micronemes , confirming correct gene modification ( Figure 1B ) . Excision of the floxed sequence was predicted to produce a truncated gene product lacking a functional catalytic domain , TMD , and HA3 tag . To assess the efficiency of DiCre-mediated gene disruption , synchronised rings of both integrant SUB2HA3:loxP clones were pulse-treated with rapamycin ( RAP ) which induces Cre recombinase activity . The parasites were examined ~44 hr later at the multinucleated schizont stage ( when SUB2 expression is maximal ) within the erythrocytic cycle of RAP treatment ( cycle 0 ) . Diagnostic PCR showed efficient excision of the floxed sub2 sequence ( Figure 1C ) , whilst western blot confirmed virtually complete loss of the HA3 signal ( Figure 1D ) and no signal was detectable by IFA in most of the RAP-treated schizonts ( Figure 1E ) . No defect in merozoite biogenesis was evident ( Figure 1—figure supplement 1 ) . These results confirmed disruption of SUB2 expression within a single erythrocytic cycle , and suggested that loss of SUB2 during intracellular development had no impact on schizont maturation . SUB2 is discharged onto the free merozoite surface to cleave its substrates ( Riglar et al . , 2011; Harris et al . , 2005; Howell et al . , 2003; Green et al . , 2006 ) . We expected egress to be unaffected by loss of SUB2 , and in accord with this time-lapse video microscopy showed no differences in rupture of mock- and RAP-treated SUB2HA3:loxP cycle 0 schizonts ( Figure 2A ) . To assess the invasive capacity of the released merozoites , highly synchronised schizonts were incubated for 4 hr with RBCs to allow egress and invasion . Newly-invaded ( cycle 1 ) rings were produced in the ΔSUB2 cultures , but at levels only ~50% of those in control cultures ( Figure 2B ) . Western blot analysis with selective antibodies of medium harvested from the ΔSUB2 cultures showed substantially decreased levels of MSP1 , AMA1 , and PTRAMP , consistent with loss of shedding ( Figure 2C , D ) . SUB2-mediated shedding during invasion is highly efficient , so antibodies specific to shed segments of MSP1 and AMA1 are invariably unreactive with newly-invaded rings ( Blackman et al . , 1996; Blackman et al . , 1990; Blackman et al . , 1991; Howell et al . , 2005 ) . In contrast , the membrane-proximal AMA1 ‘stub’ and GPI-anchored MSP119 domain that remain on the parasite surface following cleavage at the respective juxtamembrane sites are readily detected in rings ( Blackman et al . , 1990; Blackman et al . , 1991; Howell et al . , 2003; Olivieri et al . , 2011 ) . We therefore used selected antibodies to examine the cycle one rings by IFA . Monoclonal antibody ( mAb ) X509 , which recognises a shed fragment of the MSP1 complex , did not react with control rings , as expected . In contrast , the mAb strongly recognised the ΔSUB2 cycle one rings ( Figure 2E ) , indicating that in the absence of SUB2 , unshed MSP1 complex was carried into RBCs on invading merozoites . Similarly , antibodies to the AMA1 ectodomain showed stronger reactivity with ΔSUB2 rings than with control rings ( Figure 2F ) . To interrogate the global effects of SUB2 disruption , culture media harvested following rupture of control and ΔSUB2 schizonts in the presence of fresh RBCs were analysed by quantitative mass spectrometry . This revealed reduced levels of several merozoite surface proteins in the ΔSUB2 egress supernatants , consistent with their reduced shedding ( Figure 3 and Figure 3—source data 1 ) . The depleted proteins included SUB2 itself , as well as AMA1 , PTRAMP and members of the MSP1 complex ( MSP1 , 3 , 6 and 7 ) , in accord with the western blot and IFA data . Unexpectedly , the most depleted proteins also included the GPI-anchored merozoite surface proteins Pf92 , MSP2 , MSP4 and MSP5 ( Gilson et al . , 2006 ) as well as the MSP7-like protein MSRP2 , none of which are components of the MSP1 complex ( Ranjan et al . , 2011; Lin et al . , 2016; Stafford et al . , 1994; Trucco et al . , 2001; Pachebat et al . , 2001 ) . These results suggest that SUB2 is required to shed these proteins too , indicating a broader repertoire of merozoite surface substrates than previously appreciated . Collectively , these results confirmed SUB2 as the enzyme responsible for shedding of AMA1 , PTRAMP and the MSP1 complex as well as several other merozoite surface proteins , and revealed an important role for SUB2 in invasion . The findings also proved that complete shedding of merozoite surface MSP1 and AMA1 is not a prerequisite for invasion as the ΔSUB2 parasites formed rings , albeit with reduced efficiency . To explore the long-term consequences of SUB2 loss , ring-stage SUB2HA3:loxP parasites were mock- or RAP-treated , dispensed at low density into flat-bottomed microwell plates , and cultured undisturbed . Quantitation of the plaques appearing in the wells ( resulting from localised zones of RBC destruction; Thomas et al . , 2016 ) revealed that RAP-treated parasites produced significantly fewer plaques than controls ( Figure 4A , B ) . Genotyping of parasites expanded from 2 of the few plaques in the RAP-treated cultures showed that they were derived from the small fraction of parasites that failed to undergo excision upon RAP treatment ( Figure 4B and Figure 1E ) . This suggested that correctly excised parasites lacking an intact sub2 locus failed to replicate . To test this , and to establish whether the growth defect in the RAP-treated population was solely due to loss of the sub2 gene , SUB2HA3:loxP parasites were transfected with plasmid pSUB2-BSD designed for ectopic expression of a full-length synthetic SUB2 gene called sub2synth ( Child et al . , 2013 ) . The plasmid included a blasticidin ( BSD ) resistance marker and an mCherry expression cassette ( Figure 4C ) . Parallel SUB2HA3:loxP cultures were transfected instead with a control plasmid ( pDC2-mCherry-MCS ) lacking the sub2synth expression cassette . Following BSD selection , both lines were RAP-treated to disrupt the genomic sub2 locus , then examined by plaque assay . This showed that carriage of the sub2synth transgene , but not the control plasmid , compensated for loss of the genomic sub2 gene ( Figure 4D ) . Parasites from single plaques were then expanded in BSD-containing medium and the resulting clones ( all mCherry positive ) inspected by diagnostic PCR for excision of the floxed chromosomal sub2 sequence . Whilst none of 25 viable parasite clones harbouring the control plasmid had undergone excision , 2 of 5 selected clones transfected with pSUB2-BSD had undergone excision and so lacked a functional chromosomal sub2 gene ( Figure 4E ) . Episomal plasmids segregate inefficiently in Plasmodium ( O'Donnell et al . , 2001; O'Donnell et al . , 2002; van Dijk et al . , 1997 ) so are usually quickly lost upon removal of drug selection . However , upon further culture of the excised , pSUB2-BSD-harbouring parasites in the absence of BSD , mCherry expression was uniformly maintained ( Figure 4F , top right ) indicating that survival was dependent upon maintenance of the pSUB2-BSD expression plasmid . In contrast , the non-excised clones harbouring the control plasmid began to lose mCherry expression in the absence of BSD ( Figure 4F , top left image ) . These results showed that parasite viability requires a functional sub2 gene . To understand the loss of viability in ΔSUB2 parasites , we returned to examine the fate of those ΔSUB2 parasites that successfully invaded at the end of cycle 0 . Microscopy and flow cytometry revealed an arrest in intracellular development of the cycle one parasites , which - despite appearing initially normal - failed to mature and showed no signs of haemozoin production ( a byproduct of haemoglobin digestion ) ( Figure 4G and Figure 4—figure supplement 1 ) . Examination by transmission and serial block-face scanning EM identified no structural defects in the ΔSUB2 cycle one rings ( Figure 4—figure supplement 1 , Figure 4—video 1 and Figure 4—video 2 ) . We concluded that the lethal phenotype associated with loss of SUB2 arose from a reduced capacity to invade and a developmental arrest in those parasites that did invade . During invasion assays involving ΔSUB2 parasites , we noticed that the culture media were unusually red in colour , suggesting a high free haemoglobin ( Hb ) content . To explore this , mature cycle 0 schizonts of DMSO- or RAP-treated SUB2HA3:loxP parasites were incubated with fresh RBCs for 4 hr to allow egress and invasion . Levels of Hb in culture supernatants were then quantified by spectrophotometry and SDS-PAGE , and the cells examined by microscopy and flow cytometry ( Figure 5A ) . As previously noted , ring production from the ΔSUB2 schizonts was reduced compared to controls . Despite this , higher levels of extracellular Hb appeared in the ΔSUB2 culture supernatants ( Figure 5B ) . This did not derive from schizont rupture per se , as incubation of similar numbers of schizonts without addition of RBCs resulted in low levels of Hb release that did not differ between ΔSUB2 and control schizonts . Moreover , Hb release required extensive interaction between released merozoites and the host cells , since it did not occur in the presence of cytochalasin D ( cytD ) , an actin-binding drug that blocks invasion downstream of TJ formation by disrupting the parasite actinomyosin motor that drives invasion ( Miller et al . , 1979; Figure 5C and Figure 5—figure supplement 1 ) . We conjectured that those ΔSUB2 merozoites unable to invade instead interacted with target RBCs in an abortive manner that led to lysis . To investigate this , following co-incubation of schizonts with RBCs , cultures were supplemented with fluorescent phalloidin , a membrane-impermeable peptide that binds the F-actin of the RBC cytoskeleton ( Atkinson et al . , 1982; Glushakova et al . , 2010 ) , then immediately examined by live fluorescence microscopy . This revealed rounded , phalloidin-labelled cells that were ~6 fold more abundant in the ΔSUB2 cultures ( Figure 5D ) . The dimensions of these cells , their low buoyant density ( shown by a tendency to ‘float’ above intact RBCs ) and their accessibility to phalloidin , suggested that they were ‘ghosts’ derived from lysis of RBCs . We concluded that these derived from abortive invasion attempts in which interaction with ΔSUB2 merozoites had induced RBC lysis . To directly visualise the fate of target RBCs , we examined the process by live microscopy ( Figure 5E and Figure 5—video 1 ) . Initial interactions with ΔSUB2 merozoites appeared normal , with RBC deformation often followed by parasite entry and echinocytosis . Subsequently however , and in contrast to the behaviour typical of RBCs invaded by control merozoites where echinocytosis quickly resolved , RBCs invaded by ΔSUB2 merozoites often remained rounded and then lysed , as indicated by loss of differential interference contrast ( DIC ) . Lysis often occurred within ~14 min of the initial interaction and was sometimes accompanied by ejection of the merozoite . In other cases lysis took longer and was visualised by images captured over 30–60 min . However , washing and re-culture of the cycle one rings forms resulted in no further Hb release , indicating that lysis was usually rapid . Attempts to fix ΔSUB2 merozoites in the act of invasion for analysis by EM were unsuccessful , possibly due to its transient nature . Nonetheless we concluded that loss of SUB2 led in ~50% of invasion attempts to abortive invasion that culminated in rapid RBC lysis , presumably due to a defect in RBC sealing . To dissect the ΔSUB2 phenotype , we investigated the effects of mutations that directly prevent SUB2-mediated shedding of the most abundant SUB2 substrate , the MSP1 complex . We previously mapped the SUB2 cleavage site in MSP1 to the Leu1606-Asn1607 bond just upstream of the C-terminal MSP119 domain ( Blackman et al . , 1990 ) ( see also Figure 2C ) and showed that proline substitutions of the residues flanking the AMA1 SUB2 cleavage site blocks cleavage ( Olivieri et al . , 2011 ) . Based on this , transgenic parasite line iMSP1PP was generated in which DiCre-mediated excision of a floxed segment of the msp1 gene produced a partial allelic replacement , substituting the Leu-Asn cleavage motif with a Pro-Pro motif ( Figure 6A and Figure 6—figure supplement 1 ) . In a control parasite line , iMSP1LN , excision reconstituted the wild type motif . RAP-treatment of iMSP1PP and iMSP1LN parasite clones produced the expected genomic excision events ( Figure 6—figure supplement 1 ) . To examine the effects of the Leu-Asn to Pro-Pro substitution , synchronous rings were RAP- or mock-treated , matured to schizont stage , then allowed to undergo egress in the presence of RBCs . This showed no significant effects on efficiency of ring formation ( Figure 6B ) , although examination of culture supernatants following invasion showed a selective reduction in MSP1 shedding in RAP-treated iMSP1PP cultures , consistent with the predicted effects of the mutations on SUB2-mediated cleavage ( Figure 6C ) . Cycle one rings from RAP-treated iMSP1PP schizonts were strongly recognised by mAb X509 , similar to ΔSUB2 cycle one rings ( Figure 6D ) , indicating successful invasion despite reduced MSP1 shedding . The newly-invaded cycle one mutant iMSP1PP rings appeared morphologically normal , but further monitoring revealed extensive retarded intracellular development ( Figure 6E ) , and longer-term experiments confirmed a replication defect in the RAP-treated iMSP1PP parasites ( Figure 6F ) . We concluded that selective inhibition of MSP1 shedding at invasion did not affect RBC entry but led to defective intracellular parasite development similar to that although not as severe as in the ΔSUB2 mutants . The MSP1 mutagenesis data provided a plausible explanation for the impact of SUB2 loss on intracellular parasite growth , but did not explain the invasion phenotype of ΔSUB2 merozoites nor their capacity to lyse RBCs . Prior to this work , the only other experimentally-demonstrated essential SUB2 substrate was the microneme protein AMA1 . The AMA1 ectodomain comprises three globular domains linked to the TMD via a short juxtamembrane segment ( Pizarro et al . , 2005 ) . Cleavage of P . falciparum AMA1 by SUB2 occurs at the Thr517-Ser518 bond 29 residues upstream of the TMD ( Figure 2C ) , releasing domains I-III , leaving a juxtamembrane ‘stub’ bound to the merozoite surface ( Howell et al . , 2003 ) . AMA1 can additionally be shed by the parasite rhomboid protease ROM4 via cleavage within the TMD at the Ala550-Ser551 bond ( Howell et al . , 2003; Howell et al . , 2005 ) ; this can be blocked by a Tyr substitution of Ala550 , a mutation tolerated by the parasite ( Olivieri et al . , 2011 ) . To probe the biological significance of SUB2-mediated shedding of AMA1 , we conditionally modified both the SUB2 and ROM4 cleavage sites to simultaneously render them refractory to cleavage . For this , we generated transgenic parasite line iΔRΔS in which excision of a floxed ama1 gene simultaneously substitutes the Thr517-Ser518 site with a Pro-Pro motif and Ala550 with a Tyr residue ( Olivieri et al . , 2011; Figure 7 and Figure 7—figure supplement 1 ) . Replication of untreated iΔRΔS parasites was indistinguishable from parental parasites , and RAP treatment produced the expected genomic changes ( Figure 7—figure supplement 1 ) as well as the anticipated decrease in AMA1 shedding ( Figure 7A ) . However , we detected no change in the invasive capacity of the RAP-treated parasites and replication was unaffected ( Figure 7B ) . Given the absence of deleterious effects , we reasoned that direct mutagenesis of the AMA1 cleavage sites should generate viable parasites , so we used targeted homologous recombination to directly introduce the same cleavage site mutations into the ama1 gene ( Figure 7—figure supplement 1 ) . The resulting parasite line , dΔRΔS , displayed a phenotype similar to that of the RAP-treated iΔRΔS parasites , with reduced AMA1 shedding ( Figure 7A ) but no effects on invasion or replication ( Figure 7B ) . These results suggested that inhibition of AMA1 shedding per se was not responsible for the invasion and lysis phenotype observed in the ΔSUB2 mutants . Given the role of AMA1 in TJ formation , we examined whether the ΔSUB2 phenotype might in part reflect loss of AMA1 function . We generated a parasite line ( iΔAMA1 ) in which DiCre-mediated excision ablated AMA1 expression by severely truncating the gene ( Figure 7—figure supplement 2 ) . In a control line ( iAMA1_C ) , excision reconstituted a functional full-length gene ( Figure 7—figure supplement 2 ) . RAP treatment of the parasites produced the expected genomic modifications ( Figure 7—figure supplement 2 ) , with loss of AMA1 expression in mature cycle 0 schizonts of the iΔAMA1 clones ( Figure 7C , D ) . This led to complete loss of ring formation and proliferation following rupture of RAP-treated iΔAMA1 schizonts at the end of cycle 0 ( Figure 7E , F ) , confirming AMA1 essentiality . Despite the lack of invasion , levels of Hb released into the culture media were much higher in RAP-treated iΔAMA1 schizonts incubated with RBCs than in controls ( Figure 7G ) . To determine the source of the Hb , we used microscopy to visualise interactions between ΔAMA1 merozoites and RBCs . This confirmed the loss of invasion , whilst showing that interaction of ΔAMA1 merozoites with target RBCs produced extensive RBC echinocytosis ( Figure 7—video 1 ) . We did not observe rapid RBC lysis; however , target RBCs often transformed to a rounded form , remaining in this state for some time before losing Hb content ( Figure 7H ) . We concluded that loss of AMA1 not only prevented invasion but also produced host RBC lysis upon merozoite interaction , presumably due to an RBC sealing defect . We have provided the first genetic proof of SUB2 as the merozoite sheddase , and have shown that abolishing SUB2-dependent cleavage produces a complex phenotype that is lethal in at least two superficially distinct manners; merozoites that successfully complete invasion fail to develop , whilst other merozoites induce RBC lysis at or shortly following entry . Our most important conclusion is that SUB2-mediated protein shedding is essential for parasite survival . Our primary mechanistic explanation for this is that shedding is required for resealing of the RBC membrane at invasion ( Figure 8 ) . Using IFA , western blot and mass spectrometry to compare SUB2-expressing and ΔSUB2 parasite cultures , we confirmed that SUB2 is required for shedding of MSP1 , AMA1 and PTRAMP . We also identified several putative new SUB2 substrates not known to associate with any of the three previously known substrates . Some of the most prominent of these , including MSP2 , MSP4 , MSP5 and Pf92 , belong to a group of merozoite surface proteins known or predicted to possess GPI anchors ( Gilson et al . , 2006; Sanders et al . , 2005; Marshall et al . , 1998; Marshall et al . , 1997 ) . Since GPI-anchored proteins cannot be shed by rhomboid cleavage ( which occurs within TMDs ) , it is likely that shedding of each is independently catalysed by SUB2 . The other newly identified SUB2 substrate , MSRP2 , likely has no GPI anchor ( so could be peripherally associated with one of the GPI-anchored substrates ) but does undergo proteolytic cleavage around egress ( Kadekoppala et al . , 2010 ) . This wide range of SUB2 substrates was unexpected; indeed earlier work that did not benefit from access to mutants concluded that MSP2 and MSP4 are not shed at invasion ( Boyle et al . , 2014 ) . Previous extensive mutagenesis analysis of the AMA1 SUB2 cleavage site ( Olivieri et al . , 2011 ) indicated that SUB2 has a rather promiscuous substrate recognition profile , cleaving its membrane-bound substrates at a roughly similar distance from the membrane rather than at a specific amino acid sequence . Our new mass spectrometric evidence revealing a diverse range of additional putative SUB2 substrates supports that model and underlines the potential for multiple consequences of SUB2 inhibition . It was interesting to note that SUB2 depletion resulted in significantly upregulated levels in culture supernatants of a number of merozoite proteins , particularly prominent examples including a conserved protein of unknown function ( PF3D7_0520800 ) , an inner membrane complex protein ( PF3D7_1003600 ) , and the myosin A tail interacting protein PF3D7_1246400 ( MTIP ) , a component of the merozoite actinomyosin contractile system that drives invasion ( see Figure 3—source data 1 and the left-hand side of the plot in Figure 3 ) . Whilst we were initially puzzled by this , we suspect that these proteins might simply be derived from degradation of merozoites that failed to invade in the ΔSUB2 cultures; these merozoites would accumulate extracellularly and eventually disintegrate , likely leading to increased supernatant levels of merozoite proteins unrelated to the activity of SUB2 . Despite the reduction in merozoite surface protein shedding , ΔSUB2 parasites formed new rings at the end of cycle 0 , albeit at ~50% reduced efficiency . These uniformly arrested , and growth assays supported by genetic complementation proved that SUB2 is indispensable for parasite survival . To dissect the phenotype , we generated lines in which shedding of the most abundant SUB2 substrate , the MSP1 complex , was selectively inhibited . These mutants invaded with wild type efficiency but displayed a developmental defect similar to – if somewhat less severe than - that of the ΔSUB2 mutants . Given the technical difficulties in determining causality upon intracellular arrest in Plasmodium , we can only speculate on the underlying defect ( s ) . The MSP119 species remaining on the merozoite surface following SUB2-mediated shedding has been shown to be a highly stable constituent of the post-invasion membrane rearrangements involved in formation of the acidic digestive vacuole ( DV ) that is the site of haemoglobin catabolism during intraerythrocytic development ( Dluzewski et al . , 2008 ) . Formation of the DV appears to arise through endocytosis of plasma membrane constituents ( including MSP119 ) , forming a number of small endocytic vesicles that eventually fuse to form the single DV . As a result , MSP119 is the earliest known marker for the DV in the developing ring ( Dluzewski et al . , 2008 ) . We speculate that in the absence of SUB2-mediated shedding , the bulky MSP1 complex remaining on the plasma membrane of the internalised parasite , likely together with other unshed SUB2 substrates , interferes with DV biogenesis , inhibiting haemoglobin digestion and stalling growth . The lack of haemozoin in both the cycle 1 ΔSUB2 parasites and many arrested iMSP1PP mutants is consistent with this model . We suggest that the intracellular death phenotype of the ΔSUB2 mutant was more severe than that of the RAP-treated iMSP1PP mutant due to the fact that in the case of the ΔSUB2 mutant none of the multiple SUB2 substrates are shed , leading to a much more pronounced effect on DV biogenesis than that caused by lack of shedding of the MSP1 complex alone . The most dramatic phenotype associated with loss of SUB2 was extensive lysis of targeted RBCs . What could underlie this ? The mechanisms regulating PVM formation , resealing and scission during invasion by apicomplexan parasites are not understood . Numerous EM studies ( e . g . Aikawa et al . , 1978 ) indicate that , topologically , invasion resembles induced invagination of the RBC membrane , so that the host RBC cytosol is never exposed to the extracellular milieu . Conflicting with this is recent evidence that successful RBC invasion by P . falciparum merozoites involves the formation of a transient discontinuity in the host cell membrane , allowing Ca2+ flux into the RBC ( Weiss et al . , 2015; Volz et al . , 2016 ) . Similarly , an elegant time-resolved patch-clamp study in the related apicomplexan Toxoplasma gondii detected a spike in host cell membrane conductance upon apical attachment of the parasite , again consistent with pore formation ( Suss-Toby et al . , 1996 ) . Near the end of invasion , as the parasite posterior enters the nascent PVM , it seems likely that a membrane scission event seals the shrinking TJ aperture , and evidence for this too was found in the Toxoplasma study ( Suss-Toby et al . , 1996 ) . Recent work in Toxoplasma suggests that PVM scission is aided by rotation of the parasite along its apical-posterior axis immediately following entry ( Pavlou et al . , 2018 ) , and malaria merozoites have also been observed to spin post invasion ( Dvorak et al . , 1975; Yahata et al . , 2012 ) . Given these models of pore formation early in invasion and a membrane scission event at the end of the process , it seems plausible that a defect in either could cause a localised loss of host cell membrane integrity sufficiently catastrophic to result in lysis . RBC lysis associated with defective or delayed entry has been previously documented even in wild type Plasmodium ( Dvorak et al . , 1975; Yahata et al . , 2012 ) , and indeed we noticed phalloidin-labelled RBC ghosts at low frequency in our control cultures , so the process is clearly sensitive to perturbation . We suggest that RBC lysis associated with loss of SUB2 is at least in part due to an accumulation of unshed merozoite surface proteins at the narrowing TJ aperture , preventing sealing ( Figure 8 ) . That lysis does not occur in the presence of cytD , which blocks invasion at the point of TJ assembly ( Miller et al . , 1979 ) , is consistent with this . Our MSP1 mutagenesis results show that inhibition of MSP1 shedding is not alone sufficient to cause lysis , but since MSP1 is just one of many SUB2 substrates this mutant cannot recapitulate the entire ΔSUB2 phenotype , where multiple proteins remain unshed from the invading parasite surface . It is similarly possible that , among the many substrates of SUB2 , there is one that captures the entire phenotype . Genetic analysis of each of the new substrates identified here would be required to resolve that question . While we favour the above model , alternative scenarios cannot be ruled out . Disruption of the Toxoplasma rhomboid protease TgROM4 , which sheds several key parasite adhesins , resulted in ‘hyper adhesive’ parasites that bound to host cells with enhanced avidity but displayed a defect in entry ( Buguliskis et al . , 2010; Rugarabamu et al . , 2015 ) . This was likely due to excessive adhesion and the absence of release of points of contact with the host cell membrane , creating a physical obstacle to entry into the nascent PV . A similar phenomenon , leading to high levels of adhesive traction on the RBC membrane , could potentially underlie the RBC lysis induced by the ΔSUB2 mutants . SUB2 depletion could also conceivably result in a defect in rhoptry discharge , although we consider this unlikely since the ΔSUB2 merozoites induced extensive RBC echinocytosis , a phenomenon associated with normal rhoptry function in Plasmodium ( Weiss et al . , 2015 ) . The ΔSUB2 phenotype may be magnified by the fact that the SUB2 substrate AMA1 is a core component of the TJ , bridging the parasite to the RBC through interactions with RON2 , a member of a complex of rhoptry neck proteins inserted into the host cell membrane early in the invasion pathway , perhaps at pore formation ( Alexander et al . , 2005; Srinivasan et al . , 2011; Lamarque et al . , 2011; Tyler and Boothroyd , 2011; Lamarque et al . , 2014; Collins et al . , 2009 ) . The AMA1-RON2 interaction may play a role in drawing the TJ orifice closed as the parasite posterior enters the PV . If so , an attractive concept is that SUB2-mediated release of the AMA1 ectodomain may facilitate pinching off of the PVM , and so we initially considered that loss of AMA1 shedding might be primarily responsible for the ΔSUB2 lysis phenotype . We were therefore surprised to find that selective inhibition of AMA1 shedding had no impact on invasion or viability . By contrast , disruption of AMA1 expression did produce a lysis phenotype , as well as the expected loss of invasion . We reconcile these disparate observations by speculating that lysis occurs upon loss of AMA1 because the RBC membrane is compromised early in invasion by insertion of the RON complex ( which occurs independently of the presence of AMA1; Lamarque et al . , 2014; Giovannini et al . , 2011 ) but subsequent stabilisation of the TJ cannot occur . We suspect that , whilst superficially similar , the mechanism underlying RBC lysis by ΔAMA merozoites is therefore distinct from that induced by the ΔSUB2 mutant ( Figure 8 ) . A role for P . falciparum AMA1 in RBC sealing has been postulated previously , based on work using a less efficient conditional mutagenesis strategy which led to variable levels of AMA1 depletion ( Yap et al . , 2014 ) . In that study many parasites that invaded failed to develop , leading the authors to suggest that even a partial sealing defect could stall development . We concur , and although we detected no membrane defects in the ΔSUB2 cycle one rings , the developmental arrest we observed in the ΔSUB2 and MSP1 cleavage site mutants might also stem from a sealing defect insufficiently severe to cause lysis . Interestingly , neither the Yap et al . study nor other studies of Plasmodium or Toxoplasma AMA1-null mutants noted widespread host cell lysis ( Giovannini et al . , 2011; Mital et al . , 2005; Bargieri et al . , 2013 ) . We suspect three likely reasons for this . First , our conditional gene disruption system is highly efficient , with close to 100% conversion in a single erythrocytic cycle . Second , in Toxoplasma disruption of AMA1 can lead to upregulation of AMA1 paralogues absent in Plasmodium ( Lamarque et al . , 2014; Parker et al . , 2016 ) . Third , RBCs are more sensitive to lysis resulting from membrane wounding than the nucleated cells invaded by Toxoplasma , perhaps due to the absence in RBCs of endomembrane-dependent repair mechanisms ( McNeil et al . , 2003 ) . In summary , the lethal consequences of SUB2 disruption likely result from: ( 1 ) a combination of loss of shedding of multiple surface proteins , leading to defective RBC sealing; and ( 2 ) arrested post-invasion development due to inhibition of DV biogenesis and/or incorrect sealing ( Figure 8 ) . Our work identifies SUB2 as a critical mediator of host RBC integrity and parasite viability . Drug-like inhibitors of SUB2 protease activity have potential as a new class of antimalarial drug that would inhibit invasion and parasite replication . Asexual blood-stages of P . falciparum were maintained at 5–10% parasitaemia in RPMI 1640 supplemented with 0 . 5% Albumax II ( RPMI 1640-Albumax; ThermoFisher ) . Parasites were synchronised at 48–96 hr intervals using standard methodology ( Blackman , 1994 ) . Briefly , mature schizonts were enriched by centrifugation over a 70% isotonic Percoll cushion ( GE Healthcare Life Sciences ) and then allowed to invade fresh AB+ RBCs ( NHSBT ) for 1–2 hr . Following invasion , remaining intact schizonts and schizont debris were removed by centrifugation over a 70% isotonic Percoll cushion and the newly-invaded ring stages further treated with 5% ( w/v ) sorbitol for 7 min at 37°C to lyse any residual schizonts . The final ring cultures were washed and returned to culture or used as required . PCR amplicons used in plasmid cloning were generated using Fusion high fidelity DNA polymerase ( ThermoFisher ) or Platinum Taq DNA polymerase , High Fidelity ( ThermoFisher ) and purified using Qiagen PCR purification or Qiagen Gel extraction kits . ( Colours and letters in parenthesis in the left-hand column correspond to primer annotations in Figure 1 , Figure 4 , and Figure 6—figure supplement 1 , Figure 7—figure supplement 1 and Figure 7—figure supplement 2 ) . Transgenic parasite lines were generated on the background of either the 1G5DC ( Collins et al . , 2013a ) or B11 DiCre-expressing parasite lines . In all cases DNA was introduced by electroporation of purified schizonts with sterile DNA , in Amaxa Primary cell solution P3 , using a 4D-Nucleofector ( Lonza ) . For single cross-over homologous recombination , 80 μg of purified DNA was used . Drug pressure ( WR99210 , 2 . 5 nM; Sigma-Aldrich ) was applied 24 hr post transfection and maintained until viable parasites were obtained . For stable transgenic lines , parasites were subjected to multiple rounds of drug pressure as previously described ( Harris et al . , 2005 ) . For Cas9-targeted double cross-over homologous recombination , 20 μg of the guide/Cas9 plasmid and 60 μg of Sap I or Nar I linearised repair plasmid were used ( Knuepfer et al . , 2017 ) . Drug selection pressure ( 2 . 5 nM WR99210 ) was applied 24 hr following transfection and maintained for 4 d before removal and continued culture of the parasites . Plaque assays of transgenic parasites were as described previously ( Thomas et al . , 2016 ) . For limiting dilution cloning , 200 μl of culture containing parasites at an estimated 1 parasite per 200 μl and 0 . 75% haematocrit , was added to each well of a flat-bottomed 96-well plate . Wells containing single plaques were identified after 10–14 days using an inverted microscope and the parasites expanded for further analysis . Parasite pellets were lysed with 0 . 15% saponin and genomic DNA extracted using a Qiagen DNeasy Blood and Tissue Kit ( Qiagen ) . PCR screens were carried out using GoTaq Green ( Promega ) with the primer combinations described . For estimation of invasion efficiency , Percoll-enriched schizonts were added to fresh RBCs to obtain a parasitaemia of 5–10% in 4 ml at a 2% haematocrit in RPMI-Albumax . Cultures were incubated at 37°C in a shaking incubator for 4 hr . Ring-stage parasites were purified as described above and returned to culture in RPMI-Albumax . Samples were taken , fixed with 0 . 2% glutaraldehyde and stored at 4°C for flow cytometry analysis . The remaining culture was followed for 48 hr to check parasite development . Giemsa-stained thin films were prepared as required for microscopic analysis . For longer-term replication assays , cultures were synchronised as described and resulting ring-stage cultures maintained for 24 hr to mature to trophozoite stage . Parasitaemia levels were calculated and cultures adjusted to 0 . 1% parasitaemia , 2% haematocrit in a final volume of 2 ml per well of a six well plate . Samples were then taken at t = 0 , 48 , 96 , and 144 hr , fixed in 0 . 2% glutaraldehyde and stored at 4°C for flow cytometry analysis . Culture media were replaced at 96 hr and 120 hr . Parasite samples were fixed in 0 . 2% glutaraldehyde in PBS and stored at 4°C . Cells were prepared for analysis by staining with 2 x SYBR Green I nucleic acid gel stain ( Life Technologies ) for 30 min at 37°C . Labelling was stopped with an equal volume of PBS and samples analysed using a BD Fortessa flow cytometer ( BD Biosciences ) with BD FACSDiva software ( BD Biosciences ) . Total RBC numbers were calculated using forward- and side-scatter whilst fluorescence was detected using the 530/30 blue detection laser . Fluorescence intensity was used to distinguish uninfected from infected RBCs , low fluorescence indicating uninfected cells and gating fixed accordingly . Data were analysed using FlowJo . Thin films of parasite cultures were made on glass slides , then air dried and stored under dessicant at −80°C . Slides were thawed , fixed in 4% paraformaldehyde for 30 min then permeabilized in 0 . 1% Triton X-100 in phosphate buffered saline ( PBS ) for 10 min prior to blocking in 3% ( w/v ) BSA in PBS overnight . Antibody incubations were carried out for 30 min at 37°C in a humidified chamber followed by washing twice for 5 min each in PBS . All antibodies were diluted into 3% ( w/v ) BSA in PBS . For microscopic imaging , samples were mounted in Vectashield containing DAPI ( Vector laboratories ) . Images were acquired using a Nikon Eclipse Ni microscope with a 100x Plan Apo λ HA 1 . 45 objective , with a Hamamatsu C11440 digital camera and processed using Fiji . Synchronous ring-stage SUB2HA3:loxP parasites were mock -or RAP-treated , then ~44 hr later schizonts were Percoll-enriched and added to fresh RBCs for 4 hr to allow egress and invasion . Without an intervening centrifugation step , a sample of the culture was supplemented with Hoechst 33342 ( 2 μg/ml; ThermoFisher ) and Alexafluor 488 Phalloidin ( diluted 1:50; ThermoFisher ) and immediately applied to a viewing chamber for live imaging as described previously ( Collins et al . , 2013b ) . Z-stack images were acquired at 2 μm intervals using a Nikon Eclipse Ni microscope with a 100x Plan Apo λ HA 1 . 45 objective and a Hamamatsu C11440 digital camera . The number of phalloidin-labelled ghosts per image was counted manually for each treatment . DMSO- or RAP-treated mature SUB2HA3:loxP schizonts were Percoll-enriched and resuspended in fresh RPMI 1640-Albumax . Schizont suspensions were mixed with an equal volume of either RPMI 1640-Albumax ( control; schizonts only ) or RBCs at a 2% haematocrit in RPMI 1640-Albumax ( co-culture; schizonts plus RBCs ) . Two additional control samples were set up comprising a similar volume of only RBCs ( 2% haematocrit ) in RPMI 1640-Albumax mixed with an equal volume of RPMI 1640-Albumax ( RBC controls ) . RBCs from one of these samples were immediately pelleted by centrifugation , the supernatant removed and 200 μl water added to obtain hypotonic lysis of the RBCs . Complete lysis was ensured by subjecting the sample to repeated freeze thawing . RPMI 1640-Albumax was then added to return the sample to the original volume . This sample acted as total RBC lysate control . The schizont only control samples , the co-culture samples and the second RBC control sample were incubated with gentle shaking for 4–6 hr at 37°C to allow egress and/or invasion . In some experiments the medium was supplemented with cytochalasin D ( 10 μM ) or 0 . 5% DMSO ( vehicle only ) as a control . All samples were then clarified by centrifugation and the culture medium filtered through a 2 μm filter . The samples of culture medium were then subjected to serial 2-fold dilutions in RPMI 1640-Albumax , and the haemoglobin content relative to that of the total RBC lysate determined by absorbance at 405 nm ( Snell and Marini , 1988 ) using a Spectramax M multi-mode microplate reader ( Molecular Devices ) . The pelleted cells from the schizont-containing samples were centrifuged over Percoll cushions to remove residual schizonts and the newly-invaded rings returned to culture . Samples of these were taken immediately as well as 18 hr and 24 hr later for flow cytometry and microscopic analysis . Visualisation of egress was as described previously ( Thomas et al . , 2018 ) . Briefly , mature schizonts were Percoll-enriched then returned to culture for 4 hr in RPMI 1640-Albumax supplemented with compound 2 ( 4-[7-[ ( dimethylamino ) methyl]−2- ( 4-fluorphenyl ) imidazo[1 , 2-α]pyridine-3-yl]pyrimidin-2-amine , C2; 1 μM ) to arrest and synchronise the schizonts in a highly mature , pre-egress stage . The parasites were then pelleted , washed twice in pre-warmed , gassed RPMI 1640-Albumax medium without C2 , then finally pelleted at 1800 x g for 1 min and immediately resuspended at ~1×107 cells/μl in pre-warmed gassed medium without C2 . The parasites were introduced by capillary flow into a pre-warmed viewing chamber , made by adhering a 22 × 64 mm borosilicate glass coverslip to a microscope slide ( Collins et al . , 2013b ) , then at once imaged on a 37°C heated stage on a Zeiss Axio Imager M1 microscope , using a EC Plan-Neofluar 1006/1 . 3 oil immersion DIC objective fitted with an AxioCam MRm camera . Images were collected at 5 s intervals for 30 mins , annotated and exported as QuickTime movies using Axiovision 3 . 1 . To image invasion , Percoll-enriched , highly synchronous mature schizonts from RAP- or DMSO-treated cultures were added to fresh RBCs at a 2% haematocrit in pre-warmed , gassed RPMI 1640-Albumax to achieve a parasitaemia of ~10% . The cells were introduced into a pre-warmed viewing chamber , transferred to the heated stage ( maintained at 37°C ) of a Nikon Eclipse Ni-E wide field microscope and imaged with a Hamamatsu C11440 digital camera using a Nikon N Plan Apo 100 × 1 . 45 NA oil immersion objective . DIC images were acquired at 1 s intervals for 30 min , exported as ND2 files and annotated using Fiji ( Image J version 2 ) . Supernatants of DMSO-and RAP-treated SUB2HA3:loxP schizonts samples following egress and invasion in Albumax-free RPMI 1640 medium were subjected to SDS-PAGE on 16 . 5% Mini-PROTEIN Tris-Tricine gels ( BioRad ) . Reduced and alkylated proteins from band slices excised from the entire separation profile were in-gel digested at 37°C overnight with 100 ng trypsin ( modified sequencing grade , Promega ) . Supernatants were dried in a vacuum centrifuge and resuspended in 0 . 1% TFA . Peptides were loaded onto an Ultimate 3000 nanoRSLC HPLC ( Thermo Scientific ) coupled to a 2 mm x 0 . 3 mm Acclaim Pepmap C18 trap column ( Thermo Scientific ) at 15 μl/min prior to elution at 0 . 25 μl/min through a 50 cm x 75 μm EasySpray C18 column into an Orbitrap Fusion Lumos Tribrid Mass Spectrometer ( Thermo Scientific ) . A gradient of 2–32% acetonitrile in 0 . 1% formic acid over 45 min was used , prior to washing ( 80% acetonitrile , 0 . 1% formic acid ) and re-equilibration . The Orbitrap was operated in data-dependent acquisition mode with a survey scan at a resolution of 120 , 000 from m/z 300–1500 , followed by MS/MS in TopS mode . Dynamic exclusion was used with a time window of 20 s . The Orbitrap charge capacity was set to a maximum of 1 x e6 ions in 10 ms , whilst the LTQ was set to 1 x e4 ions in 100 ms . Raw files were processed using Maxquant 1 . 3 . 0 . 5 and Perseus 1 . 4 . 0 . 11 . A decoy database of reversed sequences was used to filter false positives , at a peptide false detection rate of 1% . Mature schizonts enriched from DMSO- and RAP-treated SUB2HA3:loxP cultures were incubated with fresh RBCs for 4 hr to allow invasion , then newly-invaded rings isolated as described . Either immediately or following further culture for 20 hr , the parasites were washed , fixed with 2 . 5% glutaraldehyde 4% formaldehyde in 0 . 1 M phosphate buffer ( PB ) , then further washed in 0 . 2 M phosphate buffer ( PB ) , embedded in 4% low gelling temperature agarose and cut into 1 mm3 blocks ( Hanssen et al . , 2010 ) . For TEM , sample blocks were washed 4 × 15 min in PB and post-fixed in 1% osmium tetroxide/1 . 5% potassium ferrocyanide for 1 hr at 4°C . After further washes in PB at room temperature , the blocks were stained in 1% tannic acid for 45 min and quenched in 1% sodium sulphate for 5 min . The blocks were then washed 4 × 5 min in distilled water and dehydrated through an ethanol series ( 70–100% , 2 × 10 min each ) . Finally , the sample blocks were embedded by incubating in propylene oxide ( PO ) for 10 min , then overnight in a 1:1 PO:Epon 812 ( TAAB ) resin mixture , followed by 3 changes of pure Epon resin over 2 d , and baking at 60°C for 24 hr . For TEM image collection , 80 nm sections were stained with lead citrate and imaged in a Tecnai Spirit BioTwin ( Thermofisher Scientific ) transmission electron microscope . For SBF-SEM , samples were embedded in agarose as above , then processed using a protocol adapted from the NCMIR protocol ( https://ncmir . ucsd . edu/sbem-protocol ) . Briefly , the sample blocks were washed 4 × 15 min in 0 . 1 M PB and post-fixed in 2% osmium tetroxide/1 . 5% potassium ferrocyanide for 1 hr at 4°C , washed 5 × 3 min in water ( also after the following steps up to the dehydration ) , incubated in 1% w/v thiocarbohydrazide for 20 min before a second staining with 2% osmium tetroxide for 30 min , followed by incubation overnight in 1% aqueous uranyl acetate at 4°C . The sample blocks were then stained with Walton’s lead aspartate for 30 min at 60°C and dehydrated through an ethanol series on ice ( 25–100% , 10 min each ) . Finally , the sample blocks were embedded as above , except Durcupan ( TAAB ) resin was used and the baking step was for 48 hr . Embedded samples were trimmed and mounted on pins using conductive epoxy glue ( Russell et al . , 2017 ) . SBF-SEM images were collected using a 3View 2XP system ( Gatan Inc ) , mounted on a Sigma VP scanning electron microscope ( Zeiss ) . Images were collected at 1 . 9 kV , with a 30 μm aperture , at 6 Pa chamber pressure , and a 2 μs dwell time . The datasets were 69 . 63 × 69 . 63×13 μm , and 69 . 63 × 69 . 63×12 . 6 μm in xyz , consisting of 260 and 252 serial images of 50 nm thickness and a pixel size of 8 . 5 × 8 . 5 nm . The total volume of the datasets was ~63 , 028 μm3 and ~61 , 089 μm3 . SBF-SEM movies were processed in Fiji to adjust brightness and contrast , binned to 4096 × 4 , 096 pixels , and then converted to AVI , before exporting from Quicktime Pro as 950 × 950 pixel H . 264 movies . All graphs of experimental data and statistical analysis were generated using GraphPad Prism 8 . 0 . Statistical analysis methods for each experiment are stated in the corresponding figure legend . Further information and request for resources and reagents should be directed to and will be fulfilled by corresponding contact Michael J . Blackman ( mike . blackman@crick . ac . uk ) . All new plasmid and parasite lines generated in this study can be requested from the corresponding author and will be provided subject to a completed Material Transfer Agreement .	Malaria kills or disables hundreds of millions of people across the world , especially in developing economies . The most severe form of the disease is caused by Plasmodium falciparum , a single-cell parasite which , once inside a human host , forces its way into red blood cells to feed on a protein called haemoglobin . This invasion relies on P . falciparum being engulfed by the membrane of the red blood cell , which then seals off to form a compartment inside the cell where the parasite can feed and multiply . Invasion takes less than 30 seconds , and it involves P . falciparum losing the coat of proteins that covers its surface . An enzyme calls SUB2 cleaves or cuts off these proteins , but exactly why and how the shedding takes place during infection is still unclear . To investigate , Collins , Hackett et al . deactivated the gene which codes for SUB2 , and examined how mutant P . falciparum would survive and multiply . Without the enzyme , the parasites failed to shed many of their proteins , including some that were not previously known to be removed by SUB2 . The majority of the genetically modified parasites also failed to invade red blood cells . In particular , most of the host cells ruptured , suggesting that the protein coat needs to be discarded for the engulfing process to be completed properly . When the enzyme-free mutants did manage to make their way into a red blood cell , they starved to death because they could not digest haemoglobin . SUB2 and surface coat shedding therefore appears to be essential for the parasite to survive . P . falciparum is fast becoming resistant to the many drugs that exist to fight malaria . New treatments that target SUB2 may therefore help in combatting this deadly disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2020	The malaria parasite sheddase SUB2 governs host red blood cell membrane sealing at invasion
Neural systems use homeostatic plasticity to maintain normal brain functions and to prevent abnormal activity . Surprisingly , homeostatic mechanisms that regulate circuit output have mainly been demonstrated during artificial and/or pathological perturbations . Natural , physiological scenarios that activate these stabilizing mechanisms in neural networks of mature animals remain elusive . To establish the extent to which a naturally inactive circuit engages mechanisms of homeostatic plasticity , we utilized the respiratory motor circuit in bullfrogs that normally remains inactive for several months during the winter . We found that inactive respiratory motoneurons exhibit a classic form of homeostatic plasticity , up-scaling of AMPA-glutamate receptors . Up-scaling increased the synaptic strength of respiratory motoneurons and acted to boost motor amplitude from the respiratory network following months of inactivity . Our results show that synaptic scaling sustains strength of the respiratory motor output following months of inactivity , thereby supporting a major neuroscience hypothesis in a normal context for an adult animal . For the brain to work properly , neurons must employ compensatory mechanisms to retain information in synapses and to maintain normal circuit function . These compensatory mechanisms are considered homeostatic if they regulate firing frequency and/or circuit activity around a set-point , but they also may oppose disturbances to neuronal function without being firmly homeostatic ( Turrigiano , 2012 ) . Homeostatic and compensatory plasticity maintains neuronal function through a suite of synaptic and intrinsic mechanisms ( Pozo and Goda , 2010; Schulz and Lane , 2017; Turrigiano , 2011 ) , with the ultimate goal being preservation of normal network function . The most prominent mechanisms used to counter disturbances in neuronal activity include synaptic scaling , a slow and global form of synaptic plasticity where excitatory synaptic strength increases or decreases equally across all synapses ( Turrigiano et al . , 1998 ) , modulation of presynaptic neurotransmitter release ( Davis , 2013 ) , and altering voltage-gated membrane conductances ( e . g . Na+ , K+ , and Ca2+ channels ) ( Desai et al . , 1999; Ransdell et al . , 2012; Wilhelm et al . , 2009 ) . Manipulations that perturb neuronal activity in vitro are , by far , the most common experimental approaches to evoke homeostatic plasticity ( Williams et al . , 2013 ) , leaving many open questions as to what normal scenarios require homeostatic mechanisms to regulate circuit output and behavior . Only few physiological examples exist . In developing rodents , neurons in the visual cortex undergo synaptic scaling at the onset of visual experience ( Desai et al . , 2002 ) . Also within the visual system , neurons in the retinotectal circuit of the aquatic frog , Xenopus laveis , homeostatically decrease their intrinsic excitability to compensate for developmental increases in excitatory synaptic inputs ( Pratt and Aizenman , 2007 ) . Additionally , plasticity in the dynamic regulation of synaptic strength seems to compensate for variability in neuron number in the crustacean stomatogastric ganglion ( Daur et al . , 2012 ) . However , rather than uncovering responses to normal , expected physiological challenges , most investigations of homeostatic plasticity dramatically perturb neuronal activity with artificial ( e . g . pharmacological and genetic manipulations ) or pathological ( e . g . sensory denervation/stimulation , injury ) modalities . While these synthetic challenges evoke striking compensatory or homeostatic responses in neurons in vivo ( Aizenman et al . , 2003; Braegelmann et al . , 2017; Echegoyen et al . , 2007; Frank , 2014; Gonzalez-Islas and Wenner , 2006; Hengen et al . , 2013; Kline et al . , 2007; Knogler et al . , 2010; Lambo and Turrigiano , 2013; Rajman et al . , 2017 ) , how stabilizing mechanisms activated by these kinds of disturbances relate to non-pathological physiological scenarios is unclear , especially in mature animals . Animals with life histories involving profound disturbances to neuronal activity may provide new insights into the physiological necessity of such mechanisms underlying stability of neurons in functional circuits . Here , we take advantage of an adult animal ( American bullfrogs , Lithobates catesbeianus ) that normally undergoes drastic and prolonged reductions in neuronal activity to understand if compensatory mechanisms commonly evoked during artificial deprivation of neuronal firing occur in an environmentally relevant setting . We demonstrate that a well-described mechanism of homeostatic plasticity , up-scaling of excitatory synapses , occurs in motoneurons innervating a primary respiratory muscle from bullfrogs after 2 months in a submerged-aquatic , overwintering habitat: a natural environment that induces complete respiratory motor inactivity ( Santin and Hartzler , 2017 ) . Strikingly , we further identify that increased excitatory synaptic strength onto these motoneurons enhances population motoneuron output from the respiratory network , thereby acting to preserve respiratory motor drive to critical respiratory muscles under conditions when breathing would be obligatory at warm temperatures after 2 months without breathing movements ( Santin and Hartzler , 2016a ) . As failure to stabilize motor output from the brainstem to respiratory muscles has fatal consequences , these results implicate up-scaling of excitatory synapses on respiratory motoneurons as a critical mechanism for an adult vertebrate experiencing prolonged bouts of neural inactivity in a natural context . American bullfrogs often overwinter in ice-covered ponds with no need to use their lungs for breathing during cold winter months due to adequate skin gas exchange at low temperatures ( Tattersall and Ultsch , 2008 ) . Although frogs retain a relatively high locomotor capacity during overwintering submergence ( Tattersall and Boutilier , 1999 ) , important muscles of the respiratory apparatus , buccal floor constrictors and glottal dilators , are inactive during cold-submergence ( Santin and Hartzler , 2017 ) . Despite months of respiratory motor inactivity , upon forced emergence at warm temperature bullfrogs immediately exhibit breathing movements , ventilate to match resting metabolic demands , increase ventilation during exposure to low oxygen , and generate respiratory motor output from the brainstem ( Santin and Hartzler , 2016a; Santin and Hartzler , 2016b ) . Largely normal ventilatory behaviors after months of motor inactivity led us to hypothesize that preservation of respiratory motoneuron output from the brainstem to breathing muscles may rely on slow acting , global mechanisms that compensate for neuronal inactivity ( i . e . homeostatic plasticity ) . To test this hypothesis , we backfilled vagal motoneurons that innervate a primary respiratory muscle , the glottal dilator involved in regulating airflow into and out of the lung in frogs ( Gans et al . , 1969 ) , with fluorescent dye ( Figure 1 ) . We then measured synaptic and intrinsic neuronal properties from fluorescently labeled respiratory motoneurons in brainstem slices at 23°C ( a temperature requiring lung breathing ) in control bullfrogs and those that experienced ~2 months of respiratory motor inactivity during overwintering-like conditions . This allowed us to determine the extent to which winter inactivity leads to expression of compensatory mechanisms in respiratory motoneurons . Increases in excitatory synaptic transmission occur during inactivity/AMPA-glutamate receptor blockade in vitro ( Fong et al . , 2015; Turrigiano et al . , 1998 ) and in vivo ( Hengen et al . , 2013; Knogler et al . , 2010 ) . Therefore , we first sought to determine whether winter inactivity results in a compensatory increase in excitatory synaptic transmission by assessing the quantal amplitude ( i . e . the postsynaptic response to spontaneous release of neurotransmitter single vesicles recorded in voltage clamp , termed miniature excitatory post synaptic currents [mEPSCs] ) . Consistent with this hypothesis , the amplitude and charge transfer of mEPSCs mediated by AMPA-glutamate receptors were increased in respiratory motoneurons after overwintering inactivity ( Figure 2A–D ) . In contrast , we did not observe changes in mEPSC frequency ( Figure 2E ) , suggesting a similar number of glutamate release sites onto respiratory motoneurons in brainstem slices . Although the mEPSC amplitude is prone to dendritic filtering , charge transfer ( i . e . mEPSC area ) is more robust against space clamp errors ( Spruston et al . , 1993 ) . Thus , increases in both amplitude and charge transfer of the mEPSC suggest an enhancement of post-synaptic AMPA receptor function rather than changes in dendritic filtering properties of motoneurons ( Han and Stevens , 2009 ) . Consistent with this assertion , parameters influencing and influenced by dendritic filtering , neuronal input resistance and mEPSC rise time , respectively , did not differ between control and winter inactivity motoneurons ( Figure 2F–G ) . These results indicate that respiratory motoneurons have enhanced excitatory synaptic strength after winter inactivity presumably by up-regulating postsynaptic function of AMPA-glutamate receptors . Increased excitatory synaptic strength of neurons responding to reductions in activity typically follows a multiplicative scaling relationship in dissociated culture ( Fong et al . , 2015; Turrigiano et al . , 1998 ) and in vivo ( Knogler et al . , 2010; Lambo and Turrigiano , 2013 ) , implying that all measurable synapses increase in strength relatively by the same extent . Thus , synaptic scaling maintains relative synaptic strength in the face of disturbances in neuronal activity . This is observed experimentally as a right-shift in the cumulative distribution of mEPSC amplitudes induced by inactivity that can be down-scaled mathematically by dividing the entire distribution of mEPSCs by a common scaling factor . The scaling factor is derived from the slope of a linear rank order plot of mEPSC amplitudes from control and activity-deprived distributions . In contrast to synaptic scaling , a rank order plot following long-term potentiation ( LTP ) - a rapid and synapse-specific form of synaptic plasticity- is better approximated by an exponential curve ( Gainey et al . , 2009 ) . Do increases in the mEPSC amplitude scale multiplicatively following winter inactivity , an ecologically relevant , long-term perturbation of neuronal activity ? When rank ordering an equal number of mEPSC amplitudes from control and winter inactivity vagal respiratory motoneurons ( black dots; 50 mEPSCs from 16 neurons in each group ) , the plot is well-fit by a linear regression ( r2 = 0 . 99; Figure 3A; dashed blue line ) , rather than an exponential curve ( r2 = 0 . 73; Figure 3A; dashed black line ) . If the slope of this line is greater than 1 ( i . e . , the unity line; Figure 3A; solid gray line ) , it suggests multiplicative scaling has occurred . We detected a slope of 1 . 51 . The slope of the line from the regression is equal to the scaling factor , implying that all measurable excitatory synapses increased by a factor of 1 . 51 during winter inactivity . Indeed , when the right-shifted cumulative distribution of mEPSC amplitudes from winter inactivity motoneurons ( Figure 3B; solid blue line; Kolmogorov-Smirnov test control vs . winter inactivity; p=2 . 2×10−16 ) is divided by the scaling factor , the ‘down-scaled’ winter inactivity distribution ( Figure 3B; dashed blue line ) overlaps the control mEPSC amplitude distribution ( Figure 3B; solid green line; Kolmogorov-Smirnov test control vs . scale winter inactivity; p=0 . 1403 ) . Therefore , excitatory synapses on inactive respiratory motoneurons ‘scale up’ during winter inactivity . In addition to synaptic scaling , reduced activity may influence the function of intrinsic ion channels , including Na+ , K+ , Ca2+ , and non-selective cation channels , to oppose reductions in neuronal and network activity . Compensatory changes in the intrinsic membrane currents carried by ion channels during inactivity typically manifest as enhanced excitability in response to injected current ( Desai et al . , 1999; Echegoyen et al . , 2007; Lambo and Turrigiano , 2013; Wilhelm et al . , 2009 ) . Therefore , we evaluated action potential firing in respiratory motoneurons to determine whether intrinsic changes , in addition to synaptic scaling , may be involved in the compensatory response to winter inactivity . To assess intrinsic excitability , we measured firing frequency of respiratory motoneurons in response to step increases in injected current . Figure 4A shows example recordings of responses to current injections from control ( green traces; top ) and winter inactivity ( blue traces; bottom ) motoneurons . We did not observe differences in firing frequency at any amount of current ( Figure 4B ) . The frequency-current ( F-I ) gain did not differ between both groups of motoneurons ( Figure 4B–C ) . Additionally , resting membrane potential did not differ between control and winter inactivity motoneurons ( control: −57 . 62 ± 1 . 74 mV vs . winter inactivity: −57 . 13 ± 1 . 41 mV; p=0 . 8269; T30 = 0 . 2206 ) . These results suggest that these motoneurons do not enhance their excitability in response to winter inactivity . Thus far , our results indicate that up-scaling of AMPA glutamate-receptors increases the synaptic strength of respiratory motoneurons in response to winter inactivity . How may this affect respiratory behavior immediately following overwintering ? In bullfrogs and other vertebrate animals , central pattern generating ( CPG ) networks in the brainstem underlie the neural activity of respiratory movements ( Baghdadwala et al . , 2015; Feldman et al . , 2013 ) . This respiratory CPG activity is transmitted to cranial and spinal motoneurons , in part , via excitatory glutamatergic synapses ( Greer et al . , 1991; Kottick et al . , 2013 ) to produce the motor output that activates respiratory muscles . We hypothesized that synaptic scaling on respiratory vagal motoneurons receiving respiratory CPG input could increase the reliance of AMPA-receptor transmission of the vagal respiratory motor outflow in bullfrogs after winter inactivity . To test this possibility , we extracellularly recorded respiratory-related vagal motoneuron population discharge from the cut nerve root ( i . e . fictive breaths exiting the brainstem via the axons of vagal motoneurons ) in rhythmically-active , in vitro brainstem-spinal cord preparations ( Figure 5A ) ; a preparation that produces respiratory motor output similar to that observed in intact bullfrogs ( Hedrick , 2005 ) . In rhythmic respiratory preparations , synaptic properties of motoneurons ( and premotoneurons ) can be determined by assessing relative changes in the amplitude of the integrated fictive breath after application of receptor antagonists , while the activity of the rhythm generating circuits can be assessed through the frequency of fictive breaths ( for example Greer et al . , 1991; Johnson et al . , 2002 ) . Using the same in vitro preparation from bullfrogs , Chen and Hedrick ( 2008 ) showed that a low concentration of AMPA-receptor antagonist does not influence the amplitude of fictive breaths , while the frequency is more sensitive to inhibition of AMPA-receptors . To test the hypothesis that winter inactivity enhances the reliance of AMPA-receptor transmission on vagal motoneurons and to explore the possibility of altered AMPA-receptors in rhythm generating circuits , we measured the amplitude and frequency of fictive-breaths via vagal motoneurons to assess differences in the sensitivity to an AMPA receptor antagonist , DNQX , between control and winter inactivity bullfrogs . To be consistent with our results demonstrating up-scaling of AMPA receptors at the level of the vagal motoneuron after winter inactivity ( Figures 2–3 ) , the amplitude fictive breath carried by vagal motoneurons should be more sensitive to inhibition of AMPA receptors . First , we analyzed the amplitude of the integrated motoneuron population discharge of the vagus nerve ( cranial nerve X; CN X ) during lung-related respiratory bursts ( Figure 5A ) to infer whether or not scaling of AMPA-receptors on vagal motoneurons increased the reliance of AMPA-receptors for transmitting respiratory output . An inherent limitation to measuring the amplitude of the extracellular integrated motoneuron population discharge from the cut nerve root in vitro is that only relative comparisons can be made . This precludes us from comparing absolute fictive breath amplitudes at baseline between control and winter inactivity frogs . Therefore , we rely on relative changes in burst amplitude in response to a low concentration of DNQX that did not silence respiratory bursts ( 4 μM ) to infer differences in the reliance on AMPA-receptors for producing ‘normal , ’ baseline respiratory motoneuron population amplitude in each group of frogs . In control bullfrogs , AMPA-receptor inhibition did not change the amplitude and area of the respiratory-related CN X population discharge ( Figure 5B , C–D; green traces and bars ) , corroborating previous results in bullfrogs ( Chen and Hedrick , 2008 ) . This was not the case for bullfrogs following winter inactivity . In stark contrast , CN X population output underwent a reduction in amplitude and area during exposure to AMPA-receptor blockade ( Figure 5B , C–D , blue traces and bars ) and tended to recover toward baseline values following ~1 hr of DNQX washout ( Figure 5—figure supplement 1 ) . These results indicate that full transmission of respiratory CPG activity to vagal motoneurons requires an increase in synaptic strength following winter inactivity . With respect to fictive breathing frequency , brainstem-spinal cord preparations from winter inactivity bullfrogs produced an elevated number of respiratory bursts at baseline compared to control bullfrogs ( Figure 5B & E2 ) . During exposure of the brainstem-spinal cord preparations to a low concentration of DNQX , both control and winter inactivity bullfrogs decreased fictive-breathing frequency by ~80% ( Figure 5 E1 ) , suggesting AMPA receptor-sensitive components of the respiratory rhythm generator did not differ . Collectively , our findings imply that up-scaling of AMPA-glutamate receptors on vagal motoneurons preserves respiratory motor output because , in its absence ( i . e . , AMPA-receptor inhibition ) , the amplitude of vagal motoneuron population discharge was approximately 40% less than baseline on average , while respiratory motor amplitude was unaffected by the same concentration of AMPA receptor blocker in control bullfrogs . The mechanisms underlying how stable neuronal function emanates from inherently unstable components have been of intense interest to neuroscientists for over two decades ( Abbott and LeMasson , 1993; Siegel et al . , 1994; Turrigiano et al . , 1994 ) . There has been an explosion of mechanisms - postsynaptic , presynaptic , cell-autonomous , non-neuronal , etc - generated by in vitro , in vivo , and in silico approaches that paint an immensely complex picture of the genesis and maintenance of stable neuronal function ( O'Leary et al . , 2014; Schulz and Lane , 2017; Stellwagen and Malenka , 2006; Turrigiano , 2012 ) . A sobering reality is that only a few studies experimentally relate mechanisms of homeostatic plasticity to the stabilization of behaviorally relevant circuit function [e . g . ( Gonzalez-Islas and Wenner , 2006; Knogler et al . , 2010; Lane et al . , 2016 ) . Of such studies , most opt for tractability at the expense of reality by using drastic experimental perturbations that most organisms will never experience . Although these mechanistic studies are critical , it is worth considering how animals experiencing analogous situations utilize those mechanisms for survival in their environments . After all , it is the animal in its natural environmental which natural selection acts upon . Because respiratory motor output of bullfrogs is silent under submerged-overwinter conditions ( Santin and Hartzler , 2017 ) , they provided an ideal platform to test whether mechanisms that stabilize neuronal output over chronic time scales respond to prolonged inactivity experienced normally by an animal . In a sense , this serves as a natural analogue for experiments that dramatically reduce neural activity via pharmacological tools or injury because bullfrogs do not use neural mechanisms of ventilatory control for potentially several months during overwintering submergence . We found that respiratory vagal motoneurons underwent up-scaling of excitatory synapses in response to winter inactivity . This result is intriguing because up-scaling commonly occurs in response to experimental decreases in neuronal activity and/or synaptic transmission ( Fong et al . , 2015; Lambo and Turrigiano , 2013; Turrigiano et al . , 1998 ) , indicating that up-scaling of AMPA-receptors in response to inactivity-related stimuli is a physiologically relevant response in an animal normally experiencing extremely low levels of neuronal activity . Strikingly , we also observed that vagal motoneuron population output driven by respiratory CPG activity was ~40% less than baseline when we , presumably , blocked the effects of scaling with a sub-saturating concentration of AMPA-receptor antagonist after winter inactivity . This contrasts with no effect of AMPA-receptor block on respiratory motor amplitude in control bullfrogs ( Figure 5 ) . Given that amplitude of population motoneuron firing corresponds with drive to respiratory muscles in vivo ( Sakakibara , 1984b ) , up-scaling seems to ensure adequate motor output to critical breathing muscles . Since lung ventilation satisfies gas exchange requirements immediately after winter inactivity in intact , unrestrained bullfrogs ( Santin and Hartzler , 2016a ) , our findings suggest that scaling of AMPA receptors on respiratory motoneurons may contribute to normal ventilation after months without lung breathing . Several lines of evidence suggest this is the case . First , we observed an increase in mEPSC amplitude and charge transfer without changes in neuronal input resistance and mEPSC rise time , indicating that increased function of AMPA-glutamate receptors , rather than differences in electrotonic properties , explain the increase in mEPSC amplitude ( Figure 2 ) ( Han and Stevens , 2009 ) . Whether post-synaptic enhancement of AMPA-receptor function ( via elevated expression or post-translational modification ) or a presynaptic synaptic mechanism ( Liu et al . , 1999 ) causes the larger mEPSC remains an open question . Second , ranked ordered mEPSCs from control and winter inactivity distributions were well-fit by a linear regression , a hallmark of synaptic scaling , instead of an exponential curve , which describes the rank order relationship following LTP ( Figure 3A ) ( Gainey et al . , 2009 ) . Third , dividing the winter inactivity mEPSC amplitude distribution by the scaling factor obtained from the linear fit of ranked order mEPSC amplitudes produced a ‘down-scaled’ distribution that overlays the control mEPSC amplitude distribution ( Figure 3B ) ( Turrigiano et al . , 1998 ) . Points two and three imply that AMPA-receptors scaled slowly and globally throughout the duration of winter inactivity , rather than rapidly through LTP-like mechanisms during the time between removal of the frog from the simulated winter environment and ~3 hr later when the recordings were made . However , the timing of induction and completion of up-scaling in vagal respiratory motoneurons during the course of the 2-month period without breathing is unresolved . Finally , increased synaptic strength enhanced the contribution of AMPA-receptors for producing respiratory motoneuron discharge . Unlike control bullfrogs , vagal motoneurons receiving respiratory CPG input reduced motoneuron population amplitude by ~40% after winter inactivity upon exposure to a low concentration of AMPA-receptor antagonist ( Figure 5 ) . Consistent with a compensatory or homeostatic response , this result suggests that respiratory output leaving the brainstem through the vagus nerve would have been absolutely smaller in bullfrogs after winter inactivity if scaling did not occur . Since motoneuron population discharge increases proportionally with outflow to respiratory muscles in vivo ( Sakakibara , 1984b ) , we collectively interpret our results to mean that scaling of AMPA-receptors on respiratory motoneurons is part of a strategy used to preserve respiratory motor output that produces appropriate lung breathing after winter inactivity . Although the amplitude of vagal motoneuron output became increasingly reliant on AMPA-receptors following winter inactivity , upregulation of AMPA-receptors probably does not underlie the facilitation of fictive-breathing frequency that we observed ( Figure 5E present study , Santin and Hartzler , 2016b ) . This implies that respiratory rhythm generating circuits use distinct mechanisms of compensation to enhance the frequency of respiratory-related discharge following winter inactivity . Although the mechanisms are presently unknown , several , potentially interacting , processes could underlie enhanced burst frequency in vitro after winter inactivity . These mechanisms may include , but are not limited to ( 1 ) decreasing the inhibitory GABA/glycinergic tone within rhythm generating circuits ( Straus et al . , 2000 ) , ( 2 ) altering the neuromodulatory state of rhythm generating circuits , and ( 3 ) decreasing the O2 sensitivity of the respiratory control system that may act to tonically depress fictive lung bursts ( Santin and Hartzler , 2016b; Winmill et al . , 2005 ) . Although the mechanisms leading to enhanced fictive lung burst frequency remain untested , it is apparent that a rich repertoire of compensatory mechanisms may ensure reliable action of the different components of the respiratory control system after months of inactivity . Exciting questions remain for homeostatic plasticity that can only be understood using animals in natural environments; mainly , how do such compensatory mechanisms operate in natural contexts ? An inability to understand principles linking induction and maintenance of these compensatory mechanisms with normal physiological challenges makes it difficult to imagine that these mechanisms can be extrapolated and understood in the context of human pathologies for which they may play a role ( Beck and Yaari , 2008; D'Amico et al . , 2014; Rajman et al . , 2017 ) . Recently , factors associated with physiological state ( e . g . sleep-wake ) have been suggested to induce synaptic scaling ( Diering et al . , 2017 ) implying that neuronal circuits in intact animals use homeostatic and compensatory mechanisms that extend beyond the simple paradigms in which they are commonly studied . The respiratory motor network of bullfrogs following winter inactivity provides a tractable and ecologically relevant model to address major questions regarding preservation of nervous system output . For example , how do neurons ‘know’ when to scale their synapses during physiological challenges ( or even when to stop scaling ) , do specific or multiple interacting stimuli trigger synaptic scaling during natural perturbations in vivo , and why do different components of a behavioral control system apparently use distinct compensatory mechanisms ? Only considering mechanisms in the context of realistic challenges encountered by the animals in which they exist can the plethora of mechanisms underlying brain stability be brought to life . Experiments were approved by the Wright State University Institutional Animal Care and Use Committee ( protocol number 1047 ) . Two groups of adult , female bullfrogs , Lithobates catesbeianus , were used in this study: ( 1 ) control frogs maintained at 22°C ( n = 13 ) and ( 2 ) frogs experiencing an overwintering-like environment ( n = 11 ) . Control bullfrogs were maintained in a plastic tank containing 22°C aerated water , provided crickets two times per week , could access wet and dry areas , and were kept on a 12 hr:12 hr light:dark cycle . Control frogs were maintained in this environment for at least 1 week before experiments . Winter inactivity frogs were kept in a plastic tank under the same conditions as control frogs for ~1 week before water temperature was gradually cooled to 2°C over 6 weeks ( ~3 . 3°C decrease per week ) in a walk-in , temperature-controlled environmental chamber . Frogs were fed twice per week during the cooling phase until water temperature reached ~7°C , and then food was withheld because frogs no longer ate . The light:dark cycle was gradually shifted from 12 hr:12 hr to 10 hr:14 hr over the 6-week cooling phase to simulate day length changes that occur during the winter . Once water temperature reached 2°C , air access was denied using a plastic screen placed in the tank . After 8 weeks of submergence , experiments commenced . To generate brainstem slices containing vagal motoneurons , the brainstem-spinal cord was dissected as previously described ( Santin and Hartzler , 2016b ) . The dissection was performed in bullfrog artificial cerebrospinal fluid ( aCSF; concentrations in mM: 104 NaCl , 4 KCl , 1 . 4 MgCl2 , 7 . 5 glucose , 40 NaHCO3 , 2 . 5 CaCl2 and 1 NaH2PO4 , and gassed with 90% O2 , 1 . 3%CO2 , balance N2; pH = 7 . 8; CO2/pH values reflect normal for bullfrogs ) . The glottal dilator muscle gates airflow into and out of the lungs during ventilation in anuran amphibians ( Gans et al . , 1969 ) and receives its innervation from the laryngeal branch of the vagus nerve that arises the fourth rootlet of the IX-X complex in anurans ( Stuesse et al . , 1984; Yamaguchi et al . , 2003; Zornik and Kelley , 2007 ) . The fourth root of the IX-X nerve is composed of axons mostly projecting to laryngeal muscles ( Simpson et al . , 1986 ) . To label motoneurons involved in lung ventilation , following dissection of the brainstem-spinal cord , the 4th branch of the IX-X cranial nerve root was isolated from the rest of the root using fine , spring scissors and then cutting the first three branches close to their exit point from the brain . A similar preparation has been used in Xenopus laevis to prepare slices containing laryngeal motoneurons ( Yamaguchi et al . , 2003 ) . The brainstem-spinal cord was then pinned to a 6 mL Sylard ( Dow Corning , Midland , MI ) -coated dish filled with aCSF . The 4th branch of the IX-X complex was drawn into a glass pipette using suction . aCSF was removed from the pipette using polyethylene 50 tubing connected to syringe and was replaced with 2–3 μL of 10% tetramethylrhodamine-dextran , 3000 MW ( ThermoFisher Scientific , Waltham , MA ) . The dye was allowed to diffuse for 3 hr while the brain was superfused with gassed aCSF at 10 mL min−1 . We found that 3 hr was sufficient to label motoneurons ( Figure 1 ) . Following the 3 hr incubation period , we glued the brainstem to an agar block and then cut 300-μM-thick brainstem slices using a Vibratome tissue slicer ( Leica Microsystems , Buffalo Grove , IL ) in cold , gassed aCSF as previously described ( Santin and Hartzler , 2016b ) . Given the complex organization of the cranial motor nuclei in amphibians ( Matesz and Székely , 1978 ) , we selected slices with a high probability to contain laryngeal motoneurons that drive the glottal dilator based on the following anatomical rationale . The 4th root of the IX-X complex contains the axons innervating the glottal dilator in anurans , but this root also contains axons projecting motoneuron other peripheral targets . However , there is less overlap in the caudal brainstem near the obex ( Stuesse et al . , 1984 ) . Furthermore , laryngeal motoneurons are morphologically distinct from gastric and cardiac vagal motoneurons ( double the soma diameter of the long axis ) that reside in the same area ( Matesz and Székely , 1996 ) . To maximize the likelihood that we recorded from laryngeal motoneurons that innervate the glottal dilator muscle , we ( 1 ) used brainstem slices within 600 μm rostral to the obex ( Stuesse et al . , 1984 ) and ( 2 ) selected labeled neurons with a long soma diameter >20 μm as these neurons are not visceral motoneurons ( Matesz and Székely , 1996 ) . Slices were then transferred to the 0 . 5 mL recording chamber and were superfused with gassed aCSF at 1–2 mL min−1 during experiments . Preparation and neuron identification was the same for both groups of bullfrogs in this study . Borosilicate glass pipettes were back-filled with intracellular solution containing ( in mM ) : 110 potassium gluconate , 2 MgCl2 , 10 Hepes , 1 Na2 -ATP , 0 . 1 Na2-GTP , 2 . 5 EGTA , pH 7 . 2 with KOH , and positioned over an AgCl2-coated Ag wire with a resistance of 2–4 MΩ . The recording chamber was located under a fixed-stage microscope with a ( Nikon , Elgin , IL ) where the slice was visualized at 4x magnification with a Nikon Cool Snap ( Nikon ) to roughly identify the location of the X motor nucleus . Individual neurons were identified in this area using at 60X and labeling with tetramethylrhodamine-dextran was determined by fluorescence imaging ( Lambda LS Xenon Lamp House with liquid light guide , Lambda 25 mm excitation filter wheel with SmartShutter , and Lambda 10–3 controller; Sutter Instrument Company , Novato , CA; C-FL G-E/C TRITC Filter Block DM 565 , EX 540/25 ( 528-553 ) , BA 620/60 ( 590-650 ) . The recording electrode was placed near the neuron using a Burleigh micromanipulator ( PCS 5000; Thorlabs , Newton , NJ ) while applying positive pressure through the pipette . The pipette offset was zeroed before contacting the neuron . When the pipette touched the neuron , positive pressure was removed and slight negative pressure was applied by mouth until the formation of a gigaohm seal . Rapid negative pressure was applied by mouth to break the gigaohm seal and obtain whole-cell electrochemical access . Data were low-pass filtered at 2 kHz and acquired at 10 kHz for current clamp experiments and 100 kHz for voltage-clamp experiments with an Axopatch 200B amplifier , Digidata 1440A A/D converter , and Molecular Devices P10 Clampex software ( Molecular Devices , Sunnydale , CA ) . Current and voltage-clamp recordings were analyzed offline using LabChart 8 ( AD Instruments Inc . , Colorado Springs , CO ) . All voltages from voltage- and current-clamp experiments were corrected for a liquid junction potential of −12 mV ( pipette relative to the bath ) . Neurons with membrane voltages more negative than −45 mV and that contained >50 mV action potentials were used in analysis . The brainstem-spinal cord was dissected the same way as described for the preparation of the brainstem slices and pinned in a 6 mL , Sylgard-coated dish and superfused at 13 mL min−1 . Spinal nerve II ( SN II; the hypoglossal nerve in anuran amphibians ) and cranial nerve X ( CN X; vagus ) contain branches that innervate the respiratory muscles of amphibians; therefore , spontaneous , rhythmic activity recorded through these cranial nerves corresponds with respiratory rhythm/pattern generator central nervous system activity that drives breathing in intact frogs ( Sakakibara , 1984a ) . SN II was pulled into a borosilicate glass suction electrode , and another glass electrode was attached to the caudal portion of CN X to maximize coverage of the 4th root that supplies the glottis . Glass electrodes were pulled using a two-stage micropipette puller ( PC-10; Narishige , East Meadow , NY ) , broken to size to fit each rootlet , and then fire polished . Nerve activity was amplified ×1000 using differential amplifiers ( DP-311; Warner Instruments , Hamden , CT ) , filtered ( 100–1000 Hz ) , full-wave rectified , integrated ( time constant , 60 ms ) and recorded using the Powerlab 8/35 data acquisition system ( ADInstruments Inc . , Colorado Springs , CO ) . Data shown here are the integrated traces . For voltage-clamp experiments , average amplitude ( current measurement from baseline to peak ) , area ( integral of the mESPC ) , rise time ( time from baseline to peak ) , and frequency of mEPSCs ( mEPSC per minute ) ( Figure 2 ) were analyzed from one minute of gap free recording following exposure to TTX using the peak analysis function on LabChart 8 ( ADInstruments Inc . , Colorado Springs , CO ) . We rejected events below 7 . 5 pA , as this value is about double the background noise and detection was unreliable . All events were inspected by eye to ensure the software detected mEPSCs . For rank ordering mEPSC amplitudes and construction of cumulative probability histogram ( Figure 3 ) , we used the first 50 mEPSCs from the minute of data obtained from each neuron . The amplitude average of the first 50 mEPSCs for each neuron was within ~1–2 pA of the average obtained using 1 mine of data . The winter inactivity distribution was downscaled by dividing all values in the distribution by the slope of the line of the rank order relationship ( i . e . the scaling factor ) . We only included scaled values that were above the noise threshold of 7 . 5 pA , as mEPSC amplitudes below this value are not represented in the control distribution and would produce erroneous results in statistical analysis ( Kim et al . , 2012 ) . For current-clamp experiments , input resistance was calculated as slope of the voltage-current relationship in response to negative current injections ( −150 pA to 0 pA in 50 pA steps , 0 . 5 s steps ) . Membrane potential is reported as the voltage at the beginning of the step protocol . To establish the F-I gain , we averaged the firing frequency across the 0 . 5 s step from 0 pA to 1000 pA for each neuron . F-I gain for each neuron was also determined by taking the slope of the F-I relationship . For extracellular nerve recordings from the in vitro brainstem-spinal cord preparation , activity on the vagus nerve was classified as a respiratory burst if it occurred near-synchronously with the hypoglossal and if they were ~1 s in duration . We analyzed 10 min of burst data in control before application of 4 μM DNQX , the last 10 min of burst data after 40 min of exposure to 4 μM DNQX , and 10 min of data ~ 1 hr after washing DNQX . Burst properties were determined using LabChart 8 ( ADInstruments Inc . , Colorado Springs , CO , USA ) . Fictive-breath amplitude was analyzed as the maximum height subtracted from the baseline at the start of each burst of the integrated neurogram . Fictive-breath area was obtained from the area under the curve of the integrated nerve burst . Fictive-breath properties in DNQX and washout are expressed as a percent change from baseline or a percent of baseline . The number of neurons ( n = 16 per group ) and/or animals ( nine control frogs and seven winter inactivity frogs ) studied in synaptic and firing frequency experiments are consistent with other studies determining compensatory changes in neurophysiological properties in response to activity perturbations ( e . g . Wilhelm et al . , 2009; Knogler et al . , 2010; Lambo and Turrigiano , 2013 ) . For in vitro brainstem experiments of rhythmic respiratory activity in response to AMPA-receptor inhibition , four brainstems per group were used . As fictive respiratory burst amplitude has been shown to be insensitive to low concentrations of AMPA receptor inhibition ( Chen and Hedrick , 2008 ) , any changes in this insensitivity compared to control ( i . e . deviations from no change ) following the simulated winter environment would have been apparent in our ‘before and after drug’ experimental design . Similar sample sizes have been used previously for understanding synaptic transmission in the respiratory network from various species when the relative changes induced by receptor antagonists were large ( Greer et al . , 1991; Johnson et al . , 2002 ) . For comparisons of unpaired data that were normally distributed , we used a two-tailed , unpaired t test . If variance between the two groups differed , a Welch’s correction was applied to the t test . Unpaired data that failed to meet the assumption of normality were compared using the Mann Whitney test on ranked data . Within preparation comparisons for recovery from DNQX were analyzed using a paired t test as this test has been shown to be robust for small sample sizes when the correlation coefficient is large ( de Winter , 2013 ) . For the mean F-I relationship of each group , we compared the mean slopes between each group with an analysis of covariance ( ANCOVA ) . These analyses were performed using Graphpad Prism 6 . 01 ( Graphpad Software , San Diego , CA ) . Cumulative distributions were compared with the Kolmogorov-Smirnov test using R ( R Development Core Team , 2014 ) to compute exact p values . Detailed statistical information for each figure is provided in a table ( Supplementary file 1 ) . Error bars are presented as ±standard error of the mean unless otherwise specified . Statistical significance was accepted when p<0 . 05 .	Neurons in the brain communicate using chemical signals that they send and receive across junctions called synapses . To maintain normal behavior over time , circuits of neurons must reliably process these signals . A variety of nervous system disorders may result if they are unable to do so , as may occur when neural activity changes as a result of disease or injury . The processes underlying the stability of a neuron’s synapses is referred to as “homeostatic” synaptic plasticity because the changes made by the neuron directly oppose the altered level of activity . In one form of homeostatic plasticity , known as synaptic scaling , neurons modify the strength of all of their synapses in response to changes in neural activity . There is substantial experimental evidence to show that in young animals , neurons that communicate using a chemical called glutamate undergo synaptic scaling in response to artificial changes in activity . It had not been directly shown that synaptic scaling protects the neural activity of adult animals in their natural environments , in part , because neural activity in most healthy animals generally only goes through small changes . However , the neurons in the brain that cause the breathing muscles of bullfrogs to contract are ideal for studying homeostatic plasticity because they are naturally inactive for several months when frogs hibernate in ponds during the winter . During this time , the bullfrogs do not need to use their lungs to breathe because enough oxygen passes through their skin to keep them alive . Santin et al . have now observed synaptic scaling of glutamate synapses in individual bullfrog neurons that had been inactive for two months . Further experiments that examined the activity of the breathing control circuit in the brainstem provided evidence that synaptic scaling leads to sufficient amounts of neural activity that would activate the breathing muscles when frogs emerge from hibernation . Therefore neural activity after prolonged , natural inactivity relies on synaptic scaling to preserve life-sustaining behavior in frogs . These results open up new questions: mainly , how do synaptic scaling and other forms of homeostatic plasticity operate in animals as they experience normal variations in neural activity ? Determining how homeostatic plasticity works normally in an animal will help us to understand what happens when plasticity mechanisms go wrong , as is thought to occur in several human nervous system diseases including nervous system injury , Alzheimer’s disease , and epilepsy .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	Synaptic up-scaling preserves motor circuit output after chronic, natural inactivity
Our visual memory percepts of whether we have encountered specific objects or scenes before are hypothesized to manifest as decrements in neural responses in inferotemporal cortex ( IT ) with stimulus repetition . To evaluate this proposal , we recorded IT neural responses as two monkeys performed a single-exposure visual memory task designed to measure the rates of forgetting with time . We found that a weighted linear read-out of IT was a better predictor of the monkeys’ forgetting rates and reaction time patterns than a strict instantiation of the repetition suppression hypothesis , expressed as a total spike count scheme . Behavioral predictions could be attributed to visual memory signals that were reflected as repetition suppression and were intermingled with visual selectivity , but only when combined across the most sensitive neurons . The everyday act of viewing the things around us leaves us with memories of the things that we have encountered . Under the right conditions , this type of ‘visual recognition memory’ can be quite remarkable . For example , after viewing thousands of images , each only once and only for a few seconds , we can determine with high accuracy the specific images that we have viewed ( Brady et al . , 2008; Standing , 1973 ) . Additionally , we can remember not just the objects that we’ve seen , but also the specific configurations and contexts we saw them in ( Brady et al . , 2008 ) , suggesting that our brains store these memories with considerable visual detail . Where and how are visual memories stored and where and how is the percept of visual memory signaled ? One candidate mechanism for signaling visual memory percepts is the adaptation-like response reduction that occurs in high-level visual brain areas with stimulus repetition , known as ‘repetition suppression’ ( Fahy et al . , 1993; Li et al . , 1993; Miller and Desimone , 1994; Riches et al . , 1991; Xiang and Brown , 1998 ) . Consistent with that proposal , individual viewings of a novel image produce response reductions in inferotemporal cortex ( IT ) that can last tens of minutes to hours ( Fahy et al . , 1993; Xiang and Brown , 1998 ) . Signaling visual memories in this way is attractive from a computational perspective , as it could explain how IT supports visual identity and visual memory representations within the same network . That is , insofar as visual representations of different images are reflected as distinct patterns of spikes across the IT population ( Figure 1; DiCarlo et al . , 2012; Hung et al . , 2005 ) , this translates into a population representation in which visual information is reflected by the population vector angle ( Figure 1 ) . If it were the case that visual recognition memories were reflected by changes in the total numbers of spikes or equivalently population response vector length , this could minimize interference when superimposing visual memories and visual identity representations within the same network ( Figure 1 ) . While attractive , there are also reasons to question whether visual memory percepts manifest purely as repetition suppression in IT cortex . For example , following many repeated image exposures ( e . g . hundreds to thousands ) , IT neurons exhibit tuning sharpening ( Anderson et al . , 2008; Freedman et al . , 2006 ) , and a subset of neurons reflect tuning peak enhancement ( Lim et al . , 2015; Woloszyn and Sheinberg , 2012 ) , and these changes could happen during single-exposure memory as well . Similarly , in the case of highly familiar images , neurons in a brain area that lie beyond IT , perirhinal cortex , are reported to signal familiarity with increases ( as opposed to decreases ) in firing rate ( Tamura et al . , 2017 ) and highly familiar faces produce larger perirhinal fMRI BOLD responses as compared to faces that are unfamiliar ( Landi and Freiwald , 2017 ) . In humans , tests of the hypothesis that limited-exposure visual memory percepts are supported by repetition suppression signals have produced mixed results , with some studies providing support ( Gonsalves et al . , 2005; Turk-Browne et al . , 2006 ) and others refuting the hypothesis ( Ward et al . , 2013; Xue et al . , 2011 ) . Additionally , studies have implicated factors beyond overall response strength in limited-exposure familiarity , including the repeatability of human fMRI response patterns across exposures ( LaRocque et al . , 2013; Xue et al . , 2010 ) and synchronization between gamma band oscillations and spikes in monkey hippocampus ( Jutras et al . , 2013 ) . Notably , while a number of studies have investigated limited-exposure repetition suppression effects in IT at the resolution of individual-units ( De Baene and Vogels , 2010; Li et al . , 1993; McMahon and Olson , 2007; Ringo , 1996; Sawamura et al . , 2006; Verhoef et al . , 2008; Xiang and Brown , 1998 ) , no study to date has attempted to determine whether these putative visual memory signals can in fact account for visual memory behaviors . To evaluate the hypothesis that repetition suppression in IT accounts for familiarity judgments during a visual memory task , we trained two monkeys to view images and report whether they were novel ( had never been seen before ) or were familiar ( had been seen exactly once ) , across a range of delays between novel and familiar presentations . To explore the IT representation of visual memory on both correct and error trials , we parameterized the task such that visual memories were remembered over a timescale of minutes within experimental sessions that lasted approximately one hour . We found that while both monkeys displayed characteristic forgetting functions and reaction time patterns , these behavioral patterns were not well-predicted by a spike count decoder that embodied the strictest interpretation of the repetition suppression hypothesis . These behavioral patterns were better accounted for by a linear read-out that weighted each neuron proportional to the amount of visual memory information reflected in its responses . While compelling , the robustness with which visual memories are stored also presents a challenge to investigating their underlying neural correlates . Ideally , investigations of the neural signals supporting a behavior are made in a context where a task is parametrically varied from easy-to-challenging , and one can evaluate the degree to which behavioral sensitivities and behavioral confusions are reflected in neural responses ( Parker and Newsome , 1998 ) . Following on visual recognition memory studies that demonstrate a relationship between the time that images are viewed and how well they are remembered ( Brady et al . , 2009; Potter and Levy , 1969 ) , we increased task difficulty by reducing image viewing time from the 2–3 s used in the classic human visual recognition memory studies to 400 ms . To titrate task difficulty within this regime , we explored a range of delays between novel and repeated presentations . In these experiments , two monkeys performed a task in which they viewed images and indicated whether they were novel or familiar with an eye movement response . Monkeys initiated each trial by fixating a point at the center of the screen , and this was followed by a brief delay and then the presentation of an image ( Figure 2a ) . After 400 ms of fixating the image , a go cue appeared , indicating that the monkeys were free to make their selection via a saccade to one of two response targets ( Figure 2a ) . Correct responses were rewarded with juice . While the first image presented in each session was always novel , the probability of subsequent images being novel versus familiar quickly converged to 50% . Novel images were defined as those that the monkeys had never viewed before ( in the entire history of training and testing ) whereas familiar images were those that had been presented only once , and earlier in the same session . A representative set of images can be found in Figure 2—figure supplement 1 . Delays between novel and familiar presentations ( Figure 2b ) were pseudorandomly selected from a uniform distribution , in powers of two ( n-back = 1 , 2 , 4 , 8 , 16 , 32 and 64 trials corresponding to mean delays of 4 . 5 s , 9 s , 18 s , 36 s , 1 . 2 min , 2 . 4 min , and 4 . 8 min , respectively ) . To prevent confusion , we emphasize that our usage of the term ‘n-back’ refers to the numbers of trials between novel and familiar presentations , in contrast to the usage of this term in other studies that required a same/different comparison between the current stimulus and a stimulus presented a fixed number of trials back ( e . g . 2-back ) in a block design ( e . g . Cornette et al . , 2001 ) . The monkeys’ performance on this task was systematic , as illustrated by smoothly declining ‘forgetting functions’ , plotted as the proportion of trials that images were reported familiar as a function of n-back ( i . e . the number of trials between novel and familiar presentations; Figure 3a , c ) . When familiar images were immediately repeated ( n-back = 1 ) , both monkeys most often called them familiar ( proportion chose familiar = 0 . 98 and 0 . 94; Figure 3a , c ) . Similarly , when images were novel , monkeys were unlikely to call them familiar ( proportion chose familiar = 0 . 13 and 0 . 07; Figure 3a , c ) . Between these two extremes , the proportion of familiar reports systematically decreased as a function of n-back ( Figure 3a , c ) . In monkey 1 , performance at 32 and 64 back fell below chance ( 32-back = 0 . 46 , 64-back = 0 . 27 , chance = 0 . 50 ) , indicating that this animal most often reported that familiar images repeated after these longer delays were novel ( Figure 3a ) . In monkey 2 , performance at 32 and 64 back remained above chance ( 32-back = 0 . 76 , 64-back = 0 . 54 ) , indicating higher performance in this animal as compared to monkey 1 ( Figure 3c ) . We also analyzed reaction times for novel and familiar trials , parsed by correct and error trial outcomes . Reaction times were measured relative to the onset of the go cue ( which appeared 400 ms after stimulus onset ) . We found that mean reaction times on correctly reported familiar trials systematically increased as a function of n-back , or equivalently , reaction times on correct trials increased with increasing difficulty ( Figure 3b , d red ) . Conversely , reaction times on error trials decreased as a function of n-back , or equivalently , reaction times on error trials decreased with increasing difficulty ( Figure 3b , d , cyan ) . In both animals , this led to an x-shaped pattern in the mean reaction times on familiar trials when plotted as a function of n-back . On novel trials , reaction times mimicked the pattern observed for the low n-back familiar cases in that reaction times were faster on correct as compared to error trials ( Figure 3b , d ) . From what underlying process might these x-shaped reaction time patterns arise ? As is the case for nearly any task , behavioral performance can be thought of as the outcome of passing a signal ( in this case a memory signal ) through a decision process . The x-shaped patterns that we observed differ from the patterns reported for tasks that are well-accounted for by the standard drift diffusion model ( DDM ) of decision making , such as the dot-motion-direction task ( Gold and Shadlen , 2007 ) . In agreement with standard DDM predictions , reaction times on correct trials increased as task performance decreased ( i . e . with n-back ) . However , reaction times on error trials decreased with n-back whereas the standard DDM predicts that reaction times will be matched on correct and error trials ( and thus reaction times on error trials should increase with n-back as well ) . While it is the case that extensions to this framework can predict reaction time asymmetries ( Ratcliff and McKoon , 2008 ) , additional parameters are required for it to do so , and these additions make it less well-suited for the purposes of this study ( focused on evaluating the plausibility that IT visual memory signals can quantitatively account for visual memory behavior ) . We have , however , determined that these x-shaped reaction time patterns can be captured by a very simple , low-parameter extension to the signal detection theory framework , as proposed by ‘strength theory’ ( Murdock , 1985; Norman and Wickelgren , 1969 ) . Like signal detection theory , strength theory proposes that a noisy internal variable ( ‘memory strength’ ) is compared to a criterion to determine whether an image is novel or familiar ( Figure 4 , left ) . Strength theory also predicts that during a visual memory task , reaction times will be inversely related to the distance of this variable from the criterion , loosely analogous to a process in which increased certainty produces faster responses ( Figure 4 , middle ) . This leads to the qualitative prediction that when images are repeated with short n-back , memories are strong , and this will produce reaction times that are faster on correct as compared to error trials ( Figure 4 , red vs . blue ) . In contrast , at long n-back , memories are weak , and this will produce reaction times that are slower on correct as compared to error trials ( Figure 4 , green vs . purple ) . The combined consequence of strong and weak memories is an x-shaped pattern . In sum , the reproducible patterns reflected in both monkeys’ forgetting functions , along with their reaction time patterns , place non-trivial constraints on the candidate neural signals that account for single-exposure visual memory behavior . The x-shaped patterns of reaction times that we observe cannot be accounted for by a standard drift diffusion process , but they can , in principle , be captured by the predictions of strength theory . However , a successful description of the neural signals supporting single-exposure visual memory behavior requires identifying a neural signal whose sensitivity to the elapsed time between initial and repeated presentations of an image matches the sensitivity reflected in the monkeys’ behavior . As monkeys performed this task , we recorded neural responses from IT using multi-channel probes acutely lowered before the beginning of each session . For quality control , recording sessions were screened based on their neural recording stability across the session , their numbers of visually responsive units , and the numbers of behavioral trials completed ( see Materials and methods ) . The resulting data set included 15 sessions for monkey 1 ( n = 403 units ) , and 12 sessions for monkey 2 ( n = 396 units ) . Both monkeys performed many hundreds of trials during each session ( ~600–1000 , corresponding to ~300–500 images each repeated twice ) . The data reported here correspond to the subset of images for which the monkeys’ behavioral reports were recorded for both novel and familiar presentations ( e . g . trials in which the monkeys did not prematurely break fixation during either the novel or the familiar presentation of an image ) . We began by considering the proposal that the signals differentiating novel versus familiar presentations of images were systematically reflected as response decrements with stimulus repetition ( i . e . ‘repetition suppression’ ) . As a first , simple illustration of the strength of these putative single-exposure memory signals , shown in Figure 5a is a plot of the grand mean firing rate response of all 799 units parsed by n-back , plotted as a function of time relative to stimulus onset . This plot reveals a fairly systematic decrement in the response with repetition that diminished with time since the novel presentation . We quantified the magnitude of suppression as the decrement in the area under each n-back trace relative to the novel trace , computed 150–400 ms after stimulus onset ( Figure 5b ) . Consistent with a visual memory signal that degrades ( or forgets ) with time , immediate stimulus repetition resulted in a decrement in the response of ~11% and suppression magnitudes decreased systematically with n-back . Also , qualitatively consistent with the repetition suppression hypothesis was the finding that when the same analysis was isolated to the units recorded from each monkey individually , repetition suppression was stronger in the monkey that was better at the task ( monkey 2; Figure 5c–d ) . To quantitatively assess whether IT neural signals could account for the monkeys’ behavioral reports , we applied two types of linear decoding schemes to the IT data . The first , a spike count classifier ( SCC ) , is an instantiation of the strictest form of the repetition suppression hypothesis in that it differentiated novel versus familiar responses based on the total number of spikes across the IT population ( i . e . every unit in the population received a weight of 1 ) . The second , a Fisher Linear Discriminant ( FLD ) , is an extension of the SCC that allows for IT units to be differentially weighted and allows for weights to be positive as well as negative ( corresponding to repetition suppression and enhancement , respectively ) . Because the neural data collected in any individual recording session had too few units to fully account for the monkeys’ behavior ( e . g . near 100% correct for 1-back familiar images ) , we concatenated units across sessions to create a larger pseudopopulation , where responses were quantified 150–400 ms following stimulus onset . When creating this pseudopopulation , we aligned data across sessions in a manner that preserved whether the trials were presented as novel or familiar as well as their n-back separation . More specifically , the responses for each unit always contained sets of novel/familiar pairings of the same images , and pseudopopulation responses across units were always aligned for novel/familiar pairs that contained the same n-back separation . Because different images were used in each session , aligning images this way implicitly assumes that the total numbers of spikes are matched across different images , the data recorded in any one session is a representative sample of those statistics , and that the responses of the units recorded in different sessions are uncorrelated . When the number of images in a session exceeded the number required to construct the pseudopopulation , a subset of images were selected randomly , and we confirmed that our main results did not change for different random selections . In the case of the pooled data , the resulting pseudopopulation consisted of the responses from 799 neurons to 107 images presented as both novel and familiar ( i . e . 15 , 15 , 16 , 17 , 17 , 15 and 12 trials at 1 , 2 , 4 , 8 , 16 , 32 and 64-back , respectively ) . We begin by illustrating our procedure for computing neural predictions of the behavioral forgetting functions and reaction time patterns with the FLD weighted linear read-out , applied to the data pooled across the two subjects . We then present a more systematic comparison between different decoders applied to each monkey’s individual data . To compute neural predictions for behavioral forgetting functions , we began by training an FLD linear decoder to discriminate the same images presented as novel versus as familiar ( Figure 6a ) using the data corresponding to all n-backs simultaneously . The FLD training procedure assigned a weight to each neuron proportional to the amount of linearly-separable visual memory information reflected in its responses ( i . e . it’s d’; see Materials and methods ) , and a single criterion value to parse the combined , weighted population responses for novel versus familiar predictions . A final parameter specified the size of the IT population under consideration ( detailed below ) . Shown in Figure 6b are the neural estimates of the distributions of memory signal strength at each n-back , computed across many iterations of the cross-validated linear classifier training and testing procedure for the best sized population ( n = 799 units ) . As expected , we found that the weighted population response strengths were largest for novel images ( Figure 6b , black ) and were weakest for familiar images presented as immediate repeats ( Figure 6b , red ) . Between these two extremes , we observed a continuum of strengths loosely organized according by n-back ( Figure 6b , rainbow ) . Finally , a neural prediction for the forgetting function was computed as the fraction of each distribution that fell on the ‘familiar’ side of the criterion differentiating novel versus familiar predictions ( Figure 6c ) . This analysis revealed a high degree of alignment between the neural prediction at each n-back and behavior , including high performance for familiar images presented at low n-back , performance at mid-range n-back that fell off with a similar sensitivity , and performance at the longest n-back ( 64 ) that fell below chance ( Figure 6c ) . Similarly , neural predictions for novel images were well-aligned with the monkeys’ behavioral reports ( Figure 6c , ‘N’ ) . To produce neural predictions for reaction times , we turned to strength theory ( Figure 4 ) . Shown in Figure 6d is the first step required for making those predictions: a plot of the neural predictions for the proportions of ‘correct’ and ‘error’ trials , plotted as a function of n-back . Note that the correct predictions simply replicate the forgetting function shown in Figure 6c , and the error predictions are simply those same values , subtracted from 1 . While these plots directly follow from Figure 6c , we include them to illustrate that they qualitatively reflected an inverted version of the monkeys’ behavioral reaction time plots , including an x-shaped pattern . To determine the degree to which these qualitative relationships quantitatively predict the monkeys’ reaction times , we examined the relationship between the proportions plotted in Figure 6d and the monkeys’ mean reaction times , and found it to be approximately linear ( Figure 6e ) . We thus fit a two parameter linear function to convert the neural predictions of these proportions into reaction times ( Figure 6e , black line ) . The resulting neural predictions were largely aligned with the monkeys’ mean reaction times ( Figure 6f ) , including increasing reaction times as a function of n-back on correctly reported familiar trials , decreasing reaction times as a function of n-back on familiar trials in which the monkeys’ made errors , and the characteristic x-shaped pattern . Additionally , shorter mean reaction times for novel images on correct versus error trials were largely well-predicted by the neural data . One important step in the procedure , not detailed above , involved determining the appropriate IT population size for making neural and behavioral comparisons . Because there really wasn’t a way to do this a priori , we applied a fitting approach in which we computed the mean squared error ( MSE ) between the actual forgetting functions and their neural predictions at a range of population sizes , including simulated extensions of our population up to sizes 50% larger than the maximal size we recorded ( Figure 6—figure supplement 1 ) . The existence of a minimum in these plots follows from the fact that they depict the error between the neural prediction and the behavioral forgetting function ( as opposed to overall neural population d’ for this task , which continued to increase with increasing population size; Figure 6h ) . When too few units were included in the population , neural d’ was too low and high performance at low n-back was underestimated ( Figure 6g , left inset ) . In contrast , when too many units were included in the population , neural population d’ was too high and performance at low n-back was over-saturated ( Figure 6g , right inset ) . Additionally , for populations that were too large , performance fell off with n-back with a slope that was too steep . Of interest was the question of whether a global alignment of behavioral and neural sensitivity produced an accurate neural prediction of the shape for forgetting function with n-back . In the case of the FLD applied to the pooled data , the best population size fell near the maximal size of the total number of units that we recorded ( n = 799 , Figure 6g , red dot ) . The analyses presented thus far were computed based on spike count windows 150–400 ms following stimulus onset . A complementary plot illustrates how the position of the spike count window relative to stimulus onset impacted the best MSE ( across all population sizes ) for spike count windows 150 ms wide ( Figure 6i ) . Consistent with the arrival of a visual memory signal that is delayed relative to onset but remains relatively constant thereafter , error was high for windows that began earlier than 150 ms following stimulus onset and then saturated . This suggests that the 150–400 ms position of the spike count window used to analyze the data throughout this report was a reasonable selection . As a final step for our procedure , we determined a measure of prediction quality for both the forgetting function and reaction time patterns . Our measure benchmarked the MSE between the behavioral patterns and neural predictions by the worst-possible fit given that our procedure involves a global alignment of behavioral and neural data ( Figure 6g ) . The upper bound of our measure , 100% ‘prediction quality’ ( PQ ) , reflects a neural prediction that perfectly replicates the behavioral data . The lower bound ( 0% PQ ) was computed as the MSE between the actual behavioral function and a predicted forgetting function that took the shape of a step , matched for global performance ( percent correct across all conditions; Figure 6c , f , dotted ) . The rationale behind the step is that under a reasonable set of assumptions ( i . e . that performance as a function of n-back should be continuous , have non-positive slope , and be centered around chance ) , a step reflects the worst possible fit of the data . Finally , PQ was calculated as the fractional distance of the MSE between these two benchmarks . In the case of the FLD applied to the pooled data , PQ was 94% for the forgetting function and 86% for the reaction time data ( Figure 6c , f ) . We emphasize that these numbers reflect the quality of generalized neural predictions to the behavioral reports , as these neural predictions were not fit directly to the behavioral data in a manner not already accounted for by the PQ measure . Our methods for determining predictions of the SCC decoder differed only in the algorithm used to combine the spike counts across the population into a measure of memory strength ( Figure 6b ) . In the case of the SCC , the weight applied to each unit was 1 , and the training procedure determined a single criterion value to parse the total population spike counts into novel versus familiar predictions . The same cross-validated procedure used for the FLD was applied to the SCC to determine distributions analogous to those depicted in Figure 6b . When applied to the data pooled across the two monkeys , the best sized SCC decoded population was 559 units ( Figure 7a ) . Additionally , we found that while the SCC was a better predictor of behavior than the FLD for smaller sized populations ( less than 400 neurons ) , the FLD was a better predictor of behavior overall ( Figure 7a ) . Examination of a plot of overall population d’ as a function of population size ( Figure 7b ) reveals that the minimal error fell at the same population d’ for both decoding schemes , consistent with the notion that our procedure involved a global matching of overall performance between the behavioral and neural data . The fact that the lowest MSE differed between the two decoding schemes reflects differences in the shapes of the neural predictions following global performance matching . Figure 7b also reveals systematically better global performance of the SCC as compared to the FLD for matched sized populations , which is likely a consequence of the fact that a smaller number of parameters are fit with the SCC read-out and the estimation of FLD weights is a noisy process . A comparison of SCC and FLD MSE plots isolated to each monkey’s data revealed that the FLD decoder was a better predictor of behavior in both individuals ( Figure 7c–d ) . Why was the FLD weighted linear decoder a better predictor of the behavioral forgetting function ? This was because the spike count decoding scheme under-predicted memory strength , particularly at the longest delays . While this is discernable in plots of the raw alignment of the behavioral and neural data for each monkey plotted with the same conventions as Figure 6f ( Figure 7—figure supplement 1a–b ) , it is more easily observed in a visualization of the data in which the proportion of familiar choices for both the behavioral data and neural predictions are plotted after subtracting the false alarm rate for the novel images ( Figure 7e ) , thus producing plots analogous to the suppression plots presented in Figure 5c–d . For example , in monkey 1 , the SCC decoder predicted that the monkey would report 64-back familiar images as familiar at a rate lower than the false alarm rate for novel images , whereas the actual forgetting function reflected a small amount of remembering after a 64-back delay ( Figure 7e ) . Similarly , in monkey 2 , the SCC predicted rate of remembering at 64-back under-predicted the actual rate reflected in the behavior ( Figure 7e ) . In contrast , the FLD better predicted the behavior across all n-back in both animals ( Figure 7e ) . Lower MSE for the FLD as compared to SCC translated into higher neural PQ in each monkey ( Figure 7e – labeled; not shown for the pooled data: SCC PQ = 83% , FLD PQ = 94% ) . The same behavioral and neural comparisons , plotted with the same conventions as Figure 6c and Figure 6f , are shown in Figure 7—figure supplement 1 . We note that while the FLD PQ was lower in monkey two as compared to monkey 1 ( monkey 1 FLD PQ = 92% , monkey 2 FLD PQ = 70% ) , this was not due to a lower MSE of the fits in monkey 2 ( Figure 7c–d ) but rather due to the fact that the forgetting function for monkey two better resembled the step benchmark for computing PQ , thus reducing the PQ dynamic range ( Figure 7—figure supplement 1a–b ) . Together , these results suggest that a weighted linear read-out was a better description of the transformation between IT neural signals and single-exposure visual memory behavior than a total spike count decoding scheme . The results presented above suggest that the SCC under-predicted memory strength as a function of n-back whereas the FLD prediction was more accurate . At least two different scenarios might lead to this result . First , it could be the case that visual memory may be reflected as net repetition suppression in some units and net repetition enhancement in others ( across all n-back ) . In this scenario , the FLD would preserve both types of memory information ( by assigning positive and negative weights for enhancement and suppression , respectively ) , whereas these two types of effects would cancel in a SCC decoding scheme , resulting in information loss . Alternatively , it might be the case that the repetition suppression hypothesis is approximately correct insofar as the IT units that carry visual memory signals systematically reflect visual memory with net repetition suppression , however , repetition suppression may be stronger at longer n-back for some units than others . In this scenario , better FLD behavioral predictions would result from preferentially weighting the neurons with the strongest ( by way of longest lasting ) visual memory signals . As described below , our results suggest that the latter scenario is a better description of our data . To distinguish between these two scenarios , we began by examining the distributions of unit d’ as a proxy for the FLD decoding weights . In both monkeys , the unit d’ means were significantly shifted toward positive values ( Figure 8a–b; Wilcoxon sign rank test , monkey one mean = 0 . 05 , p=6 . 810−17; monkey two mean = 0 . 12 , p=1 . 910−41 ) . In both monkeys , units with negative d’ were also present ( proportion of negative units for monkey 1 = 32%; monkey 2 = 19% ) , although from raw d’ values alone , the degree to which negative d’ resulted from reliable net repetition enhancement versus from noise is unclear . A comparison of the mean responses to novel as compared to familiar images for each unit revealed that very few units with negative d’ had statistically distinguishable responses ( bootstrap statistical test; criterion p<0 . 01; monkey 1: positive d’ units = 14; negative d’ units = 3; monkey 2: positive d’ units = 75; negative d’ units = 2 ) . While a screen of p<0 . 01 can under-estimate the contributions of a unit to population performance , additional analyses , described below , confirm that negative d’ units made a measurable but modest contribution to the differences between the SCC and FLD behavioral predictions . To understand how these unit d’ measures combined to determine behavioral predictions , we performed an analysis to determine the minimal number of ‘best’ d’ IT units required to predict behavior . The general idea behind this analysis is that if it were the case that strong signals were carried by a small subpopulation of units , error should plateau quickly when only best units are included . We thus compared FLD behavioral prediction error trajectories for the pooled data ( to maximize the numbers of directly measured units ) when subsets of units were randomly sampled ( our typical procedure ) versus when the top-ranked d’ units were selected via a cross-validated procedure ( i . e . based on the training data; Figure 8c , left ) . We also converted these MSE measures into prediction quality estimates ( Figure 8d , right ) . We found that 400 top-ranked IT units were required to achieve the same prediction quality as 800 randomly sampled units , suggesting that FLD behavioral predictions rely on visual memory signals that are distributed across approximately half of the IT population . The absence of a contribution from the lower-ranked 50% of the IT population could not be attributed to non-responsiveness , as nearly all the units ( 759/799 , 95% ) produced statistically significant stimulus-evoked responses that differed from the pre-stimulus baseline period ( bootstrap statistical test; criterion p<0 . 01; comparison of spike count windows ( −150–0 ) ms versus ( 75 – 225 ) ms relative to stimulus onset ) . Why did the FLD produce better behavioral predictions than the SCC ( Figure 7a ) ? To address this question , we repeated the top-ranked analysis for the SCC . Specifically , we performed a cross-validated procedure in which units were ranked by their signed d’ as described above for the ranked FLD , but within the top-ranked units , spikes were summed to produce behavioral predictions ( Figure 8d ) . One can envision this as a binary classifier where the top-ranked units each receive a weight of 1 whereas the remaining units each receive a weight of 0 . Surprisingly , the ranked-SCC decoder also peaked at 400 units and performed nearly as well as the ranked-FLD ( ranked SCC PQ for 400 units = 91% , Figure 8d; ranked FLD PQ for 400 units = 94% , Figure 8d ) . This suggests that within the subset of 50% top-ranked IT units , spikes could largely be summed to make behavioral predictions . What happens when the 50% bottom-ranked units are added to each type of decoder ? Addition of bottom-ranked units had no impact on the ranked-FLD ( Figure 8c right , ‘All units’ ) . This suggests that the FLD largely disregards the lower 50% ranked units when making behavioral predictions . In contrast , the introduction of the lower 50% ranked units detrimentally impacted ranked-SCC behavioral predictions ( ranked SCC PQ for best 50% of units = 91%; for all units = 76%; Figure 8d , right ) . This is presumably because the SCC does not have a weighting scheme and was thus forced to incorporate them . When parsed by the sign of d’ for the lower-ranked units , addition of lower-ranked , positive d’ units reduced ranked-SCC behavioral predictions from 91% to 81% , and further addition of negative d’ units reduced behavioral predictions to 76% ( Figure 8d , right ) . Returning to the two scenarios presented at the beginning of this section , these results suggest that better FLD as compared to SCC behavioral predictions could largely be attributed to the FLD preferentially weighting the neurons with the strongest ( by way of longest lasting ) visual memory signals , as opposed to the inability of the SCC to appropriately weight reliable , mixed sign modulation ( i . e . mixtures of repetition suppression and enhancement ) . Together , these results suggest that largely accurate behavioral predictions could be attributed to ~50% of IT units whose memory signals were reflected as repetition suppression , and within this top-ranked subpopulation , spike counts could largely be summed . These results also show that while the lower ranked units had a detrimental impact on the ability of the spike count decoder to produce accurate behavioral predictions , a weighted linear decoder largely disregarded these otherwise confounding responses . As a complementary consideration , we also examined the impact of visual selectivity on the size of the population required to account for behavior . Hypothetically , if only a small fraction of IT units were activated in response to any one image , a large population would be required to support robust visual memory behavioral performance . Because our data only include the response to each image twice ( once as novel and repeated as familiar ) , and measures of visual selectivity ( e . g . ‘sparseness’ ) produce strongly biased estimates with limited samples ( Rust and DiCarlo , 2012 ) , we applied a simulation-based approach to determine how visual selectivity impacted the population size required to make accurate behavioral predictions . The general idea behind this analysis is to compare the best population size for our intact data with a simulated version of our data in which visual memory signals have been kept intact but visual selectivity has been removed . To perform this analysis , we began by creating a simulated ‘replication’ population designed to match the image selectivity , memory signal strength , and grand mean spike count response for each unit we recorded , followed by the introduction of Poisson trial variability ( see Materials and methods ) . This simulated population produced FLD behavioral prediction error trajectories that were highly similar to the intact population , both when computed with the regular FLD ( Figure 9a , gray versus black ) , as well as with the ranked-FLD ( Figure 9b , gray versus black ) , suggesting that the simulation was effective at capturing relevant aspects of the raw data . Next , we created a simulated ‘visual-modulation-removed’ version of each unit in which the memory signal strength ( as a function of n-back ) and the grand mean spike count response ( across all conditions ) were preserved , but visual selectivity was removed ( see Materials and methods ) . Conceptually , one can think about this simulation as creating a version of each unit with pure selectivity for visual memory in the absence of visual modulation . The FLD behavioral prediction error trajectory of the visual-modulation-removed population fell faster than the replication population and took on approximately the same MSE as the intact population with only 479 ( as compared to 800 ) units for the regular FLD ( Figure 9a , red ) and only 159 ( as compared to 400 ) units for the ranked-FLD ( Figure 9b , red ) . These results suggest that visual selectivity resulted in a substantial increase in the number of units required to account for behavioral performance within the FLD decoding scheme . In sum , at least two factors combined to determine that a large number of FLD decoded IT units ( ~800 ) were required to accurately predict single-exposure behavioral performance . First , the visual memory signals that combined to produce largely accurate behavioral predictions were limited to ~50% of the IT population . Second , as a consequence of visual selectivity , the presentation of an image only activated a subset of units , thus increasing the population size required for robust neural performance that was capable of generalizing to new images . As a final , complementary set of analyses , we focused on the neural correlates of the differences in behavioral patterns reflected between the two animals . From the results presented above , we can infer that this is not a straightforward relationship: while the animal that was better at the task ( monkey 2 , Figure 3a , c ) had stronger average repetition suppression ( Figure 5c–d ) , fewer units were also required to account that animal’s behavior ( 500 versus 800 , Figure 7c–d ) . This suggests that differences in behavioral performance between the two monkeys does not simply reflect two populations that are matched in size but contain neurons whose visual memory signals differ in average strength . For deeper insights into the differences between animals , we performed an analysis in which we attempted to predict each monkey’s behavioral forgetting functions from the other monkey’s neural data using the FLD decoder ( Figure 10a–b ) . For both monkeys , the minimal error ( as a function of population size ) was lower when behavioral and neural data came from the same monkey as compared to when they were mixed between monkeys ( Figure 10a–b , red versus black dots ) and this translated to better PQ when behavioral and neural data came from the same animal versus when they came from different animals ( Figure 10c ) . Figure 10c illustrates the alignment of the behavioral forgetting functions and their neural predictions , after subtracting the false alarm rate for novel images ( similar to 7e ) , shown for the cases in which behavioral and neural data came from the same animal and when they were crossed . In the case of monkey 1 , the neural prediction from the same animal largely captured the pattern of forgetting with n-back , whereas the neural data from monkey two predict a shape that was too flat . In other words , FLD applied to the neural data from monkey two predicted a similar amount of forgetting across a wide range of n-back and this pattern was inconsistent with the steeper fall-off in that same range reflected in the behavior of monkey 1 ( Figure 10c , monkey 1 ‘Cross’ ) . Similarly , the neural data collected from monkey one reflected a considerable amount of forgetting at higher n-back , whereas the behavioral data from monkey two were more flat in this range . This led to a discrepancy between the behavioral data and neural predictions when aligned around the novel image prediction ( Figure 10c , monkey 2 ‘Cross’ ) . While our study was limited to only two subjects and thus lacked the power to establish individual differences , the better alignment of behavioral and neural data within subjects versus across subjects is an effective demonstration that signal strength and population size cannot simply be traded off to fit any possible behavioral function . Additionally , these results provide added support of the hypothesis that single-exposure visual memory behaviors are in fact reflected in the neural responses of IT cortex . This study was designed to test the hypothesis that the signals supporting single-exposure visual recognition memories , or equivalently answers to the question , ‘Have I seen that image before ? ” , are reflected as decrements in the responses of neurons in IT with stimulus repetition ( Fahy et al . , 1993; Li et al . , 1993; Miller and Desimone , 1994; Riches et al . , 1991; Xiang and Brown , 1998 ) . Prior to this study , this hypothesis had received mixed support from human fMRI studies ( Gonsalves et al . , 2005; Turk-Browne et al . , 2006; Ward et al . , 2013; Xue et al . , 2011 ) and was largely untested at the resolution of individual neurons . We found that a strict interpretation of the repetition suppression hypothesis in the form of counting the total numbers of spikes across the IT population provided an incomplete account of single-exposure visual memory behavior ( Figure 7 ) , whereas a weighted linear read-out of IT provided reasonably accurate predictions of the rates of forgetting as a function of time ( Figure 6c , Figure 7e ) , as well as mean reaction time patterns ( Figure 6f; Figure 7—figure supplement 1 ) . Additionally , behavioral predictions could be attributed to IT visual memory signals that were reflected as repetition suppression ( Figure 8 ) and were intermingled with visual selectivity ( Figure 9 ) , but only when combined across the most sensitive 50% of IT units ( Figure 8c–d ) . Our study was focused on changes in IT that follow a single image exposure , and the net repetition suppression that we observed is qualitatively consistent with earlier reports ( Fahy et al . , 1993; Li et al . , 1993; Miller and Desimone , 1994; Riches et al . , 1991; Xiang and Brown , 1998 ) . Net repetition suppression has also been reported following exposure of IT neurons to the same images hundreds or thousands of times ( Anderson et al . , 2008; Baker et al . , 2002; Freedman et al . , 2006; Lim et al . , 2015; Meyer et al . , 2014; Woloszyn and Sheinberg , 2012 ) . However , the suppression that we observed was transient ( ~5 min ) , whereas the suppression that follows many repeated image exposures is much longer lasting . Some studies have reported repetition enhancement in IT for images that are highly familiar , particularly when an image falls at the peak of a neuron’s tuning function and the neuron in question is excitatory ( Lim et al . , 2015; Woloszyn and Sheinberg , 2012 ) . In our study , we found no evidence that net repetition enhancement contributed to behavioral predictions . At the next stage of processing in the medial temporal lobe , perirhinal cortex , there are indications that following many repeated exposures , the sign of familiarity modulation may flip from net suppression to net enhancement ( Landi and Freiwald , 2017; Tamura et al . , 2017 ) . In contrast , following a limited number of exposures , neurons in a region now attributed to perirhinal cortex have been reported to exhibit repetition suppression ( Li et al . , 1993; Miller et al . , 1991 ) . Future work will be required to determine the effects of image familiarity in IT and perirhinal cortex as images transition from novel to highly familiar . Notably , when monkeys are engaged in a task that involves both stimulus repetition as well as a same/different judgment about repeated stimuli , heterogeneous combinations of repetition enhancement and suppression are observed in IT and perirhinal cortex ( Miller and Desimone , 1994; Pagan et al . , 2013; Vogels and Orban , 1994 ) . These results may reflect the fact that the responses of neurons in these brain areas reflect mixtures of the signals supporting visual memory , attention , and decision processes . In fact , considerable evidence supports the notion that the task a subject is engaged in at the time of viewing will have an impact on what will be remembered ( reviewed by Chun and Turk-Browne , 2007 ) . In our study , the targets were present at stimulus onset for the first monkey but delayed until the go cue ( 400 ms ) in the second animal , and poorer performance of monkey one in this task may reflect divided attention between the visual image and the targets . The neural correlates of explicit visual memory reports have been investigated in the human brain using PET ( Vandenberghe et al . , 1995 ) and fMRI ( Gonsalves et al . , 2005; Turk-Browne et al . , 2006; Ward et al . , 2013; Xue et al . , 2010 ) . A number of factors might contribute to the discrepancy between our study and human fMRI studies that fail to find a relationship between repetition suppression magnitudes in high-level visual brain areas and explicit visual memory reports ( Ward et al . , 2013; Xue et al . , 2011 ) . For example , one implication of our results is that near-single unit resolution is required to determine how to appropriately weight IT units to account for single-exposure visual memory behaviors . In contrast , measures that average the responses across large numbers of neurons result in an information loss that cannot fully be recovered ( e . g . via a multi-voxel pattern analysis ) . Another factor that may contribute to differences between our results and those studies is a distinct difference in experimental design: our study correlates repetition suppression and behavioral reports on the same trial , whereas these studies correlate repetition suppression to a second viewing of an image with the behavioral report about remembering during a third viewing . The rationale behind the fMRI design is a desire to dissociate memory processes from the processes involved in decision making and response execution . In our study , we were focused on evaluating the plausibility that the signal supporting visual memory behavioral reports is reflected in IT cortex , as opposed to the plausibility that memory signals are reflected in IT in the absence of a subject being engaged in a memory task . The consistent ( positive ) sign of the linear weights recovered across IT units suggests that our results cannot be accounted for by motor responses , as the task required the monkeys to saccade to two different targets to report novel versus familiar predictions and a motor account would require that all the IT neurons were tuned for the same target ( e . g . ‘upward’ for monkey one and ‘downward’ for monkey 2 ) . Finally , differences between our study and those reports could also arise from differences between species , analogous to the differences reported between monkey IT and human LOC for changes in the representations of highly familiar images as measured with fMRI ( Op de Beeck et al . , 2006; Op de Beeck et al . , 2008 ) . Our results suggest that visual memory signals are reflected as repetition suppression in the majority of IT units and that reports of whether an image has been seen before can be predicted by counting the numbers of spikes across the top half of the repetition suppressed IT subpopulation ( Figure 8e ) . One question not addressed in our experiments is how this type of decoding scheme could tease apart changes in total numbers of spikes due to stimulus repetition from changes in spike numbers due to other variables , such as contrast , luminance , object size , and potentially object identity ( Chang and Tsao , 2017 ) . In principle , the brain could address this by relying on neurons that are sensitive to visual memory but insensitive to these other types of variables . Future work will be required to investigate these issues . Analysis of our reaction time patterns parsed by trial outcome ( correct/error ) revealed a characteristic x-shaped pattern ( Figure 3 ) at odds with the predictions of standard models of decision making such as standard instantiations of the drift diffusion model . Extensions of the drift diffusion framework have been proposed in which reaction time asymmetries on correct versus error trials can be accounted for by adding per-trial noise in the decision variable drift rate or the decision variable start position ( Ratcliff and McKoon , 2008 ) . Our task was not designed to differentiate between these and other similar models , but rather to test the hypothesis that signals reflecting single-exposure visual memories are found in IT cortex . As such , we opted for the much simpler , lower-parameter description suggested by strength theory ( Murdock , 1985; Norman and Wickelgren , 1969 ) . The inverted relationship between proportion correct and reaction time captured by strength theory can loosely be thought of as a signature of confidence ( e . g . when performance is higher , reaction times are faster ) , however , the drawback of strength theory is that it lends little biophysical insight into how this process might happen in the brain . Our study provides important constraints on models of the decision making process for single-exposure memory tasks , and should constrain future work in which this process is investigated more completely . In this study , we adjusted the task parameters such that images were forgotten over minutes within sessions that lasted approximately one hour . This included reducing the viewing time from the longer durations used in previous human behavioral experiments ( 2–3 s ) to ~400 ms . Our results suggest that forgetting rates are well-aligned between behavioral reports and IT neural signals within this regime . Will longer timescale memories be reflected by signals in IT as well ? That remains to be seen . It could be the case that IT reflects single-exposure visual memories across all behaviorally-relevant timescales , alternatively , it could be the case that the signals reflecting single-exposure memories across longer timescales ( e . g . hours and days ) are only reflected in higher brain areas such as perirhinal cortex and/or the hippocampus . A related issue is the question of where and how single-exposure visual memories are stored in the brain . Crucially , it is important to recognize that it does not necessary follow from the fact that a particular brain area reflects a memory signal , that it must be the locus at which storage occurs . It is likely the case that the visual memory signals that we observe are at least partially the consequence of the cumulative adaptation-like processes that happen within IT and within brain areas preceding IT . What is less clear is whether these signals also reflect contributions from higher brain areas as well . Similarly , a computational description of the learning rule ( s ) that accurately capture the changes in the brain that follow a single image exposure remain to be determined . While important first steps toward those computational descriptions have been proposed ( Androulidakis et al . , 2008; Lulham et al . , 2011 ) they have yet to be tested in deep neural network architectures that approximate the patterns of neural activity reflected in the visual system ( e . g . Yamins et al . , 2014 ) . All behavioral training and testing was performed using standard operant conditioning ( juice reward ) , head stabilization , and high-accuracy , infrared video eye tracking . Stimuli were presented on an LCD monitor with an 85 Hz refresh rate using customized software ( http://mworks-project . org ) . As an overview of the monkeys’ task , each trial involved viewing one image for at least 400 ms and indicating whether it was novel , ( never seen before ) or familiar ( seen exactly once prior ) with an eye movement to one of two response targets . Images were never presented more than twice ( once as novel and then as familiar ) during the entire training and testing period of the experiment . Trials were initiated by the monkey fixating on a red square ( 0 . 25° ) on the center of a gray screen , within a square window of ±1 . 5° , followed by a 200 ms delay before a 4° stimulus appeared . The monkeys had to maintain fixation of the stimulus for 400 ms , at which time the red square turned green ( go cue ) and the monkey made a saccade to the target indicating that the stimulus was novel or familiar . In monkey 1 , response targets appeared at stimulus onset; in monkey 2 , response targets appeared at the time of the go cue . In both cases , targets were positioned 8° above or below the stimulus . The association between the target ( up vs . down ) and the report ( novel vs . familiar ) was swapped between the two animals . The image remained on the screen until a fixation break was detected . The images used in these experiments were collected via an automated procedure that explored and downloaded images from the internet , and then scrubbed their metadata . Images smaller than 9696 pixels were not considered . Eligible images were cropped to be square and resized to 256256 pixels . An algorithm removed duplicate images . The resulting database included 89 , 787 images . Within the training and testing history for each monkey , images were not repeated . A representative sample of a subset of 49 images are presented in Figure 2—figure supplement 1 . The specific random sequence of images presented during each session was generated offline before the start of the session . The primary goal in generating the sequence was to select trial locations for novel images and their repeats with a uniform distribution of n-back ( where n-back = 1 , 2 , 4 , 8 , 16 , 32 and 64 ) . This was achieved by constructing a sequence slightly longer than what was anticipated to be needed for the day , and by iteratively populating the sequence with novel images and their repeats at positions selected from all the possibilities that remained unfilled . Because the longest n-back locations ( 64 ) were the most difficult to fill , a fixed number of those were inserted first , followed by systematically working through the insertion of the same fixed number at each consecutively shorter n-back ( 32 , 16 … ) . In the relatively rare cases that the algorithm could not produce that fixed number at each n-back , it was restarted . The result was a partially populated sequence in which 83% of the trials were occupied . Next , the remaining 17% of trials were examined to determine whether they could be filled with novel/familiar pairs from the list of n-back options ( 64 , 32 , 16-back … ) . For the very small number of trials that remained after all possibilities had been extinguished ( e . g . a 3-back scenario ) , these were filled with ‘off n-back’ novel/familiar image pairs and these trials were disregarded from later analyses . ‘Forgetting functions’ ( Figure 3a , c and Figure 6c ) were computed as the mean proportion of trials each monkey selected the familiar target , across all trials and all sessions . Because behavioral outcome is a binary variable , error was estimated by computing the mean performance trace for each session , and then computing the 97 . 5% confidence interval as 2 . 2standard error of those traces . Mean reaction times ( Figure 3b , d and Figure 6f ) were computed as means across all trials and sessions , and 97 . 5% confidence intervals were computed as 2 . 2standard error of those same values . The activity of neurons in IT was recorded via a single recording chamber in each monkey . Chamber placement was guided by anatomical magnetic resonance images in both monkeys . The region of IT recorded was located on the ventral surface of the brain , over an area that spanned 5 mm lateral to the anterior middle temporal sulcus and 14–17 mm anterior to the ear canals . Recording sessions began after the monkeys were fully trained on the task and after the depth and extent of IT was mapped within the recording chamber . Combined recording and behavioral training sessions happened 4–5 times per week across a span of 5 weeks ( monkey 1 ) and 4 weeks ( monkey 2 ) . Neural activity was recorded with 24-channel U-probes ( Plexon , Inc , Dallas , TX ) with linearly arranged recording sites spaced with 100 μm intervals . Continuous , wideband neural signals were amplified , digitized at 40 kHz and stored using the Grapevine Data Acquisition System ( Ripple , Inc . , Salt Lake City , UT ) . Spike sorting was done manually offline ( Plexon Offline Sorter ) . At least one candidate unit was identified on each recording channel , and 2–3 units were occasionally identified on the same channel . Spike sorting was performed blind to any experimental conditions to avoid bias . A multi-channel recording session was included in the analysis if: ( 1 ) the recording session was stable , quantified as the grand mean firing rate across channels changing less than 2-fold across the session; ( 2 ) over 50% of neurons were visually responsive ( a loose criterion based on our previous experience in IT ) , assessed by a visual inspection of rasters; and ( 3 ) the number of successfully completed novel/familiar pairs of trials exceeded 100 . In monkey 1 , 21 sessions were recorded and 6 were removed ( 2 from each of the 3 criterion ) . In monkey 2 , 16 sessions were recorded and 4 were removed ( 1 , 2 and 1 due to criterion 1 , 2 and 3 , respectively ) . The sample size ( number of successful sessions recorded ) was chosen to approximately match our previous work ( Pagan et al . , 2013 ) . Because the data recorded in any individual session ( on 24 channels ) corresponded to a population too small to provide a full account of behavioral performance , we combined data across sessions into a larger pseudopopulation ( see Results ) . We compared the ability of four different linear decoders to predict the monkeys’ behavioral responses from the IT pseudopopulation data . Spikes were counted in a window 150–400 ms following stimulus onset with the exception of Figure 6i , where spikes were counted in a 150 ms bin at sliding positions relative to stimulus onset . For all decoders , the population response x was quantified as the vector of simultaneously recorded spike counts on a given trial . To ensure that the decoder did not erroneously rely on visual selectivity , the decoder was trained on pairs of novel/familiar trials in which monkeys viewed the same image ( regardless of behavioral outcome and for all n-back simultaneously ) . Here we begin by describing each decoder , followed by a description of the cross-validated training and testing procedure that was applied in the same manner to each one . All four decoders took the general form of a linear decoding axis: fx= wTx+b where w is an N-dimensional vector ( and N is the number of units ) containing the linear weights applied to each unit , and b is a scalar value . What differed between the decoders was how these parameters were fit . To estimate performance for larger sized populations than those we recorded , we computed quantified how the mean and standard deviation of the distributions depicted in Figure 6b , as well as the value of the criterion , grew as a function of population size ( Figure 6—figure supplement 1 ) . For both the SCC and FLD , the trajectories of the means and the criterion were highly linear as a function of population size ( Figure 6—figure supplement 1a–b , left ) , whereas the standard deviations plateaued ( Figure 6—figure supplement 1a–b , right ) . We modeled the population response distributions at each n-back ( Figure 6b ) as Gaussian , and we estimated the means and standard deviations of each distribution at different population sizes by extending the trajectories computed from our data to estimates at larger population sizes ( Figure 6—figure supplement 1 dotted lines ) . This process was similar in spirit but differed in detail for each decoder . In the case of the SCC , the mean population response was computed as the grand mean spike count across the population , and consequently did not grow with population size ( Figure 6—figure supplement 1a , left ) . We extended these trajectories with a simple linear fit to the values computed from the data . In contrast , the trajectory corresponding to standard deviation decreased as a function of population size ( Figure 6—figure supplement 1a , right ) and to extend these trajectories , we fit a two-parameter function: ( 5 ) SCC_sd ( x ) = ( ∑1xab ) 1/bwhere x corresponds to population size and the parameters a and b were fit to the data . In the case of the FLD , the population mean was computed as a weighted sum and grew linearly with population size ( Figure 6—figure supplement 1b , left ) . We extended these trajectories with a linear fit to the values computed from the data . In contrast , the trajectories corresponding to the population standard deviations for each n-back grew in a nonlinear manner ( Figure 6—figure supplement 1b , right ) , and we extend them by fitting the 2-parameter function: ( 6 ) FLD_sd ( x ) = ( ax ) bwhere x corresponds to population size and the parameters a and b were fit to the data . For both the SCC and FLD decoders and their threshold variants , we computed behavioral predictions for larger sized populations by replacing the histograms in Figure 6b with Gaussians matched for the means and standard deviations determined by the extended trajectories , relative to the extended estimate for the criterion . To measure the prediction quality of the neural predictions for both the forgetting function and reaction time patterns , we developed a measure that benchmarked the MSE between the behavioral patterns and neural predictions by the worst-possible fit given that our procedure involves a global alignment of behavioral and neural data ( Figure 6g ) . The worst-possible fit was computed as a step function , under the assumptions that performance as a function of n-back should be continuous , have non-positive slope , and be centered around chance . For example , the average proportion correct for the monkey’s pooled behavioral forgetting function ( Figure 6g ) was 84% , and the benchmark was thus assigned as 84% proportion chose familiar for every n-back , and 16% for the novel images . Prediction quality was computed as: ( 7 ) PQ= 100*MSEneural-MSEbenchmarkMSEneuralwhere MSEneural and MSEbenchmark correspond to the MSE between the actual behavioral forgetting function and the neural prediction or the benchmark , respectively . To produce prediction quality estimates for reaction times ( Figure 6f ) , the benchmark forgetting function was passed through the same procedure as the neural prediction to produce benchmarked reaction time predictions ( Figure 6f , dotted ) . PQ was then computed as described in Equation 7 . To estimate the impact of visual selectivity on population size ( Figure 9c ) , we compared FLD and ranked-FLD behavioral prediction error trajectories ( as a function of population size ) for two simulated versions of our data: one that ‘replicated’ each unit and another that corresponded to ‘visual modulation removed’ ( Figure 9 ) . For these simulations , the strength of the visual memory signal for each unit was measured at each n-back as the mean proportional change in the spike count response for the same images presented as novel versus as familiar across all image pairs , and visual memory modulation was modeled as multiplicative . In the case of the ‘replicated’ simulation , the novel and familiar responses to each image were determined by considering the average response to that image when it was novel versus familiar , and adjusting that quantity based on the proportional decrement computed for each n-back . For example , if the proportional decrement at 1-back for a unit was 10% and the unit responded to one image with an average ( across the novel/familiar presentations ) of 6 spikes , the replicated prediction for the novel and familiar presentation would be 6 . 32 spikes and 5 . 69 spikes , respectively ( for a total difference of 0 . 63 spikes ) . If the same unit responded to a different image at 1-back with an average of 3 spikes , the replicated prediction would be 3 . 16 spikes and 2 . 84 spikes for novel and familiar images , respectively ( for a total difference of 0 . 32 spikes ) . The process was repeated for each image by applying the proportional decrement determined for the n-back at which it was presented . These predictions were then converted into spike counts by applying Poisson trial variability . As a verification that this simulation captured the relevant aspects of the data , we compared its FLD behavioral prediction error trajectory to the error trajectory of the intact data ( Figure 9c , gray versus black ) . In the case of the ‘visual modulation removed’ simulation , the process was similar but instead of considering the actual response of the unit to a particular image , visual memory modulation was applied to the grand mean spike count across all images for that unit . A response prediction for each image was determined by applying the proportional decrement determined for the n-back at which it was presented around the grand mean spike count . These predictions were then converted into spike counts by applying Poisson trial variability . Unit d’ was calculated , for each unit , as the difference in the mean responses to the set of images presented as novel versus the set presented as familiar , divided by the average standard deviation across the two sets ( Figure 8a–b ) . To determine the fraction of units that produced responses that differed between novel versus familiar images or between the pre-stimulus and stimulus-evoked period , we computed p-values to evaluate the statistical significance of the observed differences in the mean values via a bootstrap procedure . On each iteration of the bootstrap , we randomly sampled the true values from each population , with replacement , and we computed the difference between the means of the two newly created populations . We computed the p value as the fraction of 1000 iterations on which the difference was flipped in sign relative to the actual difference between the means of the full data set ( Efron and Tibshirani , 1998 ) .	As we go about our daily lives , we store visual memories of the objects and scenes that we encounter . This type of memory , known as visual recognition memory , can be remarkably powerful . Imagine viewing thousands of images for only a few seconds each , for example . Several days later , you will still be able to distinguish most of those images from previously unseen ones . How does the brain do this ? Visual information travels from the eyes to an area of the brain called visual cortex . Neurons in a region of visual cortex called inferotemporal cortex fire in a particular pattern to reflect what is being seen . These neurons also reflect memories of whether those things have been seen before , by firing more when things are new and less when they are viewed again . This decrease in firing , known as repetition suppression , may be the signal in the brain responsible for the sense of remembering . Meyer and Rust have now tested this idea by training macaque monkeys to report whether images on a screen were new or familiar . The monkeys were very good at remembering the images they had seen more recently , although they tended to forget some of the images with time . Then , the rate at which the monkeys forgot the images was compared with the rate at which repetition suppression disappeared in inferotemporal cortex . The results showed that the total number of firing events in this region was not a great predictor of how long the monkeys remembered images . However , a decrease in the number of firing events for a particular subset of the neurons did predict the remembering and forgetting . Repetition suppression in certain inferotemporal cortex neurons can thus account for visual recognition memory . Brain disorders and aging can both give rise to memory deficits . Identifying the mechanisms underlying memory may lead to new treatments for memory-related disorders . Visual recognition memory may be a good place to start because of our existing knowledge of how the brain processes visual information . Understanding visual recognition memory could help us understand the mechanisms of memory more broadly .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Single-exposure visual memory judgments are reflected in inferotemporal cortex
Proteins of the secretin family form large macromolecular complexes , which assemble in the outer membrane of Gram-negative bacteria . Secretins are major components of type II and III secretion systems and are linked to extrusion of type IV pili ( T4P ) and to DNA uptake . By electron cryo-tomography of whole Thermus thermophilus cells , we determined the in situ structure of a T4P molecular machine in the open and the closed state . Comparison reveals a major conformational change whereby the N-terminal domains of the central secretin PilQ shift by ∼30 Å , and two periplasmic gates open to make way for pilus extrusion . Furthermore , we determine the structure of the assembled pilus . Secretins form multimeric pores through the outer membrane of Gram-negative bacteria ( Averhoff , 2009; Korotkov et al . , 2011; Burkhardt et al . , 2012 ) . They are the central secretion conduits for proteins and virulence factors in type II and III secretion systems ( T2SS/T3SS ) and are essential for extrusion of type IV pili ( T4P ) ( Martin et al . , 1993 ) and transport of some bacteriophages ( Korotkov et al . , 2011 ) . In addition , secretins are key components of DNA transport systems , which mediate uptake of free DNA from the environment , referred to as natural transformation ( Schwarzenlander et al . , 2009 ) . The ability to take up DNA is one of the major mechanisms of horizontal gene transfer ( Domingues et al . , 2012 ) and enables organisms to adapt rapidly to changing environments ( Averhoff , 2009 ) . This process is also fundamental for adaptation of pathogenic bacteria to human hosts and the acquisition of multi-drug resistance ( Domingues et al . , 2012 ) . In many Gram-negative bacteria , such as Thermus , and also in many major human pathogens such as Neisseria , Pseudomonas , and Vibrio , DNA uptake is linked to the T4P machinery ( Wolfgang et al . , 1998; Graupner et al . , 2000; Seitz and Blokesch , 2013 ) ( Figure 1 ) . To investigate the structure and function of this system , we chose the thermophilic bacterium Thermus thermophilus HB27 , which exhibits the highest transformation rates known to date ( Koyama et al . , 1986 ) , and due to the thermostability of its proteins is a convenient model for structural studies . 10 . 7554/eLife . 07380 . 003Figure 1 . Schematic of the T4P machinery in T . thermophilus . The type IV pilus machinery is a heterooligomer , formed from at least 10 different proteins . The PilQ secretin ( orange ) forms a channel in the outer membrane for secretion of the pilus-forming protein PilA4 ( green ) , which is processed by the prepillin peptidase PilD ( grey ) ( Friedrich et al . , 2002; Schwarzenlander et al . , 2009 ) . The membrane protein PilW ( light orange ) plays a role in DNA transport , PilQ assembly , and pilus extrusion ( Rumszauer et al . , 2006; Schwarzenlander et al . , 2009 ) . The dimeric complex PilC ( red ) is located in the inner membrane and is essential for pilus formation ( Friedrich et al . , 2002; Karuppiah et al . , 2010 ) . PilM ( light brown ) , PilN ( dark brown ) , and PilO ( beige ) are suggested to form the inner membrane assembly platform and connect the periplasmic and cytoplasmic sides of the complex ( Rumszauer et al . , 2006; Schwarzenlander et al . , 2009; Karuppiah and Derrick , 2011; Karuppiah et al . , 2013 ) . The cytoplasmic ATPases PilF ( bright yellow ) and PilT1/PilT2 ( pale yellow ) drive pilus extension and retraction , respectively ( indicated with red arrows ) ( Rose et al . , 2011; Salzer et al . , 2014b ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07380 . 003 Pili are several micron-long flexible filaments ( Craig and Li , 2008 ) that can generate forces of over 100 pN ( Maier et al . , 2004 ) . T4P are grouped together in a class based on the production and secretion of the major pilin protein PilA4 ( Thermus nomenclature ) , thousands of copies of which form the helical pilus ( Craig et al . , 2004; Schwarzenlander et al . , 2009 ) . The T4P is the only known bacterial pilus that can be retracted rapidly ( Maier et al . , 2004 ) to enable motility and adherence ( Merz et al . , 2000 ) , major contributors to bacterial virulence ( Hahn , 1997 ) . Assembly and disassembly of the pilus is driven by the AAA-ATPases ( ATPases associated with diverse cellular activities ) PilF ( extension ) and PilT1/PilT2 ( retraction ) ( Salzer et al . , 2014b ) . It has been suggested that mature PilA4 assembles into pili extending from the inner membrane by action of PilF ( Collins et al . , 2013; Salzer et al . , 2014b ) . The outer membrane channel of the T4P machinery is formed by the dodacemeric ∼1 MDa secretin complex PilQ ( Burkhardt et al . , 2011 ) ( Figure 1 ) . Other proteins , in particular PilM , PilN , and PilO , are hypothesized to be a central part of the pilus assembly platform and may couple the cytoplasmic and periplasmic sides of the T4P machinery ( Karuppiah et al . , 2013 ) . Some proteins of the complex have been implicated to play a dual role in both pilus assembly and natural competence ( Friedrich et al . , 2002; Averhoff and Friedrich , 2003; Friedrich et al . , 2003; Rumszauer et al . , 2006 ) . Recent results indicate that T4P themselves are not directly involved in DNA uptake ( Burkhardt et al . , 2012; Salzer et al . , 2014a ) . T4P are essential for pathogenesis by mediating adhesion , biofilm formation , and twitching motility ( Burrows , 2012 ) . Thus , both secretins and T4P play important roles in virulence of different pathogenic bacteria , which has fostered their use as new targets for drug development ( Baron , 2010 ) . To date , there is no information on the in situ structure of either the T4P machinery or DNA translocator . Determining structures of T4P complexes in whole bacterial cells is therefore of paramount importance and will enable further study of bacterial resistance and disease . Electron cryo-tomography ( cryoET ) has the unique ability to determine protein structures in cells at molecular resolution . We have applied cryoET to whole T . thermophilus HB27 cells , in order to visualize the T4P machinery in situ . We determine the helical structure of the pilus and find that the secretin complex PilQ is a central dynamic component of this system . CryoET and subtomogram averaging of the T4P machinery with and without pili reveal a ∼30 Å conformational change as the gates in the complex open . T . thermophilus has an unusual cell architecture with deep surface clefts , formed by invaginations of the outer membrane ( Figure 2A ) . By cryoET , these clefts are seen to be constrictions that run around the cell body ( Figure 2D ) . Distal to the most polar outer membrane ring , numerous fibrous and straight pili extend from the cell ( Figure 2A–D ) . The pili are clearly associated with large protein complexes crossing the ∼70 nm periplasm ( Figure 2B , C ) . This distribution is in line with previous fluorescence and electron microscopy data , which demonstrate the polar localization of PilQ ( Seitz and Blokesch , 2013 ) and pili ( Salzer et al . , 2014c ) . 10 . 7554/eLife . 07380 . 004Figure 2 . Cell morphology and pili of T . thermophilus . ( A–C ) Tomographic slices through T . thermophilus cells show invaginations in the outer membrane and large protein complexes crossing the periplasm ( white arrowheads ) , which are associated with pili . Scale bar = 500 nm in A , 100 nm in B and C . ( D ) Volume rendering shows the distribution of pili ( multi-coloured ) , protruding from the outer membrane ( pale yellow ) . A concentric invagination of the outer membrane is indicated ( white arrowheads ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07380 . 004 The periplasm of T . thermophilus cells is extraordinarily wide ( Quintela et al . , 1995; Castan et al . , 2002 ) and too dense to select subvolumes for subtomogram averaging reliably . Therefore , cells were treated with 100 mM ethylenediaminetetraacetic acid ( EDTA ) and pipetted in order to render the outer membrane leaky ( Caston et al . , 1988 ) . This had the desired effect of depleting the periplasm of most small proteins . Sample preparation by this method also removed the pilus from the complex , the empty T4P machinery was nonetheless still clearly visible ( Figure 3A , B ) . We determined by subtomogram averaging the structure of the entire complex and found features distinct from those of the T2SS and T3SS secretins ( Marlovits et al . , 2006; Hodgkinson et al . , 2009; Reichow et al . , 2010 ) . The resolution obtained by averaging ∼4000 particle subvolumes was ∼35 Å ( Figure 3—figure supplement 1 ) , most likely limited by the inherent flexibility of the complex ( Burkhardt et al . , 2011 ) and the difficulty of correcting precisely the contrast transfer function ( CTF ) for thick specimens . 10 . 7554/eLife . 07380 . 005Figure 3 . Structure of the T4P machinery in the closed state . ( A and B ) Tomographic slices of T . thermophilus cells show large protein complexes crossing the periplasm in the absence of pili ( white arrowheads ) . Scale bars = 100 nm . ( C ) Resulting subtomogram average ( left panel ) and its 2D projection ( centre ) are compared to the previously determined projection map of isolated and stained PilQ ( right panel ) ( Burkhardt et al . , 2011 ) . The contrast of the stained PilQ has been inverted . This image was originally published in The Journal of Biological Chemistry . Janin Burkhardt , Janet Vonck , and Beate Averhoff . Structure and Function of PilQ , a Secretin of the DNA Transporter from the Thermophilic Bacterium T . thermophilus HB27 . JBC . 2011; 286:9977–9984 , the American Society for Biochemistry and Molecular Biology . The putative N0–N5 domains of PilQ ( Burkhardt et al . , 2012 ) are marked . ( D ) 3D surface rendering of the average reveals that PilQ has a periplasmic vestibule closed at both ends by two gates . Additional protein densities distinct from PilQ ( green arrowheads; C1 = proximal to the cytoplasmic membrane , P1 = central periplasmic ring 1 , P2 = central periplasmic ring 2 ) are also shown . OM , outer membrane; PG , peptidoglycan; CM , cytoplasmic membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 07380 . 00510 . 7554/eLife . 07380 . 006Figure 3—figure supplement 1 . Fourier shell correlation curves for subtomogram averages . Resolution estimates were based on conventional Fourier shell correlation ( FSC ) measurements and the 0 . 5 criterion in IMOD . Calculations for the closed state of the T4P machinery ( orange , 35 Å ) , open state with pilus extended ( yellow , 45 Å ) , and the pilus ( green , 32 Å ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07380 . 006 Subtomogram average maps show the central protein channel ( ∼35 nm long and ∼15 nm wide ) made up of several ring-shaped domains inserted into the outer membrane ( Figure 3C , left panel ) . We compared a 2D projection of the channel part of the subtomogram average with projections of purified , negatively stained PilQ only ( Burkhardt et al . , 2011 ) ( Figure 3C , central and right panels ) . PilQ was seen to consist of a C-terminal trapezoid ‘cone structure’ with staggered rings in the N-terminal domain ( Burkhardt et al . , 2011 ) , in excellent agreement with our in situ structure . Moreover , the new cryoET structure shows additional protein densities extending from the putative N0 domain of PilQ through the peptidoglycan layer ( P1 and P2 ) to the cytoplasmic membrane ( C1 ) ( Figure 3D ) . Candidate proteins include PilW , which is associated with the inner and outer membranes and is essential for the outer membrane localization of PilQ ( Rumszauer et al . , 2006 ) , and PilO/PilN heterodimers that could connect PilQ to the ATPases by PilM in the cytoplasm ( Karuppiah and Derrick , 2011; Karuppiah et al . , 2013 ) ( Figure 1 ) . These proteins are most likely connected to one another by flexible domains that are not well contrasted in the subtomogram average . A longitudinal slice through the complex reveals that PilQ has two gates , which are closed in the absence of a pilus ( Figure 3C , left panel and Figure 3D , right panel ) . Gate 1 is formed by the ‘cone’ in the outer membrane and gate 2 by the N1 domain at the base of PilQ , enclosing an empty periplasmic vestibule ( Figure 3D , right panel ) . The C-terminal ‘cone’ has been shown to form a sodium dodecyl sulfate ( SDS ) -stable sub-domain ( Burkhardt et al . , 2011 ) , thus , it is plausible that gate 1 is responsible for maintaining the integrity of the cell membrane in the closed state . A second gate formed by the N-terminal domains has not been observed in other secretins ( Korotkov et al . , 2011 ) . The N-terminus of PilQ forms an unusual βββαββ fold , different from the conserved ring-building βαββα folds ( Burkhardt et al . , 2012 ) . Thus , we hypothesize that this motif may form part of gate 2 . To determine the structure of the T4P machinery in the open state , the pipetting step during sample preparation was omitted ( ‘Materials and methods’ ) , which reduced the shearing forces and kept the assembled pilus intact ( Figure 4A , B ) . Close inspection of the pilus shows a periodic structure ( Figure 4C ) , which was suitable for subtomogram averaging . Since the ice surrounding the pili was thin ( ∼100 nm ) , it was possible to apply CTF-correction . The structure of the pilus was determined at ∼32 Å resolution by averaging 740 subvolumes ( Figure 3—figure supplement 1 ) . Power spectra calculated from a single tomographic slice or from the subtomogram average revealed a repeat distance of ∼4 . 9 nm ( Figure 4C , lower panels ) . The T . thermophilus pilus forms a right-handed helix of ∼3 nm diameter ( Figure 4D ) , which is different from the previously determined electron cryo-microscopy structure of the isolated Neisseria gonorrhoeae T4P ( Craig et al . , 2006 ) . Diameters of pili can vary considerably and pilin proteins have limited sequence similarity ( Craig and Li , 2008 ) , which likely accounts for this difference . The result may prompt a reassessment of the functional roles of different T4P in cells . Due to the more challenging sample preparation method , the structure of the open T4P machinery with the pilus extended was determined from only ∼300 particles at ∼45 Å resolution ( Figure 3—figure supplement 1 ) . This resolution is sufficient to reveal the central protein channel complex with the ∼3 nm pilus protein density in the centre ( Figure 4E ) . An extensive conformational change is evident that shifts PilQ domains N0–N3 away from N4/N5 by ∼30 Å towards the cytoplasmic membrane . This change opens the periplasmic vestibule to make way for the pilus ( Figures 4E , 5A , B and Video 1 ) . The shape and dimensions of the pore change from a tapered form ranging in width from ∼4 nm ( at N5 ) to ∼8 nm ( at N2/N3 ) in the closed state ( Figure 3C ) , to a roughly constant 7 nm-wide channel in the open state ( Figure 4E ) . In comparison , the closed state of Neisseria meningitidis PilQ is ∼9 nm wide tapering to a point ( Collins et al . , 2004 ) , which would be sufficient to accommodate the wider Neisseria pilus ( Collins et al . , 2003; Craig et al . , 2006 ) after a comparably large conformational change . Additional conformational changes and shifts also occur in protein densities P1 , P2 , and C1 ( Figure 5A ) . In the open state , we observe an extra protein density in the cytoplasm ( Figure 4E , yellow arrowheads and Video 1 ) , which we speculate could be PilF , linked to the inner membrane platform via PilM ( Karuppiah and Derrick , 2011 ) ( Figure 1 ) . Homologous proteins have been shown to interact with the cytoplasmic ATPases , stimulating their activity ( Lu et al . , 2013 , 2014 ) . Because EDTA treatment caused a depletion of periplasmic protein , we cannot exclude the possibility that some proteins may have been removed from the T4P machinery . However , we also averaged complexes from a tomogram of the two cells shown in Figure 2 , which contained both open ( with pili ) and closed ( without pili ) complexes and where the periplasmic protein density was considerably higher . These subtomogram averages ( Figure 5—figure supplement 1 ) show clearly that the same large conformational changes occur in the T4P machinery , irrespective of the degree of periplasmic protein depletion , and hence of the effect of EDTA . 10 . 7554/eLife . 07380 . 007Figure 4 . Structure of the T4P and the assembled machinery in the open state . ( A and B ) Tomographic slices show close-up views of the T4P machinery with assembled pili ( white arrowheads ) . Scale bar = 50 nm . ( C ) Upper panel , a tomographic slice of the pilus shows that the structure is periodic . Scale bar = 20 nm . A slice through the subtomogram average ( inset ) shows the repeat more clearly . Scale bar = 5 nm . Lower panels , power spectra of the tomographic slice ( left ) and the average ( right ) depict layer lines ( orange arrows ) at a distance of 1/ ( 49 Å ) from the equator ( green arrow ) , corresponding to a helical pitch of 4 . 9 nm . The contrast has been inverted . ( D ) Subtomogram average of the ∼3 nm wide T . thermophilus pilus ( green , top panel ) , compared to the previously determined ∼6 nm wide structure from N . gonorrhoeae ( blue , bottom panel , EMDB 1236 ) ( Craig et al . , 2006 ) . ( E ) Subtomogram average ( left panel ) and 3D surface rendering ( right panel ) of the T4P machinery in the open state with the central pilus ( green ) . The putative N0–N5 ( Burkhardt et al . , 2012 ) domains of PilQ are marked . Additional protein densities distinct from PilQ ( green arrowheads; C1 = proximal to the cytoplasmic membrane , P1 = central periplasmic ring 1 , P2 = central periplasmic ring 2 ) are also shown . Compared to the closed state of the complex , additional protein densities ( left panel , yellow arrowheads ) are visible in the cytoplasm . See Video 1 for details . OM , outer membrane; PG , peptidoglycan; CM , cytoplasmic membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 07380 . 00710 . 7554/eLife . 07380 . 008Figure 5 . Changes between the open and closed state of the T4P machinery and its distribution in situ . ( A and B ) Comparisons between the PilQ components of the T4P machinery reveal large conformational changes whereby both gates open and domains N0–N3 ( now shown in blue for both states ) shift by ∼30 Å towards the cytoplasm on pilus extrusion . Green arrowheads indicate additional protein densities ( C1 = proximal to the cytoplasmic membrane , P1 = central periplasmic ring 1 , P2 = central periplasmic ring 2 ) . In ( B ) the structure of the T4P has been docked into the open state . OM , outer membrane; PG , peptidoglycan; CM , cytoplasmic membrane . ( C and D ) Docking subtomogram averages ( purple ) into the tomographic volume of a cell reveals the distribution of the closed T4P machinery in situ with respect to the outer membrane ( pale yellow ) and cytoplasmic membrane ( blue ) . See Video 2 for details . ( E ) Averaged histogram of nearest-neighbour distance between protein complexes , calculated from 9 cells , with a total of 332 data points . Error bars indicate the standard deviation of the frequency distribution for each minimal distance . DOI: http://dx . doi . org/10 . 7554/eLife . 07380 . 00810 . 7554/eLife . 07380 . 009Figure 5—figure supplement 1 . Subtomogram averages from cells with high periplasmic protein content . Subtomogram averages were calculated for the open and closed state of the T4P machinery from the tomogram shown in Figure 2A–C , where the periplasm contains high-protein levels . The resolution of the subtomogram averages is limited due to the low number of protein complexes available from a single tomogram . Therefore , data for PilQ and protein P2 ( green arrowhead ) only are shown , which indicate essentially the same large conformational change in domains N0–N3 ( blue ) as seen in Figure 5A , B . DOI: http://dx . doi . org/10 . 7554/eLife . 07380 . 00910 . 7554/eLife . 07380 . 010Video 1 . Comparison of the T4P machinery in the open and closed state . The video was generated by morphing the two subtomogram averages in ImageJ . DOI: http://dx . doi . org/10 . 7554/eLife . 07380 . 010 Docking the structures of the closed complex back into the tomographic volume reveals their three-dimensional distribution in the cell ( Figure 5C , D and Video 2 ) . We find that the T4P machinery tends to be tightly packed , with an inter-particle distance of 30–40 nm ( Figure 5E ) . When the pili were depleted by pipetting , each cell contained on average 33 ± 19 closed complexes . However , T . thermophilus assembles ∼6 pili per cell ( Salzer et al . , 2014a , 2015 ) , suggesting that ∼80% of the complexes are not involved in pilus formation under standard growth conditions ( Salzer et al . , 2014c ) . We speculate that these idle complexes may form a second class of transporter , which may be active in DNA uptake or protein secretion . 10 . 7554/eLife . 07380 . 011Video 2 . Distribution of T4P complexes in situ . The video shows the rendered tomographic volume of a T . thermophilus cell . The outer membrane ( pale yellow ) , peptidoglycan ( orange ) , and cytoplasmic membrane ( blue ) are shown with the closed T4P complexes ( purple ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07380 . 011 Taken together , our results demonstrate that the DNA translocator protein PilQ forms a dynamic central component of the T4P machinery in T . thermophilus . The length of the complex across the wide periplasm may be an adaptation to the thermophilic environment that Thermus thrives in . However , core components of the T4P machinery are conserved in bacteria ( Nudleman and Kaiser , 2004 ) , and thus , we speculate that the overall architecture may be similar . Our findings will enable further structure-function studies of the proteins that comprise this elaborate and important macromolecular machine . T . thermophilus HB27 was grown in TM+ medium ( 8 g/l tryptone , 4 g/l yeast extract , 3 g/l NaCl , 0 . 6 mM MgCl2 0 . 17 CaCl2 ) ( Oshima and Imahori , 1971 ) . Cells from a 24-hr pre-culture were transferred onto TM+ plates ( containing 2% [wt/vol] agar ) and incubated under humid conditions for 48 hr at 68°C . To determine the structure of the closed complex , cubes of agar with growing T . thermophilus cells were cut out and placed into buffer containing 20 mM Tris pH 7 . 4 , 100 mM EDTA and gently agitated for 1 hr at room temperature . Samples were mixed 1:1 with 10 nm protein A-gold ( Aurion , Wageningen , The Netherlands ) as fiducial markers and applied to glow-discharged R2/2 Cu 300 mesh holey carbon-coated support grids ( Quantifoil , Jena , Germany ) by gentle pipetting . For the structure of the open complex , cells were treated with EDTA as above , then protein A-gold was added and grids dipped into the solution without the pipetting step . Grids were blotted for ∼4 s in a humidified atmosphere and plunge-frozen in liquid ethane in a home-made device . Grids were maintained under liquid nitrogen and transferred into the electron microscope at liquid nitrogen temperature . Tomograms were typically collected from +60° to −60° at tilt steps of 2° and 5–7 μm underfocus , using either a Tecnai Polara or Titan Krios microscope ( FEI , Hillsboro , USA ) , both equipped with field-emission guns operating at 300 kV and Quantum energy filters ( Gatan , Pleasanton , USA ) operated at a slit width of 20 eV . Both instruments were fitted with K2 Summit direct electron detector cameras ( Gatan , Pleasanton , USA ) . Dose fractionated data ( 3–5 frames per projection image ) were collected using Digital Micrograph ( Gatan , Pleasanton , USA ) at a nominal magnification of 34 , 000× ( corresponding to a pixel size of 0 . 66 nm ) in the Polara or at 33 , 000× ( corresponding to a pixel size of 0 . 42 nm ) in the Krios . The total dose per tomogram was ∼140e−/Å2 . Tomograms were aligned using gold fiducial markers and volumes reconstructed by weighted back-projection using the IMOD software ( Boulder Laboratory , Boulder , USA ) ( Kremer et al . , 1996 ) . Contrast was enhanced by non-linear anisotropic diffusion ( NAD ) filtering in IMOD ( Frangakis and Hegerl , 2001 ) . Segmentation was performed using AMIRA ( FEI , Hillsboro , USA ) . Data collected at 34 , 000× and 8 μm underfocus on the Tecnai Polara were used to calculate the subtomogram averages shown in Figure 5—figure supplement 1 . Subtomogram averages of the T4P machinery shown in all other figures were calculated from data collected at 33 , 000× and 7 μm underfocus on the Titan Krios microscope . Coordinates corresponding to the outer membrane and inner membrane domains of the complex were marked manually in IMOD ( Kremer et al . , 1996 ) . Subvolumes from twice-binned tomograms were then extracted from NAD filtered ( Frangakis and Hegerl , 2001 ) data and an initial alignment and averaging performed in SPIDER ( Frank et al . , 1996 ) . This average was used as a reference for alignment and refinement using PEET ( Nicastro et al . , 2006 ) . We have previously shown that T . thermophilus PilQ is a dodecamer by biochemical analysis ( Burkhardt et al . , 2011 ) , which is supported by single-particle electron microscopy of N . meningitidis PilQ ( Collins et al . , 2003 , 2004 ) . Therefore , 12-fold symmetry was applied to the core complex by 30° ( 360°/12 subunits ) rotation of each subvolume prior to the alignment search . The final averages were obtained from 3984 particles for the closed complex and 312 particles for the open complex , using a mask drawn around PilQ . Any duplicates due to oversampling were removed in PEET ( Nicastro et al . , 2006 ) . Due to the larger sample size , data for the closed complex were replaced by unfiltered tomograms . For 2D comparisons between states ( Figure 3C , left panel and Video 1 ) , the NAD filtered version with low-contrast noise removed is shown . For the pilus , 740 subvolumes of ∼2 nm length were selected in IMOD ( Kremer et al . , 1996 ) , from unbinned CTF-corrected tomograms collected at 5 μm underfocus on the Titan Krios microscope . Subvolumes were aligned and averaged with a cylindrical mask and any duplicates due to oversampling were removed in PEET ( Nicastro et al . , 2006 ) . Resolution estimates were obtained using conventional ‘even/odd’ Fourier shell correlation ( FSC ) , applying the 0 . 5 FSC criterion . A mask was drawn around the protein to exclude membrane and peptidoglycan from this estimate . The averages in Figure 5—figure supplement 1 were smoothed by Gaussian filtering in UCSF Chimera , which was used to draw all surface views and remove low-contrast background noise using the ‘hide dust’ tool ( Pettersen et al . , 2004 ) . The morph shown in Video 1 was produced in ImageJ ( Schneider et al . , 2012 ) . All subtomogram averages were uploaded to the EMDataBank ( http://www . emdatabank . org ) with ID codes 3021 ( closed state , filtered ) , 3022 ( closed state , unfiltered ) , 3023 ( open state , filtered ) and 3024 ( pilus ) . Images of pili were cut out of a twice-binned tomographic slice and the unbinned subtomogram average ( shown in Figure 4C ) using helixboxer in EMAN2 ( Tang et al . , 2007 ) . Power spectra were then calculated using the Iterative Helical Real Space Reconstruction ( IHRSR++ ) software ( Egelman , 2007 ) . The distance between PilQ complexes was determined with a MATLAB ( Mathworks , California , USA ) script ( Gold et al . , 2014 ) . The centroid coordinates of complexes selected for subtomogram averaging were loaded into MATLAB and the distances calculated in an iterative for for-loop according to Pythagoras' theorem . This was performed for 332 closed PilQ complexes , combined from 9 different cells . Averaged histograms were calculated to depict the mean frequency of occurrence for each minimal distance . To account for the different numbers of PilQ complexes in each data set , the mean frequency was calculated as a percentage .	Gram-negative bacteria can cause serious diseases in humans , such as cholera and bacterial meningitis . These bacteria are surrounded by two membranes: an inner membrane and an outer membrane . Proteins called secretins are components of several large molecular complexes that are embedded within the outer membrane . Some secretin-containing complexes form pores in the bacterial membranes and allow molecules to pass in or out of the cell . Some secretins also form part of the machinery that allow Gram-negative bacteria to grow fibre-like structures called type IV pili . These pili help bacteria that cause infections to move and stick to host cells , where they can also trigger massive changes in the host cells' architecture . Multiple copies of a secretin protein called PilQ form a channel in the outer membrane of the bacteria that allows a type IV pilus to grow out of the surface of the cell . The pilus can then hook the bacteria onto surfaces and other cells . There is evidence to suggest the type IV pilus machinery is involved in the uptake of DNA from other bacteria , an important but poorly understood process that has contributed to the spread of multi-drug resistance . Now , Gold et al . have used a cutting-edge technique called ‘electron cryo-tomography’ to analyse the three-dimensional structure of the machinery that builds the type IV pili in the membranes of a bacterium called Thermus thermophilus . This analysis revealed that , similar to many other channel complexes , the PilQ channel can be ‘open’ or ‘closed’ . When pili are absent , the channel is closed , but the channel opens when pili are present . Further analysis also revealed the structure of an assembled pilus . Next , Gold et al . studied the open state of the type IV pilus in more detail and observed that a region of each of the PilQ proteins moves a considerable distance to make way for the pilus to enter the central pore . These results will pave the way for future studies of type IV pili and other secretin-containing complexes and underpin efforts to investigate new drug targets to combat bacterial infections .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "short", "report", "structural", "biology", "and", "molecular", "biophysics" ]	2015	Structure of a type IV pilus machinery in the open and closed state
ANE syndrome is a ribosomopathy caused by a mutation in an RNA recognition motif of RBM28 , a nucleolar protein conserved to yeast ( Nop4 ) . While patients with ANE syndrome have fewer mature ribosomes , it is unclear how this mutation disrupts ribosome assembly . Here we use yeast as a model system and show that the mutation confers growth and pre-rRNA processing defects . Recently , we found that Nop4 is a hub protein in the nucleolar large subunit ( LSU ) processome interactome . Here we demonstrate that the ANE syndrome mutation disrupts Nop4’s hub function by abrogating several of Nop4’s protein-protein interactions . Circular dichroism and NMR demonstrate that the ANE syndrome mutation in RRM3 of human RBM28 disrupts domain folding . We conclude that the ANE syndrome mutation generates defective protein folding which abrogates protein-protein interactions and causes faulty pre-LSU rRNA processing , thus revealing one aspect of the molecular basis of this human disease . Ribosomes are essential for life . The fundamental cellular process of ribosome assembly requires the coordinated action of all three RNA polymerases , over 200 biogenesis factors , and a number of small nucleolar RNAs ( Thomson et al . , 2013; Woolford and Baserga , 2013; Fernández-Pevida et al . , 2015 ) . In yeast , ribosome biogenesis initiates in the nucleolus with the transcription of the 35S polycistronic pre-ribosomal RNA ( rRNA ) precursor by RNA polymerase I . The 35S pre-rRNA undergoes a number of cleavage and modification events to give rise to the mature 18S , 5 . 8S and 25S rRNAs . Mutations that partially disrupt ribosome assembly or function are often deleterious and can lead to disease in humans . Collectively , these diseases of ribosome biogenesis are called ribosomopathies . Ribosomopathies are caused by mutations in proteins that function in all stages of ribosome assembly ( McCann and Baserga , 2013; Armistead and Triggs-Raine , 2014 ) . A homozygous missense mutation in the nucleolar protein RBM28 causes the ribosomopathy , alopecia , neurological defects and endocrinopathy ( ANE ) syndrome ( Nousbeck et al . , 2008 ) . Five affected children of a consanguineous kindred displayed degrees of baldness , mental retardation , motor deterioration , and reduced pituitary gland function in their second decade of life ( Nousbeck et al . , 2008; Warshauer et al . , 2015 ) . The mutation segregated in the family in an autosomal recessive manner and was mapped to a single leucine to proline amino acid substitution at position 351 ( L351P ) of RBM28 . This residue resides in the first α-helix of its third RNA recognition motif ( RRM3; Figure 1A , Figure 1—figure supplement 1 ) . ANE syndrome was classified as a ribosomopathy because RBM28 is localized to the nucleolus and because patient fibroblasts showed reduced numbers of ribosomes ( Damianov et al . , 2006; Nousbeck et al . , 2008 ) . While the L351P mutation is predicted to disrupt the first α-helix of RRM3 and thereby impair RBM28 function , it is not known how this single amino acid substitution disrupts the normal function of RBM28 in the nucleolus . Specifically , what is the molecular basis of ANE syndrome pathogenesis ? 10 . 7554/eLife . 16381 . 003Figure 1 . The ANE syndrome mutation confers a growth defect in yeast . ( A ) The leucine that is mutated in ANE syndrome is highly conserved . Top: Diagram of the domain structure for human RBM28 and its yeast ortholog , Nop4 . The boxes represent RNA Recognition Motifs ( RRMs ) . Arrowheads indicate the approximate location of the mutated amino acid , L351P in humans and L306P in yeast . Bottom: Multiple sequence alignment of the portion of RRM3 containing the mutated leucine . Shaded amino acids in are conserved . A box outlines the conserved leucine that is mutated in ANE syndrome . ( B ) Schematic of the yeast strain used for testing the ANE syndrome mutation in Nop4 . Endogenous Nop4 was placed under the control of the inducible GAL4 promoter in haploid yeast . FLAG-tagged unmutated Nop4 WT or Nop4 L306P was expressed constitutively from the p414GPD plasmid . ( C ) Nop4 WT and Nop4 L306P are expressed at equivalent levels from the yeast expression vector p414GPD-3xFLAG-GW . The depletion of endogenous Nop4 protein was confirmed by western blot using an HRP-conjugated monoclonal antibody against the 3xHA tag . Expression of Nop4 WT or Nop4 L306P from p414GPD-3xFLAG-GW was analyzed by western blot using a monoclonal antibody against the 3xFLAG tag . As a loading control , a western blot using α-Mpp10 was performed . The expression levels of Nop4 WT and Nop4 L306P relative to Mpp10 were quantitated and normalized to Nop4 WT: Nop4 WT = 1 , Nop4 L306P = 0 . 96 . EV = empty vector . The arrows indicate the expected bands . The arrowhead indicates an Mpp10 species only observed when yeast are grown in galactose and raffinose . The asterisk denotes degradation . ( D ) The ANE syndrome mutation in Nop4 impairs growth on solid medium . Serial dilutions of yeast expressing the indicated Nop4 constructs were grown on solid medium for 3 days at 30°C and 37°C or for 5 days at 23°C and 17°C . Three biological replicates were performed starting with transformation of the plasmids into the yeast strain . ( E ) The Nop4 ANE syndrome mutation impairs growth in liquid medium . Yeast expressing the indicated Nop4 constructs were transferred from SG/R-Trp to SD-Trp and 23°C to deplete the endogenous Nop4 . Growth was monitored for 48 hr by measuring the absorbance at OD600 . The log2 of the OD600 was plotted over time and the slope was used to estimate the doubling time . Four biological replicates were performed starting with transformation of the plasmids into the yeast strain . DOI: http://dx . doi . org/10 . 7554/eLife . 16381 . 00310 . 7554/eLife . 16381 . 004Figure 1—figure supplement 1 . Multiple sequence alignment of RRM3 from RBM28 . The amino acids corresponding to RRM3 from human RBM28 , yeast Nop4 and RBM28 from M . mulatta , M . musculus , X . tropicalis and D . rerio were aligned . The conserved structural elements of a canonical RRM are indicated above the alignment . The shaded amino acids are conserved . The conserved leucine that is mutated in ANE syndrome is outlined with a box . DOI: http://dx . doi . org/10 . 7554/eLife . 16381 . 00410 . 7554/eLife . 16381 . 005Figure 1—figure supplement 2 . RBM28 complements the growth defect in yeast observed upon depletion of its essential ortholog , Nop4 . ( A ) Nop4 and RBM28 are expressed from the yeast expression vector p414GPD-3xFLAG-GW . Expression of Nop4 ( 78 kDa ) or RBM28 ( 86 kDa ) from p414GPD-3xFLAG-GW was analyzed by western blot using a monoclonal antibody against the 3xFLAG tag . As a loading control , a western blot using α-Mpp10 was performed . The arrowhead indicates an Mpp10 species only observed when yeast are grown in galactose and raffinose . The expression levels of Nop4 and RBM28 relative to Mpp10 were quantitated and normalized to Nop4: Nop4 = 1 , RBM28 = 0 . 4 . ( B ) RBM28 complements growth on solid medium . Serial dilutions of yeast expressing either Nop4 or RBM28 were grown on solid medium for 3 days at 30°C and 37°C or for 5 days at 23°C and 17°C . DOI: http://dx . doi . org/10 . 7554/eLife . 16381 . 00510 . 7554/eLife . 16381 . 006Figure 1—figure supplement 3 . Amino acid sequence alignment of human RBM28 and its yeast ortholog Nop4 . The complete amino acid alignment of human RBM28 protein and its yeast ortholog , Nop4 , was created using ClustalX . The conserved leucine that is mutated to proline in ANE syndrome , L351 in RBM28 and L306 in Nop4 , is circled in black . RRM domains are indicated by the labeled boxes . Below the alignment , the conservation score for each column is plotted . A high score indicates high conservation whereas a low score indicates low conservation . DOI: http://dx . doi . org/10 . 7554/eLife . 16381 . 006 Nop4 , the yeast ortholog of RBM28 , is required for assembly of the large ribosomal subunit ( LSU; Bergès et al . , 1994; Sun and Woolford , 1994; Nousbeck et al . , 2008 ) . Recently , the LSU processome interactome revealed that Nop4 functions as a hub protein , interacting with many more proteins than average within the LSU processome ( McCann et al . , 2015 ) . We hypothesized that introduction of the orthologous ANE syndrome mutation into Nop4 ( L306P ) would disrupt Nop4’s function as a hub protein and therefore disrupt LSU assembly in the nucleolus . We demonstrate that introduction of the ANE syndrome mutation into Nop4 disrupts growth and pre-rRNA processing in yeast and abrogates several , but not all , of its protein-protein interactions . Surprisingly , the C-terminal half of Nop4 , where the ANE syndrome mutation occurs , is necessary and sufficient for hub protein function , cell growth and pre-rRNA processing . Consistent with these findings , circular dichroism and NMR reveal that the ANE mutation in RRM3 of the Nop4 human ortholog , RBM28 , disrupts folding of the entire domain , not just the first α-helix . Together , these results suggest that the molecular basis of ANE syndrome lies in defective protein folding that reduces protein interactions and the function of RBM28 as a hub protein , resulting in pre-rRNA processing defects in the nucleolus . Our goal was to elucidate the molecular basis of the ribosomopathy ANE syndrome , which is attributed to a single amino acid substitution , L351P , in RBM28 ( Nousbeck et al . , 2008 ) . Human RBM28 can complement the growth defect in the yeast , Saccharomyces cerevisiae , when its ortholog , the essential Nop4 protein , is depleted ( Figure 1—figure supplement 2; Kachroo et al . , 2015 ) . Therefore , we used yeast genetics to pinpoint the molecular basis of ANE syndrome . A ClustalX alignment of the yeast Nop4 and human RBM28 amino acid sequences permitted identification of the orthologous ANE syndrome mutation in Nop4 ( Figure 1—figure supplement 3 ) . Over their entire length , the amino acid sequences of Nop4 and RBM28 are ~26% identical and 34% similar , and both contain four RRMs ( Figure 1A ) . The ANE syndrome mutation in human RBM28 , L351P , is within the third RRM . Inspection of the alignment revealed an orthologous leucine in the third RRM of yeast Nop4 , which we mutated to proline to introduce the ANE syndrome mutation into Nop4 ( L306P; Figure 1A; Figure 1—figure supplement 1 ) . We determined that the ANE syndrome mutation in Nop4 ( L306P ) impaired yeast growth . We generated a strain where endogenous NOP4 is under the control of a galactose-inducible , glucose-repressible promoter and tagged with a triple-HA epitope ( Figure 1B ) . Unmutated ( wild type; WT ) Nop4 or Nop4 L306P protein is tagged with a triple-FLAG epitope and constitutively expressed from a plasmid ( p414GPD ) . Western blotting of total protein demonstrated that after growth of this strain in glucose for 48 hr at 23°C , the endogenous Nop4 was reduced to undetectable levels and plasmid-borne Nop4 WT and Nop4 L306P were expressed at comparable levels ( Figure 1C ) . Serial dilutions of strains bearing the plasmids: empty vector ( EV ) , Nop4 WT and Nop4 L306P were spotted onto plates containing glucose and incubated at 30°C , 37°C , 23°C and 17°C . At all tested temperatures , depletion of Nop4 ( EV ) conferred a severe growth defect relative to growth of Nop4 WT ( Figure 1D ) . The L306P mutation impaired growth at all temperatures tested compared to WT , although the defect was not as severe as that observed with the EV control ( Figure 1D ) . To confirm our findings , we analyzed growth in liquid medium at 23°C and estimated the doubling time for each strain . Endogenous Nop4 was depleted and the growth of strains bearing the plasmids: EV , Nop4 WT or Nop4 L306P was monitored for 48 hr . Similar to growth on solid medium , Nop4 L306P exhibited a moderate growth defect in liquid culture , doubling every 7 . 8 hr , compared to WT , which doubled every 4 . 8 hr; however , the defect was not as severe as that observed with the EV control , which doubled every 20 . 5 hr ( Figure 1E ) . The ANE syndrome mutation , L306P in yeast Nop4 , also disrupts pre-rRNA processing . As growth defects caused by mutation of a nucleolar protein are often indicative of ribosome biogenesis defects , we tested whether the growth defects conferred by Nop4 L306P were due to disruption of ribosome biogenesis . Previously , it has been shown that the mature 25S rRNA and the 27S and 7S pre-rRNA precursors are severely reduced in yeast depleted of Nop4 ( Figure 2A; Bergès et al . , 1994; Sun and Woolford , 1994 ) . To determine whether Nop4 L306P similarly disrupts production of the 25S rRNA , total RNA was harvested from strains bearing plasmids expressing no Nop4 ( empty vector; EV ) , Nop4 WT or Nop4 L306P and depleted of endogenous Nop4 for 0 and 48 hr . The 25S and 18S rRNAs were visualized by ethidium bromide staining , quantified and the ratio of 25S/18S , a measure of the relative levels of the mature rRNAs , was calculated and normalized to Nop4 WT for each time point ( Figure 2B top panels ) . The observed decrease in the 25S/18S ratios correlated with the trend of the observed growth defects . The EV control , which had the most severe growth defect , also had the most severe reduction in 25S/18S ratio levels in comparison to Nop4 WT . Nop4 L306P conferred a moderate growth defect and a moderate , but statistically significant , reduction in the 25S/18S rRNA ratio ( Figure 2C ) , consistent with reduced 25S levels . 10 . 7554/eLife . 16381 . 007Figure 2 . The ANE syndrome mutation disrupts pre-rRNA processing in yeast . ( A ) Simplified diagram depicting the pre-rRNA processing steps in yeast . The pre-rRNA is transcribed as a 35S polycistronic precursor . The external transcribed spacers ( 5´ and 3´ ETS ) and the internal transcribed spacers ( ITS1 and 2 ) are removed through a number of cleavage steps to produce the mature 18S , 5 . 8S and 25S rRNAs . Oligonucleotide probe e , which is complementary to ITS2 and detects all 27S and 7S pre-rRNAs ( indicated on top line ) , was used for northern blotting . ( B ) The ANE syndrome mutation in Nop4 impairs pre-rRNA processing in yeast . Top panel: Ethidium bromide staining of total RNA extracted from yeast expressing no Nop4 ( empty vector; EV ) , Nop4 WT or Nop4 L306P after depletion of endogenous Nop4 for the indicated time . Bottom panel: Northern blots of total RNA using radio-labeled oligonucleotide probe e to detect 35S , 27S , and 7S pre-rRNAs and an oligonucleotide probe complementary to Scr1 as a loading control . ( C ) The ratios of the mature rRNAs ( 25S/18S ) , the ratios of the precursors ( 27S/35S and 7S/35S ) and the ratios of the precursors to the loading control Scr1 ( 35S/Scr1 , 27S/Scr1 and 7S/Scr1 ) were calculated from four replicate experiments and were plotted with error bars representing the standard deviation . The significance of the ratios of Nop4 depleted yeast ( empty vector; EV ) or Nop4 L306P compared to WT was evaluated using one-way ANOVA . **indicates a p value < 0 . 0001 . indicates a p value < 0 . 001 . indicates a p value <0 . 01 . NS = not significant . Four biological replicates were performed . DOI: http://dx . doi . org/10 . 7554/eLife . 16381 . 00710 . 7554/eLife . 16381 . 008Figure 2—source data 1 . Quantitation and statistical analyses for Figure 2C . DOI: http://dx . doi . org/10 . 7554/eLife . 16381 . 008 Since the L306P mutation resulted in a reduction of the 25S/18S ratio , we determined whether the L306P mutation had an effect on pre-rRNA processing . Northern blot analysis of total RNA harvested from strains expressing no Nop4 ( EV ) , Nop4 WT or Nop4 L306P after depletion of endogenous Nop4 for 0 and 48 hr was performed using an oligonucleotide probe in ITS2 and an oligonucleotide probe against the loading control Scr1 ( Figure 2A ) . The ratios of 27S/35S and 7S/35S pre-rRNAs as well as the ratios of the precursors to the loading control , Scr1 , were quantified and normalized to Nop4 WT . Similar to the 25S/18S ratios , the pre-rRNA processing defects mirror the growth defects . Depletion of Nop4 ( EV ) resulted in a severe reduction of 27S and 7S levels , with a concomitant decrease in the 27S/35S , 7S/35S , 27S/Scr1 and 7S/Scr1 ratios , indicative of an ITS1 processing defect , as has been previously observed ( Figure 2B , C; Bergès et al . , 1994; Sun and Woolford , 1994 ) . The Nop4 L306P mutant showed an intermediate growth defect and also displayed an intermediate , but statistically significant , ITS1 processing defect as indicated by reduced 27S/35S , 7S/35S and 7S/Scr1 ratios ( Figure 2B , C ) . As the LSU processome interactome revealed that Nop4 functions as a hub protein ( McCann et al . , 2015 ) , we tested whether the ANE mutation in Nop4 abrogates protein-protein interactions using a directed yeast two-hybrid ( Y2H ) assay ( Figure 3A ) . Nop4 WT and Nop4 L306P were expressed at comparable levels as prey fusion proteins from the Y2H vector , pACT2 , in PJ69-4α ( Figure 3B ) . Yeast expressing either of these prey proteins or no Nop4 ( empty vector; EV ) were co-transformed with the Y2H bait vector , pAS2-1 , encoding 5 Nop4-interacting proteins ( Table 1; McCann et al . , 2015 ) , including Nop4 itself , and tested for interaction by serial dilution on the indicated selective medium ( Figure 3C ) . 10 . 7554/eLife . 16381 . 009Figure 3 . The ANE syndrome mutation in Nop4 disrupts protein-protein interactions . ( A ) Schematic of Y2H analysis . Nop4 WT or Nop4 L306P were cloned into the prey vector ( pACT2 ) while five Nop4 interacting proteins ( Noc2 , Mak5 , Nop4 , Nsa2 and Dbp10 ) were cloned into the bait vector ( pAS2-1 ) . Each bait was individually co-transformed into the yeast strain PJ69-α with empty vector ( EV ) , Nop4 WT or Nop4 L306P prey and spotted onto medium to confirm the presence of both Y2H plasmids ( SD-Leu-Trp ) and onto medium to test for protein-protein interactions ( SD-Leu-Trp-His + 6 mM 3-AT ) . ( B ) Nop4 WT and Nop4 L306P are expressed at equivalent levels from the Y2H vector pACT2 . Total protein was extracted from PJ69-4α yeast transformed with EV or expressing Nop4 WT or Nop4 L306P from the Y2H prey vector , pACT2 . Nop4 WT and Nop4 L306P are expressed as fusions with the GAL4 activation domain and a 3xHA tag . Protein extracts were separated by SDS-PAGE and analyzed by α-HA western blot . As a loading control , a western blot using α-Mpp10 was performed . The expression levels of Nop4 WT and Nop4 L306P relative to Mpp10 were quantitated and normalized to Nop4 WT: Nop4 WT = 1 , Nop4 L306P = 1 . 1 ( C ) Y2H analysis by serial dilution reveals that the ANE syndrome ( L306P ) mutation disrupts some Nop4 protein-protein interactions . Two biological replicates of a subset of interacting proteins were performed starting with co-transformation of the bait and prey plasmids into the Y2H strain . ( D ) The ANE syndrome ( L306P ) mutation reduces protein-protein interactions as determined by co-immunoprecipitation . Yeast extract was generated from yeast expressing either Nop4 WT or Nop4 L306P and one of its interacting partners and incubated with α-FLAG resin . Co-immunoprecipitations were assessed by α-HA western blot . The expected molecular weights of the Nop4 interacting proteins are: Dbp10 = 113 kDa , Mak5 = 87 kDa , Noc2 = 82 kDa , Nop4 = 78 kDa and Nsa2 = 30 kDa . ( E ) The ratio of co-purified 3xFLAG tagged Nop4 WT or Nop4 L306P to co-immunoprecipitated 3xHA tagged interacting partner was calculated from three replicate experiments and plotted with error bars representing the standard deviation . The significance of the co-immunoprecipitation ratio of Nop4 L306P compared to WT for each interacting partner was evaluated using a t-test . *indicates a p value < 0 . 0001 . indicates a p value < 0 . 001 . indicates a p value <0 . 01 . NS = not significant . Three biological replicates were performed . DOI: http://dx . doi . org/10 . 7554/eLife . 16381 . 00910 . 7554/eLife . 16381 . 010Figure 3—source data 1 . Quantitation and statistical analyses for Figure 3E . DOI: http://dx . doi . org/10 . 7554/eLife . 16381 . 01010 . 7554/eLife . 16381 . 011Table 1 . Nop4 interacts with 23 large subunit assembly factors with high confidence . The Nop4 interacting proteins were identified by yeast two-hybrid and were assigned a confidence score in ( McCann et al . , 2015 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16381 . 011Nop4 Interacting PartnerConfidence Score ( from McCann et al . , 2015 ) Nop492%Loc192%Ebp285%Nop1285%Nsa285%Mak584%Cgr170%Cic170%Has170%Noc270%Nop1370%Nsr170%Rrp1270%Rrp1470%Mak2168%Dbp1063%Drs163%Nop1663%Nug163%Prp4363%Spb463%Tma1663%Nog153% The presence of the ANE syndrome mutation ( L306P ) disrupted the interaction between Nop4 and 4 of the 5 interacting partners we tested ( Figure 3C ) , including Nop4 itself . While all 5 bait proteins interacted with Nop4 WT , as indicated by growth on SD-Leu-Trp-His + 6 mM 3-AT , Mak5 , Nop4 , and Nsa2 did not interact with Nop4 L306P , as no growth was observed ( Figure 3C ) . Noc2 did interact with Nop4 L306P , but growth was reduced compared to WT . In contrast , Dbp10 interaction with Nop4 was unaffected by the L306P mutation ( Figure 3C ) . Thus , the presence of the ANE syndrome mutation ( L306P ) disrupts Nop4 interaction with a subset of Nop4’s interacting proteins ( Mak5 , Nop4 , Nsa2 and Noc2 ) by Y2H , suggesting that the ANE syndrome mutation disrupts Nop4 function as a hub protein in the LSU processome . To confirm our findings , we utilized a co-immunoprecipitation method developed to validate Y2H datasets ( McCann et al . , 2015 ) to assay for changes in protein-protein interactions in the presence of the ANE syndrome mutation ( L306P ) . Nop4 WT and Nop4 L306P were expressed as 3xHA fusion proteins from the modified yeast expression vector p414GPD-3xFLAG and the 5 Nop4-interacting proteins were expressed as 3xFLAG fusion proteins from the modified yeast expression vector p415GPD-3xHA ( Mumberg et al . , 1995; McCann et al . , 2015 ) . The plasmids were co-transformed into yeast , immunoprecipitations were performed with anti-FLAG resin and the co-purifying Nop4 proteins were visualized by Western blotting with an antibody to the 3xHA tag ( Figure 3D ) . The ratio of co-purifying 3xHA-Nop4 or 3xHA-Nop4 L306P to co-immunoprecipitated 3xFLAG-interacting protein was calculated and normalized to WT for each interacting partner ( Figure 3E ) . Interestingly , 3xHA-Nop4 L306P co-purified significantly less efficiently than 3xHA-Nop4 WT with all interacting partners assayed , except Dbp10 , as was observed by Y2H . Thus , the ANE syndrome mutation disrupts or reduces a subset of Nop4 protein-protein interactions . We hypothesized that Nop4 RRM3 may be important for protein binding since it contains the ANE syndrome mutation ( L306P ) that , when present , abrogates interaction with a subset of Nop4 interacting proteins ( Figure 1A; Figure 3C–E ) . Although RRM domains typically bind RNA , there are several published examples of RRMs that bind proteins rather than RNA ( Fribourg et al . , 2003; Lau et al . , 2003; Selenko et al . , 2003; Bono et al . , 2004; Kadlec et al . , 2004 ) . To determine the contribution of the 4 RRMs to Nop4’s function as a protein-binding hub , we divided Nop4 into two fragments . One fragment contained RRM1 and RRM2 ( Nop4 RRM 1–2 ) , and the second fragment contained RRM3 and RRM4 ( Nop4 RRM 3–4; Figure 4A ) . We also attempted to determine if RRM3 alone was sufficient to mediate protein-protein interactions , however , RRM3 can not be stably expressed from the yeast two-hybrid vector ( data not shown ) . 10 . 7554/eLife . 16381 . 012Figure 4 . RRM3 and RRM4 of Nop4 mediate protein-protein interactions . ( A ) Schematic representation of Nop4 RRM domains and the N- and C-terminal fragments containing RRMs 1 and 2 ( RRM 1–2; residues 1–250 ) or 3 and 4 ( RRM 3–4; residues 252–685 ) , respectively . ( B ) Nop4 WT and the Nop4 fragments are differentially expressed from the Y2H vector pACT2 . Total protein was extracted from PJ69-4α yeast transformed with empty vector ( EV ) or expressing Nop4 WT ( 78 kDa ) , Nop4 RRM 1–2 ( 28 . 3 kDa ) or Nop4 RRM 3–4 ( 49 . 4 kDa ) from the Y2H prey vector , pACT2 . Nop4 WT , Nop4 RRM 1–2 and Nop4 RRM 3–4 are expressed as fusions with the GAL4 activation domain and a 3xHA tag . Protein extracts were separated by SDS-PAGE and analyzed by α-HA western blot . As a loading control , a western blot using α-Glucose-6-Phosphate Dehydrogenase ( G-6-PDH ) was performed . The expression levels of Nop4 WT , Nop4 RRM 1–2 and Nop4 RRM 3–4 relative to G-6-PDH were quantitated and normalized to Nop4 WT: Nop4 WT = 1 , Nop4 RRM 1–2 = 7 . 4 , Nop4 RRM 3–4 = 0 . 29 . ( C ) Y2H analysis demonstrates that Nop4 RRM 3–4 mediates protein-protein interactions . Nop4 WT and the Nop4 fragments described in ( a ) were tested as preys for interaction with 23 Nop4 interacting proteins as baits . The baits are labeled for the empty vector ( EV ) control plate . Growth on selective medium ( SD-Leu-Trp-His + 6 mM 3-AT ) indicates an interacting bait-prey pair . Two biological replicates were performed starting with the transformation of the bait and prey plasmids into the Y2H strains . DOI: http://dx . doi . org/10 . 7554/eLife . 16381 . 012 We found that Nop4 RRM 3–4 mediated the protein-protein interactions observed in the LSU processome interactome using a directed Y2H assay . Full-length Nop4 WT , Nop4 RRM 1–2 and Nop4 RRM 3–4 were expressed as prey fusion proteins ( Figure 4B ) and tested with 23 Nop4 interacting proteins ( Table 1; McCann et al . , 2015 ) expressed as bait fusions ( James et al . , 1996 ) . Each prey was mated against each bait in an array-based Y2H assay and interactions were identified by growth on selective medium ( Figure 4C; de Folter and Immink , 2011; McCann et al . , 2015 ) . As was previously observed in the LSU processome interactome , Nop4 WT interacted with the majority of the defined Nop4 interacting proteins after two weeks ( Figure 4C; McCann et al . , 2015 ) . To our surprise , Nop4 RRM 1–2 did not interact with any of the Nop4 interacting proteins . In contrast , Nop4 RRM 3–4 interacted with the majority of the defined interacting set of proteins , similar to the full-length Nop4 WT ( Figure 4C ) . Thus , Nop4 RRMs 3 and 4 are necessary and sufficient for these protein-protein interactions and are thereby likely to mediate Nop4’s function as a hub protein in the LSU processome . Nop4 RRM 3–4 is also sufficient to complement the growth defect observed upon Nop4 depletion . We constitutively expressed either full length Nop4 WT , Nop4 RRM 1–2 or Nop4 RRM 3–4 from plasmids in a yeast strain in which endogenous Nop4 was depleted ( Figure 1B ) . Western blotting of total protein demonstrated that plasmid-borne , FLAG-tagged Nop4 WT , Nop4 RRM 1–2 and Nop4 RRM 3–4 were expressed , albeit at very different levels ( Figure 5A ) . Serial dilutions of strains bearing the plasmids: empty vector ( EV ) , Nop4 WT , Nop4 RRM 1–2 and Nop4 RRM 3–4 were spotted onto plates containing glucose and incubated at 30°C , 37°C , 23°C and 17°C . As expected , EV did not complement the growth defect at any temperature whereas Nop4 WT complemented at all temperatures ( Figure 5B ) . Like EV , Nop4 RRM 1–2 did not complement the growth defect . However , Nop4 RRM 3–4 complemented the growth defect at 23°C and 17°C and partially complemented at 30°C ( Figure 5B ) . 10 . 7554/eLife . 16381 . 013Figure 5 . RRM3 and RRM4 of Nop4 are necessary and sufficient to complement the growth defect due to Nop4 depletion . ( A ) Nop4 WT and the Nop4 fragments are differentially expressed from the yeast expression vector p414GPD-3xFLAG-GW . Total protein was extracted from YPH499 GAL::3xHA-NOP4 yeast transformed with empty vector ( EV ) or expressing Nop4 WT ( 78 kDa ) , Nop4 RRM 1–2 ( 28 . 3 kDa ) or Nop4 RRM 3–4 ( 49 . 4 kDa ) from the yeast expression vector , p414GPD-3xFLAG-GW . Protein extracts were separated by SDS-PAGE and analyzed by α-FLAG western blot . As a loading control , a western blot using α-Mpp10 was performed . The expression levels of Nop4 WT , Nop4 RRM 1–2 and Nop4 RRM 3–4 relative to Mpp10 were quantitated and normalized to Nop4 WT: Nop4 WT = 1 , Nop4 RRM 1–2 = 4 . 9 , Nop4 RRM 3–4 = 0 . 27 . ( B ) Serial dilutions of yeast expressing the indicated Nop4 fragments were grown on solid medium for 3 days at 30°C and 37°C or for 5 days at 23°C and 17°C . ( C ) Yeast expressing the indicated Nop4 fragments were transferred from SG/R-Trp-Leu to SD-Trp-Leu to deplete the endogenous Nop4 . Growth was monitored for 24 hr at 30°C by measuring the absorbance at OD600 . The log2 of the OD600 was plotted over time and the slope was used to estimate the doubling time . Three biological replicates were performed starting with transformation of the plasmids into the yeast strain . ( D ) Yeast expressing the indicated Nop4 fragments were transferred from SG/R-Trp-Leu to SD-Trp-Leu to deplete the endogenous Nop4 . Growth was monitored for 48 hr at 23°C by measuring the absorbance at OD600 . The log2 of the OD600 was plotted over time and the slope was used to estimate the doubling time . Three biological replicates were performed starting with transformation of the plasmids into the yeast strain . DOI: http://dx . doi . org/10 . 7554/eLife . 16381 . 01310 . 7554/eLife . 16381 . 014Figure 5—figure supplement 1 . Nop4 RRM 1–2 fails to complement even when targeted to the nucleus . ( A ) Expression of the Nop4 fragments from the yeast two-hybrid vector pACT2 ensures nuclear targeting . Yeast expressing 3xHA-Nop4 WT , 3xHA-Nop4 RRM 1–2 or 3xHA-Nop4 RRM 3–4 from the yeast two-hybrid vector , pACT2 , after depletion of endogenous Nop4 for 48 hr at 23°C were stained with DAPI ( blue ) and HA antibodies ( green ) . The immunofluorescence microscopy images were merged . Nop4 WT , Nop4 RRM 1–2 and Nop4 RRM 3–4 localize primarily to the nucleus . ( B ) Serial dilutions of yeast expressing the indicated Nop4 fragments from pACT2 were grown on solid medium for 3 days at 30°C and 37°C or for 5 days at 23°C and 17°C . DOI: http://dx . doi . org/10 . 7554/eLife . 16381 . 014 As an additional test , we analyzed complementation in liquid medium at 30°C and 23°C . Endogenous Nop4 was depleted and the growth of strains expressing no Nop4 ( empty vector; EV ) , Nop4 WT , Nop4 RRM 1–2 and Nop4 RRM 3–4 , was monitored for 24 hr at 30°C or 48 hr at 23°C ( Figure 5C , D ) . Similar to results on solid medium , Nop4 RRM 3–4 did not significantly complement the growth defect at 30°C but did complement the growth defect at 23°C , whereas Nop4 RRM 1–2 did not complement at either temperature ( Figure 5C , D ) . The failure of Nop4 RRM 1–2 to complement the growth defect is not due to aberrant localization of the protein fragment . Nop4 RRM 1–2 also fails to complement when expressed from the yeast two-hybrid vector , pACT2 , which ensures targeting of the fragment to the nucleus ( Figure 5—figure supplement 1A , B ) , suggesting that this domain is not essential for Nop4 function . In contrast , the ability of Nop4 RRM 3–4 to complement the growth defect in both solid and liquid medium suggests that the essential function of Nop4 is mediated through RRMs 3 and 4 . Expression of Nop4 RRM 3–4 is sufficient to partially restore pre-rRNA processing . To determine whether complementation of the growth defect is due to rescue of the pre-rRNA processing defect , total RNA was harvested from strains depleted of endogenous Nop4 for 48 hr at 23°C or for 24 hr at 30°C and bearing plasmids expressing no Nop4 ( EV ) , Nop4 WT , Nop4 RRM 1–2 or Nop4 RRM 3–4 . The 25S and 18S rRNAs were visualized by ethidium bromide staining , quantified and the ratio of 25S/18S was calculated and normalized to WT for each time point ( Figure 6A , B ) . Complementation of growth correlated with the rescue of pre-rRNA processing . As expected , Nop4 WT restored the 25S/18S ratio compared to EV at both 23°C and 30°C . Nop4 RRM 1–2 did not complement growth and did not rescue the 25S/18S ratio at either temperature ( Figure 6A , B ) . In contrast , Nop4 RRM 3–4 was sufficient to significantly rescue the 25S/18S ratio compared to EV at 23°C , but not at 30°C ( Figure 6B ) , consistent with a restoration of 25S levels at 23°C . 10 . 7554/eLife . 16381 . 015Figure 6 . Nop4 RRM 3–4 is necessary and sufficient to complement the pre-rRNA processing defect after Nop4 depletion . ( A ) Top panel: Total RNA extracted from yeast expressing the indicated Nop4 fragment after depletion of endogenous Nop4 for 24 hr at 30°C or for 48 hr at 23°C was visualized by ethidium bromide staining . Bottom panel: Northern blot analysis of total RNA using oligonucleotide probe e , which is complementary to a region of ITS2 of the pre-rRNA . As a loading control , we used an oligonucleotide complementary to the Scr1 RNA . Three biological replicates were performed . ( B ) The ratios of the mature rRNAs ( 25S/18S ) and the ratios of the precursors to the loading control Scr1 ( 35S/Scr1 , 27S/Scr1S and 7S/Scr1 ) were calculated from three replicate experiments and were plotted with error bars representing the standard deviation . The significance of the ratios of 25S/18S , 35S/Scr1 , 27S/Scr1 and 7S/Scr1 of Nop4 WT , Nop4 RRM 1–2 or Nop4 RRM 3–4 compared to EV was evaluated using one-way ANOVA . *indicates a p value < 0 . 0001 . indicates a p value < 0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 16381 . 01510 . 7554/eLife . 16381 . 016Figure 6—source data 1 . Quantitation and statistical analyses for Figure 6B . DOI: http://dx . doi . org/10 . 7554/eLife . 16381 . 016 To further assess the rescue of pre-rRNA processing , we also examined pre-rRNA processing by northern blotting using an oligonucleotide probe in ITS2 ( Figure 2A ) . Depletion of Nop4 ( EV ) led to a reduction in the 27S and 7S pre-rRNAs but did not affect the levels of the 35S pre-rRNA , as has been observed before ( Bergès et al . , 1994; Sun and Woolford , 1994; Qiu et al . , 2008 ) . Expression of Nop4 WT restored pre-rRNA processing and the levels of the 27S and 7S pre-rRNA intermediates compared to the EV control at both 23°C and 30°C ( Figure 6A , B ) . Strikingly , expression of Nop4 RRM 3–4 , but not Nop4 RRM 1–2 , significantly restored the levels of the 27S and 7S intermediates at 23°C but failed to rescue at 30°C ( Figure 6B ) . Thus , pre-rRNA processing parallels the observed growth complementation and suggests that the essential function of Nop4 in ribosome assembly is mediated through the protein binding RRMs , RRM3 and RRM4 . As the ANE syndrome mutation in yeast Nop4 causes pre-rRNA processing defects and reduces protein-protein interactions with a subset of proteins that we tested , we hypothesized that the mutation causes a structural change in RRM3 . To test this , we analyzed WT and mutant human RBM28 RRM3 domains ( amino acids 330–419 ) by circular dichroism ( CD ) and NMR . We used RBM28 RRM3 because Nop4 RRM3 was not soluble . We found that the L351P mutation disrupts the backbone structure of the RBM28 RRM3 domain . The CD spectra of WT and L351P RBM28 RRM3 showed notable differences in ellipticities [θ] at 208 , 215 and 222 nm ( Figure 7A ) , indicating that the amino acid substitution reduced α-helical and β-sheet content . The CD spectrum of the L351P mutant protein with a minimum only at ~200 nm indicated the presence of random coil . We confirmed the disruption of domain folding by NMR . 15N-HSQC of WT RBM28 RRM3 showed well dispersed resonances ( Figure 7B ) , indicating the presence of both α-helices and β-strands , as expected for an RRM domain ( Nagai et al . , 1990 ) . In contrast , the majority of resonances in the spectrum of L351P ANE syndrome mutant RRM3 were clustered around 8 . 0~8 . 5 ppm in the proton dimension , demonstrating that the mutant protein backbone is disordered . A homology model of the human RBM28 RRM3 , based on an NMR structure of mouse RBM28 RRM3 , indicates that L351 is buried within the core of the domain ( Figure 7C ) . Mutation to proline would be expected to disrupt helix α1 and the overall tertiary structure of this RRM . 10 . 7554/eLife . 16381 . 017Figure 7 . The ANE syndrome mutation , L351P , in human RBM28 disrupts RRM3 domain structure . ( A ) Circular dichroism spectra of WT human RBM28 RRM3 ( black ) and L351P mutant ( red ) . Four technical replicates were performed . ( B ) 15N-HSQC spectra of WT hRBM28 RRM3 ( amino acids 330–419 ) ( black ) and L351P mutant ( red ) are superimposed and plotted at the same contour level . In addition to clustering of resonances around 8 . 0~8 . 5 ppm in the proton dimension , dispersion of glutamine and asparagine side chains ( 7 . 0~7 . 8 ppm in the 1H dimension and 111~114 ppm in the 15N dimension ) is reduced considerably , consistent with protein backbone disruption . Thirty-two technical replicates were performed . ( C ) Ribbon diagram of a homology model of human RBM28 RRM3 . The model including residues 330–419 was generated using the Phyre2 server ( Kelley et al . , 2015 ) , and the best template was a solution structure of mouse RBM28 RRM3 ( 90% sequence identity with human RBM28 , PDB ID 1X4H ) . L351 is shown with red space-filling spheres and typical RNA interacting residues in RNP motifs are colored yellow . DOI: http://dx . doi . org/10 . 7554/eLife . 16381 . 017 We determined that the molecular pathogenesis of ANE syndrome ( L351P in the nucleolar protein , RBM28 ) results from disrupted RRM3 protein structure that reduces its function as a hub protein in the LSU processome and causes defects in pre-rRNA processing . Introducing the ANE syndrome mutation into RRM3 of the yeast ortholog , Nop4 ( L306P ) , causes growth and pre-rRNA processing defects , as well as reduced association with interacting protein partners . RRMs 3 and 4 of Nop4 alone are necessary and sufficient for Nop4 protein-protein interactions and for yeast growth and pre-rRNA processing . Biophysical methods indicate that RRM3 containing the ANE syndrome mutation is unfolded . Taken together , these results provide evidence that ANE syndrome is a ribosomopathy with a molecular defect in nucleolar steps of ribosome biogenesis . The finding that an essential function of Nop4 is to mediate protein binding was unexpected . Nop4 has been cross-linked to pre-rRNA ( Bergès et al . , 1994; Sun and Woolford , 1994; Granneman et al . , 2011 ) , leading to the expectation that its 4 RRMs would be essential for RNA binding . We show by Y2H analysis that the C-terminal half of Nop4 ( RRMs 3 and 4 ) mediates protein-protein interactions and hub protein function in the LSU processome . Strikingly , this half of Nop4 is necessary and sufficient to complement the growth and pre-rRNA processing defects observed upon depletion of endogenous Nop4 . As the ANE syndrome mutation occurs in RRM3 and disrupts Nop4 protein-protein interactions , ANE syndrome is therefore likely a disease of altered protein interaction rather than of RNA binding . We are now in a position to determine which interactions are critical for Nop4 function and how disruptions of those specific interactions contribute to ANE syndrome pathogenesis . Nevertheless , Nop4 is undoubtedly an RNA-binding protein . Nop4 binds RNA in vitro ( Sun and Woolford , 1997 ) , co-immunoprecipitates the 27S and 7S pre-rRNAs and crosslinks to the 25S within the pre-rRNA in vivo ( Granneman et al . , 2011 ) . Additionally , all 4 RRMs are important for Nop4 function as mutations in any of the RRMs disrupt growth and LSU assembly at 37°C ( Sun and Woolford , 1997 ) . Although RRMs 1 and 2 are not required for Nop4’s hub protein function or sufficient for growth , they may bind the pre-rRNA . Furthermore , while RRMs 3 and 4 mediate protein binding , the possibility that they may also bind RNA is not precluded . How does an RRM facilitate protein binding and thus hub protein function ? Several examples of RRMs mediating interactions with other proteins have been identified including in the U2AF35-U2AF65 , the U2AF65-SF1 , the Snu17-Bud13 , and the Y14-Mago complexes ( Kielkopf et al . , 2001; Fribourg et al . , 2003; Lau et al . , 2003; Selenko et al . , 2003; Tripsianes et al . , 2014 ) . An RRM is comprised of a 4-stranded β-sheet , which forms the primary RNA binding interface , and 2 α-helices ( Maris et al . , 2005; Cléry et al . , 2008 ) . The α-helices of the RRM often mediate interaction between protein pairs , leaving the β-sheet accessible for RNA binding ( Kielkopf et al . , 2001; Selenko et al . , 2003; Tripsianes et al . , 2014 ) . Alternatively , in the case of the Y14-Mago complex , the interaction is through the β-sheet , which precludes RNA binding ( Fribourg et al . , 2003; Lau et al . , 2003 ) . RRMs also can serve as oligomerization domains . In the case of Human antigen R ( HuR ) , the third RRM promotes dimerization through its α-helices ( Scheiba et al . , 2014 ) . These examples highlight the possibility that RRMs 3 and 4 of Nop4 may mediate protein binding and hub function through more than one of its structural motifs . The ANE mutation lies in an α-helix and the mutation disrupts not only the α-helix , but unfolds the RRM tertiary structure . Since the mutation abrogates some but not all protein-protein interactions , a subset of Nop4 interactions may be mediated by unstructured peptide elements in RRM3 or by the RRM4 domain . Fibroblasts from patients with ANE syndrome have ribosome levels reduced to approximately 60% of controls ( Nousbeck et al . , 2008 ) , consistent with the modest defects in ribosome biogenesis that we can now associate with the ANE syndrome mutation . In the yeast model system , the mutation confers reduced growth and mild pre-rRNA processing defects when compared to the Nop4 null ( EV ) . Thus , the ANE syndrome mutation is a hypomorphic allele , consistent with its autosomal recessive inheritance ( Nousbeck et al . , 2008 ) . This hypomorphism may , in part , explain how the ANE syndrome mutation is compatible with life , as the presence of the mutation leads to a partially functional RBM28 protein . The amino acid sequence of Nop4 was obtained from the Saccharomyces Genome Database ( www . yeastgenome . org ) . The amino acid sequence of human RBM28 ( accession number NP_060547 ) was obtained from the Protein Database ( http://www . ncbi . nlm . nih . gov/protein/ ) . The amino acid sequences of RBM28 from M . mulatta , M . musculus , X . tropicalis and D . rerio were obtained from Uniprot ( www . uniprot . org ) . Amino acid alignments were determined using either ClustalX ( Jeanmougin et al . , 1998 ) or MegAlign Pro version 12 . 2 . 0 from DNASTAR . Madison , WI . A GAL::3HA-NOP4 strain was generated in the parental strain YPH499 ( MATa ura3-52 lys2-801 ade2-101 trp1-Δ63 his3-Δ200 leu2-Δ1 ) as described in ( Charette and Baserga , 2010 ) that expresses 3HA-tagged Nop4 from the endogenous locus when grown in medium containing galactose but represses endogenous Nop4 expression when grown in medium containing glucose . The strain was confirmed by western blot using α-HA-HRP ( Roche , Indianapolis , Indiana ) . NOP4 was shuttled into the Gateway-modified yeast expression vector p414GPD-3xFLAG-GW ( TRP1 ) or the Gateway-modified Y2H prey vector pACT2 ( LEU2 ) and RBM28 was shuttled into p414GPD-3xFLAG-GW ( TRP1 ) by Gateway cloning ( Life Technologies ) as in ( Charette and Baserga , 2010 ) . Site-directed mutagenesis to introduce the L306P missense mutation was performed using a Change-IT kit ( Affymetrix , Santa Clara , California ) . RBM28 , Nop4 WT and Nop4 L306P were all fully sequenced by either the W . M . Keck Foundation facility at the Yale School of Medicine or by GENEWIZ , Inc . Expression of RBM28 , Nop4 WT and Nop4 L306P from p414GPD-3xFLAG-GW or pACT2 was analyzed by western blot using either α-3xFLAG-HRP ( Sigma , St . Louis , Missouri ) or α-HA-HRP ( Roche , Indianapolis , Indiana ) . As a loading control , a western blot using α-Mpp10 ( Dunbar et al . , 1997 ) was performed . The Nop4 fragments in Figure 4A were cloned into a Gateway Entry vector ( pDONR221 ) and subsequently shuttled into the Y2H prey vector ( pACT2 ) or the yeast expression vector p414GPD-3xFLAG-GW by Gateway cloning ( Life Technologies ) as in ( Charette and Baserga , 2010 ) . All clones were fully sequenced by either the W . M . Keck Foundation facility at the Yale School of Medicine or by GENEWIZ , Inc . Expression of the Nop4 fragments from p414GPD-3xFLAG-GW and pACT2 was analyzed by western blot using either α-3xFLAG-HRP ( Sigma , St . Louis , Missouri ) or α-HA-HRP ( Roche ) . As a loading control , a western blot using α-G-6-PDH ( Sigma ) or using α-Mpp10 ( Dunbar et al . , 1997 ) was performed . For analysis of the complementation by RBM28 of the growth defect conferred by Nop4 depletion , the YPH499 GAL::3xHA-NOP4 yeast strain was transformed with either empty vector ( EV ) or plasmids expressing Nop4 or RBM28 . For serial dilutions , 0 . 2 mL of cells at an OD600 of 1 were resuspended in 1 mL water , diluted 1/10 and spotted onto SG/R-Trp or SD-Trp . Cells were incubated at 30°C or 37°C for 3 days or at 23°C or 17°C for 5 days . Two biological replicates were performed starting with transformation of the plasmids into the yeast strain . For analysis of the effect of the ANE syndrome mutation ( L306P ) on growth , the YPH499 GAL::3xHA-Nop4 yeast strain was transformed with either empty vector ( EV ) , or plasmids expressing Nop4 WT or Nop4 L306P . For serial dilutions , 0 . 2 mL of cells at an OD600 of 1 were resuspended in 1 mL water , diluted 1/10 and spotted onto medium containing 2% w/v galactose and 2% w/v raffinose and lacking tryptophan ( SG/R-Trp ) or onto medium containing 2% w/v glucose ( dextrose ) and lacking tryptophan ( SD-Trp ) . Cells were incubated at 30°C or 37°C for 3 days or at 23°C or 17°C for 5 days . Four biological replicates were performed starting with transformation of the plasmids into the yeast strain . For analysis in liquid medium , the GAL::3xHA-NOP4 yeast strain transformed with empty vector ( EV ) or expressing either 3xFLAG-tagged Nop4 WT or Nop4 L306P from the p414GPD vector was depleted of endogenous Nop4 by first growing cultures to mid-log phase ( OD600 = 0 . 4–0 . 8 ) in SG/R-Trp at 30°C and then transferring the cultures to the non-permissive ( SD-Trp ) medium and 23°C . The cells were maintained in mid-log phase ( OD600 < 0 . 8 ) by dilution of the culture with fresh SD-Trp media . Growth was monitored by OD600 measurement for 48 hr . Three biological replicates were performed starting with transformation of the plasmids into the yeast strain . For analysis of the complementation by the Nop4 fragments of the growth defect conferred by Nop4 depletion , the YPH499 GAL::3xHA-NOP4 yeast strain was co-transformed with empty p415GPD-3xHA-GW and either empty p414GPD-3xFLAG-GW vector ( EV ) or p414GPD-3xFLAG-GW expressing the Nop4 fragments . For serial dilutions , 0 . 2 mL of cells at an OD600 of 1 were resuspended in 1 mL water , diluted 1/10 and spotted onto SG/R-Trp-Leu or SD-Trp-Leu . Cells were incubated at 30°C or 37°C for 3 days or at 23°C or 17°C for 5 days . Two biological replicates were performed starting with transformation of the plasmids into the yeast strain . For analysis in liquid medium of the complementation by the Nop4 fragments of the growth defect conferred by Nop4 depletion , the YPH499 GAL::3xHA-NOP4 yeast strain expressing one of the 3xFLAG-tagged Nop4 fragments was depleted of endogenous Nop4 by first growing cultures to mid-log phase ( OD600 = 0 . 4–0 . 8 ) in SG/R-Trp-Leu at 30°C and then transferring the cultures to the non-permissive ( SD-Trp-Leu ) medium and growing at either 30°C or 23°C . The cells were maintained in mid-log phase ( OD600 < 0 . 8 ) by dilution of the culture with fresh SD-Trp-Leu media . Growth was monitored by OD600 measurement for 24 hr at 30°C or for 48 hr at 23°C . Three biological replicates were performed starting with transformation of the plasmids into the yeast strain . For analysis of the complementation of the growth defect conferred by Nop4 depletion by the Nop4 fragments expressed from pACT2 , the YPH499 GAL::3xHA-NOP4 yeast strain was transformed with either pACT2 Nop4 WT , pACT2 Nop4 RRM 1–2 or pACT2 Nop4 RRM 3–4 . For serial dilutions , 0 . 2 mL of cells at an OD600 of 1 were resuspended in 1 mL water , diluted 1/10 and spotted onto SG/R- Leu or SD- Leu . Cells were incubated at 30°C or 37°C for 3 days or at 23°C or 17°C for 5 days . Two biological replicates were performed starting with transformation of the plasmids into the yeast strain . For analysis of the effect of the ANE syndrome mutation , the YPH499 GAL::3xHA-NOP4 yeast strain was transformed with EV or plasmids expressing either Nop4 WT or Nop4 L306P from the p414GPD-3xFLAG vector . The strain was depleted of endogenous Nop4 as described above . Cells ( 20 mL ) at an OD600 of ~0 . 5 were collected from each culture after 0 and 48 hr of growth at 23°C . Total RNA was extracted as described in ( Dunbar et al . , 1997 ) . For analysis of the mature rRNAs , 5 μg of total RNA per sample was separated by electrophoresis on a 1% agarose gel . RNA was visualized by ethidium bromide staining , and the bands were quantified by densitometric analysis using ImageJ ( Schneider et al . , 2012 ) . For northern blot analysis , 3 μg of total RNA per sample was separated by electrophoresis on a 1% agarose/1 . 25% formaldehyde gel , transferred to a nylon membrane ( Hybond-XL , GE Healthcare , Buckinghamshire , England ) and detected by hybridization with radiolabelled oligonucleotide e ( 5´ – GGCCAGCAATTTCAAGT – 3´ ) and radiolabelled oligonucleotide Scr1 ( 5’ – CGTGTCTAGCCGCGAGGAAGGATTTGTTCC – 3’ ) as described in ( Wehner and Baserga , 2002; Qiu et al . , 2014 ) . The 35S , 27S and 7S pre-rRNAs and the Scr1 RNA were quantified using a Biorad Personal Molecular Imager . The ratios of 27S or 7S to the 35S pre-rRNA and the ratios of the 35S , 27S or 7S to Scr1 were calculated . Four biological replicates were performed for each experiment . GraphPad PRISM was used to calculate the means of the ratios and plotted with error bars ( SD ) . Significance compared to the WT control was determined using one-way ANOVA . For analysis of the complementation by the Nop4 fragments of the pre-rRNA processing defect , the YPH499 GAL::3xHA-NOP4 yeast strain transformed with EV or expressing one of the 3xFLAG-tagged Nop4 fragments was depleted of endogenous Nop4 as described . Cells ( 20 mL ) at an OD600 of ~0 . 5 were collected from each culture after either 24 hr of growth at 30°C or after 48 hr of growth at 23°C . Total RNA was extracted as described in ( Dunbar et al . , 1997 ) . For northern blot analysis , 3 μg of total RNA per sample was separated by electrophoresis on a 1% agarose/1 . 25% formaldehyde gel , transferred to a nylon membrane ( Hybond-XL , GE Healthcare , Buckinghamshire , England ) and detected by hybridization with radiolabelled oligonucleotide probe e ( 5´ – GGCCAGCAATTTCAAGT – 3´ ) , which is complementary to ITS2 of the yeast pre-rRNA , and probe Scr1 ( 5´-CGTGTCTAGCCGCGAGGAAGGATTTGTTCC-3´ ) , which is complementary to the RNA Scr1 , as described in ( Wehner and Baserga , 2002 ) . The 7S , 27S and 35S pre-rRNA species were quantified on a Biorad Personal Molecular Imager , and the ratios of 7S , 27S or 35S to Scr1 were calculated . Three biological replicates were performed for each experiment . GraphPad PRISM was used to calculate the means of the ratios and plotted with error bars ( SD ) . Significance compared to the EV control was determined using one-way ANOVA . For Y2H analysis to determine the effect of the ANE syndrome mutation ( L306P; Figure 3 ) on protein-protein interactions , a subset of the Nop4 interacting proteins identified in ( McCann et al . , 2015 ) were expressed from the Y2H bait vector ( pAS2-1 ) and were co-transformed with either empty pACT2 vector , Nop4 WT or Nop4 L306P into the yeast strain PJ69-4α . Co-transformed yeast were serially diluted by resuspending 0 . 2 mL of cells at an OD600 of 1 in 1 mL water , diluting 1/10 and spotting onto medium selecting for the presence of both plasmids ( SD-Leu-Trp ) and medium selecting for Y2H interactions [SD-Leu-Trp-His + 6 mM 3-Amino-1 , 2 , 4 triazole ( 3-AT ) ] . Cells were incubated at 30°C for 7 days . Two biological replicates of a subset of interacting proteins were performed starting with co-transformation of the bait and prey plasmids into the Y2H strain . For Y2H analysis of Nop4 fragments ( Figure 4 ) , Nop4 WT and the Nop4 fragments were shuttled into the Y2H prey vector ( pACT2 ) by Gateway ( Invitrogen ) recombination and individually transformed into the yeast strain PJ69-4a . The Nop4 interacting proteins identified in ( McCann et al . , 2015 ) were shuttled into the Y2H bait vector ( pAS2-1 ) and transformed into the yeast strain PJ69-4α as an array . All baits were mated against all preys in a semi-high-throughput Y2H matrix screen ( de Folter and Immink , 2011 ) . The mated yeast were transferred to SD-Leu-Trp plates to select for diploids bearing both the bait and prey vectors . Diploids were then transferred to the selective medium: SD-Leu-Trp-His + 6mM 3-AT . Growth on selective medium greater than that of the negative control after 2 weeks was indicative of an interacting bait-prey pair . Two biological replicates were performed starting with the transformation of the bait and prey plasmids into the Y2H strains . A subset of the Nop4 interacting proteins identified in ( McCann et al . , 2015 ) were expressed from p414GPD-3xFLAG-GW and were co-transformed with either p415GPD-3xHA-GW Nop4 WT or Nop4 L306P into the yeast strain YPH499 ( MATa ura3-52 lys2-801 ade2-101 trp1-Δ63 his3-Δ200 leu2-Δ1 ) . The resulting transformed strains were grown in medium containing 2% dextrose and lacking leucine and tryptophan ( SD-Leu-Trp ) at 30°C . Negative control strains were only transformed with p415GPD-3xHA-GW clones and were grown in medium containing 2% dextrose and lacking leucine ( SD-Leu ) at 30°C . For each co-immunoprecipitation , 20 mL of cells at an OD600 of ~0 . 5 was collected , washed with water and resuspended in NET2 ( 20 mM Tris-HCl , pH 7 . 5 , 150 mM NaCl , 0 . 01% Nonidet P-40 ) with 1x HALT protease inhibitors ( Thermo Fisher Scientific , Rockford , Illinois ) . Cells were lysed with 0 . 5-mm glass beads . The lysate was cleared by centrifugation at 15 , 000g for 10 min at 4°C . Aliquots of 500 µL of lysate were incubated with α-FLAG beads ( Sigma ) for 1 hr at 4°C . The beads were washed five times with NET2 and resuspended in 25 µL SDS loading dye . Immunoprecipitates were separated on 4–12% Bis-Tris PAGE and transferred to a PVDF membrane . Western blot analysis with α-HA ( Abcam , Cambridge , Massachusetts ) and α-FLAG-HRP ( Sigma ) was performed . The protein bands were quantified using ImageJ ( Schneider et al . , 2012 ) and the ratio of HA to FLAG was calculated . GraphPad PRISM was used to plot the means of the ratios with error bars ( SD ) . Significance compared to the WT control for each interacting protein was determined using a t-test . Three biological replicates were performed . For analysis of the Nop4 fragment localization , the YPH499 GAL::3xHA-NOP4 yeast strain was transformed with either pACT2 Nop4 WT , pACT2 Nop4 RRM 1–2 or pACT2 Nop4 RRM 3–4 and endogenous Nop4 was depleted for 48 hr at 23°C as described above . For immunofluorescence , 50 mL of cells at an OD600 of ~0 . 5 was collected , washed with water , resuspended in Fixing buffer ( 100 mM Sucrose , 5% paraformaldehyde ) and incubated at room temperature for 45 min . The cells were then washed three times with Buffer B ( 100 mM K2HPO4 pH 7 . 5 , 1 . 2 M sorbitol ) , resupsended in 1 ml of Spheroplasting buffer ( 100 mM K2HPO4 pH 7 . 5 , 1 . 2 M sorbitol , 30 mM β-mercaptoethanol ) containing lyticase ( Sigma ) at 800 U/mL and incubated for 8 min at 30°C . The reaction was stopped by adding 5 mL of ice-cold Buffer B . The yeast were washed once with ice-cold Buffer B , resuspended in 1 ml of Buffer B and 500 µL were plated into each well of a 12-well plate containing a poly-D-lysine coated cover glass . Cells were incubated for 1 hr at 4°C , washed once with Buffer B and permeabilized overnight in 70% ethanol at -20°C . Fixed and permeabilized cells incubated with mouse anti-HA . 11 ( BioLegend , San Diego , California ) diluted 1:1000 for 90 min at room temperature . The secondary antibody ( Alexa Fluor 488 donkey anti-Mouse; Life Technologies , Carlsbad , California ) was used at a dilution of 1:1000 and was incubated for 1 hr at room temperature . Cover glasses were mounted with Prolong Gold containing DAPI ( Life Technologies ) and cells were imaged on a wide-field , epifluorescence microscope using a x100 oil-immersion objective ( Carl Zeiss ) . E . coli codon-optimized cDNAs encoding the wild type ( WT ) and L351P mutant human RBM28 RRM3 ( 330-419 ) were obtained by gene synthesis ( Genewiz , Inc . ) . The cDNAs were subcloned into pSMT3 with an N-terminal His6-SUMO tag . WT and L351P human RBM28 RRM3 domains were overexpressed in E . coli strain BL21-CodonPlus ( DE3 ) -RIL ( Agilent Technologies , Santa Clara , California ) at 20 °C overnight after induction with 0 . 5 mM IPTG . The cells were collected by centrifugation , and pellets were resuspended in lysis buffer ( 50 mM Tris-HCl , pH 8 . 0 , 500 mM NaCl ) and stored at −80 °C until use . Cells expressing WT human RBM28 RRM3 domain were disrupted by sonication . The soluble fraction was applied to a Ni-NTA agarose column and thoroughly washed with lysis buffer containing 20 mM imidazole . The target SUMO fusion protein was eluted with lysis buffer containing 400 mM imidazole . The fusion protein was cleaved overnight with 0 . 2 mg of Ulp1 protease . The cleaved fusion protein sample was applied to a HiLoad 16/60 Superdex 75 column ( GE Healthcare ) equilibrated with lysis buffer containing 1 mM Tris ( 2-carboxyethyl ) phosphine hydrochloride ( TCEP ) . The eluted fractions containing WT hRBM28 RRM3 protein were pooled and applied to a Ni-NTA agarose column again to remove released SUMO protein . The protein sample was dialyzed against a buffer containing 50 mM Tris-HCl , pH 8 . 0 , 100 mM NaCl and 0 . 5 mM TCEP and purified further using a HiTrap Q HR anion-exchange column ( GE Healthcare ) . Bound proteins were eluted using a linear gradient from 0 . 05 to 1 M NaCl in 50 mM Tris-HCl , pH 8 . 0 and 0 . 5 mM TCEP . Peak fractions containing WT hRBM28 RRM3 were pooled and concentrated . L351P mutant human RBM28 RRM3 domain was purified by the same procedure as WT protein up to the first Ni-NTA agarose column . The eluted SUMO fusion protein was cleaved overnight with Ulp1 protease in conjunction with dialysis into 50 mM Tris-HCl , pH 8 . 0 , 100 mM NaCl and 0 . 5 mM TCEP . The cleaved fusion protein sample was applied to a HiTrap Q HR anion-exchange column and eluted with a linear gradient from 0 . 1 to 1 M NaCl in 50 mM Tris-HCl , pH 8 . 0 and 0 . 5 mM TCEP . The eluted fractions containing L351P hRBM28 RRM3 protein were pooled and reapplied to a Ni-NTA agarose column . The protein was concentrated and purified further using a HiLoad 16/60 Superdex 75 column equilibrated with lysis buffer containing 0 . 5 mM TCEP . Two peaks of L351P hRBM28 RRM3 eluted from the Superdex 75 column . Because the peak eluting at 58 . 5 ml contained many contaminating proteins , only the fractions containing the peak eluting at 72 . 2 ml were pooled and concentrated . The CD spectra of WT and L351P mutant human RBM28 RRM3 domains were measured on a JASCO J-810 CD spectrometer at room temperature . For each sample ( 200 μL in a 0 . 1 cm light-path cell ) , four scans were accumulated in the wavelength range of 190–260 nm with a 0 . 2 nm step size . Protein samples were 100 μg/mL in 20 mM Na phosphate buffer , pH 7 . 0 , 100 mM NaCl and 0 . 2 mM TCEP . The raw CD data were adjusted by subtracting a buffer blank . Four technical replicates were performed . 15N-/13C-labeled WT and L351P human RBM28 RRM3 domains were prepared as described above , except that E . coli cultures were grown in M9 medium containing appropriate isotopes . The protein samples for NMR experiments were 0 . 4 mM in 20 mM sodium phosphate buffer , pH 7 . 0 , 100 mM NaCl , 0 . 2 mM TCEP and 10% ( v/v ) D2O . 15N-HSQC spectra were collected on a Varian Inova 60 MHz magnet installed with a cryo-probe at 298 K . The data were processed and plotted using NMRPipe ( Delaglio et al . , 1995 ) and NMRViewJ ( Johnson , 2004 ) . Thirty-two technical replicates were performed .	ANE syndrome is a rare genetic disease that causes many problems including hair loss , mental retardation and a failure to develop normally during puberty . A study of 5 boys in the same family that were all born with the condition revealed that the disease is caused by a small change ( or mutation ) in a protein called RBM28 . While little is known about the role of human RBM28 , it is known that the equivalent protein in yeast – known as Nop4 – plays a critical role in forming a network of proteins needed to assemble ribosomes , the machines that make proteins . McCann et al . investigated how such a small mutation in human RBM28 could cause disease and whether this involves interrupting the assembly of ribosomes . The experiments show that introducing the same mutation into yeast Nop4 impaired the ability of Nop4 to form the network of proteins needed for ribosomes to assemble . This ultimately restricted the growth of the yeast . Further experiments revealed that the mutation also alters the shape of the human RBM28 protein . The main challenges for the future are to find out whether human RBM28 plays a similar role in ribosome assembly as the yeast protein , and to work out how disrupting ribosome assembly could lead to the symptoms of ANE syndrome .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "structural", "biology", "and", "molecular", "biophysics" ]	2016	The molecular basis for ANE syndrome revealed by the large ribosomal subunit processome interactome
Entry and extrusion of cations are essential processes in living cells . In alkaliphilic prokaryotes , high external pH activates voltage-gated sodium channels ( Nav ) , which allows Na+ to enter and be used as substrate for cation/proton antiporters responsible for cytoplasmic pH homeostasis . Here , we describe a new member of the prokaryotic voltage-gated Na+ channel family ( NsvBa; Non-selective voltage-gated , Bacillus alcalophilus ) that is nonselective among Na+ , Ca2+ and K+ ions . Mutations in NsvBa can convert the nonselective filter into one that discriminates for Na+ or divalent cations . Gain-of-function experiments demonstrate the portability of ion selectivity with filter mutations to other Bacillus Nav channels . Increasing pH and temperature shifts their activation threshold towards their native resting membrane potential . Furthermore , we find drugs that target Bacillus Nav channels also block the growth of the bacteria . This work identifies some of the adaptations to achieve ion discrimination and gating in Bacillus Nav channels . Ion selectivity is a defining feature of ion channels . While the structural and biophysical determinants of K+ selectivity are well described ( Doyle et al . , 1998; Jiang et al . , 2003 ) , those of Na+ and Ca2+ are unresolved . To fill this knowledge gap , the voltage-gated sodium channels ( Nav channels ) from bacteria have been extensively studied by structural biologists . To date , two full-length ( NavAb and NavRh ) ( Payandeh et al . , 2011 , 2012; Zhang et al . , 2012 ) and three pore-only ( NavMs , NavAe and NavCt ) ( McCusker et al . , 2012; Tsai et al . , 2013; Shaya et al . , 2014 ) prokaryotic Na+ channel crystal structures have been solved . The full-length structures demonstrate that four identical subunits , each containing 6-transmembrane segments , assemble together to form a functional channel . The first four transmembrane segments form the voltage sensor ( S1–S4 ) while the fifth and sixth transmembrane segments ( S5 , S6 ) form the pore domain . The selectivity filter , which forms critical interactions with the permeating hydrated ions , defines the pore and is scaffolded by two pore helices ( P1 and P2 ) from each subunit . The dipole created by the helices , and an acidic residue from each subunit , create an electronegative region that attracts cations . However , the molecular arrangement that enables Na+-selectivity in prokaryotes might differ from mammalian voltage-gated Na+ channels . Thus , it is not clear if structural features that determine ion selectivity in homotetrameric prokaryotic NaV reflect the functionally heterotetrameric eukaryotic sodium channels . The first prokaryotic voltage-gated sodium channel , cloned from Bacillus halodurans C-125 , was called NaChBac ( Na+ Channel of Bacteria ) ( Ren et al . , 2001 ) . Since then , NaV channels from at least 9 bacterial species have been functionally characterized . All full-length channels exhibit Na+-selectivity ( Ren et al . , 2001; Ito et al . , 2004; Koishi et al . , 2004; Irie et al . , 2010; Ulmschneider et al . , 2013 ) . Most bacteria require the ion-motive forces provided by H+ or Na+ ions that move through the MotAB or MotPS stator to power the rotary motor proteins to drive the motion of the flagella , ( Figure 1A ) ( Ito et al . , 2004; Fujinami et al . , 2009 ) . In soda lakes and other extremely H+-poor ( pH 9–12; 1 nM—1 pM [H+] ) and Na+-rich environments ( up to 500 mM ) , alkaliphilic prokaryotes couple motility and ion transporters to Na+ rather than H+ . Growth and motility of the alkaliphilic bacterium Bacillus alcalophilus AV ( Vedder , 1934 ) are better supported in K+ rather than Na+ conditions ( Terahara et al . , 2012 ) . In alkaliphilic Bacillus , either ion can drive the flagellar motor , and a single mutation in the MotS protein was identified that converted the naturally nonselective MotPS ( Na+ or K+ ) stator into a Na+-selective stator . Presumably , bacterial Nav channels provide a source of Na+ ions that drives the stators and maintains ion homeostasis . However , it is unclear what condition or stimulus would open these channels at the very hyperpolarized resting membrane potentials ( Ψrest ≈ −180 mV ) found in bacteria . 10 . 7554/eLife . 04387 . 003Figure 1 . The alkaliphilic Bacillus cation cycle and the relationship between bacteria Nav homologs . ( A ) A diagram depicting the membrane proteins involved in the Na+ cycle of Bacillus alcalophilus . The cation/proton antiporters , including Mrp antiporters , catalyze net proton accumulation in the cytoplasm in cells that are extruding protons during respiration . Na+ re-entry in support of pH homeostasis is achieved by Na+ solute symporters and through the voltage-gated channel , NsvBa . ( B ) Relatedness within functionally characterized members of the bacterial sodium channel superfamily . A rooted phylogenic tree analysis of bacterial Nav channels calculated by using the CLUSTALW program ( http://clustalw . genome . jp ) . Branch lengths are proportional to the sequence divergence ( scale bar = 0 . 2 substitutions per amino acid site ) . ( C ) An alignment of the selectivity filters from various Na+ , K+ , and nonselective ion channels . The pore region from the indicated ion channel families ( italic ) were aligned using the ClustalW multiple sequence alignment program applying the default color scheme with <60% conservation of character: Hydrophobic ( blue ) ; polar ( green ) ; glutamine , glutamate , aspartate ( magenta ) . Special amino acids are designated with their own color: glycine ( orange ) ; proline ( yellow ) and tyrosine or histidine ( cyan ) . The barrels indicate the regions spanning the pore helices found in the Shaker ( K+-selective ) , NaK and NavRh crystal structures . DOI: http://dx . doi . org/10 . 7554/eLife . 04387 . 00310 . 7554/eLife . 04387 . 004Figure 1—figure supplement 1 . A comparison of the voltage-dependence of the bacterial Nav channels . ( A ) Left , Example traces from HEK 293T cells expressing sodium channels , NsvBa ( black ) , NaChBac ( gray ) and NavBp ( purple ) . Currents were activated by 0 . 5 s ( NsvBa and NaChBac ) or 1 s NavBp prepulses of increasing potential followed by a test pulse to −20 mV . ( B ) Resulting conductance-voltage and inactivation-voltage relationships were measured by plotting the average prepulse peak current converted to conductance , and reduction of test pulse peak current , respectively as a function of prepulse potential ( n = 6–10 , Error = ±SEM ) and fit to a sigmoid equation . ( C ) Inactivation rate ( τinact ) -voltage relationship was measured by fitting the current decay to a single exponential equation for NsvBa ( Black ) , NaChBac ( gray ) and NavBp ( purple ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04387 . 004 Here , we determine the permeation and gating properties of a voltage-gated ion channel ( NsvBa ) from Bacillus alcalophilus AV . While related to members of the Nav superfamily ( Figure 1B ) , NsvBa is nonselective among cations . Using mutagenesis , we demonstrate that the nonselective filter from NsvBa can be converted into one that is more selective for Na+ or Ca2+ , and that these features can be conferred onto the NaChBac filter . In addition to selectivity , we characterize the voltage- , pH- and temperature-dependence of Bacillus Nav channels . We find that a combination of higher temperature and pH are required to reduce the activation threshold of channel opening in Bacillus Nav channels , which is unique to this family of sodium channels . We also characterize Nav current antagonism by drugs that impaired the growth and motility of alkaliphilic Bacillus species at corresponding concentrations . These findings shed light on the biophysical requirements for ion channel selectivity , pharmacology , biochemical adaptations among Bacillus species , and the evolution of voltage-gated Na+ channels . Among the first alkaliphilic extremophiles described in the literature , the gram-positive , rod-shaped Bacillus alcalophilus AV was initially isolated from human feces ( Vedder , 1934 ) . We cloned NsvBa and generated plasmids for mammalian cell expression to enable measurement by patch clamp methods . Currents from NsvBa-transfected HEK293T cells were robust ( ≈119 pA/pF in 150 mM extracellular [Na+] ) with voltage-dependent activation and inactivation similar to NaChBac ( Figure 1—figure supplement 1A , B ) . The time constant of NsvBa Na+ current inactivation ( τinact ) measured at 0 mV was 42 ms , ∼2 times faster than NaChBac ( 78 ms ) but ∼6 times slower than NavMs ( 7 ms , Figure 1—figure supplement 1C ) . The sequence TLESWxxG is conserved in the selectivity filters of prokaryotic Na+ channels ( Figure 1C ) . Although the sequence of NsvBa is homologous to other prokaryotic Nav channels , the selectivity filter sequence ( TLDSWGSG ) deviates at the high field-strength site ( SiteHFS most extracellular site; see Discussion ) with an aspartate residue replacing glutamate ( D186 ) and a flexible glycine at an extracellular position ( G189 ) . When aligned , G189 and G191 form a ‘GxG motif’ commonly found in K+-selective ( Kv ) and nonselective ion channels ( e . g . , TRP and CNG channels , Figure 1C ) . To determine the selectivity of the NsvBa channel , we patch clamped transiently-transfected HEK cells and measured voltage-dependent currents in the presence of monovalent alkali ions ( Li+ , Na+ , K+ , Rb+ , Cs+ ) or divalent alkaline earth metals ( Mg2+ , Ca2+ , Sr2+ , Ba2+ ) ( Figure 2A , B ) . Compared to the relatively Na+-selective NaChBac channel , NsvBa was nonselective and all cations but Cs+ and Mg2+ permeated the pore ( Figure 2C , D ) . NsvBa Na+ and K+ single channel conductances were equivalent ( 30 ± 3 pS and 36 ± 3 pS , respectively , Figure 2—figure supplement 1 ) , indicating that the channel does not distinguish between these ions . 10 . 7554/eLife . 04387 . 005Figure 2 . Comparison of cation selectivity between the nonselective NsvBa and Na+-selective NaChBac channels . ( A and C ) Representative current traces from NsvBa channel ( A ) or from the NaChBac channel ( C ) showing the first 0 . 25 s of 0 . 5 s activations from −180 mV holding potential: Top , 150 mM Na+ was substituted with an equal concentration of the indicated monovalent ions; Bottom , 110 mM Na+ was substituted for equal concentrations of the indicated divalent cations . ( B and D ) Resulting current-voltage relationships measured for the conditions tested in ( A ) and ( C ) . NsvBa: n = 5–9 , NaChBac: n = 6–9; Error = ±SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 04387 . 00510 . 7554/eLife . 04387 . 006Figure 2—figure supplement 1 . Na+ and K+ are highly conductive through the nonselective NsvBa channel . ( A ) Example single channel opening events recorded from Na+ and K+ conditions in the inside-out patch configuration . ( B ) Average current amplitudes plotted over the indicated potentials ( n = 4 , Error = ±SEM ) . The conductance for the K+ ( 36 pS ) and Na+ ( 30 pS ) conditions was derived from the slope of a linear fit of the current amplitudes . DOI: http://dx . doi . org/10 . 7554/eLife . 04387 . 006 Given the 61% shared amino acid identity between NaChBac and NsvBa channels , we were able to introduce the NaChBac selectivity filter ( TLESWASG ) into the NsvBa channel to examine two relevant residues , D186E and G189A . These changes conferred Na+-selectivity onto the channel and abrogated all other monovalent and divalent conductances ( Figure 3A , B ) . When tested separately , the single mutation D186E was sufficient for Na+-selectivity ( Figure 3—figure supplement 1A , B ) , whereas the single mutation of G189A substantially reduced , but did not entirely abolish , the permeability of K+ and Rb+ ( Figure 3—figure supplement 1C , D ) . Furthermore , the G189A mutant channel still conducted Ca2+ and Ba2+ divalent ions , whereas the D186E mutation did not . These data suggest that a glutamate at SiteHFS selects Na+ among other cations while the ‘GxG motif’ likely provides an extracellular K+ or Rb+ ion coordination site , possibly involving the glycine backbone carbonyls as found in the selectivity filter of the potassium and NaK channels ( Doyle et al . , 1998; Jiang et al . , 2002; Alam and Jiang , 2009a; Alam and Jiang 2009b ) . Next we sought to determine whether the NsvBa filter sequence could be transferred into Na+-selective NaChBac . When the Na+-selectivity filter of NaChBac ( TLESWASG ) was mutated at two positions ( E192D:A195 G ) to conform to the NsvBa nonselective filter ( TLDSWGSG ) , ion selectivity was identical to that of NsvBa ( Figure 3C , D ) . This finding illustrates the mutual portability of selectivity between the voltage-gated cation channels in Bacillus species halodurans and alcalophilus AV . 10 . 7554/eLife . 04387 . 007Figure 3 . Reciprocal substitutions of the NsvBa and NaChBac filters transfers cation selectivity . ( A , B ) The selectivity mutant NsvBa channel containing the NaChBac selectivity filter sequence TLESWASG ( D186E: G189A ) or the ( C , D ) mutant NaChBac channel containing the NsvBa selectivity sequence TLDSWGSG ( E192D:A195G ) . ( A and C ) Representative current traces from the mutant channels under the same conditions described in Figure 2 . ( B and D ) Resulting current-voltage relationships measured for the mono- and divalent conditions . ( n = 4–9 for both channels , Error = ±SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04387 . 00710 . 7554/eLife . 04387 . 008Figure 3—figure supplement 1 . The effects of the single mutations D186E and G189A in the NsvBa selectivity filter . ( A and C ) Representative current traces recorded from the NsvBa selectivity filter mutations D186E ( sequence TLESWGSG ) and G189A ( sequence TLDSWASG ) in response to 10 mV depolarizing activation steps from −140 mV in ( top ) monovalent or ( bottom ) divalent extracellular conditions . ( B and D ) Resulting current–voltage relationships measured for the mono- ( 150 mM ) or divalent ( 110 mM ) conditions . Current amplitudes are normalized to either the maximum inward current recorded in the 150 mM or 110 mM Na+ condition ( n = 4–9 cells for both channels , Error = ±SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04387 . 008 The selectivity for Na+ ions in vertebrate Nav channels is attributed to an asymmetric ring of 4 amino acids ( Asp , Glu , Lys , and Ala: DEKA ) contributed by each of the pore-lining loops of the 4 domains ( Catterall , 2012 ) . Voltage gated calcium channels ( CaVs ) are thought to achieve Ca2+-selectivity by a symmetric ring of 4 glutamate residues ( EEEE ) , each contributed by one domain of the polypeptide ( Hess et al . , 1986 ) . In contrast , prokaryotes achieve Na+-selectivity from an apparent 4-fold symmetry of acidic residues , each from a subunit in the homomer . In previous studies , we demonstrated that the Na+-selective NaChBac ( TLESWASG ) filter could be converted into one that prefers the divalents Ba2+ and Ca2+ by introducing acidic residues into three positions in the filter sequence ( TLDDWADG ) ( Yue et al . , 2002 ) . This filter sequence also was grafted into the NavAb channel ( called CavAb ) , shown to be more Ca2+-selective , and the high-resolution structure determined ( Tang et al . , 2014 ) . We tested the LDDWADG mutated filter ( S187D:G189A:S190D ) in the NsvBa channel and confirmed that it was divalent permeant ( Ca2+ , Sr2+ , Ba2+ ) but it also had measurable permeability to monovalent cations ( including K+ and Rb+ ) , demonstrating that Ca2+-preference achieved by this filter ( PCa/PNa∼30 , Figure 4; Figure 4—figure supplement 1 ) is much lower than mammalian CaV channels ( PCa/PNa ≳ 1000 ) ( Tsien et al . , 1987 ) . Rather , this selectivity is more analogous to some members of the TRP channel family , such as TRPV5 , TRPV6 ( Owsianik et al . , 2006; Wu et al . , 2010 ) . A comparison of the filter mutations effects on relative permeability of the NaChBac and NsvBa channels are summarized in Figure 4 and listed in Figure 4—figure supplement 2 . We also attempted to convert the NsvBa channel into a K+-selective channel by changing the selectivity filter sequence ( TLTSWGSG and TLTSWGYG ) , but these channels either did not express on the plasma membrane or did not conduct cations under our experimental conditions ( data not shown ) . 10 . 7554/eLife . 04387 . 009Figure 4 . Summary of the relative permeability of cations from selectivity filter mutations . The relative permeability of monovalent and divalent cations against sodium for each channel tested . Values are listed in and Figure 4—figure supplement 1 . The relative permeability ( Px/PNa ) was estimated using the Goldman-Hodgkin-Katz equation . Dashed lines indicate the lower limit of Px/PNa detection under our experimental conditions ( See ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04387 . 00910 . 7554/eLife . 04387 . 010Figure 4—figure supplement 1 . NsvBa can be converted into a divalent cation-selective channel . The selectivity mutant NsvBa channel was replaced with the CaVBac selectivity filter sequence TLDDWADG ( S187D:G189D:S190D ) . ( A ) Representative current traces from the mutant channels under the same conditions described in Figure 2 . ( B ) Resulting current-voltage relationships measured for the mono and divalent conditions . ( n = 4–8 for each channel , Error = ±SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04387 . 01010 . 7554/eLife . 04387 . 011Figure 4—figure supplement 2 . Reversal potentials ( Erev ) measured at steady state with calculated relative permeability ( Px/PNa ) for bacterial Nav channels . DOI: http://dx . doi . org/10 . 7554/eLife . 04387 . 011 Alkaliphilic Bacillus are estimated to have very negative resting membrane potentials ( Ψrest ≈ −180 mV ) , although membrane potentials in bacteria are measured from voltage- sensitive dye studies and variability within populations can be large . Nevertheless , the activation threshold for Bacillus sodium channels is ≈ −40 mV , which is extremely depolarized relative to estimates of Ψrest . Since alkaliphilic bacteria live in high pH environments , we tested whether their sodium channel gating shifted as a function of pH . As shown for the Na+-selective channel from Bacillus pseudofirmus OF4 ( NavBp ) ( Ito et al . , 2004 ) , Na+ currents from NaChBac and NsvBa are also modulated by high extracellular pH ( Figure 5 ) . When extracellular pH was increased from 7 . 4 to 9 . 4 , the peak current increased twofold to fourfold and the steady state voltage-dependence was negatively shifted by 28–34 mV ( Figure 5—figure supplement 1 ) . Basic extracellular pH alone is probably insufficient to reduce this substantial energy barrier to activate these channels from Ψrest = −180 mV ( ≈−3 . 2 kcal/mol ) . Thus additional influences are required to bring Ψrest and V1/2 closer together . 10 . 7554/eLife . 04387 . 012Figure 5 . The bacterial Nav channels are modulated by extracellular alkaline pH ( pHo ) . Left , representative traces recorded from one cell expressing the NsvBa ( black ) , NaChBac ( gray ) and hNav1 . 1 ( green ) channels in 150 mM NaCl conditions with the external pH adjusted to 6 . 4 , 7 . 4 , 8 . 4 and 9 . 4 . Currents were activated by +10 mV steps from a holding potential of −140 mV ( NsvBa and NaChBac ) or −120 mV ( hNav1 . 1 ) . Right , resulting voltage current relationship normalized to the peak current measured in the 7 . 4 pHo condition ( n = 4–7 for each channel , Error = ±SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04387 . 01210 . 7554/eLife . 04387 . 013Figure 5—figure supplement 1 . Nav steady state voltage-dependence of activation ( V1/2 ) measured in different extracellular pH ( pHo ) conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 04387 . 013 Many alkaliphilic Bacillus species growth rates are temperature-dependent ( 30–60°C ) . Thus we tested the effects of temperature ( 20–37°C ) at neutral and basic extracellular pH ( 7 . 4 and 9 . 4 , respectively ) on the Na+ currents conducted by Bacillus NsvBa , NaChBac and NavBp channels ( Figure 6A–C ) . At neutral pH , we observed that increasing the temperature shifted the voltage dependence of activation for these channels by −19 to −24 mV ( −62 , −55 , −51 mV respectively ) . When tested together , temperature and basic pH effects on V1/2 for NsvBa , NaChBac and NavBp were additive , converging on ≈−100 mV ( −95 , −102 , −100 mV respectively ) . Importantly , we determined that the NsvBa remains a non-selective channel at higher pH and temperature , although the Px/PNa for K+ and Ca2+ did increase slightly 2–3 times ( Figure 6—figure supplement 1 ) . To quantify the temperature sensitivity of these channels , the relationship between the peak current during a voltage ramp and temperature was fit to a linear equation to determine the 10-degree temperature coefficient ( Q10 ) . The Q10 for NaChBac and NavBp peak currents was 3 . 5–4 . 4 at neutral pH and 3 . 5–4 . 1 at basic pH . In contrast , the voltage-dependence of activation of the hNav1 . 1 channel was less temperature- ( Q10 = 1 . 2 and 1 . 4 , Figure 6D ) and pH- ( Δ V1/2 < 8 mV , Figure 5—figure supplement 1 ) sensitive . Thus , pH and temperature-induced increases of the peak current and reduction of the voltage-dependence of activation are distinct from eukaryotic Nav channels . 10 . 7554/eLife . 04387 . 014Figure 6 . Temperature and pH dependence of sodium channels . Left Example INa traces conducted by ( A ) NsvBa , ( B ) NaChBac , ( C ) NavBp and ( D ) Nav1 . 1 channels , when the extracellular saline ( pH = 7 . 4 or 9 . 4 ) was heated from 20 to 37°C . Channels were activated by a 0 . 5 Hz voltage ramp . Voltage ramps were applied for different durations to compensate for different channel kinetics of activation and inactivation: NsvBa and NaChBac ( 100 ms ) ; NavBp ( 200 ms ) and Nav1 . 1 ( 10 ms ) . Right , Arrhenius plots with resulting peak current ( plus symbols ) and V1/2 ( open squares ) are graphed as a function of temperature . The peak currents were fit to a linear equation and the resulting slope ( Peak Q10 ) given for both external pH conditions ( n = 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04387 . 01410 . 7554/eLife . 04387 . 015Figure 6—figure supplement 1 . The effect of temperature and pH on NsvBa selectivity . ( A ) Top , Representative NsvBa current traces recorded at 34⁰C under different extracellular cationic conditions with a pH = 9 . Currents were activated by increasing potentials from −10 mV to 70 mV from a holding potential of −140 mV . Bottom , Corresponding current–voltage relationships measured under the conditions listed above ( n = 4 or 5 for each condition , Error = ±SEM ) . Peak inward currents were normalized to the Na+ current measured either in 110 or 150 mM extracellular solution . Erev was calculated by fitting the current from 0 mV to 60 mV to a linear equation , and determining the potential at zero current . When corrected for liquid junction potential differences in the salines , Erev under these conditions were 40 mV for 150 mM Na+; 36 mV for 110 mM Na+; 41 mV for 110 mM Ca2+ and 43 mV for 150 mM KCl . ( B ) The calculated relative permeability for cations measured at 22⁰C and 34⁰C . DOI: http://dx . doi . org/10 . 7554/eLife . 04387 . 015 The stimuli that depolarize these bacteria from −180 mV to the more depolarized range where voltage-gated channels activate ( −40 to −100 mV , depending on pH and temperature ) , are not known . We speculate that Ψrest declines as bacterial pumps are starved for internal H+ or Na+ to supply the hyperpolarizing extrusion pumps . Internal [Na+] would be rapidly recharged by activation of the voltage-gated monovalent cation channels . Since cation entry into alkaliphilic bacteria is at least partially dependent on Nav channels , we hypothesized that Nav channel antagonists would attenuate bacterial growth . Under voltage clamp , we observed that sodium current from Bacillus channels NavBp , NaChBac and NsvBa are blocked by known Nav channel antagonists , the local anesthetic lidocaine , the anti-hypertensive nifedipine , and the anti-estrogen tamoxifen , with similar potencies ( Figure 7A , B and Figure 7—figure supplement 1 ) . When these drugs were introduced into the culture media , growth of Bacillus species alcalophilus and pseudofirmus were severely impaired as measured by spectroscopic absorbance ( Figure 7C , D ) . The measured half-inhibitory concentrations ( IC50 ) of sodium current and bacterial growth were within a half-log unit ( Figure 7—figure supplement 1 ) , suggesting that growth inhibition was not an off-target effect . We also examined the effect of these drugs on two matrices of Bacillus motility: tumbling frequency and swim speed . tamoxifen , nifedipine and lidocaine increased tumbling frequency and decreased swim speed of B . pseudofirmus ( Figure 7E , Figure 7—figure supplement 1 ) . However , B . alcalophilus swim speed was not delayed by the three drugs and only tamoxifen and nifedipine increased tumbling frequency ( EC50 = 70 μM and 375 μM , respectively ) . 10 . 7554/eLife . 04387 . 016Figure 7 . Antagonism of Nav channels blocks the growth and motility of alkaliphilic Bacillus . ( A ) Example Na+ currents from NaChBac , NsvBa and NavBp channels in the presence of vehicle control ( ≤0 . 1% DMSO ) and extracellularly applied drugs . Currents were activated by a 0 . 2 Hz train of 500 ms depolarizations to 0 mV from −140 mV . ( B ) The resulting NsvBa ( open circles ) and NaChBac ( filled triangles ) Na+ current block-drug relationship by tamoxifen , nifedipine and lidocaine ( n = 4–6 cells per concentration , Error = ±SEM ) . ( C ) The effect of drugs on the time course of bacterial growth . ( D ) The resulting drug antagonism of growth by Bacillus alcalophilus ( open circles ) and Bacillus pseudofirmus ( filled squares ) inferred by light spectroscopy 600 λ absorbance ( n = 4 growth trials , Error = ±SEM ) . Tamoxifen is a fluorescent compound which significantly absorbs 600 λ light at concentrations above 30 μM , indicated by a hashtag ( # ) . ( E ) The effect of nifedipine , tamoxifen , and lidocaine on Bacillus alcalophilus ( open circles ) and pseudofirmus ( filled squares ) tumbling and swim speed ( n = 4 trials , Error = ±SEM ) . Statistical significance of motility in untreated and drug-treated cells using a Student's t-test and indicated by an asterisk ( p < 0 . 05 ) ( * ) . EC50 and IC50 values for all assays are listed in Figure 7—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04387 . 01610 . 7554/eLife . 04387 . 017Figure 7—figure supplement 1 . The effect of Nav antagonist on INa , bacteria motility and growth by drugs: lidocaine , tamoxifen and nifedipine . DOI: http://dx . doi . org/10 . 7554/eLife . 04387 . 017 We have functionally characterized NsvBa , a cation nonselective voltage-gated ion channel from Bacillus alcalophilus . Our findings demonstrate that this family of prokaryotic voltage-gated Na+ channels are not exclusively Na+-selective and that the filter sequence that controls passage of cations into the bacteria are transferable within the Bacillus genus . Through a series of mutations , we showed that the selectivity of Bacillus Nav channels can be converted from nonselective , to relatively higher Na+ or Ca2+ selectivity , suggesting that the filter is readily mutable in evolution to adapt to ionic conditions . These highly adaptable selectivity filters are critical for the Bacillus alkaliphiles , allowing for the habitation of various cation-rich environments in which they presumably evolved . It is anticipated that the ion-specificity of at least some antiporters will parallel that of the major coupling ions for the voltage-gated channels that provide a significant amount of their cytoplasmic substrate . Although no Na+ ions reside within the filter of the wild type NavAb crystal structures , 3 Na+ coordinating sites were proposed: glutamate side chains form a high-field-strength site ( SiteHFS ) near the extracellular end of the filter , while backbone carbonyls of Leu and Thr comprise central ( SiteCEN ) and inner sites ( SiteIN ) ( Payandeh et al . , 2011 ) . In the CavAb structure , 3 hydrated Ca2+ ions were found coordinated within the filter and 3 sites ( Sites 1–3 ) proposed within the filter . Site 1 , the one closest to the extracellular surface , is formed by 4 Asp carboxyl side chains , equivalent to positions Ser 196 in NaChBac and Ser 190 in NsvBa . Site 2 ( equivalent to SiteHFS in the wt NavAb crystal structure ) is formed by four side chain carboxyls and four backbone carbonyls from equivalent residues of Glu 192 in NaChBac and Asp 186 in NsvBa . Our data suggest that the shorter side chain of the NsvBa Asp acidic residue at site 2 is correlated with decreased Na+-selectivity , but Na+ selectivity can be artificially endowed when replaced by a Glu ( D186E ) ( Figure 3—figure supplement 1 ) . In agreement with this interpretation , a recent report demonstrates that when the longer side chain Glu 192 is mutated to the shorter Asp in NaChBac ( E192D ) , the channel becomes less selective for Na+ among Ca2+ and K+ ions ( Finol-Urdaneta et al . , 2014 ) . The full-length bacteria NaV from marine α-proteobacterium Rickettsiales HIMB114 ( NavRh ) was crystallized containing the selectivity filter TLSSWET- . Although the NavRh channel did not function when heterologously expressed in mammalian and insect cells , the filter was found to be Na+-selective when grafted into the NaChBac channel . It is surprising that although the glutamates within the NavRh and NavAb filters originate at distinct positions ( See Figure 1 ) , the carboxyl side chains from both channels occupy the same positions in space ( Zhang et al . , 2012 ) . Thus , it is likely that selectivity for Na+ over other cations is partly achieved by the geometry of the arrangement of the carboxylates at Site 2 ( SiteHFS ) within the filter of bacterial Na+ channels . High-resolution crystal structures of Nav channels in which sodium ions are resolved are needed to confirm this hypothesis . Among all sequenced living species , naturally occurring nonselective voltage-gated ion channels are rare . It is not clear whether cation-selective or nonselective filters arose first within prokaryotes . Analysis comparing domains of bacteria and mammalian Nav channels indicate that Na+-selectivity was independently acquired in these channels families ( Liebeskind et al . , 2013 ) . Bacterial Nav channels are functionally more related to homotetrameric CatSper and Kv channels and thus may not be direct ancestors of mammalian Nav channels . Our results are consistent with this interpretation . Not surprisingly , we were not able convert Bacillus Nav channels to K+-selective channels by mutating the filter sequence , presumably since the pore architecture between Kv and bacterial Nav channels ( e . g . lumen size , filter length and number of pore helices ) are structurally divergent . Our results suggest that the ‘GxG motif’ found in the NsvBa filter evolved convergently within the bacterial Nav family and is only distantly related to K+-selective Kv channels . A recent report on the expression and characterization of Na+ channel homologs from the invertebrate sea anemone Nematostella vectensis ( NvNaV2 . 1 ) , revealed a heterotetrameric Na+ channels bearing noncanonical selectivity site ( DEEA ) that was not selective among K+ and Na+ ions ( Gur Barzilai et al . , 2012 ) . Thus it appears that the homomeric prokaryotic NsvBa and functionally-heteromeric eukaryotic NvNaV2 . 1 channels are examples of evolution of Na+-selectivity to non-selectivity that occurred independently within prokaryotes and metazoans . In excitable eukaryotic tissues such as nerves and muscle , the half-activation threshold ( V1/2 ) of voltage-gated sodium channels is within ∼80 mV of their cellular resting membrane potential ( Ψrest ∼ -60 to −90 mV ) . At first glance , this appears to contrast with the 140 mV difference between the Ψrest and V1/2 of alkaliphilic Bacillus . But as we have shown , increasing alkalinity and temperature shift bacterial Nav V1/2's to approximately −100 mV . At temperatures approaching 40°C , our patch recordings became unstable , thus limiting the temperature range we tested to <37°C . If the sodium current sensitivity is extrapolated to 60°C , the voltage dependence would likely be well within 80 mV of the most negative bacterial resting membrane potential . Since bacterial Ψrest is a function of metabolic state ( Na+/H+ pumping ) and bacteria occupy more variable ionic environments than neurons , the larger difference between ‘optimal’ a Ψrest and V1/2 makes physiological sense . The high temperature and pH sensitivity of bacterial Nav channels is similar to TRP and some Kv channels ( Patapoutian et al . , 2003; Yang and Zheng , 2014 ) , and is not shared by vertebrate Nav channels . We suspect that the fidelity of neuronal firing in metazoans , which depend on their NaV channels , led to evolution of less sensitivity to pH and temperature , perhaps by selection of the residues that become exposed to waters within the voltage-sensing S4 and selectivity filter ( Chowdhury et al . , 2014 ) . In contrast , we propose that alkaliphilic bacteria , which dwell in high temperature ( 35–60°C ) and high pH ( 9–11 ) environments , employ homomeric Nav channels for an entirely different purpose , that of rapidly adjusting internal sodium concentration for metabolic control . This would thus impact growth rates of these Bacillus species as we have observed . Concentrations of lidocaine , nifedipine , and tamoxifen that inhibit heterologously-expressed bacterial Nav channels ( NavBp and NsvBa ) also block the growth of native Bacillus species ( B . pseudofirmus and B . alcalophilus ) . Although all three drugs affected motility for B . pseudofirmus , only tamoxifen and nifedipine increased tumbling frequency of B . alcalophilus . These results suggest that in B . alcalophilus there may be other transporters or channels besides NsvBa that provide a cation source to drive flagellar motion . These finding demonstrate that block of Na+ entry via Nav , which disrupts the sodium cycle , can impair Bacillus motility and growth . The prokaryotic Nav from Magnetococcus marinus ( NavMs ) , was recently crystalized with a brominated drugs bound and the proposed binding site validated ( Bagneris et al . , 2014 ) . Thus voltage-gated channels found in pathogenic Bacillus ( e . g . Bacillus cereus and anthracis ) could potentially be antimicrobial targets . HEK 293T cells were transiently transfected with mammalian cell expression plasmid pTracer CMV2 containing either NsvBa or NaChBac genes . The NCBI GenBank accession number for NsvBa is JX399518 . 1 and is annotated as a K+ ion transporter ( BalcAV3624 ) . Cells were seeded onto glass coverslips , and placed in a perfusion chamber for experiments where extracellular conditions could be altered . With the exception of the experiments described in Figure 5 , Figure 2—figure supplement 1 and Figure 5—figure supplement 1 , the pipette electrode solution contained ( in mM ) : NMDG ( 90 ) , NaCl ( 20 ) , HEPES ( 10 ) , EGTA ( 5 ) , CaCl2 ( 0 . 5 ) and pH was adjusted to 7 . 4 with HCl . When testing the relative permeability of monovalent cations , the bath solution contained ( in mM ) : X-Cl ( 150 ) , HEPES ( 10 ) , EGTA ( 5 ) , CaCl2 ( 0 . 5 ) and the pH was adjusted with X-OH , where the X is the indicated monovalent cation . When testing the relative permeability of divalent cations , the bath solution contained: X-Cl2 ( 110 ) , HEPES ( 10 ) , EGTA ( 5 ) , CaCl2 ( 0 . 5 ) and the pH was adjusted with X- ( OH ) 2 , where the X is the indicated divalent cation . For the experiments in Figure 5 , extracellular saline contained NaCl ( 150 ) , HEPES ( 10 ) , CaCl2 ( 2 ) and pH was adjusted with CsOH so that the concentration of permeant Na+ ion remained constant at high pH . All saline solutions were adjusted to 300 mOsm ( ±5 ) with mannitol , if needed . Data were analyzed by Igor Pro 7 . 00 ( Wavemetrics , Lake Oswego , OR ) . Residual leak ( >−100 pA ) and capacitance were subtracted using a standard P/4 protocol . Current-voltage-relationships were fit with the following equation:I=V−Erev{1+exp ( V−V12RT/F ) where I is current and Erev is the extrapolated reversal potential . Erev was used to determine the relative permeability of monovalent cation X to Na ( Px/PNa ) according to the following equation ( Hille , 1972; Sun et al . , 1997 ) . PxPNa=αNaeαxe[exp ( ΔErevRT/F ) ]where ΔErev , αx , R , T and F are the reversal potential , effective activity coefficients for cation x ( i , internal and e , external ) , the universal gas constant , absolute temperature , and the Faraday constant , respectively . For these pseudo-bionic conditions , we are assuming that the internal NMDG is impermeant—it has a similar ionic radius as cesium , which is impermeant in all of the Nav channels we tested . The effective activity coefficients ( αx ) were calculated using the following equations:αx=γx[X]where γx is the activity coefficient and [X] is the concentration of the ion . For calculations of membrane permeability , activity coefficients ( γ ) was calculated using the Debye-Hückel equation: 0 . 74 , 0 . 76 , 0 . 72 , 0 . 71 , 0 . 69 , 0 . 35 , 0 . 29 , 0 . 27 and 0 . 27 correspond to Na+ , Li+ , K+ , Rb+ , Cs+ , Mg2+ , Ca2+ , Sr2+ and Ba2+ respectively . To determine the relative permeably of divalent cations to Na+ , the following equation was used:PxPNa={αNai[exp ( ErevFRT ) ][exp ( ErevFRT ) +1]}4αxe Erev for each cation condition was corrected to the measured liquid junction potentials ( −4 . 4 to 3 . 4 mV ) . In some cationic conditions , no inward ( negative ) voltage-dependent currents could be activated , but Erev was measured as ≤−4 mV . In these cases , the lower limit of Px/PNa was reached ( 0 . 1 ) due to low levels of endogenous chloride and nonselective currents in HEK cells effecting Erev . The voltage-dependence of channel activation and inactivation was fit to the equation:Inorm=Imax− ( Inorm−Imin ) / ( 1+exp[V−V12k] ) where Imax and Imin are the maximum and minimum current values , V is the applied voltage , V1/2 is the voltage at half activation , and k is the slope factor . For the single channel conductance measurements , experimental conditions were the same except that the pipette and bath saline solutions were switched . The equation for the exponential fits used in to estimate the rate of inactivation was:f ( x ) =B+A exp[ ( 1τ ) x]where τ is the time constant . For the experiments performed in Figure 6 , where extracellular pH and/ or temperature were altered , the pipette electrode solution contained ( in mM ) : CsMES ( 100 ) , HEPES ( 10 ) , NaCl ( 15 ) , EGTA ( 6 ) , TAPS ( 5 ) , MgCl2 ( 2 ) , CaCl2 ( 3 ) and pH was adjusted to 6 . 4–9 . 4 with CsOH or HCl . For temperature-controlled experiments , the perfusate was heated and cooled at rate of 0 . 4–2°C / s using a Warner TC-344B heater controller and Warner SHM-6 solution heater while bath temperature was monitored using a thermistor placed in close proximity to the recording electrode . A linear fit of the peak currents during the voltage ramp was used to determine Q10 , as described by the Arrhenius equation:Q10=[R2R1]10/ ( T2−T1 ) Where R = rate and T = temperature B . alcalophilus and B . pseudofirmus OF4 were grown in malate yeast extract ( MYE ) or KMYE ( 50 mM K2CO3 instead of Na2CO3 in MYE ) medium with shaking at 37°C for 6 hr . Cells were suspended in 1 . 0 ml of the MYE or KMYE medium ( with or without 10 , 30 , 100 , or 300 µM nifedipine ) , and incubated at 37°C for 10 min . Microscopic observation was carried out immediately by the hanging drop method using a Leica DMLB100 dark field microscope ( 400× ) and Leica DC300F camera , Leica IM50 version 1 . 20 software ( Leica Geosystems , Tokyo ) , and recorded with Display capture ARE software ( http://www . vector . co . jp/soft/win95/art/se221399 . html ) . The swimming speed of 40 individual cells ( swimming for more than 15 s ) , and swimming fractions of more than 50 individual cells were measured by 2D movement measurement capture 2D-PTV software ( DigimoCo . , Ltd . ) . All results shown are the averages of three independent experiments . Drug antagonism of bacteria growth was assessed using the equation:Abcontrol−AbdrugAbcontrol×100where Abcontrol and Abdrug are the maximum spectroscopic absorbance at 600λ measured within 16 hr of bacterial growth in untreated and drug treated conditions , respectively . The growth inhibition-drug concentration relations were fit using the Hill equation described above .	Life essentially runs on electricity: electrical signals cause nerve cells to fire , heart muscles to contract and allow organisms to sense the world around them . These signals are triggered by the movement of positively-charged ions—such as sodium , potassium and calcium—moving into a cell through special ion channels in the cell membrane , which can open and close in response to changes in the voltage across the cell membrane . With few exceptions , voltage sensitive ion channels usually only let one type of ion pass into the cell . But how do ion channels discriminate amongst ions and how did they acquire this ability during evolution ? To address these questions , researchers have studied a family of sodium channels from bacteria for the past decade . Here DeCaen et al . describe a new member from this ion channel family from a bacterium called Bacillus alcalophilus . This ion channel does not discriminate between positively-charged ions and B . alcalophilus needs this ion channel for it to dwell in environments that have high levels of potassium or sodium . DeCaen et al . demonstrate that these ion channels can be made selective for sodium or calcium with as little as two small changes in the gene that encodes the ion channel . Furthermore , making similar genetic mutations in related ion channel genes from other Bacillus species has the same effect . DeCaen et al . suggest that Bacillus ion channel genes are easily adapted to function in a variety of environmental conditions with different levels of positively-charged ions . Thus it is easier for Bacillus channels to evolve to be selective for different ions . Bacillus bacteria divide rapidly in warm to hot temperatures and under alkaline pH . DeCaen et al . demonstrate that both of these conditions make Bacillus ion channels easier to open in response to voltage . In addition , DeCaen et al . demonstrate that Bacillus ion channels can be targeted by drugs that impair the ability of the bacteria to grow . These findings—together with other work that revealed where drug molecules bind to ion channels—could potentially guide efforts to develop treatments for illnesses caused by other Bacillus strains , which include anthrax and some forms of food poisoning .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "microbiology", "and", "infectious", "disease" ]	2014	Ionic selectivity and thermal adaptations within the voltage-gated sodium channel family of alkaliphilic Bacillus
Brief periods of sleep loss have long-lasting consequences such as impaired memory consolidation . Structural changes in synaptic connectivity have been proposed as a substrate of memory storage . Here , we examine the impact of brief periods of sleep deprivation on dendritic structure . In mice , we find that five hours of sleep deprivation decreases dendritic spine numbers selectively in hippocampal area CA1 and increased activity of the filamentous actin severing protein cofilin . Recovery sleep normalizes these structural alterations . Suppression of cofilin function prevents spine loss , deficits in hippocampal synaptic plasticity , and impairments in long-term memory caused by sleep deprivation . The elevated cofilin activity is caused by cAMP-degrading phosphodiesterase-4A5 ( PDE4A5 ) , which hampers cAMP-PKA-LIMK signaling . Attenuating PDE4A5 function prevents changes in cAMP-PKA-LIMK-cofilin signaling and cognitive deficits associated with sleep deprivation . Our work demonstrates the necessity of an intact cAMP-PDE4-PKA-LIMK-cofilin activation-signaling pathway for sleep deprivation-induced memory disruption and reduction in hippocampal spine density . Sleep is a ubiquitous phenomenon and most species , including humans , spend a significant time asleep . Although the function of sleep remains unknown , it is widely acknowledged that sleep is crucial for proper brain function . Indeed , learning and memory , particularly those types mediated by the hippocampus , are promoted by sleep and disrupted by sleep deprivation ( Havekes et al . , 2012a; Abel et al . , 2013; Whitney and Hinson , 2010 ) . Despite the general consensus that sleep deprivation impairs hippocampal function , the molecular signaling complexes and cellular circuits by which sleep deprivation leads to cognitive deficits remain to be defined . The alternation of wakefulness and sleep has a profound impact on synaptic function , with changes observed in synaptic plasticity and transmission ( Havekes et al . , 2012a; Abel et al . , 2013; Tononi and Cirelli , 2014 ) . This relationship has led to the development of influential theories on the function of sleep ( Tononi and Cirelli , 2006; Pavlides and Winson , 1989 ) . Recent imaging suggests that dendritic structure is dynamic , especially during development , with alterations in spine numbers correlating with changes in sleep/wake state ( Maret et al . , 2011; Yang and Gan , 2012 ) . However , the impact of sleep deprivation or sleep on synaptic structure in the hippocampus in the context of memory storage or synaptic plasticity has not been examined . This is an important issue , as such structural changes in ensembles of synapses have been shown to play a critical role in memory storage ( Caroni et al . , 2012; Vogel-Ciernia et al . , 2013 ) . The formation of associative memories increases the number of dendritic spines in area CA1 of the hippocampus ( Leuner et al . , 2003 ) . Also , the induction of long-term potentiation ( LTP ) , a cellular correlate of memory storage ( Mayford et al . , 2012 ) , is associated with an increase in spine density in cultured hippocampal neurons ( Oe et al . , 2013 ) . In addition to a critical function during development ( Gurniak et al . , 2005 ) , cofilin plays an essential role in synapse structure by mediating both the enlargement and pruning of dendritic spines ( Rust , 2015; Bamburg , 1999; Bosch et al . , 2014 ) . The activity of cofilin is negatively regulated by phosphorylation . Specifically , phosphorylation of serine 3 of cofilin suppresses its depolymerizing and F-actin severing activity ( Bamburg , 1999 ) . Importantly , increased cofilin activity can lead to the depolymerization and severing of F-actin , which in turn results in the shrinkage and loss of spines ( Rust , 2015; Zhou et al . , 2004; Davis et al . , 2011; Shankar et al . , 2007 ) . Hippocampal cofilin phosphorylation levels are increased after the induction of long-term potentiation ( LTP ) ( Rex et al . , 2010; Chen et al . , 2007; Briz et al . , 2015 ) , and during memory consolidation ( Fedulov et al . , 2007; Suzuki et al . , 2011 ) . Additionally , elevated cofilin activity in the hippocampus was recently implicated in abnormal spine structure and function in mutant mice with altered chromatin remodeling ( Vogel-Ciernia et al . , 2013 ) . Here we show for the first time that 5 hr of sleep deprivation leads to the loss of dendritic spines of CA1 , but not CA3 , neurons in the dorsal hippocampus . The spine loss in CA1 neurons was accompanied by reductions in dendrite length . This process was readily reversed by sleep , with just 3 hr of recovery sleep normalizing this spine loss and dendrite length . The molecular mechanisms underlying these negative effects of sleep deprivation were shown to target cofilin , whose elevated activity could contribute to spine loss . Indeed , suppression of cofilin activity in hippocampal neurons prevented the structural , biochemical , and electrophysiological changes as well as the cognitive impairments associated with sleep loss . The elevated cofilin activity is caused by the activity of the cAMP degrading phosphodiesterase-4A5 isoform ( PDE4A5 ) , which suppresses activity of the cAMP-PKA-LIMK pathway . Genetic inhibition of the PDE4A5 isoform in hippocampal neurons restores LIMK and cofilin phosphorylation levels and prevents the cognitive impairments associated with sleep loss . Thus changes in the cAMP-PDE4-PKA-LIMK-cofilin signaling pathway in the adult hippocampus underlie the cognitive deficits associated with sleep loss . These observations provide a molecular model for the notion that prolonged wakefulness reduces structural signaling and negatively impacts dendritic structure , which is then restored with sleep . To determine whether short periods of sleep loss affect dendritic structure in the hippocampus , we used Golgi staining to examine the length of dendrites and number of dendritic spines in the mouse hippocampus following 5 hr of sleep deprivation , a period of sleep loss that is known to impair selectively hippocampus-dependent memory consolidation and synaptic plasticity ( Havekes et al . , 2012a; Abel et al . , 2013; Graves et al . , 2003; Vecsey et al . , 2009; Havekes et al . , 2014 ) . Analyses of Golgi-impregnated CA1 neurons ( Figure 1A ) indicated that sleep deprivation significantly reduced the apical/basal spine density ( Figure 1B; spine numbers per dendrite , NSD: 1 . 42 ± 0 . 03 , SD: 1 . 17 ± 0 . 02; Student’s t-test , p=0 . 0002 ) and dendrite length ( Figure 1C; NSD: 1198 . 4 ± 31 . 6 , SD: 984 . 2 ± 29 . 8 µm; Student’s t-test , p=0 . 0012 ) . This decrease in spine density and dendrite length was observed in both apical and basal dendrites ( Figure 1—figure supplement 1A , B ) . To complement our Golgi studies , we conducted an additional experiment in which individual CA1 neurons in hippocampal slices from sleep deprived and non-sleep deprived mice were labeled using the DiI method as described ( Seabold et al . , 2010 ) . In line with our Golgi studies , we found that sleep deprivation significantly reduced the total number of spines on apical dendrites of CA1 neurons ( Figure 1D; NSD: 1 . 0 ± 0 . 03 , SD: 0 . 84 ± 0 . 04 Student’s t-test , p=0 . 033; Figure 1E; NSD: 1 . 0 ± 0 . 06 , SD: 0 . 86 ± 0 . 02 Student’s t-test , p=0 . 03 ) . 10 . 7554/eLife . 13424 . 003Figure 1 . Sleep deprivation reduces spine numbers and dendrite length in CA1 neurons of the hippocampus . ( A ) Representative images of Golgi-impregnated dendritic spines of CA1 pyramidal neurons from sleep deprived ( SD ) and non-sleep deprived ( NSD ) mice . Scale bar , 5 µm . ( B ) Sleep deprivation reduces the spine density of apical/basal dendrites of CA1 neurons ( n = 5–6 , Student’s t-test , p=0 . 0002 ) . ( C ) Sleep deprivation decreases apical/basal dendrite length of CA1 neurons ( n = 5–6 , Student’s t-test , p=0 . 0012 ) . ( D , E ) Comparative analyses of spine numbers in the second-third branch of apical dendrites of CA1 neurons reveal a significant reduction as a result of sleep deprivation using either the DiI labeling method ( n = 3–4 , Student’s t-test , p=0 . 03 ) or Golgi analyses ( n = 5 , Student’s t-test , p=0 . 03 ) . Importantly , for the comparison of the two methods we focused on the second and third branch of the apical dendrites . See also the Materials and methods section . ( F ) Sleep deprivation reduces the number of all spine types in apical/basal dendrites of CA1 neurons ( n = 5–6 , Student’s t-test , p<0 . 005 ) . ( G ) Sleep deprivation reduces spine density of apical/basal dendrites between 60 and 150 µm away from the soma of CA1 neurons ( n = 5–6 , Student’s t-test , p<0 . 005 ) . ( H ) Sleep deprivation reduces apical/basal spine density in branch 3–9 of CA1 neurons ( n = 5–6 , Student’s t-test , p<0 . 005 ) . NSD: non-sleep deprived , SD: sleep deprived , Values represent the mean ± SEM . p<0 . 05 , *p<0 . 005 , by Student’s t test . See also Figure 1—figure supplement 1 and 2 for separate Golgi analyses of apical and basal spine numbers . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 00310 . 7554/eLife . 13424 . 004Figure 1—figure supplement 1 . Sleep deprivation decreases spine density and dendrite length in both basal and apical dendrites of CA1 neurons . ( A ) Sleep deprivation reduces the spine density of both basal and apical dendrites in CA1 neurons ( 5–6 animals per group , 5 neurons per animal ) . ( B ) Sleep deprivation decreases basal and apical dendrite length of CA1 neurons ( 5–6 animals per group , 5 neurons per animal ) . ( C ) Sleep deprivation reduces the number of all spine types in basal dendrites in CA1 neurons ( 5–6 animals per group , 5 neurons per animal ) . ( D ) Sleep deprivation reduces the number of all spine types with exception of filopodia spines in apical dendrites of CA1 neurons ( 5–6 animals per group , 5 neurons per animal ) . ( E ) Sleep deprivation reduces spine density of basal dendrites between 60 and 150 µm away from the soma ( 5–6 animals per group , 5 neurons per animal ) . ( F ) Sleep deprivation reduces spine density of apical dendrites between 60 and 150 µm away from the soma ( 5–6 animals per group , 5 neurons per animal ) . ( G ) Sleep deprivation decreases spine density in the third to sixth branch of basal dendrites of hippocampal CA1 neurons ( 5–6 animals per group , 5 neurons per animal ) . ( H ) Sleep deprivation decreases spine density in the third to ninth branch of apical dendrites in hippocampal CA1 neurons ( 5–6 animals per group , 5 neurons per animal ) . NSD: non-sleep deprived , SD: sleep deprived . Values represent the mean ± SEM . p<0 . 05 , p<0 . 01 , *p<0 . 0001 by Student’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 00410 . 7554/eLife . 13424 . 005Figure 1—figure supplement 2 . Sleep deprivation does not reduce spine density and dendrite length in both basal and apical dendrites of CA3 neurons . ( A ) Sleep deprivation does not alter the spine density of basal and apical dendrites in CA3 neurons ( 6 animals per group , 4 neurons per animal , Student’s t test p>0 . 28 ) . ( B ) Sleep deprivation does not alter basal and apical dendrite length of CA3 neurons ( 6 animals per group , 4 neurons per animal , Student’s t test , p>0 . 37 ) . ( C ) Sleep deprivation does not change the number of any spine type in basal dendrites of CA3 neurons ( 6 animals per group , 4 neurons per animal , Student’s t test , p>0 . 31 ) . ( D ) Sleep deprivation does not affect the number of any spine type in apical dendrites of CA3 neurons ( 6 animals per group , 4 neurons per animal , Student’s t test , p>0 . 31 ) . ( E ) Sleep deprivation does not alter the spine density of basal dendrites at any distance from the soma ( 6 animals per group , 4 neurons per animal , Student’s t test , p>0 . 05 ) . ( F ) Sleep deprivation does not impact spine density of apical dendrites at any distance from the soma ( 6 animals per group , 4 neurons per animal ) . ( G ) Sleep deprivation does not affect the number of spines of basal dendrites at any branch number ( 6 animals per group , 4 neurons per animal , Student’s t test , p>0 . 05 ) . ( H ) Sleep deprivation does not affect the number of spines of apical dendrites at any branch number with exception of the first apical branch ( 6 animals per group , 4 neurons per animal , Student’s t test , p>0 . 05; branch 1 Student’s t test , p<0 . 05 ) . NSD: non-sleep deprived , SD: sleep deprived . Values represent the mean ± SEM . p<0 . 05 , by Student’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 005 Subtype-specific apical/basal spine analyses of the Golgi impregnated neurons revealed a significant decrease for all spine subtypes in sleep-deprived mice ( Figure 1F , for all spine types , Student’s t-tests p<0 . 005 , for separate apical and basal spine analyses see Supplementary Figure 1C , D ) . Sleep deprivation causes the greatest reduction in apical/basal spine density between 60 µm and 150 µm from the soma ( Figure 1G , for separate apical and basal spine analyses see Figure 1—figure supplement 1E , F ) . This region corresponds to the middle range of the dendritic branch ( third to ninth branch orders , Figure 1H ) where the primary input from CA3 is located ( Neves et al . , 2008 ) , suggesting that the hippocampal Schaffer collateral pathway is particularly vulnerable to sleep loss . We next assessed whether sleep deprivation also impacted spine numbers and dendrite length of CA3 neurons . Surprisingly , in contrast to CA1 neurons , CA3 neurons were unaffected by sleep deprivation . We did not observe reductions in spine density or dendrite length of either basal or apical dendrites of any type ( Figure 1—figure supplement 2 ) . Together , these data suggest that CA1 neurons at the level of dendritic structure seem particularly vulnerable to sleep deprivation . To determine whether recovery sleep would reverse spine loss in CA1 neurons , we repeated the sleep deprivation experiment but then left the sleep-deprived mice undisturbed for three hours afterwards . This period was chosen as our previous work indicated that three hours of recovery sleep is sufficient to restore deficits in LTP caused by sleep deprivation ( Vecsey et al . , 2009 ) . In line with the electrophysiological studies , recovery sleep restored apical/basal spine numbers and dendrite length in CA1 neurons to those observed in non-sleep deprived mice ( Figure 2A , spine density of apical/basal dendrites , NSD: 1 . 23 ± 0 . 02 , RS: 1 . 29 ± 0 . 02; Student’s t-test , p>0 . 05; Figure 2B , dendrite length in μm , NSD: 1817 . 0 ± 64 . 6 , RS: 1741 . 6 ± 55 . 57; Student’s t-test , p=0 . 1721; Figure 2C , Student’s t-test , p>0 . 05 for each distance from soma; Figure 2D , Student’s t-test , p>0 . 05 for each branch number; for separate apical and basal spine analyses see Figure 2—figure supplement 1 ) with the exception of branched spines in the basal dendrites ( Figure 2—figure supplement 1C ) . Recovery sleep slightly but significantly elevated the number of filopodia spines of the apical CA1 dendrites and total spine numbers of the seventh and eighth branch of the apical and basal dendrites respectively ( Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 13424 . 006Figure 2 . Three hours of recovery sleep restores spine numbers and dendrite length of CA1 neurons in the hippocampus . ( A ) Golgi analyses indicated that three hours of recovery sleep after 5 hr of sleep deprivation restores the total number of spines per apical/basal dendrite of CA1 neurons ( n = 6 , Student’s t-test , p>0 . 05 ) . ( B ) Three hours of recovery sleep after 5 hr of sleep deprivation restores apical/basal dendrite length of CA1 neurons ( n = 6 , Student’s t-test , p=0 . 173 ) . ( C , D ) Three hours of recovery sleep restores apical/basal spine numbers at all distances from the soma ( Student’s t-test , p>0 . 05 for each distance from soma , C ) and at each branch number ( Student’s t-test , p>0 . 05 for each branch number , C ) . NSD: non-sleep deprived , RS: Sleep deprivation + recovery sleep . Values represent the mean ± SEM . See also Figure 2—figure supplement 1 for separate Golgi analyses of apical and basal spine numbers . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 00610 . 7554/eLife . 13424 . 007Figure 2—figure supplement 1 . Three hours of recovery sleep after 5 hr of sleep deprivation is sufficient to restore spine numbers and dendrite length in both basal and apical dendrites of CA1 neurons . ( A ) Recovery sleep leads to increased spine density in apical dendrite of CA1 neurons ( 6 animals per group , 4 neurons per animal , Student’s t test p<0 . 05 ) . ( B ) Recovery sleep restores basal and apical dendrite length of CA1 neurons ( 6 animals per group , 4 neurons per animal , Student’s t test p>0 . 16 ) . ( C ) Recovery sleep restores the number of all spine types in basal dendrites of CA1 neurons with the exception of branched spines ( 6 animals per group , 4 neurons per animal , Student’s t test p>0 . 15; branched spines Student’s t test p<0 . 05 ) . ( D ) Recovery sleep restores the number of all spine types in apical dendrites of CA1 neurons with the exception of branched spines which are increased by recovery sleep ( 6 animals per group , 4 neurons per animal , Student’s t test p>0 . 4; filopodia spines Student’s t test p<0 . 05 ) . ( E ) Recovery sleep restores the spine density of basal dendrites at all distances of the soma ( 6 animals per group , 4 neurons per anima , Student’s t test p>0 . 05 ) . ( F ) Recovery sleep restores the spine density of apical dendrites between at all distances from the soma ( 6 animals per group , 4 neurons per animal , Student’s t test p>0 . 05 ) . ( G ) Recovery sleep restores the spine density of basal dendrites of hippocampal CA1 neurons at all branch numbers ( 6 animals per group , 4 neurons per animal , Student’s t test p>0 . 05 ) . ( H ) Recovery sleep restores the spine density of apical dendrites of hippocampal CA1 neurons at all branch numbers with exception of the seventh branch ( 6 animals per group , 4 neurons per animal , Student’s t test p>0 . 05; branch 7 Student’s t test p<0 . 05 ) . NSD: non-sleep deprived , RS: recovery sleep . Values represent the mean ± SEM . p<0 . 05 , by Student’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 007 We hypothesized that the structural changes in the hippocampus following sleep deprivation might be related to increased activity of the actin-binding protein cofilin because increased cofilin activity can cause shrinkage and loss of dendritic spines through the depolymerization and severing of actin filaments ( Zhou et al . , 2004; Davis et al . , 2011; Shankar et al . , 2007 ) . The ability of cofilin to bind and depolymerize and sever F-actin is inhibited by phosphorylation at serine 3 ( Rust , 2015; Bamburg , 1999; Bosch et al . , 2014 ) . We therefore assessed whether sleep deprivation alters cofilin phosphorylation by Western blot analysis of hippocampus homogenates collected after 5 hr of sleep deprivation . Indeed , 5h of sleep deprivation reduced cofilin Ser-3 phosphorylation , suggesting an increase in cofilin activity in the hippocampus ( NSD: 100 . 0 ± 6 . 9%; SD: 67 . 7 ± 9 . 2%; Student t-test p=0 . 0090; Figure 3A ) . A similar effect was not evident in the prefrontal cortex ( NSD , n = 5: 100 . 0 ± 1 . 84%; SD , n = 5: 101 . 92 ± 2 . 41%; Student t-test p=0 . 54; Figure 3—figure supplement 1 ) , indicating sleep deprivation affects cofilin phosphorylation in a brain region-specific fashion . 10 . 7554/eLife . 13424 . 008Figure 3 . Increased cofilin activity in the hippocampus mediates the spine loss associated with sleep deprivation . ( A ) Five hours of sleep deprivation leads to a reduction in cofilin phosphorylation at serine 3 in the hippocampus . A representative blot is shown . Each band represents an individual animal . ( n = 13–14 , Student’s t-test p=0 . 0090 ) . ( B ) Mice were injected with pAAV9-CaMKIIα0 . 4-eGFP or pAAV9-CaMKIIα0 . 4-cofilinS3D-HA into the hippocampus to drive expression of eGFP or the mutant inactive form of cofilin ( cofilinS3D ) in excitatory neurons . This inactive mutant form of cofilin was made by substituting serine 3 for aspartic acid , which mimics a phosphoserine residue . An HA-tag was included to discriminate between mutant and endogenous cofilin . ( C ) A representative image showing that viral eGFP expression was restricted to the hippocampus . ( D–F ) CofilinS3D expression was excluded from astrocytes in area CA1 as indicated by a lack of co-labeling ( F ) between viral cofilin ( D ) and GFAP expression ( E ) . Scale bar , 100 µM . ( G ) Virally delivered cofilinS3D protein levels were approximately 75% ( blue bar ) of wild-type cofilin levels ( green bar ) . Wild-type cofilin levels were not significantly affected by expression of cofilinS3D . An HA-tag antibody was used to detect the mutant inactive form of cofilin . ( n = 4 ) . ( H ) Hippocampal cofilinS3D expression prevents spine loss in apical/basal dendrites of CA1 neurons that is associated with sleep deprivation ( n = 6 , Student’s t-test , p>0 . 05 ) . ( I ) Hippocampal cofilinS3D expression prevents the decrease in apical/basal dendritic spine length in neurons of hippocampal that is caused by sleep deprivation ( n = 6 , Student’s t-test , p>0 . 05 ) . ( J ) Sleep deprivation does not alter the number of any spine type in apical/basal dendrites of CA1 neurons in the hippocampus of mice expressing cofilinS3D ( n = 6 , Student’s t-test , p>0 . 05 ) . ( K ) Sleep deprivation does not attenuate apical/basal spine density at any distance from the soma in mice expressing cofilinS3D ( n = 6 , Student’s t-test , p>0 . 05 ) . NSD: non-sleep deprived , SD: sleep deprived . Values represent the mean ± SEM . p=0 . 0090 . Student’s t test . See also Figure 3—figure supplement 1 . For separate analyses of apical and basal spine numbers see Figure 3—figure supplement 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 00810 . 7554/eLife . 13424 . 009Figure 3—source data 1 . Sleep deprivation reduces cofilin phosphorylation in the hippocampus . The data source file contains the relative optical density values ( in arbitrary units ) of the pcofilin and cofilin western blots for each individual animal of the non-sleep deprived ( NSD ) and sleep deprived ( SD ) group . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 00910 . 7554/eLife . 13424 . 010Figure 3—figure supplement 1 . Sleep deprivation does not alter cofilin phosphorylation in the prefrontal cortex . Five hours of sleep deprivation does not lead to a reduction in cofilin phosphorylation at serine 3 in the prefrontal cortex . A representative blot is shown . Each band represents an individual animal . ( Student’s t test p>0 . 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 01010 . 7554/eLife . 13424 . 011Figure 3—figure supplement 1—source data 1 . Sleep deprivation does not alter cofilin phosphorylation in the prefrontal cortex . The data source file contains the optical density values ( in arbitrary units ) of the pcofilin and cofilin western blots for each individual animal of the non-sleep deprived ( NSD ) and sleep deprived ( SD ) group . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 01110 . 7554/eLife . 13424 . 012Figure 3—figure supplement 2 . CofilinS3D expression prevents sleep deprivation-induced reductions in spine numbers and dendrite length in both basal and apical dendrites of CA1 neurons . ( A ) In mice expressing cofilinS3D , sleep deprivation does not decrease the spine density of basal and apical dendrite of CA1 neurons ( 6 animals per group , 4 neurons per animal ) . ( B ) In mice expressing cofilinS3D , sleep deprivation does not cause a decrease in the length of basal and apical dendrite length of CA1 neurons ( 6 animals per group , 4 neurons per animal ) . ( C ) In mice expressing cofilinS3D , sleep deprivation does not cause a reduction in the total number of spines for each subtype in basal dendrites of CA1 neurons with exception of the branched spines ( 6 animals per group , 4 neurons per animal , Student’s t test p>0 . 5 , branched spines Student’s t test p<0 . 05 ) . ( D ) In mice expressing cofilinS3D , sleep deprivation does not cause a decrease in the total number of spines for each subtype in apical dendrites of CA1 neurons with exception of the branched spines ( 6 animals per group , 4 neurons per animal , Student’s t test p>0 . 5 , branched spines Student’s t test p<0 . 05 ) . ( E ) In mice expressing cofilinS3D , sleep deprivation does not reduce the spine density of basal dendrites at any distance from the soma ( 6 animals per group , 4 neurons per animal , Student’s t test p>0 . 05 ) . ( F ) In mice expressing cofilinS3D , sleep deprivation reduces spine density of apical dendrites only at a 180 µm distance from the soma ( 6 animals per group , 4 neurons per animal , Student’s t test p>0 . 05 ) . ( G ) In mice expressing cofilinS3D , sleep deprivation does not decrease the spine density in any dendritic branch of the basal dendrites of CA1 neurons ( 6 animals per group , 4 neurons per animal , Student’s t test p>0 . 05 ) . ( H ) In mice expressing cofilinS3D , sleep deprivation does not reduce the spine density in any dendritic branch of apical dendrites of hippocampal CA1 neurons ( 6 animals per group , 4 neurons per animal , Student’s t test p>0 . 05 ) . NSD: non-sleep deprived , SD: sleep deprived . Values represent the mean ± SEM . p<0 . 05 by Student’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 012 Based on these findings , we hypothesized that suppressing cofilin activity would prevent the sleep deprivation-induced changes in spine numbers of CA1 neurons . To test this hypothesis , we used a phosphomimetic form of cofilin that renders it inactive , namely cofilinS3D ( Pontrello et al . , 2012; Popow-Wozniak et al . , 2012; Meberg et al . , 1998 ) . Previous work suggested that cofilinS3D expression can inhibit endogenous cofilin activity ( Zhao et al . , 2008; Shi et al . , 2009 ) , through competition with endogenous cofilin for signalosomes where cofilin is activated by means of dephosphorylation ( Sarmiere and Bamburg , 2004; Konakahara et al . , 2004 ) . For example , cofilinS3D may compete with endogenous cofilin for binding to the cofilin-dephosphorylating phosphatase slingshot ( Konakahara et al . , 2004 ) . Importantly , cofilinS3D expression does not alter spine density under baseline conditions ( Pontrello et al . , 2012; Shi et al . , 2009 ) . We expressed either the phosphomimetic cofilinS3D or enhanced green fluorescent protein ( eGFP ) , which served as a control , in hippocampal excitatory neurons of adult male C57BL/6J mice using Adeno-Associated Viruses ( AAVs ) ( Figure 3B , C ) . A 0 . 4kb CaMKIIα promoter fragment was used to restrict expression to excitatory neurons ( Dittgen et al . , 2004 ) . Virally mediated expression of cofilinS3D was observed in excitatory neurons in all 3 major sub-regions of the hippocampus three weeks after viral injection ( Figure 3D–F ) . Western blot analyses of hippocampal lysates 3 weeks after injection showed that the level of virally delivered cofilin was roughly estimated 75% of the amount of endogenous wild-type cofilin and that the amount of wild-type cofilin per se was not substantially affected by expression of the mutant form ( Figure 3G ) . We subsequently determined whether expression of the inactive cofilinS3D prevented the loss of dendritic spines in hippocampal area CA1 caused by sleep deprivation . Analyses of Golgi-impregnated hippocampal neurons in area CA1 indicated that in cofilinS3D expressing mice sleep deprivation no longer reduced the spine density of apical/basal dendrites ( NSD: 1 . 42 ± 0 . 03; SD: 1 . 34 ± 0 . 03; Student’s t-test , p>0 . 05 Figure 3H , J , K; for separate apical and basal spine analyses see Figure 3—figure supplement 2 ) with the exception of a small but statistically significant reduction in branched spines of apical and basal dendrites ( Figure 3 , Figure Supplement C , D ) and a decrease in number of spines on apical dendrites about 180 µm away from the soma ( Figure 3—figure supplement 2E , F ) . Likewise , sleep deprivation no longer affected dendrite length ( NSD: 1283 . 0 ± 35 . 95 µm , SD: 1250 . 1 ± 41 . 19 µm; Student’s t-test , p=0 . 5612; Figure 3I , for separate apical and basal dendrite length analyses see Figure 3—figure supplement 2B ) . Together these data suggest that suppressing cofilin function in hippocampal neurons prevents the negative impact of sleep deprivation on spine loss and dendrite length of CA1 neurons . As a next step , we sought to determine whether prevention of the increase in cofilin activity in sleep-deprived mice would not only protect against the reduction in spine numbers on CA1 dendrites but also the functional impairment at the behavioral level . The consolidation of object-place memory requires the hippocampus ( Oliveira et al . , 2010; Florian et al . , 2011 ) and is sensitive to sleep deprivation ( Havekes et al . , 2014; Florian et al . , 2011; Prince et al . , 2014 ) . Therefore , we assessed whether cofilinS3D expression would prevent cognitive deficits caused by sleep deprivation in this task . Mice virally expressing eGFP or cofilinS3D were trained in this task 3 weeks after viral infection and sleep deprived for 5 hr immediately after training or left undisturbed in the home cage . Upon testing for memory the next day , sleep-deprived mice expressing eGFP showed no preference for the relocated object indicating that brief sleep deprivation impaired the consolidation of object-place memory . In contrast , mice expressing cofilinS3Dshowed a strong preference for the displaced object despite sleep deprivation ( eGFP NSD: 45 . 2 ± 6 . 4% , eGFP SD: 33 . 4 ± 2 . 0% , cofilinS3D NSD: 51 . 9 ± 2 . 9% , cofilinS3D SD: 53 . 2 ± 4 . 6%; Figure 4A ) . 10 . 7554/eLife . 13424 . 013Figure 4 . Increased cofilin activity in the hippocampus mediates the memory and synaptic plasticity deficits associated with sleep deprivation . ( A ) Mice expressing eGFP or cofilinS3D were trained in the hippocampus-dependent object-place recognition task . Half of the groups were sleep deprived for 5 hr and all mice were tested 24 hr later . Hippocampal cofilinS3D expression prevents memory deficits caused by sleep deprivation ( n = 9–10 , two-way ANOVA , effect of virus F1 , 35 = 18 . 567 , p=0 . 0001; effect of sleep deprivation F1 , 35 = 2 . 975 , p=0 . 093; interaction effect F1 , 35 = 4 . 567 , p=0 . 040; eGFP SD group versus other groups , p<0 . 05 ) . The dotted line indicates chance performance ( 33 . 3% ) . ( B , C ) Following 5 hr of sleep deprivation , long-lasting LTP was induced in hippocampal slices by application of four 100 Hz trains , 1 s each , spaced 5 min apart to the Schaffer collateral pathway . Five hours of sleep deprivation impairs long-lasting LTP in slices from mice expressing eGFP ( n = 6–7 , two-way ANOVA , effect of virus F1 , 10 = 21 . 685 , p<0 . 001 ) . In contrast , virally delivered cofilinS3D prevents sleep deprivation-induced deficits ( n = 5 , two-way ANOVA , effect of virus F1 , 8 = 0 . 016 , p>0 . 902 ) . NSD: non-sleep deprived , SD: sleep deprived . Values represent the mean ± SEM . p<0 . 05 by posthoc Dunnet’s test , p<0 . 01 by Student’s t test . See also Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 01310 . 7554/eLife . 13424 . 014Figure 4—source data 1 . CofilinS3D expression prevents memory deficits in the object-location memory task caused by sleep deprivation . The data source file contains the object exploration times for the displaced ( DO ) and non-displaced objects ( NDO1 , NDO2 ) for all individual animals of each group . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 01410 . 7554/eLife . 13424 . 015Figure 4—figure supplement 1 . CofilinS3D expression in hippocampal neurons does not affect exploratory activity , anxiety levels , or basal synaptic transmission . ( A ) Expression of the catalytically inactive cofilinS3D in hippocampal neurons does not affect the total time spent exploring objects during training in the object place recognition task ( ANOVA F1 , 35 = 1 . 026 , p=0 . 318 ) . All groups show a decrease in the total object exploration time during the training sessions ( n = 9–10 , two-way ANOVA , effect of session F2 , 70 = 54 . 060 , p = 0 . 0001; interaction effect F2 , 70 = 0 . 880 , p=0 . 419 ) . ( B ) Mice expressing cofilinS3D in hippocampal neurons spent a similar amount of time in the enclosed arm of the zero maze as eGFP expressing mice indicating normal anxiety-related behavior ( n = 7 , Student’s t test , p=0 . 632 ) . ( C ) Mice expressing cofilinS3D in hippocampal neurons had a similar number of transitions in the zero maze as eGFP expressing mice indicating normal locomotor activity ( n = 7 , Student’s t test , p=0 . 849 ) . ( D ) CofilinS3D expression did not alter the formation of short-term object location memories measured one hour after training ( n = 7–8 , Student’s t test , p=0 . 42 ) . ( E , F ) Input-output curves relating the amplitude of the presynaptic fiber volley to the initial slope of the corresponding fEPSP at various stimulus intensities are similar in slices from eGFP and cofilinS3D in slices from sleep deprived and non-sleep deprived mice ( n = 5 , eGFP NSD vs SD group , Student’s t test p=0 . 75; cofilinS3D , NSD vs SD group , Student’s t test p=0 . 17 ) . ( G , H ) Paired-pulse facilitation , a short-term form of synaptic plasticity , was not changed in slices from eGFP and cofilinS3D in slices from sleep deprived and non-sleep deprived mice ( n = 5 , eGFP NSD vs SD group two-way repeated-measures ANOVA , F1 , 8 = 0 . 393 , p=0 . 545 ) ( cofilinS3D NSD vs SD group two-way repeated-measures ANOVA , F1 , 8 = 3 . 056 , = 0 . 114 ) . NSD: non-sleep deprived , SD: sleep deprived . Values represent the mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 01510 . 7554/eLife . 13424 . 016Figure 4—figure supplement 1—source data 1 . CofilinS3D expression in hippocampal neurons does not affect exploratory activity . ( A ) The data source file contains the total object exploration times during the three training sessions for each individual animal of all four groups . ( B ) The data source file contains the time spent in the closed arms of the zero maze for each individual animal of both groups . ( C ) The data source file contains the number of transitions in the zero maze for each individual animal of both groups . ( D ) The data source file contains the object exploration times for the displaced ( DO ) and non-displaced objects ( NDO1 , NDO2 ) for each individual animal of both groups . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 01610 . 7554/eLife . 13424 . 017Figure 4—figure supplement 2 . CofilinS3A expression in hippocampal neurons attenuates the formation of long-term object-location memories but not long-term potentiation induced by spaced-four train LTP . ( A ) Mice expressing eGFP or the catalytically active cofilinS3A in hippocampal neurons were trained in the hippocampus-dependent object-place recognition task . Expression of the cofilinS3A does not affect the total time spent exploring objects during training in the object place recognition task ( ANOVA F1 , 18 = 1 . 919 , p=0 . 183 ) . Both groups show a decrease in the total object exploration time during the training sessions ( n = 10 , two-way ANOVA , effect of session F2 , 36 = 11 . 696 , p=0 . 0001; interaction effect F2 , 36 = 1 . 85 , p=0 . 172 ) . ( B ) During the test session 24 hr after training , eGFP mice preferentially explored the displaced object indicating that they successfully remembered the previous location of the individual objects . In contrast , mice expressing cofilinS3A explored all objects to a similar extent , indicative of a poor memory for the original object locations ( eGFP , 46 . 9 ± 4 . 2%; cofilinS3A , 34 . 9 ± 2 . 1%; Student’s t test , p=0 . 025 ) . ( C ) Input-output curves relating the amplitude of the presynaptic fiber volley to the initial slope of the corresponding fEPSP at various stimulus intensities are similar in slices from eGFP and cofilinS3A in slices from non-sleep deprived mice ( eGFP n = 6 , cofilinS3A n = 8 , Student’s t test p = 0 . 857 ) . ( D ) Paired-pulse facilitation , a short-term form of synaptic plasticity , was not changed in slices from eGFP and cofilinS3A in slices from non-sleep deprived mice ( n = 5–6 , eGFP vs cofilinS3A group two-way repeated-measures ANOVA , F1 , 12 = 0 . 218 , p=0 . 649 ) . ( E ) Long-lasting LTP was induced in hippocampal slices by application of four 100 Hz trains , 1 s each , spaced 5 min apart to the Schaffer collateral pathway . Virally delivered CofilinS3A expression did not alter LTP expression ( n = 5 , two-way ANOVA , effect of virus F1 , 8 = 1 . 102 , p=0 . 0325 ) . Dotted line indicates chance level performance . Values represent the mean ± SEM . p<0 . 05 by Student’s t test . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 017 Expression of the mutant form of cofilin did not affect object exploration during training , exploration of an open field or zero maze indicating that anxiety levels were unaffected by expression of cofilinS3D in the hippocampus ( Figure 4—figure supplement 1A–C ) . Moreover , using a behaviorally naïve cohort of mice , we found that cofilinS3D expression did not alter short-term object-place memory in the same task ( Figure 4—figure supplement 1D ) . Together , these findings demonstrate that cofilinS3D expression specifically prevents the cognitive deficits caused by sleep deprivation . Although we can not rule out the possibility of off-target effects of the cofilinS3D mutant , we think that these are unlikely as expression of this mutant form of cofilin reversed the effects of sleep deprivation , restoring spine loss and memory to non-sleep deprived levels while not having an effect in non-sleep deprived mice . To further define the role of cofilin in impairments in hippocampal function caused by sleep deprivation , we next determined if suppression of cofilin activity would prevent the deficits in hippocampal LTP caused by brief periods of sleep deprivation ( Havekes et al . , 2012a; Abel et al . , 2013; Vecsey et al . , 2009; Prince et al . , 2014 ) . Five hours of sleep deprivation significantly impaired long-lasting LTP induced by 4 high-frequency trains of electrical stimuli applied at 5-minute intervals ( spaced 4-train stimulation ) in hippocampal slices from mice expressing eGFP ( Figure 4B ) , confirming our previously published findings with non-injected wild-type mice ( Vecsey et al . , 2009 ) . In contrast , spaced 4-train LTP was unaffected by sleep deprivation in hippocampal slices from mice expressing the inactive cofilinS3D ( Figure 4C ) . The expression of cofilinS3D or sleep deprivation did not alter basal synaptic properties or paired-pulse facilitation ( Figure 4—figure supplement 1E-H ) suggesting that the spine loss caused by sleep deprivation specifically impairs long-lasting forms of synaptic plasticity . As a next step , we wanted to assess whether expression of a catalytically active version of cofilin ( cofilinS3A ) mimics the behavioral and synaptic plasticity phenotypes associated with sleep deprivation . Mice virally expressing eGFP or cofilinS3A were trained in the object-place memory task 3 weeks after viral infection and tested 24 hr after training . Mice expressing eGFP showed a strong preference for the relocated object while mice expressing cofilinS3A showed no preference for the object that was moved to a novel location ( eGFP: 46 . 9 ± 6 . 4% , cofilinS3A: 34 . 9 ± 2 . 1%; Figure 4—figure supplement 2B ) . The observed memory deficit could not be explained by a reduction in object exploration during the training as the total object exploration time was similar for both groups during training ( Figure 4—figure supplement 2A ) . Based on these findings , we conducted a set of electrophysiological experiments to determine whether expression of cofilinS3A is also sufficient to induce impairments in spaced 4-train LTP . CofilinS3A expression did not affect this form of L-LTP ( Figure 4—figure supplement 2E ) . The expression of cofilinS3A did not alter basal synaptic properties or paired-pulse facilitation ( Figure 4—figure supplement 2C–D ) . In summary , these data show that phosphorylation-dependent reductions in cofilin activity in hippocampal excitatory neurons prevent the decrease in hippocampal spine numbers , and also prevent the functional impairments in synaptic plasticity and behavior caused by a brief period of sleep deprivation . Furthermore , expression of constitutively active cofilin in hippocampal neurons is sufficient to mimic the memory deficits but not the synaptic plasticity impairment associated with a brief period of sleep deprivation . Sleep deprivation attenuates cAMP signaling in the hippocampus through increased levels and cAMP hydrolyzing activity of PDE4A5 ( Vecsey et al . , 2009 ) . Cofilin activity is known to be suppressed by the PKA-LIMK signaling pathway through the LIMK-mediated phosphorylation of cofilin at Ser-3 ( Lamprecht R , 2004; Nadella et al . , 2009 ) . We hypothesized that the elevation in PDE4A5 activity , associated with sleep loss , could negatively impact the cAMP-PKA-LIMK signaling pathway by enhancing cAMP degradation , thereby leading to increased cofilin activity . Based on this hypothesis , we also anticipated that blocking PDE4A5 function in hippocampal neurons would make the cAMP-PKA-LIMK pathway , which controls cofilin activity , resistant to the effects of sleep deprivation . To test this hypothesis , we engineered a catalytically inactive form of PDE4A5 ( referred to as PDE4A5catnull ) in which an aspartate group located deep within the cAMP binding pocket of PDE4A5 ( PDE4A5D577A ) , that is critical for catalytic activity , is replaced with an alanine group ( Baillie et al . , 2003; McCahill et al . , 2005 ) . Expression of PDE4A5catnull outcompetes the low levels of active , endogenous PDE4A5 from PDE4A5-containing signalosome complexes that specifically sequester it ( Houslay , 2010 ) , thereby preventing the breakdown of cAMP in the vicinity of those complexes . We used the viral approach ( Havekes et al . , 2014 ) to express PDE4A5catnull selectively in hippocampal neurons ( Figure 5A , B ) . Four weeks after viral injections , expression of PDE4A5catnull was observed in all major hippocampal subregions ( Figure 5C–E ) , and expression was excluded from astrocytes ( Figure 5F–H ) . Expression of PDE4A5catnull did not alter PDE4 activity in the hippocampus , prefrontal cortex or cerebellum ( Figure 5—figure supplement 1A–C ) . Next , we sleep deprived mice for 5 hr and assessed whether the phosphorylation of LIMK and cofilin was altered in the hippocampus . In agreement with our hypothesis , we observed that 5 hr of sleep deprivation reduced both LIMK and cofilin phosphorylation in hippocampal lysates from eGFP mice ( Figure 5I , J ) . PDE4A5catnull expression prevented the sleep deprivation-induced decreases in LIMK and cofilin phosphorylation ( Figure 5I , J ) . While expression of PDE4A5catnull fully restored the pcofilin/cofilin ratio in the hippocampus of sleep deprived mice to the levels observed under non-sleep deprivation conditions , it should be noted that phosphatases such as slingshot ( Sarmiere and Bamburg , 2004 ) may also contribute to the reduction in cofilin phosphorylation levels under conditions of sleep deprivation . Three hours of recovery sleep was sufficient to restore both LIMK and cofilin phosphorylation levels in the hippocampus ( Figure 5K , L ) . The latter observation is in line with our previous observations that a few hours of recovery sleep is sufficient to restore hippocampal synaptic plasticity ( Vecsey et al . , 2009 ) . 10 . 7554/eLife . 13424 . 018Figure 5 . Expression of catalytically inactive PDE4A5 in hippocampal neurons prevents memory deficits and alterations in the cAMP-PKA-LIMK-cofilin signaling pathway associated with sleep deprivation . ( A ) Mice were injected with pAAV9-CaMKIIα0 . 4-eGFP or pAAV9-CaMKIIα0 . 4-PDE4A5catnull-VSV into the hippocampus to drive neuronal expression of eGFP or catalytically inactive full-length PDE4A5 ( PDE4A5catnull ) . ( B ) Robust PDE4A5catnull expression was observed at the expected molecular weight , 108 kDa , in hippocampal lysates . ( C–E ) PDE4A5catnullexpression was observed in all 3 subregions of the hippocampus . ( F–H ) PDE4A5catnullwas not expressed in astrocytes reflected by a lack of co-labeling between PDE4A5catnull and GFAP expression . ( I ) Sleep deprivation causes a reduction in LIMK serine 596 phosphorylation in the hippocampus that is prevented by PDE4A5catnull expression ( n = 7–8; two-way ANOVA , effect of virus F1 , 27 = 3 . 299 , p=0 . 08; effect of sleep deprivation F1 , 27 = 6 . 124 , p=0 . 02; interaction effect F1 , 27 = 11 . 336 , p=0 . 002; eGFP SD group versus other groups p<0 . 05 ) . ( J ) Sleep deprivation causes a reduction in cofilin phosphorylation in the hippocampus that is prevented by PDE4A5catnull expression ( n = 9–10; two-way ANOVA , effect of virus F1 , 35 = 4 . 122 , p=0 . 05; effect of sleep deprivation F1 , 35 = 2 . 885 , p=0 . 1; interaction effect F1 , 35 = 9 . 416 , p=0 . 004; eGFP SD group versus other groups p<0 . 05 ) . ( K , L ) Three hours of recovery sleep after five hours of sleep deprivation restores hippocampal LIMK phosphorylation at serine 596 and cofilin phosphorylation at serine 3 to those observed in non-sleep deprived controls ( p>0 . 45 for both comparisons ) . ( M ) Mice expressing eGFP or PDE4A5catnull were trained in the hippocampus-dependent object-place recognition task and immediately sleep deprived for 5 hr after training ( SD ) or left undisturbed ( NSD ) . Hippocampal PDE4A5catnull expression prevents memory deficits caused by sleep deprivation ( n = 8–10; two-way ANOVA , effect of virus F1 , 33 = 2 . 626 , p=0 . 115; effect of sleep deprivation F1 , 33 = 2 . 311 , p=0 . 138; interaction effect F1 , 33 = 7 . 485 , p=0 . 01; posthoc Dunnet test eGFP SD group versus other groups p<0 . 05 ) . In all blots , each lane represents one individual animal . NSD: non-sleep deprived , SD: sleep deprived , SD+RS: sleep deprived plus recovery sleep . Scale bar , 100 µm . Values represent the mean ± SEM . *p<0 . 05 by posthoc Dunnet’s posthoc test . See also Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 01810 . 7554/eLife . 13424 . 019Figure 5—source data 1 . Recovery sleep following sleep deprivation restores LIMK and cofilin phosphorylation levels in the hippocampus , and expression of an inactive version of PDE4A5 in hippocampal neurons prevents memory deficits associated with sleep deprivation . ( K ) The data source file contains the relative optical density values ( in arbitrary units ) of the pLIMK and LIMK western blots for each individual animal of both the non-sleep deprived control group ( NSD ) and the group that underwent 5 hr of sleep deprivation followed by 3 hr of recovery sleep ( SD + RS ) . ( L ) The data source file contains the relative optical density values ( in arbitrary units ) of the pcofilin and cofilin western blots for each individual animal of both the non-sleep deprived control group ( NSD ) and the group that underwent 5 hr of sleep deprivation followed by 3 hr of recovery sleep ( SD + RS ) . ( M ) The data source file contains the object exploration times for the displaced ( DO ) and non-displaced objects ( NDO1 , NDO2 ) for each individual animal of each group . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 01910 . 7554/eLife . 13424 . 020Figure 5—figure supplement 1 . Expression of catalytically null PDE4A5 in the hippocampus: Catalytically inactive PDE4A5 without the unique N-terminal localization domain fails to prevent memory deficits associated with sleep loss . ( A ) PDE4A5catnull expression in hippocampal neurons did not significantly affect PDE4 activity in the hippocampus ( n = 7 , Student’s t test p=0 . 097 ) . ( B ) PDE4 activity in the prefrontal cortex was not altered by expression of the catalytically inactive PDE4A5catnull in the hippocampus ( n = 7–8 , Student’s t test p=0 . 162 ) . ( C ) PDE4 activity in the cerebellum was not changed by expression of the catalytically inactive PDE4A5catnull in the hippocampus ( n = 7–8 , Student’s t test p=0 . 293 ) . ( D ) Expression of the catalytically inactive PDE4A5catnull in hippocampal neurons did not alter the total time spent exploring objects during training in the object-place recognition task ( n = 8–10 , two-way ANOVA , effect of virus F1 , 33 = 0 . 043 , p=0 . 873 ) . All groups show a decrease in the total object exploration time during consecutive training sessions ( two-way ANOVA effect of session F2 , 66 = 32 . 777 , p=0 . 0001 ) . Mice expressing PDE4A5catnull had a slightly but non-significantly lower object exploration time during the first training session , and a slightly but non-significantly higher object exploration time during the last training session ( interaction effect F2 , 66 = 4 . 875 , p=0 . 011 , one way ANOVAs per session , p>0 . 05 ) . ( E ) Mice expressing PDE4A5catnull spend a similar time in the periphery of the open field as mice expressing eGFP in hippocampal neurons ( n = 8 , Student’s t test , p=0 . 292 ) . ( F ) Mice were injected with pAAV9-CaMKIIα0 . 4-eGFP or pAAV9-CaMKIIα0 . 4-PDE4A5catnullΔ4-VSV into the hippocampus to drive neuronal expression of eGFP or catalytically inactive full-length PDE4A5 which lacked the N-terminal domain unique for PDE4A5 ( PDE4A5catnullΔ4 ) . A VSV-tag was included to discriminate between endogenous PDE4A5 and the truncated PDE4A5catnullΔ4 . ( G ) PDE4A5catnullΔ4 protein levels in the hippocampus 4 weeks after viral injection . A sample blot probed with an isoform-nonspecific PDE4A antibody revealed the presence of both wild-type PDE4A5 protein and truncated PDE4A5catnullΔ4 protein . Probing the blot with an antibody for the HA-tag confirmed that the truncated protein is indeed the N-terminal lacking catalytically inactive PDE4A5catnullΔ4 . Each band represents an individual animal ( H ) Expression of the catalytically inactive PDE4A5catnullΔ4 lacking the N-terminal domain in hippocampal neurons did not affect total object exploration time during training in the object-place recognition task ( n = 7–9 , two-way ANOVA effect of virus F1 , 29 = 0 . 470 , p=0 . 498 ) . All groups show decreased total object exploration times during consecutive training sessions ( two-way ANOVA effect of session F2 , 58 = 13 . 597 , p=0 . 0001; interaction effect F2 , 58 = 0 . 555 , p=0 . 557 ) . ( I ) Mice expressing eGFP or the N-terminal domain lacking inactive form of PDE4A5catnullΔ4 were trained in the hippocampus-dependent object-place recognition task . Sleep deprivation causes memory deficits in both eGFP and PDE4A5catnullΔ4 mice ( n = 7–9; two-way ANOVA effect of sleep deprivation F1 , 29 = 18 . 131 , p=0 . 0001; effect of virus F1 , 29 = 1 . 064 , p=0 . 311; interaction effect F1 , 29 = 0 . 001 , p=0 . 986; eGFP NSD versus EGFP SD , posthoc Tukey’ t test p=0 . 0054; PDE4A5catnullΔ4 NSD versus PDE4A5catnullΔ4 SD , posthoc Tukey’ t test p=0 . 0037 ) . Dotted line indicates chance level performance . NSD: non-sleep deprived , SD: sleep deprived . Values represent the mean ± SEM . #p<0 . 01 by Tukeys t test . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 02010 . 7554/eLife . 13424 . 021Figure 5—figure supplement 1—source data 1 . Exploratory activity in mice expressing catalytically inactive PDE4A5 or PDE4A5Δ4 in hippocampal excitatory neurons . ( D ) The data source file contains the total object exploration times during the three training sessions for each individual animal of all four groups . ( E ) The data source file contains the time spent in the periphery and center of the open field for each individual animal of both groups . ( H ) The data source file contains the total object exploration times during the three training sessions for each individual animal of all four groups . ( I ) The data source file contains the object exploration times for the displaced ( DO ) and non-displaced objects ( NDO1 , NDO2 ) for each individual animal of each group . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 021 Because PDE4A5catnull expression in hippocampal neurons prevents changes in the cAMP-PKA-LIMK-cofilin pathway caused by sleep deprivation , we hypothesized that expression of PDE4A5catnull in hippocampal excitatory neurons would also prevent the memory deficits induced by 5 hr of sleep deprivation . Mice expressing eGFP showed a clear preference for the displaced object 24 hr after training , which was lost in animals that were deprived of sleep for 5 hr immediately after training ( Figure 5M ) . In contrast , mice expressing PDE4A5catnull showed a strong preference for the displaced object despite sleep deprivation ( Figure 5M ) . The memory rescue was not a result of altered exploratory behavior during training in the object-place recognition task ( Figure 5—figure supplement 1D ) . Furthermore , PDE4A5catnull expression did not alter anxiety levels and exploratory behavior in the open field ( Figure 5—figure supplement 1E ) . Although the catalytic unit of the 25 distinct PDE4 isoforms is highly conserved , each has a unique N-terminal localization sequence that directs isoform targeting to a specific and unique set of protein complexes ( signalosomes ) ( Houslay , 2010 ) . This allows for a highly orchestrated sequestering of cAMP signaling in specific intracellular domains rather than a general , global degradation of cAMP throughout the cell ( Houslay , 2010 ) . We therefore aimed to determine whether the rescue of memory impairments by expression of PDE4A5catnull requires the unique N-terminal domain of PDE4A5 . To answer this question , we engineered a truncated version of PDE4A5catnull that lacks the first 303 base pairs encoding the isoform unique N-terminal domain ( Bolger et al . , 2003 ) ( referred to as PDE4A5catnullΔ4 , Figure 5—figure supplement 1F ) and expressed this mutant in excitatory neurons in the hippocampus using a viral approach . As this species has no targeting N-terminus then , unlike the full length inactive PDE4A5 that displaces endogenous active PDE4A5 from its functionally relevant location in the cell and thereby increase cAMP levels localized to the sequestering signaling complex , this engineered 5’ truncated complex would simply lead to the expression of an inactive PDE4A catalytic unit unable to be targeted like the native enzyme and so unable to exert an effect on localized cAMP degradation in the functionally relevant compartment . Western blot analyses of hippocampal tissue 4 weeks after viral injection confirmed the presence of the truncated PDE4A5catnullΔ4 at the protein level using an antibody that detects all PDE4A isoforms and an antibody against the HA-tag ( Figure 5—figure supplement 1F , G ) . With a behaviorally naïve cohort of mice now expressing eGFP or PDE4A5catnullΔ4 we repeated the object-place recognition task . Brief sleep deprivation after training in the object-place recognition task resulted in a loss of preference for the displaced object in mice expressing PDE4A5catnullΔ4 ( Figure 5—figure supplement 1I ) . The inability of PDE4A5catnullΔ4 to prevent the memory deficit caused by sleep deprivation was not a consequence of altered exploration levels during training ( Figure 5—figure supplement 1H ) . This finding indicates that the memory rescue in the previous experiment was a result of the full length PDE4A5catnull being sequestered to specific signalosomes through the isoform-unique N-terminal region rather than a consequence of PDE4A5catnull being unable totarget the functionally relevant complexes sequestering full length PDE4A5 . It also indicates that displacing sequestered , active endogenous PDE4A5 in hippocampal excitatory neurons is sufficient to prevent memory deficits induced by 5 hr of sleep deprivation . Overall , these data suggest that sleep deprivation negatively impacts spine numbers by targeting the PKA-LIMK-cofilin pathway through the alterations in activity of PDE4A5 ( Figure 6 ) . 10 . 7554/eLife . 13424 . 022Figure 6 . The impact of sleep deprivation on hippocampal spine dynamics . Sleep deprivation increases PDE4A5 protein levels that cause a reduction in cAMP levels and attenuation of the PKA-LIMK signaling pathway , which results in a reduction in the phosphorylation of cofilin . Dephosphorylated cofilin can lead to spine loss . Suppressing PDE4A5 function through viral expression of a catalytically inactive PDE4A5 prevents alterations in LIMK and cofilin signaling as well as the cognitive impairments caused by sleep deprivation . Likewise , attenuating cofilin activity through viral expression of a catalytically inactive form of cofilin prevents the loss of dendritic spines , impairments in synaptic plasticity , and memory deficits associated with sleep loss . Proteins whose function is reduced after sleep deprivation are shown in blue . Proteins whose function is promoted by sleep deprivation are shown in red . DOI: http://dx . doi . org/10 . 7554/eLife . 13424 . 022 One of the major challenges in sleep research is the elucidation of molecular mechanisms and cellular circuits underlying the adverse consequences of sleep loss . Here , we use in vivo rescue experiments to define a critical molecular mechanism by which brief sleep deprivation leads to cognitive impairments . First , we show that sleep deprivation dramatically reduces spine number and dendrite length of hippocampal CA1 neurons without affecting dendritic structure of CA3 neurons . Second , we demonstrate that sleep deprivation increases cofilin activity in the hippocampus , but not the prefrontal cortex , which is a likely explanation for the reductions in CA1 spine numbers and dendrite length . Third , we find that three hours of recovery sleep restores spine number , dendrite length and cofilin phosphorylation levels to those observed in non-sleep deprived mice . Fourth , we show that suppression of cofilin activity in hippocampal excitatory neurons is sufficient to prevent sleep-deprivation-induced decreases in dendritic spines number , LTP impairments , and memory . Fifth , we demonstrate that hippocampal expression of constitutively active cofilin is sufficient to cause long-term memory deficits but not LTP impairments . Sixth , we find that suppression of PDE4A5 function through overexpression of a catalytically inactive mutant version of PDE4A5 changes LIMK and cofilin phosphorylation levels caused by sleep deprivation . Finally , we show that hippocampal suppression of PDE4A5 function prevents the negative impact of sleep deprivation on memory consolidation . Thus , our studies demonstrate that changes in the cAMP/PKA/LIMK/cofilin pathway are necessary to cause memory deficits under conditions of sleep deprivation . In light of the fact that elevated cofilin activity can lead to spine shrinkage and spine loss ( Zhou et al . , 2004; Pontrello et al . , 2012 ) , our genetic manipulations of cofilin and PDE4A5 signaling independently link impairments in synaptic plasticity and memory caused by brief sleep deprivation with the loss of dendritic spines in the hippocampus . Deficits in synaptic plasticity and memory both represent read outs of the impact of sleep deprivation on hippocampal function , but our work does not directly examine the direct relationship between synaptic plasticity and memory , a topic that has been the subject of extensive study and discussion in the literature ( Lynch , 2004; Sah et al . , 2008 ) . That care should be taken to directly relate LTP deficits with memory impairments is emphasized by our findings that expression of constitutively active cofilin is sufficient to cause memory deficits while it does not impact at least one form of L-LTP that is disrupted by sleep deprivation ( Vecsey et al . , 2009 ) . We observed a significant reduction in the total number of dendritic spines of excitatory CA1 neurons after 5 hr of sleep deprivation . This substantial decrease in dendritic spine number occurs rapidly and exceeds the fluctuations in hippocampal spine number observed across the estrus cycle ( González-Burgos et al . , 2005 ) , or the changes in spine number caused by stress ( McLaughlin et al . , 2005; Shors et al . , 2001 ) . In contrast to sleep deprivation , acute stress results in an increase rather than a decrease in CA1 spine number in male rats ( Shors et al . , 2001 ) . Even 3 to 4 weeks of chronic stress or systemic delivery of corticosterone does not alter dendritic arborization of CA1 neurons ( McLaughlin et al . , 2005 ) . It is also unlikely that other factors associated with the procedure to keep animals awake rather than sleep deprivation per se causes the spine loss as our previous work indicated that applying the exact same amount of stimulation in the waking phase ( i . e . the dark phase ) does not lead to memory impairments ( Hagewoud et al . , 2010a ) . In line with our finding of reductions in spines during sleep deprivation , work by Yang and colleagues revealed that sleep promotes dendritic spine formation in neurons activated by learning ( Yang et al . , 2014 ) . Combined with our work , these experiments suggest that sleep deprivation disrupts learning-induced changes in spines that occur during sleep . Importantly , our structural studies reveal that spine loss is reversed by recovery sleep , consistent with this idea . Thus , our work reveals a distinct , selective , and rapid effect of brief periods of sleep loss on synaptic structure . It is noteworthy that even a short period of sleep deprivation acts to trigger such a dramatic effect on neuronal structure , which is reversed by recovery sleep . Studies assessing the impact of sleep deprivation on electrophysiological properties of excitatory hippocampal neurons suggest that sleep deprivation negatively impacts long-lasting forms of LTP ( Havekes et al . , 2012a; Abel et al . , 2013 ) . In this study and our previous work ( Vecsey et al . , 2009; Prince et al . , 2014 ) , we showed that 5 hr of sleep deprivation attenuates long-lasting forms of LTP in the hippocampus . We observed that expression of an inactive mutant form of cofilin prevented the reductions in CA1 spine number , the impairment in a long-lasting form of LTP caused by sleep loss . It is interesting to note that three hours of recovery sleep not only restores spine numbers in CA1 neurons , but also hippocampal LIMK and cofilin phosphorylation levels . These findings complement our previous electrophysiological studies , in which we showed that such a short period of recovery sleep also restores deficits in LTP caused by 5 hr of sleep deprivation ( Vecsey et al . , 2009 ) . Our work reveals that PDE4A5 is a critical mediator of the impact of sleep deprivation on memory consolidation . Indeed , one reason why hippocampal area CA1 is specifically vulnerable to sleep deprivation may be the high level of PDE4A5 expression in this region ( McPhee et al . , 2001 ) . Specific PDE4 isoforms are sequestered by distinct signalosome complexes that regulate localized cAMP signaling and impart functionally distinct roles ( Houslay , 2010 ) . Impairing the function of PDE4A5 signalosomes through expression of a full length catalytically inactive form of PDE4A5 exerts a dominant negative action , phenotypically identified here as preventing the alterations in LIMK and cofilin signaling caused by sleep deprivation . This makes memory consolidation resistant to the negative impact of sleep loss . Consistent with the notion that a key functional role of the isoform-unique N-terminal region of PDE4 isoforms is the targeting to signalosomes so as to exert functionally distinct actions ( Houslay , 2010 ) , the hippocampal expression of a catalytically in active version of PDE4A5 lacking the isoform unique N-terminal domain fails to rescue the cognitive deficits associated with sleep loss . The latter observation suggests that the isoform-specific N-terminal domain of PDE4A5 targets this specific PDE isoform to signalosomes that degrade cAMP in the vicinity of complexes that are particularly sensitive to sleep deprivation such as the complexes that contain LIMK and cofilin . Consistent with this , no such dominant negative phenotype is evident in a catalytically inactive PDE4A construct engineered to lack such an N-terminal targeting region . Our data contradict the synaptic homeostasis hypothesis for sleep function . This hypothesis proposes that sleep functions to downscale synaptic strength that has increased as a result of neuronal activity and experiences during wakefulness ( Tononi and Cirelli , 2006 ) . This hypothesis has focused on explaining data from the cortex rather than the hippocampus , but one previously published study has suggested that the synaptic homeostasis hypothesis applies to the hippocampus as well ( Vyazovskiy et al . , 2008 ) . However , the hippocampus may be unique from the cortex as the hippocampus is involved in episodic memory and in much greater experience-dependent plasticity than anywhere else in the brain and thus our findings may not extend to other areas where synaptic plasticity is not as prominent . Further , the hippocampus also exhibits many distinct forms of synaptic plasticity . Here , we examine structural changes in hippocampal neurons and find that extended wakefulness leads to a loss of synaptic spines mediated by a signaling pathway involving cofilin . This suggests that prolonged wakefulness down-regulates synaptic connectivity in the hippocampus . As little as 3 hr of recovery sleep is sufficient to restore signaling through these complexes , suggesting that sleep functions to restore synaptic connectivity . Thus , the signaling pathways that mediate changes in dendritic structure are rapidly impaired by sleep loss and then can be quickly restored during recovery sleep . Lack of sleep is a common problem in our 24/7 modern society and it has severe consequences for health , overall wellbeing , and brain function ( Bryant et al . , 2004; Harrison and Horne , 2000 ) . Despite decades of research , the mechanisms by which sleep loss negatively impacts brain function have remained unknown . Our findings suggest that the cognitive impairments caused by brief sleep deprivation are a result of altered spine dynamics leading to a reduction in spine numbers . Our findings may also explain the reduction in hippocampal volume observed in an animal model of more chronic sleep restriction ( Novati et al . , 2011 ) and sleep disorders , such as primary insomnia ( Riemann et al . , 2007 ) as well as sleep apnea ( Morrell et al . , 2003 ) . Our work defining the molecular pathway through which sleep deprivation impacts memory consolidation underscores the importance of the plasticity of the neuronal cytoskeleton and reveals that rapid synaptic remodeling occurs with changes in behavioral state . Experimentally naïve C57BL/6J male mice ( 2–3 months of age; IMSR_JAX:000664 ) were obtained from Jackson laboratories at an age of 6 weeks and housed in groups of 4 with littermates on a 12 hr/12 hr light/dark schedule with lights on at 7 am ( ZT0 ) . Mice had food and water available ad libitum . In case of the viral studies , mice underwent surgery at an age of 8–12 weeks , were single housed for 5 days and then pair-housed with a littermate throughout the experiment . For the perfusion experiments , mice were single housed 1 week prior to the start of the experiment . For all experiments , mice were randomly assigned to groups and were handled for 5 days for 2 min per day . Mice were sleep deprived using the gentle stimulation method ( Vecsey et al . , 2009; Hagewoud et al . , 2010a; Vecsey et al . , 2013; Hagewoud et al . , 2010b , 2010c ) . In short , animals were kept awake by gentle tapping the cage , gently shaking the care and/or removing the wire cage top . Their bedding was disturbed in cases when mice did not respond to tapping or shaking the cage . This method of sleep deprivation has been validated by our laboratory using EEG recordings ( Meerlo et al . , 2001 ) . In the object-place recognition task , mice learn the location of 3 distinct objects and were tested for memory of the object locations 24 hr after training by displacing one of the objects . Training commenced at ZT0 or 45 min after lights on using the previously described training protocol ( Oliveira et al . , 2010; Tretter et al . , 2009 ) . Mice were trained 3 or 4 weeks after viral surgery . Object exploration levels were scored manually by the experimenter blind to treatment conditions . The zero maze and open field studies were conducted as previously described ( Tretter et al . , 2009; Havekes et al . , 2012b ) . Mice were anaesthetized using isoflurane and remained on a heating pad throughout the surgery and kept warm using a heating lamp for 5–10 min during the recovery from the anesthesia until the mouse was awake . Mice received metacam and buprenol as analgesics during and post-surgery and artificial tears ( Puralube ) were used to prevent the eyes from drying out during surgery . Two small holes were drilled in the skull at the appropriate locations using a microdrill . The virus was injected using a nanofil 33G beveled needles ( WPI ) attached to a 10 µl Hamilton syringe . A microsyringe pump ( UMP3; WPI ) connected to a mouse stereotax and controller ( Micro4; WPI ) were used to control the speed of the injections . The needle was slowly lowered to the target site over the course of 3 min and remained at the target site for 1 min before beginning of the injection ( 0 . 2 µl per minute ) . After the injection , the needle remained at the target site for 1 min and then was slowly gradually removed over a 5 min period . The coordinates for the bilateral injections are ( A/P −1 . 9 mm , L/M ± 1 . 5 mm , and 1 . 5 mm below bregma ) . After removal of the needle , a small amount of bone wax ( Lukens ) was used to close the drill holes and the incision was closed with sutures . Site-directed mutagenesis of plasmid DNA was carried out to generate PDE4A5catnull using the Stratagene QuikChange Site-Directed Mutagenesis kit , using the method in the manufacturer's instructions . N-terminal lacking PDE4A5catnull was generated using standard PCR cloning procedures and the Stratagene PfuUltra High-Fidelity DNA polymerase . Purified plasmid DNA was produced using Qiagen QIAprep kits and stored at 4°C . The pAAV9-CaMKIIα0 . 4-PDE4A5catnull-VSV , pAAV9-CaMKIIα0 . 4-PDE4A5catnullΔ4-HA , pAAV9-CaMKIIα0 . 4-CofilinS3D-HA , pAAV9-CaMKIIα0 . 4-CofilinS3A-HA and pAAV9-CaMKIIα0 . 4-eGFP were constructed by standard methods and packaged by the University of Pennsylvania viral core . Transduced cofilinS3D may compete with endogenous cofilin for binding to the cofilin-specific phosphatase slingshot , thereby leading to inactivation of the endogenous protein ( Sarmiere and Bamburg , 2004; Konakahara et al . , 2004 ) . Titers ranged from 2 . 4 × 1012 to 4 . 91 × 1013 genome copy numbers . A 0 . 4kb CaMKIIα promoter fragment ( Dittgen et al . , 2004 ) was used to restrict expression to excitatory neurons . An HA-tag was included to discriminate endogenous from virally expressed proteins . Approximately 1 µl , ( corrected for genome copy number between constructs ) was injected per hippocampus . The cAMP-specific PDE activity assays and western blots to assess sleep deprivation-induced changes in PDE4A5 levels were conducted as described ( Vecsey et al . , 2009 ) . Hippocampal tissue was lysed using a tissue ruptor ( Qiagen , Germany ) in lysis buffer ( Tris 50 mM , pH: 9 , sodium deoxycholate 1% , Sodium fluoride 50 mM , activated sodium vanadate 20 µM , EDTA 20 µM , and beta-glycerophosphate 40 µM . Additional phosphatase inhibitor cocktail ( Thermo scientific ) and protease inhibitors ( Roche , Switzerland ) were added to the freshly prepared buffer just prior to tissue lysis . Samples were centrifuged for 10 min at 13 , 000 ×g at 4°C and supernatant was collected . Protein concentration of the samples was measured using the Bradford method ( Biorad , Hercules , CA , USA ) and sample concentration was corrected using additional lysis buffer . Afterwards LDS sample buffer ( Nupage , Invitrogen , Carlsbad , CA , USA ) including 2-mercaptoethanol was added and samples were for boiled for 5 min prior to loading on Criterion TGX 18-well 4–20% gels . After electrophoreses , proteins were transferred to PVDF membrane followed by blocking for 1 hr in 5% milk in TBST or 5% BSA in TBST ( in case of cofilin antibodies ) . After blocking , the following antibodies were used GAPDH ( 1:1000 , Santa Cruz , Santa Cruz , CA , USA RRID:AB_10167668 ) , PDE4A ( 1:1000 ( 27 ) ) , PDE4A5 ( 1:1000 ( 27 ) ) , pCofilin ( 1:1000 , Cell signaling , RRID:AB_2080597 ) , Cofilin ( 1:3000 BD Transduction Laboratories , San Jose , CA , USA; RRID:AB_399515 ) , HA-tag ( 1:1000 , Roche , RRID:AB_390918 ) , VSV-G tag ( 1:1000 , Abcam , United Kingdom , RRID:AB_302646 ) , LIMK ( 1:2000 , Millipore , Billerica , MA , USA; RRID:AB_1977324 ) . Polyclonal phospho-serine 596 LIMK antibody was generated by New England Biopeptides ( Gardner , MA , USA ) using CDPEKRP ( pS ) FVKLEQ peptide . After incubation with the primary antibodies , membranes were incubated in HRP-conjugated secondary antibodies for 1 hr at room temperature ( Santa Cruz , mouse secondary antibody 1:1000 , RRID:AB_641170; Santa Cruz , rabbit secondary antibody , RRID:AB_631746 ) . The immunoreactive bands were captured on autoradiography film ( Kodak , Rochester , NY , USA ) and analyzed using ImageJ ( NIH ) . Immunohistochemistry was conducted as described previously ( Havekes et al . , 2012b; Isiegas et al . , 2008 ) . In short , animals were transcardially perfused with ice cold 4% paraformaldehyde in PBS followed by a 48 hr post fixation in 4% PFA . Coronal brain sections were cut at a thickness of 25 microns . Sections were rinsed in PBS , blocked with 5% normal serum and incubated in PBS with 0 . 1% triton and 2% normal serum with either of the following antibodies or combinations of antibodies PDE4A5 ( 1:200 , ( 27 ) ) , HA-tag ( 1:200 , Roche , RRID:AB_390918 ) , VSV-G tag ( 1:2000 , Abcam , RRID:AB_302646 ) , GFAP-alexa 488 ( 1:200 , Invitrogen , RRID:AB_143165 ) followed by the appropriate Alexa fluor-conjugated secondary antibodies ( 1:1000 Invitrogen , RRID:AB_141459 , RRID:AB_10562718 , RRID:AB_10564074 ) . Fluorescent images were analyzed using a Leica confocal microscope . After sleep deprivation mice were injected ( i . p ) with a lethal dose of morbital and perfused with phosphate-buffered saline ( PBS , 3 min at RT ) , followed by 1 . 5% PFA ( 20 min at RT ) . Brains were then removed and post-fixed in 1 . 5% PFA ( 40 min at RT ) . After post perfusion incubation in 1 . 5% PFA , each hemisphere was cut in 130 μm slices using a vibratome . Slices were collected to multiwell plates filled with PBS . After one hour incubation in room temperature , PBS was removed and slices were stained with a GeneGun ( Biorad , pressure: 100–120 psi ) using nylon filter ( Merc Millipore , 10 μm , cat . No . NY1004700 ) . DiI bullets were prepared as described ( Seabold et al . , 2010 ) . After staining , slices were incubated over night at RT in PBS . The next day slices were incubated for one hour in 4% PFA and mounted with DapiFluoromount G ( SouthernBiotech ) . Microphotographs of DiI stained apical dendrites in the stratum radiatum of CA1 area ( approx . 100 µm from cell bodies ) were performed in z-stacks using Zeiss LSM 780 ( step 0 . 3 mm , objective 63x , digital magnifications 5x , resolution 1024 × 1024 ) . Linear density ( per mm of dendrite ) and size of spines were counted using SpineMagick software . On average 155 spines were analyzed per an animal . Importantly , for the comparison of the spine numbers using golgi and diolistic staining methods we focused specifically on the second and third branch of the apical dendrites as the diolistic staining technique is suboptimal to label and analyze branches farther away from the soma . Experiments were performed in the hippocampal Schaffer collateral pathway as previously described ( Vecsey et al . , 2009; Havekes et al . , 2012b ) . Briefly , male mice injected with eGFP or cofilinS3D virus were sacrificed by cervical dislocation , and hippocampi were quickly collected in chilled , oxygenated aCSF containing 124 mM NaCl , 4 . 4 mM KCl , 1 . 3 mM MgSO4 × 7H2O , 1 mM NaH2PO4 × H2O , 26 . 2 mM NaHCO3 , 2 . 5 mM CaCl2 × 2H2O , and 10 mM D-glucose bubbled with 95% O2 / 5% CO2 . 400 μm thick transverse hippocampal slices were placed in an interface recording chamber at 28ºC ( Fine Science Tools , Foster City , CA ) . Slices were equilibrated for at least 2 hr in aCSF ( pH 7 . 4 ) . The stimulus strength was set to elicit 40% of the maximum field excitatory postsynaptic potential ( fEPSP ) amplitude . The average of the baseline initial fEPSP slope values over the first 20 min was used to normalize each initial fEPSP slope . Brains were impregnated using the Rapid Golgi stain kit ( FD Neurotechnologies Inc ) according to the instructions . Coronal sections ( 80-um thickness ) that covered the rostro-caudal axis of CA1 of the hippocampus were analyzed . The serial sections were then chosen and analyzed using a stereology-based software ( Neurolucida , v10 , Microbrightfield , VT ) , and Zeiss Axioplan 2 image microscope with Optronics MicroFire CCD camera ( 1600 × 1200 ) digital camera , motorized X , Y , and Z-focus for high-resolution image acquisition and digital quantitation in combination with a 100x objective using a sophisticated and well established method that should represent a 3D quantitative profile of the neurons sampled and prevents a failure to detect less prominent spines . Our sampling strategy is to prescreen the impregnated neurons along the anterior/posterior axis of the region of interest to see if they were qualified for analysis . Neurons with incomplete impregnation or neurons with truncations due to the plane of sectioning were not collected . Moreover , cells with dendrites labeled retrogradely by impregnation in the surrounding neuropil were excluded . We also made sure there was a minimal level of truncation at the most distal part of the dendrites; this often happens in most of the Golgi studies , likely due to the plane of sectioning at top and bottom parts of the section . The brains were cut at a 80 μm thickness . With consideration of the shrinkage factor after processing ( generally 10–25% shrinkage ) , the thickness of the section is even less , so the visualization of the spine subclass is no issue as we used a 100x Zeiss objective lens with immersion oil , which is sufficient to resolve the details or subtype of the spines for laborious counting . All analyses were conducted by an experimenter blind to treatment . Behavioral and electrophysiological data were analyzed using Student’s t-tests or two-way ANOVAs ( in some cases with repeated measures as the within subject variable ) . Dunnett’s tests were used for post-hoc analyses where needed . Biochemical data was analyzed using independent samples t-tests . The experimenter was blind to group treatment in all studies . Differenceswere considered statistically significant when p<0 . 05 . All data are plotted as mean ± s . e . m .	The demands of modern society means that millions of people do not get sufficient sleep on a daily basis . Sleep deprivation , even if only for brief periods , can impair learning and memory . In many cases , this impairment appears to be related to changes in the activity of a brain region called the hippocampus . However , the exact processes responsible for producing the effects of sleep deprivation remain unclear . During learning or forming a new memory , the connections between the relevant neurons in the brain change . Havekes et al . found that depriving mice of sleep for just five hours dramatically reduced the connectivity between neurons in the hippocampus . This reduction is caused by the increased activity of cofilin , a protein that breaks down the actin filaments that shape the connections between neurons . Havekes et al . then used a virus to introduce an inactive version of cofilin into hippocampal neurons to suppress the activity of the naturally present cofilin . This manipulation prevented both the loss of the connections between neurons and the memory deficits normally associated with sleep deprivation . Havekes et al . also found that recovery sleep leads to the re-wiring of neurons in the hippocampus . Future studies are now needed to determine how the neurons are able to re-wire themselves during recovery sleep .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Sleep deprivation causes memory deficits by negatively impacting neuronal connectivity in hippocampal area CA1
The tumor suppressor PIP3 phosphatase PTEN is phosphorylated on four clustered Ser/Thr on its C-terminal tail ( aa 380–385 ) and these phosphorylations are proposed to induce a reduction in PTEN’s plasma membrane recruitment . How these phosphorylations affect the structure and enzymatic function of PTEN is poorly understood . To gain insight into the mechanistic basis of PTEN regulation by phosphorylation , we generated semisynthetic site-specifically tetra-phosphorylated PTEN using expressed protein ligation . By employing a combination of biophysical and enzymatic approaches , we have found that purified tail-phosphorylated PTEN relative to its unphosphorylated counterpart shows reduced catalytic activity and membrane affinity and undergoes conformational compaction likely involving an intramolecular interaction between its C-tail and the C2 domain . Our results suggest that there is a competition between membrane phospholipids and PTEN phospho-tail for binding to the C2 domain . These findings reveal a key aspect of PTEN’s regulation and suggest pharmacologic approaches for direct PTEN activation . PTEN ( phosphatase and tensin homolog deleted on chromosome 10 ) suppresses cell proliferation , migration and survival by dephosphorylating the lipid second messenger phosphatidyl inositol 3 , 4 , 5-triphosphate ( PIP3 ) , thereby opposing phosphatidyl inositol-3 kinase ( PI3K ) signaling and preventing AKT protein kinase activation ( Maehama and Dixon , 1998; Myers et al . , 1998; Sun et al . , 1999; Iijima and Devreotes , 2002 ) . PTEN is one of the most frequently mutated genes in cancer , with inactivating mutations found in many solid tumor types ( Li et al . , 1997; Steck et al . , 1997 ) . The PTEN gene has been shown to be inactivated through somatic and germline mutations as well as transcriptionally suppressed through epigenetic mechanisms ( Salvesen et al . , 2001; Meng et al . , 2007; Hollander et al . , 2011 ) . Homozygous deletion of PTEN is embryonically lethal in mice while heterozygous mice are predisposed to developing tumors ( Suzuki et al . , 1998; Di Cristofano et al . , 1999; Podsypanina et al . , 1999 ) . Germline mutations in PTEN predispose people to spontaneous tumor formation as seen in Cowden syndrome ( Nelen et al . , 1997 ) . PTEN is a 47 kDa ( 403 aa ) protein composed of a ‘dual-specificity’ phosphatase domain , a C2 domain that mediates membrane association , and a 52 aa C-terminal tail ( Figure 1A ) . An X-ray structure of PTEN , lacking the apparently flexible N-terminus ( aa 1–13 ) , internal D-loop ( aa 286–309 ) and C-tail ( aa 353–403 ) , shows tight-knit interactions between the catalytic and C2 domains ( Lee et al . , 1999 ) . The PTEN protein is believed to be regulated by a variety of mechanisms including post-translational modifications , protein–protein interactions , and protein–lipid interactions ( Campbell et al . , 2003; Iijima et al . , 2004; Walker et al . , 2004; Vazquez et al . , 2006; Denning et al . , 2007; Chagpar et al . , 2009; Shenoy et al . , 2012; Song et al . , 2012 ) . Previously mapped PTEN post-translational modifications include phosphorylation ( Vazquez et al . , 2000; Torres and Pulido , 2001; Miller et al . , 2002; Al-Kouri et al . , 2005; Cordier et al . , 2012 ) , acetylation ( Okumura et al . , 2006 ) , ubiquitylation ( Wang et al . , 2007 ) , and sumoylation ( Huang et al . , 2012 ) . Phosphorylation of a cluster of Ser/Thr residues ( Ser380 , Thr382 , Thr383 , and Ser385 ) on PTEN’s C-terminal tail have received considerable attention in studies on PTEN regulation and appears to be a major modification ( Wu et al . , 2000; Vazquez et al . , 2001; Odriozola et al . , 2007; Rahdar et al . , 2009 ) . Reported to be catalyzed by the protein kinases CK2 and/or GSK3β ( Torres and Pulido , 2001; Al-Khouri et al . , 2005 ) , this phospho-cluster is mutationally sensitive in cell assays , with replacements by Ala driving PTEN to the plasma membrane ( Rahdar et al . , 2009 ) . While the molecular mechanism of how this is achieved is uncertain , altered protein–protein interactions ( Wu et al . , 2000; Vazquez et al . , 2001; Sumitomo et al . , 2004; Takahashi et al . , 2006; van Diepen et al . , 2009; Huang et al . , 2012 ) as well as conformational changes ( Odriozola et al . , 2007; Rahdar et al . , 2009 ) in PTEN have been suggested to be induced by phosphorylation . It has been unclear if phosphorylation of PTEN has a direct effect on membrane association or is indirectly mediated through other macromolecular interactions . Most PTEN phosphorylation analyses have relied on cellular transfection experiments in which replacement of the Ser and Thr residues with Ala have been indirect indicators of phosphorylation function . However , Ser and Thr have different properties from Ala and such mutations may be misleading in deducing the action of a phosphoSer/phosphoThr ( Vazquez et al . , 2000; Torres and Pulido , 2001; Al-Khouri et al . , 2005 ) . What has been lacking thus far is a biochemical analysis of purified phosphorylated PTEN in which the phosphorylation sites are well-defined in position and stoichiometry . Inherent challenges in obtaining this material include the difficulty in using kinases for introducing phosphates site-specifically into PTEN and the proposed potential for PTEN autodephosphorylation ( Zhang et al . , 2012 ) . To circumvent these issues , we employ here expressed protein ligation , a method for protein semisynthesis ( Schwarzer and Cole , 2005 ) , for generating 380 , 382 , 383 , 385-tetraphosphorylated-PTEN ( 4p-PTEN ) . Expressed protein ligation involves the production of a recombinant protein carrying a C-terminal thioester by exploiting the action of an intein , and its chemoselective ligation to an N-Cys containing synthetic peptide ( Muir et al . , 1998; Vila-Perelló and Muir , 2010 ) . Within the synthetic peptide , phosphoSer/phosphoThr can be installed using standard chemical techniques at specific locations . We show here that , relative to the unphosphorylated form , 4p-PTEN adopts a more compact conformation in which the C-tail appears to interact with the C2 domain in a fashion that reduces its affinity for lipid membranes and diminishes its catalytic activity . 10 . 7554/eLife . 00691 . 003Figure 1 . Generation of semisynthetic PTEN proteins . ( A ) PTEN is composed of phosphatase domain , a C2 domain , and a C-terminal tail that is phosphorylated multiple times within a cluster of Ser and Thr residues ( S380/T382/T383/S385 ) . ( B ) C-terminally truncated PTEN containing an intein generated thioester at its C-terminus is ligated to a synthetic PTEN tail peptide with or without phosphorylation at the S380/T382/T383/S385 cluster . The final product is full length PTEN in the phosphorylated ( 4p-PTEN ) or unphosphorylated ( n-PTEN ) form . ( C ) C-terminal tail peptides were synthesized in the unphosphorylated form or phosphorylated at S380/T382/T383/S385 . Note that the N-Cys replaces a natural Tyr in PTEN . ( D ) The ligation reaction precedes smoothly over 72 hr . ( E ) Western blot with an anti-phospho PTEN antibody reveals 4p-PTEN but not n-PTEN is phosphorylated . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 00310 . 7554/eLife . 00691 . 004Figure 1—figure supplement 1 . Schematic views of n-PTEN , 4p-PTEN and t-PTEN . n-PTEN and 4p-PTEN are full length semisynthetic proteins and contain the Tyr 379 to Cys mutation to facilitate the native chemical ligation reaction . t-PTEN contains amino acids 1–378 . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 00410 . 7554/eLife . 00691 . 005Figure 1—figure supplement 2 . MALDI spectra for PTEN tail peptides and semisynthetic PTEN proteins . ( A ) Unphosphorylated PTEN tail peptide ( expected mass: m/z 2852 ) . ( B ) Phosphorylated PTEN tail peptide ( expected mass: m/z 3172 ) . ( C ) Unphosphorylated semisynthetic PTEN ( n-PTEN ) ( expected mass: m/z 47 , 106 . 2 ) . ( D ) Phosphorylated semisynthetic PTEN ( 4p-PTEN ) ( expected mass: m/z 47 , 426 . 2 ) . The spectra for C and D were normalized to the external reference bovine serum albumin . The approximate accuracy for the mass spectrometric measurements of these protein masses is ±50 Da . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 00510 . 7554/eLife . 00691 . 006Figure 1—figure supplement 3 . Size exclusion chromatography , Y379C enzymatic characterization and autophosphatase activity of PTEN . The chromatograms and elution volumes of ( A ) n-PTEN and ( B ) 4p-PTEN purified by FPLC on a size exclusion column . ( C ) The chromatogram of protein standards on a size exclusion column ( Peak A: Thyroglobulin , MW = 670 , 000 , elution volume = 8 . 7 ml; Peak B: gamma globulin , MW = 158 , 000 , elution volume = 9 . 7 ml; Peak C: Ovalbumin , MW = 44 , 000 , elution volume = 14 . 6 ml; Peak D: Myoglobin , MW = 17 , 000 , elution volume = 16 . 3 ml; Peak E: Vitamin B12 , MW = 1350 , elution volume = 19 . 5 ml ) . ( D ) Phosphatase assays using diC6 PIP3 for WT GST-PTEN and Y379C GST-PTEN purified from E . coli . Data are reported as the mean ± the SEM of three experiments performed in duplicate . ( E ) Generation of 4p-PTEN by expressed protein ligation in the presence or absence of 25 μM VO-OHpic , a potent PTEN inhibitor monitored by western blot . ( F ) 50 ng of 4p-PTEN treated with 2 μM n-PTEN ( left ) or 1 μM CIP ( right ) . Dephosphorylation of the phospho-tail is monitored by western blot . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 006 A requirement for expressed protein ligation is the correct positioning of a Cys for the native chemical ligation reaction ( Muir et al . , 1998; Schwarzer and Cole , 2005; Vila-Perelló and Muir , 2010 ) . Initial attempts at generating semisynthetic PTEN focused on an Escherichia coli expression system . In this way , we showed that PTEN-intein fusion proteins allowed for generation of catalytically active recombinant PTEN thioester fragments ( data not shown ) . We also determined using GST-PTEN that Cys was well-tolerated at the position needed for ligation , showing no change in catalytic activity induced by Y379C mutation ( Figure 1—figure supplement 3D ) . However , expression in E . coli of the intein-fusion protein suffered from low yields of soluble PTEN protein expression ( <0 . 1 mg/l culture ) which was insufficient for our needs . We thus subcloned aa 1–378 of PTEN ( t-PTEN ) into a baculovirus plasmid for insect cell expression . Ligation with a tetraphosphorylated ( and unphosphorylated as a control ) N-Cys synthetic peptide aa 379–403 ( Figure 1C ) proceeded smoothly over 72 hr , providing 8–10 mg/l of culture purified semisynthteic PTEN protein ( Figure 1B , D ) . Tetraphosphorylated ( 4p-PTEN ) , unphosphorylated ( n-PTEN ) and C-terminally truncated PTEN ( t-PTEN ) were generated in this way ( Figure 1—figure supplement 1 ) . Semisynthetic proteins were >90% pure using Coomassie staining and their structural integrity confirmed by mass spectrometry ( Figure 1—figure supplement 2C and D ) . Western blot with commercial anti-phospho-PTEN Ab showed that 4p-PTEN , and not n-PTEN , was appropriately phosphorylated ( Figure 1E ) . It should also be noted that baculovirus systems have rarely been used for intein expression vectors ( Pradhan et al . , 1999 ) , possibly in part because of the expression of chitinase in these hosts which interferes with the standard chitin affinity purification scheme . Nevertheless , we found the presence of chitinase to be a surmountable issue by a chromatographic pre-clearing step with a bed of fibrous cellulose . We also found that High Five insect cells rather than the more typical SF9 insect cells were critical for robust expression . Despite the theoretical concern about autodephosphorylation ( Zhang et al . , 2012 ) , 4p-PTEN prepared in the presence of a PTEN phosphatase vanadate-based inhibitor was equally phosphorylated to that prepared in its absence ( Figure 1—figure supplement 3E ) . Further experiments showed that 4p-PTEN did not undergo spontaneous pSer/pThr hydrolysis over 24 hr as monitored by western blot ( Figure 1—figure supplement 3F ) . Initial analysis of 4p-PTEN catalytic activity was carried out with a soluble PIP3 substrate containing hexanoyl rather than the more physiological palmitoyl chains using a phosphate release detection assay with malachite green ( Van Veldhoven and Mannaert , 1987 ) . These studies revealed a ∼sixfold reduction in catalytic efficiency ( kcat/Km ) conferred by C-terminal phosphorylation in which 4p-PTEN shows a significantly higher soluble PIP3 Km ( Figure 2A ) . To determine the enzymatic activity with a more physiologically relevant substrate , we analyzed 4p-PTEN’s dephosphorylation of vesicle-incorporated PIP3 ( containing palmitoyl chains ) ( McConnachie et al . , 2003 ) . 3′-[32P] PIP3 was prepared by PI3-kinase and incorporated into vesicles containing unlabeled PIP3 and phosphatidylcholine ( PC ) ( McConnachie et al . , 2003 ) . For interfacial enzymes such as PTEN , the bulk concentration as well as the surface concentration of the substrate in the lipid bilayer needs to be considered ( Deems et al . , 1975; Hendrickson and Dennis , 1984; McConnachie et al . , 2003 ) . We thus explored the rate of hydrolysis of PIP3 under conditions where the surface concentration of PIP3 was held constant while varying the bulk concentration of PIP3 and carrier lipid ( phosphatidylcholine , PC ) proportionately ( bulk dilution ) . In addition , we measured PTEN activity under conditions where the surface concentration of PIP3 was varied while its bulk concentration was held constant by varying the carrier lipid PC ( surface dilution ) . As described in the ‘Materials and methods’ section , apparent Vmax and apparent Km values obtained from these experiments were fit to the equations in the ‘Materials and methods’ section associated with the equation below:V0= ( VmaxXs[S0] ) / ( iKmKs+iKm[S0]+Xs[S0] ) ( McConnachie et al . , 2003 ) in which XS is the surface concentration ( mol fraction ) of the substrate PIP3 , S0 is the bulk concentration of PIP3 , iKm is the interfacial Michaelis constant ( mol% ) and KS is the membrane dissociation constant for PTEN interaction with vesicles . Bulk dilution experiments yielded a rectangular hyperbola ( Figure 2B ) , while surface dilution experiments showed apparent substrate inhibition at higher surface concentrations of PIP3 ( Figure 2C ) . Using the lower substrate concentrations where substrate inhibition was minimal , the interfacial kinetic analysis revealed that n-PTEN and 4p-PTEN have similar iKm values with minor differences in kcat values ( Figure 2E and Figure 2—figure supplement 1C ) . However , at 1% PIP3 , the Ks value of 4p-PTEN showed a ∼threefold increase compared to that of n-PTEN , suggesting a decrease in binding affinity for the vesicle membrane when PTEN is phosphorylated ( Figure 2E ) . n-PTEN Ks and iKm values were similar to those of t-PTEN ( Figure 2—figure supplement 2 ) , indicating that the unphosphorylated tail extension does not hinder membrane interactions . Our data suggest that at low ( <1 mol% ) PIP3 concentrations which are physiologic , the rate differential between n-PTEN and 4p-PTEN is significant ( Figure 2D ) . 10 . 7554/eLife . 00691 . 007Figure 2 . Soluble substrate activity and Interfacial kinetic analysis of semisynthetic PTEN . ( A ) PTEN activity to a soluble substrate , diC6-PIP3 . ( n-PTEN: kcat = 2 . 6 ± 0 . 1 min−1 , Km = 67 ± 4 . 2 μM , kcat/Km = 0 . 038 ± 0 . 001 min−1μM−1; 4p-PTEN: kcat/Km = 0 . 005 ± 0 . 0002 min−1μM−1 ) ( B and C ) PTEN activity to palmitoyl PIP3 incorporated into phosphatidylcholine vesicles . In the bulk dilution experiment ( B ) enzymatic activity for n-PTEN and 4p-PTEN was measured at a fixed surface concentration of 1% PIP3 while the bulk concentration was varied . In the surface dilution experiment ( C ) activity was measured at a fixed bulk concentration of 50 μM PIP3 while the surface concentration was varied . ( D ) 4p-PTEN has lower activity than n-PTEN only at low PIP3 concentrations . ( E ) Summary of the interfacial kinetic analysis of n-PTEN and 4p-PTEN . Data are reported as the mean ± the SEM from three experiments performed in duplicate . Apparent Vmax values were obtained from the best fit curves from the first four points of the surface dilution experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 00710 . 7554/eLife . 00691 . 008Figure 2—figure supplement 1 . Interfacial kinetic analysis of semisynthetic PTENs . ( A ) 32P labeled PIP3 was generated as outlined in the ‘Materials and methods’ section . ( B ) PTEN activity to radiolabled palmitoyl PIP3 incorporated into phosphatidylcholine vesicles is linear with respect to enzyme concentration . ( C ) Apparent kcat and apparent Km values were obtained by nonlinear regression analysis for each form of PTEN for surface dilution experiments ( top row ) and bulk dilution experiments ( bottom row ) . Apparent kcat and apparent Km values were then fit to the equations in the ‘Materials and methods’ section to obtain the interfacial kinetic parameters in tabulated in Figure 2E and Figure 2—figure Supplement 2DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 00810 . 7554/eLife . 00691 . 009Figure 2—figure supplement 2 . Bulk and surface dilution curves of t-PTEN . ( A ) In the bulk dilution experiment enzymatic activity for t-PTEN was measured at a fixed surface concentration of 1% PIP3 while the bulk concentration was varied . ( B ) In the surface dilution experiment activity was measured at a fixed bulk concentration of 50 µM PIP3 while the surface concentration was varied . Data a represented as the mean ± the SEM of three experiments performed in duplicate . ( C ) Summary of the interfacial kinetic analysis of t-PTEN . Data are reported as the mean ± the SEM from three experiments performed in duplicate . Apparent Vmax values were obtained from the best fit curves from the first four points of the surface dilution experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 00910 . 7554/eLife . 00691 . 010Figure 2—figure supplement 3 . Anionic lipid stimulation of n-PTEN and 4p-PTEN . At a low surface concentration of PIP3 ( 0 . 01% ) incorporated vesicles , ( A ) PIP2 and ( B ) Phosphatidylserine stimulate the enzymatic activity of n-PTEN and 4p-PTEN . Data are reported as the mean ± the SEM of three experiments performed in duplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 010 To further analyze the apparent membrane affinity loss conferred by PTEN tail phosphorylation , we explored the binding interactions of semisynthetic PTENs with vesicles containing anionic phospholipids . As reported previously for unphosphorylated PTEN produced in E . coli ( McConnachie et al . , 2003 ) , we found that increasing levels of both phosphatidylserine ( PS ) and PIP2 in vesicles positively influence PIP3 dephosphorylation by both n-PTEN and 4p-PTEN ( Figure 2—figure Supplement 3 ) . We next carried out a series of PTEN-vesicle binding assays using PIP2 and PS as membrane targeting lipids by measuring differential sedimentation fractionation . These experiments demonstrated markedly reduced membrane binding for 4p-PTEN vs n-PTEN under all conditions tested ( Figure 3A , B and Figure 3—figure supplement 1 ) . Both n-PTEN and 4p-PTEN show proportionally greater binding to vesicles as concentrations of PIP2 and PS are increased , but 4p-PTEN compared with n-PTEN requires a significantly higher concentration of anionic lipid to achieve a similar level of membrane sedimentation . For example , 7 mol% PIP2 sedimented about one-fifth of 4p-PTEN , which was comparable to the amount of n-PTEN sedimented by 3 mol% PIP2 ( Figure 3A ) . With 5 mol% PS , the ratio of n-PTEN:4p-PTEN sedimented was about 10:1 ( Figure 3B ) . These results corroborate the PTEN affinity differences observed in the phosphatase assays where C-tail phosphorylation of PTEN inhibited membrane binding . 10 . 7554/eLife . 00691 . 011Figure 3 . n-PTEN and 4p-PTEN binding to large multilamellar vesicles ( LMVs ) . The percent of n-PTEN and 4p-PTEN bound to sedimented phosphatidylcholine LMVs incorporated with ( A ) PIP2 or ( B ) phosphatidylserine was determined by quantification of western blot bands . Data are reported as the mean ± the SEM of three separate experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 01110 . 7554/eLife . 00691 . 012Figure 3—figure supplement 1 . n-PTEN and 4p-PTEN binding to large multilamellar vesicles ( LMVs ) . Representative blots of the percent of n-PTEN and 4p-PTEN bound to sedimented phosphatidylcholine LMVs incorporated with ( A ) PIP2 or ( B ) phosphatidylserine . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 012 It has been proposed that C-terminal phosphorylation of PTEN may alter its conformation ( Odriozola et al . , 2007; Rahdar et al . , 2009 ) . In the course of purification of our semisynthetic PTENs by anion exchange chromatography , we made the paradoxical observation that the tetraphosphorylated PTEN reproducibly eluted at an earlier elution volume ( ∼70 ml and lower salt concentration , 90 mM NaCl ) and more sharply relative to the unphosphorylated protein ( ∼100 ml , 140 mM NaCl ) ( Figure 4A ) . Elution of n-PTEN as one large peak followed by multiple small peaks was attributed to minor heterogeneous insect cell-mediated phosphorylation in the recombinant tail moiety on and near Thr366 and Ser370 as detected with a site-specific Ab ( Figure 4—figure supplement 1B ) . A similar elution pattern was observed with t-PTEN ( Figure 4—figure supplement 1A ) . This heterogeneity is presumably collapsed under one peak in 4p-PTEN . Since 4p-PTEN has a nominal eight negative charge increase relative to n-PTEN , we were surprised that 4p-PTEN showed decreased affinity to cationic resin . One explanation for this result is that , relative to n-PTEN , 4p-PTEN undergoes a conformational change which buries its negatively charged C-tail , reducing its availability for interacting with cationic resin . The fact that clusters of charges can be more important than overall charge for protein–ion exchange resin interactions has been discussed previously ( Chung et al . , 1989; Hou et al . , 2010 ) . 10 . 7554/eLife . 00691 . 013Figure 4 . Conformational changes associated with PTEN phosphorylation . ( A ) With a gradient of 0–50% NaCl over 250 ml on an anion exchange column , 4p-PTEN elutes at ∼70 ml while n-PTEN elutes at ∼100 ml . ( B ) 2 μg of n-PTEN and 4p-PTEN were digested with varying amounts of trypsin then visualized by colloidal blue staining or by western blot with an antibody to the N- or C-terminus of PTEN . Asterisks denote bands that are in higher abundance in the digestion of n-PTEN compared to 4p-PTEN . N-terminal sequencing of these bands identifies the cleavage sites as ( ) R15 , ( ) R84 and ( * ) R161 . ( C ) Denatured 4p-PTEN treated with 0 . 5 μM alkaline phosphatase is significantly more sensitive to dephosphorylation of the tail phosphocluster compared to the native form of 4p-PTEN treated with 1 μM alkaline phosphatase . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 01310 . 7554/eLife . 00691 . 014Figure 4—figure supplement 1 . Non-tail cluster phosphorylation of PTEN expressed in High Five insect cells . ( A ) The elution pattern on anion exchange of t-PTEN is similar to that of n-PTEN ( Figure 4A ) . ( B ) Western blot with an antibody to phospho T366/S370 shows phosphorylation at these sites is removed by mutation to alanine . X-mutant contains T366A/S370A as defined in Figure 6—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 01410 . 7554/eLife . 00691 . 015Figure 4—figure supplement 2 . Native and denatured 4p-PTEN sensitivity to alkaline phosphatase . ( A ) Native folded 4p-PTEN treated with 1 μM alkaline phosphatase . ( B ) 4p-PTEN denatured by repeated freeze thaw cycles treated with 0 . 5 μM alkaline phosphatase . Removal of tail phosphorylation was monitored by western blot with an antibody to the tail phospho cluster . Data are represented as the mean ± the SEM of three experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 015 To further explore the conformation of 4p-PTEN , we compared its susceptibility to trypsin proteolysis relative to n-PTEN . 4p-PTEN appeared more protease resistant vs its unphosphorylated counterpart as seen most clearly at 25 and 50 ng trypsin ( Figure 4B ) . Based on western blots with N-terminal and C-terminal PTEN Abs , several large metastable fragments still containing C-terminal epitopes were observed in n-PTEN , indicative of enhanced protease sensitivity in the catalytic domain in the unphosphorylated protein . N-terminal sequencing identified several of these cleavage positions ( Figure 4B ) . Alkaline phosphatase sensitivity and western blotting was used to assess the exposure of the C-terminal 380–385 phospho cluster in 4p-PTEN ( Wang et al . , 2002 ) . The half-life of alkaline phosphatase-mediated dephosphorylation of 4p-PTEN under the conditions used was found to be ∼60 min ( Figure 4C and Figure 4—figure supplement 2A ) . Serendipitously , we found that freeze-thawing dilute 4p-PTEN led to apparent denaturation of the protein since it was much more susceptible to alkaline phosphatase catalyzed dephosphorylation of the tail phosphate cluster . Under the same conditions , near complete dephosphorylation was observed within 7 . 5 min . To more precisely determine this rate , we cut in half the concentration of alkaline phosphatase , and assumed that this process follows a pseudo-first-order kinetic mechanism . In this way , we estimate that native 4p-PTEN vs denatured 4p-PTEN shows a 25-fold reduced rate of alkaline phosphatase-catalyzed dephosphorylation of the tail phospho cluster ( Figure 4C and Figure 4—figure supplement 2B ) . We infer that this rate-differential suggests a closed:open 4p-PTEN conformational equilibrium of 25:1 under the conditions of this experiment . To gain more information on the structural changes induced by PTEN tail phosphorylation , we performed small angle X-ray scattering ( SAXS ) analysis of t-PTEN , n-PTEN , and 4p-PTEN . The scattering plots are shown in Figure 5A for each protein . The shape of the pair distance distribution function p ( r ) plot for n-PTEN reveals a single hump with a large shoulder region at higher r values ( Figure 5B ) . The shoulder region is indicative of a protein with an elongated shape ( Putnam et al . , 2007; Jacques and Trewhella , 2012 ) , possibly due to the presence of the extended tail . The shoulder is reduced for t-PTEN and 4p-PTEN , suggesting the tail is no longer in an extended position ( Figure 5B ) . The radius of gyration ( Rg ) and maximum dimension ( Dmax ) of the PTEN particles can be used as measurements of protein size . Rg values calculated from the Guinier and p ( r ) plots are in agreement for each respective protein . Obtained from the p ( r ) plot , the Rg for n-PTEN ( 31 . 91 +/− 0 . 054 Å ) is larger than it is for 4p-PTEN ( 27 . 16 +/− 0 . 019 Å ) and t-PTEN ( 27 . 82 +/− 0 . 083 Å ) . Dmax is also larger for n-PTEN ( 96 . 5 Å ) than it is for 4p-PTEN ( 82 . 5 Å ) and t-PTEN ( 82 . 5 Å ) ( Figure 5D ) . 10 . 7554/eLife . 00691 . 016Figure 5 . SAXS analysis for t-PTEN , n-PTEN and 4p-PTEN . ( A ) Scattering diagrams and ( B ) pair distribution function ( Pofr ) plots for t-PTEN , n-PTEN , and 4p-PTEN . ( C ) The molecular envelopes for t-PTEN , n-PTEN , and 4p-PTEN overlaid with the tailless crystal structure containing the phosphatase domain ( blue ) , C2 domain ( magenta ) , and CBRIII loop ( yellow ) . ( D ) Summary of Rg and Dmax values for t-PTEN , n-PTEN , and 4p-PTEN . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 01610 . 7554/eLife . 00691 . 017Figure 5—figure supplement 1 . Molecular envelopes of t-PTEN , n-PTEN and 4p-PTEN obtained from SAXS analysis . Molecular envelopes of t-PTEN , n-PTEN , and 4p-PTEN were generated from SAXS scattering data using the ab initio modeling program DAMMIN . Outputs from ten DAMMIN runs were averaged for each protein using DAMAVER and are shown overlaid with the tailless crystal structure with phosphatase domain shown in blue , the C2 domain shown in magenta and the CBRIII loop shown in yellow . Front , side and top views are shown for each protein . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 017 Molecular envelopes of t-PTEN , n-PTEN and 4p-PTEN were generated from the scattering data using the ab initio modeling program DAMMIN ( Putnam et al . , 2007; Franke and Svergun , 2009; Jacques and Trewhella , 2012 ) . Outputs from 10 DAMMIN runs , averaged for each protein using DAMAVER , are shown overlaid with the tailless crystal structure in Figure 5C ( Figure 5—figure Supplement 1 ) . The tailless crystal structure containing the phosphatase and C2 domains fits nicely into the two-lobed globular portion of each envelope . The molecular envelope for n-PTEN reveals an elongated extension proximal to the C2 domain ( Figure 5C ) . The molecular envelope for t-PTEN , which contains half of the 52 residue tail , shows a small extension , similar in length to that of 4p-PTEN , but perhaps not as wide . Interestingly , the 4p-PTEN phosphatase domain appears to undergo a modest change in shape relative to t-PTEN and n-PTEN ( Figure 5C and Figure 5—figure supplement 1 ) . Based on prior models as well as the proteolysis and SAXS experiments , we constructed several mutant semisynthetic PTENs to investigate possible PTEN residues that contribute to the closed conformation of 4p-PTEN ( Figure 6—figure supplement 1 ) . Semisynthetic phospho- and unphosphorylated PTEN containing four mutations K13A , R14A , R15A , and R161A ( ‘N-mutant’ ) was constructed to test the possibility that these basic residues in the N-terminus , co-localized in the crystal structure and implicated in phospholipid binding , might be critical in stabilizing the closed 4p-PTEN conformation . We also generated the penta-mutant unphosphorylated and phosphorylated PTENs containing K260A/K263A/K266A/K267A/K269A ( ‘A5-mutant’ ) and K260D/K263D/K266D/K267D/K269D ( ‘D5-mutant’ ) as neutral and charge inverted forms that probe the importance of the CBRIII loop in the C2 domain to conformational constriction . A fourth construct of semisynthetic PTEN that included deletions in the N-terminus ( aa 1–6 ) , D-loop ( aa 286–309 ) , and C-terminus ( aa 395–403 , and also contained T366A/S370A ) , ( ‘X-mutant’ ) was prepared to facilitate comparisons to the crystallized form of PTEN which possessed the same N-terminal and ‘D-loop’ deletions ( Lee et al . , 1999 ) . These semisynthetic PTEN mutants were generated analogously to that of n-PTEN and 4p-PTEN . Anion exchange chromatography showed the same paradoxical behavior for the phosphorylated forms of the N-mutant , A5-mutant , and X-mutant PTENs with faster than expected elution of these semisynthetic proteins . Of note , elution of the unligated X-mutant PTEN on anion exchange chromatography sharpened to a single peak compared to the broader distribution of unligated wt PTEN ( t-PTEN ) , presumably because mutation of Thr366/Ser370 abolishes the phosphorylation events that lead to heterogeneity ( Figure 6—figure supplement 2 ) . Moreover , N-mutant , A5-mutant , and X-mutant 4p-PTEN proteins showed nearly identical rates of dephosphorylation by alkaline phosphatase relative to wt 4p-PTEN indicating that they are each in the same ∼25:1 equilibrium favoring the closed conformation ( Figure 6A and Figure 6—figure supplement 3 ) . 10 . 7554/eLife . 00691 . 018Figure 6 . Phosphatase sensitivity and activity of 4p-PTEN and its mutants . ( A ) The rate of dephosphorylation of 4p-PTEN and its mutants was measured by quantification of bands from western blot analysis after treatment of the PTEN protein with 1 μM alkaline phosphatase . Data points are shown as the mean ± the SEM of three experiments . ( B ) PTEN activity was measured against 160 μM diC6-PIP3 substrate . ( C ) Km curve of D5 4p-PTEN mutant ( kcat = 3 . 0 ± 0 . 3 min−1 , Km = 112 ± 22 μM , kcat/Km = 0 . 027 ± 0 . 003 min−1μM−1 ) . Data points are shown as the mean ± the SEM of three experiments performed in duplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 01810 . 7554/eLife . 00691 . 019Figure 6—figure supplement 1 . Schematic view of semisynthetic PTEN mutants . PTEN mutants were generated analogously to n-PTEN and 4p-PTEN semisynthetic proteins as outlined in the ‘Materials and method’ section . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 01910 . 7554/eLife . 00691 . 020Figure 6—figure supplement 2 . Anion exchange elution pattern of the PTEN X-mutant . Mutating T366 and S370 to alanine in the unligated ( C-terminally truncated ) X-mutant removes the observance of multiple peaks in the anion exchange chromatogram . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 02010 . 7554/eLife . 00691 . 021Figure 6—figure supplement 3 . Alkaline phosphatase sensitivity of 4p-PTEN and mutant forms . 4p-PTEN and its mutants were treated with 1 μM alkaline phosphatase . Dephosphorylation of the phospho-tail cluster was monitored by western blot with an antibody to the phospho-tail . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 02110 . 7554/eLife . 00691 . 022Figure 6—figure supplement 4 . Anion exchange chromatography elution profiles of phosphorylated and unphosphorylated D5 PTEN . ( Top ) D5 n-PTEN elutes as a sharp peak around 100 ml . ( Bottom ) D5 4p-PTEN elutes as a broad peak starting at 90 ml and finishing past 156 ml . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 02210 . 7554/eLife . 00691 . 023Figure 6—figure supplement 5 . PTEN activity to diC6 PIP3 . n-PTEN and its mutants activity was measured with 160 μM soluble diC6 PIP3 . Phosphate release was monitored by Malachite green detection . Data are represented as the mean ± the SEM of three experiments preformed in duplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 023 In contrast , the D5 mutant 4p-PTEN showed a distinctive biochemical behavior . With respect to anion exchange chromatography , D5 mutant 4p-PTEN eluted broadly with much of the protein eluting at higher NaCl concentrations than the unphosphorylated D5 mutant ( Figure 6—figure supplement 4 ) . Later elution is the expected profile for a typical phosphorylated protein relative to an unphosphorylated protein compared with the paradoxical pattern for n- and 4p- semisynthetic PTENs . In addition , D5 mutant 4p-PTEN was dephosphorylated about eightfold faster than the native wt 4p-PTEN and only threefold slower than denatured wt 4p-PTEN , implying that the tail phosphate cluster is dramatically more exposed in the D5 mutant ( Figure 6A and Figure 6—figure Supplement 3 ) . We considered the possibility that Asp substitutions of the Lys residues in the D5 mutant led to overall protein destabilization and denaturation . However , D5- mutant 4p-PTEN readily processed soluble PIP3 substrate , ∼threefold faster than wt and A5-mutant 4p-PTENs ( Figure 6B ) . Moreover , the D5-mutant 4p-PTEN phosphatase activity showed saturation with soluble PIP3 ( Figure 6C ) , with catalytic parameters approaching those of n-PTEN , consistent with a more open conformation than that of wt 4p-PTEN . Taken together , these results make clear that the penta-Asp substitutions in the CBRIII loop are non-denaturing but can significantly disrupt the closed conformation of 4p-PTEN , presumably by electrostatic repulsion of the anionic tail . As reported previously ( Campbell et al . , 2003 ) , the activity of the N-mutant which contains mutations of K13 , R14 , and R15 to alanine , is reduced compared to wt ( Figure 6B and Figure 6—figure supplement 5 ) . We investigated the possibility that the synthetic C-tail tetraphosphorylated peptide ( aa 379–403 , 4p-25mer ) used in ligation could modulate the activity of truncated ( t-PTEN , aa 1–378 ) in an intermolecular fashion using the soluble PIP3 substrate phosphatase assay . In this assay , t-PTEN showed similar behavior to n-PTEN , with saturable kinetics as a function of soluble PIP3 ( compare Figures 7B and 2A ) . Of note , the C-tail phosphopeptide 4p-25mer , but not the unphosphorylated peptide n-25mer , was a potent inhibitor of t-PTEN soluble PIP3 phosphatase activity with an IC50 ∼ 1 μM ( Figure 7A ) . In the presence of 1 μM 4p-25mer , the t-PTEN phosphatase activity showed an increase in Km for soluble PIP3 ( Figure 7B ) , mimicking the behavior of 4p-PTEN . Strikingly , the D5 mutant form of t-PTEN was resistant to the inhibition by 4p-25mer at concentrations up to 10 μM of the peptide ( Figure 7C ) . Analogous to its effects on t-PTEN phosphatase activity , 4p-25mer but not n-25mer inhibited t-PTEN binding to PIP2 vesicles , although less potently presumably because of competition for binding surfaces on the body of PTEN between the vesicle lipids and the 4p-25mer tail peptide ( Figure 7D ) . Taken together , these experiments reveal that the intermolecular effects of the C-tail phosphopeptide on the PTEN body resemble those proposed for the intramolecular conformational change in 4p-PTEN . 10 . 7554/eLife . 00691 . 024Figure 7 . In trans peptide inhibition and binding of PTEN . ( A ) Tail peptide inhibition of t-PTEN with either n-25mer or 4p-25mer . ( B ) Km curve of t-PTEN in the absence ( kcat = 2 . 9 ± 0 . 1 min−1; Km = 33 ± 2 . 1 μM , kcat/Km = 0 . 088 ± 0 . 004 min−1μM−1 ) or presence ( kcat/Km = 0 . 006 ± 0 . 0003 min−1μM−1 ) of 1 μM 4p-25mer phosphopeptide . ( C ) Reduced inhibition of D5 t-PTEN mutant in the presence of 4p-25mer peptide . ( D ) Vesicle sedimentation of t-PTEN in absence and presence of 10 μM tail peptides . Data points are shown as the mean ± the SEM of three experiments performed in duplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 024 The evidence developed here points to a straightforward model for how phosphorylation regulates PTEN structure and function . Upon phosphorylation on the 380–385 Ser/Thr cluster , the PTEN C-terminal modified tail clamps down intramolecularly on the C2 domain in the vicinity of the CBRIII loop , preventing PTEN from binding the plasma membrane and reducing its catalytic action toward PIP3 ( Figure 8 ) . The combined structural data including ion exchange chromatographic behavior , trypsin protease susceptibility , alkaline phosphatase sensitivity , and SAXS analysis point to a more compact 4p-PTEN relative to n-PTEN in which the phosphates on the tail are concealed . 10 . 7554/eLife . 00691 . 025Figure 8 . Model of PTEN regulation by phosphorylation . Upon phosphorylation , PTEN adopts a more compact conformation with the phosphorylated tail condensing around the CBRIII loop and membrane binding surface of the C2 domain , preventing it from binding to the plasma membrane . When dephosphorylated , the tail of PTEN is no longer bound tightly to the C2 domain , allowing for the open PTEN protein to bind efficiently to the plasma membrane . Both phosphorylated and unphosphorylated PTEN are in the same open conformation when bound to the plasma membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 00691 . 025 While it is formally possible that the phospho-tail of PTEN could cause indirect effects on the C2 domain by binding elsewhere on the PTEN body , the accumulated evidence argues for a direct phospho-tail–C2 interaction . The SAXS results suggest that phospho-tail is in close proximity to the C2 domain . Replacement of the PTEN Lys cluster of the CBRIII loop with Asp residues ( 5D mutant ) but not Ala residues ( 5A mutant ) in 4p-PTEN renders the phospho-tail more susceptible to alkaline phosphatase hydrolysis , presumably because the anionic phosphate tail clashes with the Asp carboxylate negative charges . The intermolecular effects of the 4p-25mer C-tail on the t-PTEN body and their reduced sensitivity conferred by CBRIII loop mutation further corroborate the phospho-tail–C2 interaction . Prior studies suggest that the five clustered Lys residues of the CBRIII loop are found on the membrane binding surface of the C2 domain , and their mutation decreases PTEN binding to lipid vesicles in vitro and plasma membranes in cell transfection experiments ( Lee et al . , 1999; McConnachie et al . , 2003 ) . It is within the realm of possibility that 4p-PTEN can bind to membrane in its closed conformation , albeit with significantly reduced affinity . Our best evidence against the possibility that conformationally closed 4p-PTEN binds membrane is that the interfacial Kms ( iKms ) for 4p-PTEN , n-PTEN , and t-PTEN are all the same , within error , in dephosphorylating vesicle-bound PIP3 ( Figure 2 ) . Since the Kms for soluble PIP3 dephosphorylation are different for 4p-PTEN and n-PTEN ( Figure 2 ) , it would be surprising that the iKms would be the same if the conformation of 4p-PTEN were closed when bound to vesicle . It is plausible , however , that some closed 4p-PTEN with reduced membrane affinity is interacting with vesicle while in a catalytically impaired state , which would contribute only in a minor way to the iKm . Nevertheless , we also know how important the CBRIII loop is for membrane binding which , as discussed above , is likely abutting the phospho-tail in the closed conformation . Thus , the information available strongly supports the model in Figure 8 . We observed here that C-terminal phosphorylation of PTEN significantly reduces the enzyme’s lipid bilayer affinity and also its catalytic efficiency with soluble PIP3 substrate and at low ( <1 mol% ) PIP3 surface concentrations in vesicles . Since physiologic membrane surface concentrations of PIP3 are in the range of 0 . 001 mol% or less , C-terminal phosphorylation of PTEN should confer an important reduction in enzymatic activity in vivo ( Rahdar et al . , 2009 ) . It is interesting that the soluble PIP3 Km for 4p-PTEN vs n-PTEN is elevated whereas the interfacial Km ( iKm ) with vesicles containing PIP3 is the same for both enzyme forms . Both the trypsin protease pattern and SAXS analysis of 4p-PTEN vs n-PTEN reveal apparent structural/dynamic changes within the catalytic domain upon phosphorylation that may account for the altered phosphatase properties . Such conformational changes in the catalytic domain upon phosphorylation may come from direct interactions with the tail , though they could also be transmitted allosterically through tail-C2 binding since the C2 domain makes intimate contacts with the catalytic domain . It should be noted that the CBRIII loop , while critical for vesicle interaction , is dispensable for soluble PIP3 substrate processing , as shown here and previously ( McConnachie et al . , 2003 ) . Thus , there are at least some differences between interactions of PTEN with soluble PIP3 and membrane-embedded PIP3 . Results from previous experiments involving co-immunoprecipitation of the co-transfected PTEN tail and body as separate pieces hinted at the potential of a conformational change induced by phosphorylation ( Rahdar et al . , 2009 ) . However , in contrast to the findings here , the co-immunoprecipitation of the PTEN tail and body was disrupted by the altered residues in the N-mutant and A5-mutant PTENs . Since the N-mutant and A5-mutant 4p-PTENs showed the same apparent closed equilibrium constant as wt 4p-PTEN , this suggests that there may be differences in this trans interaction identified in mammalian cell extracts vs the closed conformation of the intact 4p-PTEN molecule analyzed in the current study . Indeed , we found higher than expected apparent affinity between the 4p-25mer and the t-PTEN body ( IC50 1 μM ) and more complete inhibition ( >95% ) of the lipid phosphatase activity of t-PTEN by the 4p-25mer with soluble PIP3 substrate than would be predicted based on results with 4p-PTEN ( 4p-PTEN:n-PTEN activity was ∼6:1 ) . The ∼25:1 equilibrium favoring the closed conformation of 4p-PTEN based on alkaline phosphatase susceptibility seems relatively small compared with an apparent Kd of ∼1 μM for the 4p-25mer-t-PTEN intermolecular interaction . Such results may suggest structural differences in the intermolecular complex vs the intramolecular conformational change . While there are several plausible explanations for this , one interesting possibility is that there may be energetic strain associated with achieving conformational closure in the intramolecular case of 4p-PTEN not associated with a phospho-tail–t-PTEN intermolecular interaction . Long-range intramolecular protein conformational switching induced by phosphorylation has been observed in several well-established cases including CrkL , Src , and SHP-1/2 ( Lu et al . , 2001; Rosen et al . , 1995; Sicheri et al . , 1997; Xu et al . , 1997; Zhang et al . , 2003 ) . Each of these examples involves an SH2 domain interacting with a phosphotyrosine . We propose a Src-like model for PTEN , in which a cluster of pSer/pThr drives an apparently long distance intramolecular binding interaction to deactivate the enzyme . How 4p-PTEN may be reactivated in vivo remains an important question . There are presumably cellular phosphatase enzymes that can dephosphorylate phospho-PTEN or ligands which can bind allosterically to phospho-PTEN that may promote conformational opening . In contrast to prior suggestions based on trans experiments , we see no evidence of the autodephosphorylation of 4p-PTEN ( Zhang et al . , 2012 ) . As we are reliant on western blots with an antibody that may recognize states of depleted phosphorylation of the tail , partial autodephosphorylation cannot be completely ruled out by our experiments . However , the stability of the closed conformation of 4p-PTEN over the dozens of hours of expressed protein ligation points to resistance of the tail phosphates to PTEN enzymatic removal . Because of the structural concealment of the 380–5 phosphoSer/Thr residues in the closed 4p-PTEN conformation , such autodephosphorylation may be especially disfavored . There are several biomedical implications of the findings here . Altered protein–protein interactions or efficiency of ubiquitylation may be influenced by phosphorylation-mediated conformational closure of phospho-PTEN . While PTEN is a tumor suppressor and can be mutated in cancer , it is often wild type but expressed at low levels . Direct stimulation of cellular phospho-PTEN by pharmacologic agents could prove to be effective as an anti-cancer therapy . A related approach has been explored to activate the tumor suppressor p53 ( Foster et al . , 1999 ) and the apoptotic protein procaspase ( Gray et al . , 2010 ) . It may be possible to find small molecules that bind specifically to the phospho-tail and prevent its intramolecular engagement with the C2 domain or bind somewhere on the PTEN body and stabilize the open PTEN conformation allosterically . The soluble PIP3 substrate dephosphorylation assay with 4p-PTEN should provide a means for screening for such activators . Alternatively , inhibitors of CK2 and/or GSK3β protein kinase , the enzymes responsible for PTEN C-tail phosphorylation , may be effective in targeting PIP3/Akt-driven tumors . This investigation also highlights the use of intein-mediated protein thioester formation in an insect cell expression system for investigation of post-translational modifications . Multi-site phosphorylation has been elegantly studied in the TGFβ signaling axis using protein semisynthesis previously ( Huse et al . , 2001; Wu et al . , 2001 ) , but these experiments did not require insect cell expression of an intein-fusion protein . Our efforts here suggest that baculovirus expression systems are an attractive option , rather than a last resort , for expressed protein ligation with challenging eukaryotic proteins . All lipids were from Avanti Polar Lipids ( Alabaster , AL ) . Antibodies were from Santa Cruz Biotechnology ( Dallas , TX ) ( SC-6818 ) and Novus Biological ( Littleton , CO ) ( NBP1-4136 and NBP1-44 , 412 ) . MESNA was from Sigma ( St . Louis , MO ) . All Fmoc-amino acids were from EMD ( Billerica , MA ) . All peptides were synthesized on a PS3 peptide synthesizer from Protein Technologies ( Tuscon , AZ ) or by hand using Fmoc based standard solid phase peptide synthesis . PTEN C-terminally truncated at residue 378 was first subcloned into the pTXB1 vector from NEB which contains the GyrA intein from the organism Mycobacterium xenopi . Tyr 379 was mutated to a Cys to facilitate the intein mediated cleavage reaction . The PTEN-intein-cbd DNA sequence was then subcloned into the pFastBac1 baculovirus entry vector and the subsequent baculovirus was generated . The PTEN-intein-cbd fusion protein was expressed in HighFive insect cells . The fusion protein in 40 ml of cell lysate from 1 l of cell culture was first passed over a 10 ml bed of fibrous cellulose ( Whatman ) to remove viral chitinase , then bound to a 6 ml bed of chitin beads from NEB in a gravity flow chromatography column from Bio-Rad ( 2 . 5 cm diameter ) . The chitin bead column with fusion protein bound was then washed with 250 ml of washing buffer ( 50 mM HEPES pH 7 . 6 , 250 mM NaCl , 0 . 1% Triton X-100 ) . C-terminally truncated PTEN ( t-PTEN ) was generated by DTT cleavage of the fusion protein , producing t-PTEN at yields of ∼8–10 mg per liter of cell culture . Full length semisynthetic PTEN was then generated on the chitin column by adding 400 mM MESNA and 2 mM of C-terminal peptide buffered with 3 ml of 50 mM HEPES ( pH 7 . 2 ) , 150 mM NaCl . Based on the post ligation yield of PTEN protein it is estimated that there is a maximum PTEN-thioester concentration of ∼80 μM in the ligation reaction . The ligation reactions were carried out for 48–72 hr at room temperature and monitored by SDS PAGE . Upon completion of the ligation reaction the ligation mixture was eluted from the chromatography column with 15 ml of dialysis buffer ( 50 mM Tris pH 8 . 0 , 150 mM NaCl , 10 mM DTT ) and subsequently dialyzed into 4 l of dialysis buffer for a period of 48 hr with multiple buffer exchanges in a dialysis cassette ( Slidealyzer ) with a 12K MWCO in order to remove excess unreacted peptide . Proteins were then concentrated following dialysis to >5 mg/ml ( >100 μM ) . We estimate that there is <10 μM of residual unligated peptides at this stage . Due to large dilutions ( >1000-fold ) of the semisynthetic enzyme for enzymatic and other biochemical assays , small amounts of residual contaminating peptide remaining after dialysis would not be expected to interfere with any assays . Semisynthetic PTEN proteins produced in this way yields 8–10 mg of protein per liter of insect cell culture with the desired modifications on the C-terminus at purities of >90% based on Coomassie stained SDSPAGE . The semisynthetic protein was further purified for SAXS and soluble substrate assays by anion exchange chromatography ( monoQ ) using an AKTA FPLC from GE Healthcare . Proteins were purified with a gradient of 0–50% Buffer B over 250 ml at a flow rate of 1 . 0 ml/min ( Buffer A: 50 mM Tris pH 8 . 0 , 10 mM DTT; Buffer B: 50 mM Tris pH 8 . 0 , 1 . 0 M NaCl , 10 mM DTT ) . After FPLC purification by anion exchange chromatography and concentration the final yield of semisynthetic PTEN protein was 2–3 mg per liter of cell culture with estimated purity >95% by coomassie stained SDS-PAGE . Size-exclusion chromatography was carried out with a Superdex 200 column in the following buffer: 50 mM Tris pH8 . 0 , 150 mM NaCl , 10 mM DTT . Radiolabeled PIP3 was generated as previously described ( McConnache et al . , 2003 ) . Briefly , PIP3 labeled with 32Pi at the three position of the inositol ring was generated by incubating PI3K ( Echleon ) and PIP2:PS ( 1:1 ) vesicles in the presence of 250 mCi 32P-ATP , 1 mM ATP and 2 . 5 mM MgCl2 in PI3K assay buffer ( 25 mM HEPES pH 7 . 6 , 120 mM NaCl and 1 mM EGTA ) . After a Bligh-Dyer extraction of PIP3 , thin-layer chromatography ( TLC ) showed radiolabeled PIP3 to be the major product ( Figure 2—figure supplement 3 ) . TLC solvent conditions: ( CHCl3:Acetone:MeOH;Acetic Acid:H20 ) ; ( 70:20:50:20:20 ) . Lipid phosphatase assays were modified from those already described in the literature ( McConnachie et al . , 2003 ) . 32P radiolabeled PIP3 was incorporated into vesicles containing unlabeled PIP3 , phosphatidylcholine ( PC ) , and/or phosphatidylserine ( PS ) and/or PIP2 by sonication of dried lipids hydrated in the presence of PTEN assay buffer ( 50 mM Tris pH 8 . 0 , 150 mM NaCl , 10 mM DTT , 1 mM EGTA ) . Lipids were sonicated in 100 μl volumes in glass test tubes at room temperature until the solution clarified . Vesicles made in this way are 30–50 nm in diameter . Vesicles were used in assays within 15 min of being made . In assay reactions , the ratio of the number of vesicles to the number of PTEN molecules was maintained at 4:1 or greater . Ovalbumin ( 0 . 05 mg/ml ) was used to stabilize the PTEN protein in the assays . 25 μl reactions were initiated by the addition of vesicle substrate and incubated at 30°C for 3 min . The reaction was quenched with 3 M perchloric acid . Hydrolyzed 32Pi was then separated from 32P-PIP3 by a Bligh–Dyer extraction . The aqueous phase was then treated with 1% ammonium molybdate and the resulting phosphate–molybdate complex was extracted with toluene:isobutanol ( 1:1 ) . This organic phase was then counted using a Beckman scintillation counter . Analysis of the kinetic parameters of the semisynthetic PTEN proteins were determined in accordance with the procedures pioneered by Dennis and coworkers and previously performed for recombinant PTEN produced in E . coli ( Deems et al . , 1975; Hendrickson and Dennis , 1984; McConnachie et al . , 2003 ) . With this type of analysis , the initial velocity of an interfacial enzyme follows the equation below:V0= ( VmaxXs[S0] ) / ( iKmKs+iKm[S0]+Xs[S0] ) Two types of experiments were performed , bulk dilution ( BD ) and surface dilution ( SD ) . In BD experiments , the surface concentration of PIP3 was held constant and bulk concentration of PIP3 was varied by varying the concentration of PIP3 and the carrier lipid PC proportionately . In SD experiments , the bulk concentration of PIP3 was held constant and the surface concentration was varied by varying the amount of PC . In both types of experiments rectangular hyperbolas were obtained with apparent Vmax and apparent Km values . Apparent Vmax and apparent Km values were then fit to the equations below to determine the kinetic variables for each PTEN protein . ikm= ( Vmax SD/Vmax BD−1 ) XsKs=Km BD ( Xs/iKm+1 ) kcat=Vmax SD[ET] Large Multilamillar Vesicles ( LMVs ) containing various amounts of PC , PS and/or PIP2 were generated by vigorously vortexing dried lipids that were hydrated in the presence of PTEN buffer for 5 min in 1 ml volumes . The LMVs were then incubated with different forms of PTEN protein for 30 min at 25⁰C . The vesicles and bound protein in 50 μl volumes were then pelleted at 180 , 000g using a Beckman ultracentrifuge for 2 hr . The supernatant was removed from the vesicle pellet , the pellet washed with buffer , then boiled in 10% SDS loading dye and run on SDS-PAGE . The amount of PTEN protein that bound to the LMVs was then visualized by western blot using an anti-PTEN antibody from Santa Cruz Biotechnologies ( SC-6818 ) . The amount bound was quantified using Carestream Media image quantification software . For tail peptide competition assays the amount of tail peptide used was 10 μM . 2 μg of semisynthetic PTEN in 20 μl reactions volumes was digested with varying amounts of trypsin ( Promega , V511A ) for 10 min at 37°C in PTEN assay buffer . Reactions then were quenched with SDS loading dye and run on SDS-PAGE . The digestion fragments were visualized by Colloidal Blue stain from Invitrogen ( LC6025 ) or by western blot ( antibodies SC-6818 or NBP1-44 , 412 ) . Trypsin digestion products were run on a 10% SDS-PAGE gel and transferred to a PVDF membrane . The membrane was then stained with Coomassie stain . The bands of interest were cut out of the membrane and analyzed by N-terminal Edman degradation sequencing at the JHMI Synthesis and Sequencing Facility . 50 ng of semisynthetic phosphorylated PTEN and its mutants were dephosphorylated in the presence of 1 μM alkaline phosphatase from NEB ( CIP ) for varying periods of time in phosphatase assay buffer ( 50 mM Tris pH 8 . 0 , 20 mM NaCl , 25 μM MgCl2 and 10 mM DTT ) at room temperature in 20 μl . Dephosphorylation of PTEN was monitored by western blot with an antibody to the phospho-tail cluster ( NBP1-4136 ) . The fraction of phospho-PTEN remaining was determined using Carestream Media image quantification software . PTEN activity to a water soluble substrate ( diC6 PIP3 ) was determined by measuring the release of inorganic phosphate with a malachite green ( Van Veldhoven and Mannaert , 1987 ) detection kit from R and D Biosystems . 25 μl reactions were allowed to proceed for 5–10 minutes at 30°C in assay buffer ( 50 mM Tris pH 8 . 0 , 10 mM BME ) before being quenched by malachite green reagent . Amounts of PTEN used per data point ranged from 0 . 5 to 20 μg . Reactions were shown to be linear with respect to time and enzyme concentration in the ranges used . For in trans peptide inhibition assays , the amount of tail peptide used ( quantified by amino acid analysis ) is indicated in the figure legend . It is unlikely that any peptide ligated to t-PTEN given the low concentrations of peptide used , short reaction times and the low reactivity of the DTT-thioester toward native chemical ligation . In fact , with the vesicle pulldowns that used 10 μM phosphopeptide and t-PTEN , there is no evidence of ligation observed by western blot . SAXS experiments were performed at Brookhaven National Laboratories at the National Synchrotron Light Source ( NSLS ) , beamline X9 using a MarCCD detector , located 3 . 4 m from the sample . Data for each protein sample was collected in triplicate . All samples were in PTEN assay buffer . 20 μl of each sample was continuously passed through a capillary tube exposed to a 400 × 200 μm X-ray beam and data recorded for 30s . Normalization for beamline intensity , buffer subtraction and merging of data were carried out using proc . py software developed by the beamline staff ( Allaire and Yang , 2011 ) . SAXS data analysis was carried out using software from the ATSAS program suite . The radius of gyration ( Rg ) was calculated using a Guinier approximation with the program PRIMUS ( Konarev et al . , 2003 ) . The pair distribution function P ( r ) and the maximum particle dimension ( Dmax ) were determined using GNOM ( Svergun et al . , 1992 ) . Ten ab initio models were generated for each protein using DAMMIN , then averaged using DAMAVER ( Volkov and Svergun , 2003; Putnam et al . , 2007; Franke and Svergun , 2009 ) . The resulting molecular envelopes were fit with the tailless crystal structure in PyMOL . Figures of the models were made using PyMOL .	PTEN is an enzyme that is found in almost every tissue in the body , and its job is to stop cells dividing . If it fails to perform this job , the uncontrolled proliferation of cells can lead to the growth of tumors . PTEN stops cells dividing by localizing at the plasma membrane of a cell and removing a phosphate group from a lipid called PIP3: this sends a signal , via the PI3K pathway , that suppresses the replication and survival of cells . Three regions of PTEN are thought to be central to its biological functions: one of these regions , the phosphatase domain , is directly responsible for removing a phosphate group from the lipid PIP3; a second region , called the C2 domain , is known to be critical for PTEN binding to the cell membrane; however , the role of third region , called the C-terminal domain , is poorly understood . Many proteins are regulated by the addition and removal of phosphate groups , and PTEN is no exception . In particular , it seems as if the addition of phosphate groups to four amino acid residues in the C-terminal domain can switch off the activity of PTEN , but the details of this process have been elusive . Now , Bolduc et al . have employed a variety of biochemical and biophysical techniques to explore this process , finding that the addition of the phosphate groups reduced PTEN’s affinity for the plasma membrane . At the same time , interactions between the C-terminal and C2 domains of the PTEN cause the shape of the enzyme to change in a way that ‘buries’ the residues to which the phosphate groups have been added . In addition to offering new insights into PTEN , the work of Bolduc et al . could help efforts to identify compounds with clinical anti-cancer potential .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2013	Phosphorylation-mediated PTEN conformational closure and deactivation revealed with protein semisynthesis
The morphogenesis of tissues , like the deformation of an object , results from the interplay between their material properties and the mechanical forces exerted on them . The importance of mechanical forces in influencing cell behaviour is widely recognized , whereas the importance of tissue material properties , in particular stiffness , has received much less attention . Using Caenorhabditis elegans as a model , we examine how both aspects contribute to embryonic elongation . Measuring the opening shape of the epidermal actin cortex after laser nano-ablation , we assess the spatiotemporal changes of actomyosin-dependent force and stiffness along the antero-posterior and dorso-ventral axis . Experimental data and analytical modelling show that myosin-II-dependent force anisotropy within the lateral epidermis , and stiffness anisotropy within the fiber-reinforced dorso-ventral epidermis are critical in driving embryonic elongation . Together , our results establish a quantitative link between cortical tension , material properties and morphogenesis of an entire embryo . Morphogenesis and organ formation rely on force distribution and tissue material properties , which are often heterogeneous and evolve over time . Forces are generated through a group of relatively well-conserved molecular motors associated with the cytoskeleton , among which , myosin II linked to actin filaments is the most prevalent during epithelial morphogenesis ( Vicente-Manzanares et al . , 2009 ) . The spatial distribution and dynamics of myosin II greatly influence morphogenetic processes ( Levayer and Lecuit , 2012 ) . In particular , the asymmetric distribution of the actomyosin network and its pulsatile behaviour define the direction of extension during Drosophila germband elongation ( Bertet et al . , 2004; Blankenship et al . , 2006 ) , Drosophila renal tubule formation ( Saxena et al . , 2014 ) or Xenopus mesoderm convergent extension ( Shindo and Wallingford , 2014 ) . The implications of mechanical forces on cell behavior have been intensively investigated ( Zhang and Labouesse , 2012; Heisenberg and Bellaïche , 2013 ) , but many fewer studies have considered the impact of tissue material properties in vivo , except for their influence on cell behaviour in vitro ( Kasza , 2007 ) . Embryonic elongation in C . elegans represents an attractive model for studying morphogenesis , as it offers single-cell resolution and powerful genetic analysis . During its elongation , the embryo evolves from a lima-bean shape to a typical cylindrical shape with a four-fold increase in length , without cell migration , cell division , or a notable change in embryonic volume ( Sulston et al . , 1983; Priess and Hirsh , 1986 ) ( Figure 1a ) . This process requires the epidermal actomyosin cytoskeleton , which acts mostly in the lateral epidermis ( also called seam cells ) , while the dorso-ventral ( DV ) epidermal cells may remain passive ( Appendix 1 ) ( Wissmann et al . , 1997; 1999; Shelton et al . , 1999; Piekny et al . , 2003; Diogon et al . , 2007; Gally et al . , 2009; Chan et al . , 2015; Vuong-Brender et al . , 2016 ) . Indeed , the non-muscle myosin II is concentrated in seam cells; in addition , short disorganized actin filaments , which favour actomyosin contractility , are present in seam cells but not in the DV epidermis , where they instead form parallel circumferential bundles ( Figure 1b–d ) ( Gally et al . , 2009; Priess and Hirsh , 1986 ) . The actomyosin forces are thought to squeeze the embryo circumferentially , thereby increasing the hydrostatic pressure and promoting embryo elongation in the antero-posterior ( AP ) direction ( Priess and Hirsh , 1986 ) ( Figure 1e ) . 10 . 7554/eLife . 23866 . 003Figure 1 . Overview of C . elegans embryonic elongation . ( a ) Embryonic elongation in C . elegans is driven in part by epidermal actomyosin contractility and in part by muscle contractions . The length of the embryo is used for staging: 2-fold ( 2F ) stage means roughly 2-fold increase in length from the beginning of elongation . Representative stages are shown; anterior to the left , dorsal up . ( b , c , d ) Actin filament organization at the 1 . 3F , 1 . 5F and 1 . 7F stages , respectively , visualized with an ABD::GFP marker . Actin filaments progressively organize into circumferential parallel bundles in DV cells ( arrows ) , arrowheads point to seam cells . Note that the integrated ABD::GFP marker shows some cell to cell variation in expression . ( b’ , b” , c’ , c” , d’ , d” ) Close-up images of actin pattern in DV cells ( from the area in the white rectangle ) and seam cells ( from the area in the pink rectangle ) , respectively , of the images in ( b ) , ( c ) and ( d ) respectively . ( e ) Actomyosin forces squeeze the embryo circumferentially to make it elongate in the antero-posterior direction . ( f , g ) Endogenous distribution of the two non-muscle myosin II isoforms visualized with the CRISPR GFP-labelled myosin heavy chains NMY-2 ( f ) and NMY-1 ( g ) . Arrowheads point to seam cells , which are delineated by the junctional marker DLG-1::RFP . DOI: http://dx . doi . org/10 . 7554/eLife . 23866 . 003 Although the published data clearly implicate myosin II in driving elongation , they raise a number of issues . First , myosin II does not show a polarized distribution ( Figure 1f , g ) , nor does it display dynamic pulsatile foci at this stage; hence , it is difficult to account for the circumferential squeezing . Moreover , force measurements are lacking to establish that the actomyosin network does squeeze the embryo circumferentially . Second , a mechanical continuum model is needed to explain how the embryo extends preferentially in the AP direction . To address those issues , we used laser ablation to map the distribution of mechanical stress ( i . e the force per unit area ) and to assess tissue stiffness ( i . e . the extent to which the tissue resists deformation ) in the embryonic epidermis . We then correlated the global embryonic morphological changes with these physical parameters . Finally , we developed continuum mechanical models to account for the morphological changes . Altogether , our data and modelling demonstrate that the distribution of forces in the seam cells and the stiffness in the DV epidermis must be polarized along the circumferential axis ( or DV axis ) to drive elongation . To measure the stress distribution on the actin cortex , we used laser nano-ablation , which has now become a standard method to assess forces exerted in cells , to sever the actin cytoskeleton and to observe the shape of the opening hole ( Figure 2a ) . We visualized actin with a GFP- or mCherry-labelled actin-binding-domain protein ( ABD ) expressed in the epidermis ( Gally et al . , 2009 ) ( Figure 1b–d ) . We adjusted the region of interest to cut within one cell , restricting our analysis to the early phase of elongation ( ≤1 . 7F; for staging , see Figure 1 legend ) . 10 . 7554/eLife . 23866 . 004Figure 2 . Physical model using the shape of the cut opening at equilibrium to measure the ratio of stress to Young modulus . ( a ) The GFP-labelled actin cortex of HYP7 dorsal epidermal cell at the 1 . 7F stage before ( 0 s ) , 1 . 4 s and 10 s after laser severing; the cut made along the AP direction and was 5 μm in length . Double arrowheads indicate the distance between cut borders , which increases with time . ( b ) Model of epidermal cells as an infinite elastic plane under biaxial stress in the AP and DV directions after an incision of length l . The final shape of the cut opening is an ellipse . The cut opening in the DV direction ( after an incision along the AP direction ) depends on the stress along the DV direction and the Young modulus ( see text ) . ( c ) The opening depends on myosin II activity: comparison of the cut response in the seam cell H1 between wild-type ( WT ) and let-502 ( sb118 ) /Rho kinase mutant embryos at a stage when muscles start to twitch ( around 1 . 5F ) . The average value and standard error are reported . Time zero , moment of the cut; DV and AP show direction of opening . Two-tailed t-test , **p=410−7 between WT DV and let-502 ( sb118 ) DV , p=410−6 between WT AP and let-502 ( sb118 ) AP; N , number of embryos examined . DOI: http://dx . doi . org/10 . 7554/eLife . 23866 . 00410 . 7554/eLife . 23866 . 005Figure 2—figure supplement 1 . Evolution of the distance between the cut borders ( the minor axis of the cut opening ) versus time . The average value and standard error are reported . Time 0 is the moment of the cut . Solid lines show the single exponential fit with an initial width of cut opening of 0 . 6 μm ( see Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23866 . 00510 . 7554/eLife . 23866 . 006Figure 2—figure supplement 2 . Calcium imaging of ablated embryos . ( a ) Scheme for measuring the calcium sensor GCaMP3 mean fluorescence intensity at positions 9 μm and 17 μm away from the cut site ( red bar ) . The intensity was averaged over a region of interest ( ROI ) of 1 . 5 × 1 . 5 μm2 . A , anterior; P , posterior; D , dorsal; V , ventral . ( b , c ) Calcium level over time for an embryo typical of the majority of ablated embryos , in which the cortex is probably disrupted locally ( b ) , or of the minority of ablated embryos , in which the wounding response is probably more severe ( c ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23866 . 006 We observed two types of ablation responses ( see Materials and methods ) . In the first ( accounting for >80% of the cases ) , the opening hole within the actin cytoskeleton reached equilibrium in less than 10 s , and resealed within less than 2 min ( Figure 2—figure supplement 1 , Video 1 ) . In such embryos , actin occasionally accumulated around the cut borders but not around cell borders ( Video 1 ) . Imaging calcium levels , which can rise after laser wounding ( Xu and Chisholm , 2011; Razzell et al . , 2013; Antunes et al . , 2013 ) , showed either no change or a localized increase ( Video 2 , Figure 2—figure supplement 2a , b ) . In the second ablation reponse , an actin ring accumulated around the cell borders during the repair process ( Video 3 ) and a calcium wave propagated to nearby epidermal cells ( Video 4; Figure 2—figure supplement 2c ) . Embryos showing the first response continued to develop and hatched , whereas those showing the second response arrested their development and eventually died . In all subsequent studies , we only took into account the first type of response , which should correspond to a local cortex disruption . 10 . 7554/eLife . 23866 . 007Video 1 . Local disruption of actin cortex with laser ablation , visualized with the actin marker ( ABD::mCHERRY ) expressed under the epidermal promoter lin-26 . 0 s time corresponds to the first picture after the laser cut . The yellow line shows the cut region . DOI: http://dx . doi . org/10 . 7554/eLife . 23866 . 00710 . 7554/eLife . 23866 . 008Video 2 . Local disruption of actin cortex with laser ablation does not induce noticeable change in calcium level . The calcium sensor GCaMP3 was expressed under the epidermal promoter dpy-7 ( which provides strong expression in the dorso-ventral cells ) . 0 s time corresponds to the first picture after the laser cut . The yellow line shows the cut region . This is the same embryo as that shown in Video 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 23866 . 00810 . 7554/eLife . 23866 . 009Video 3 . Wound-healing response after laser ablation visualized with the actin marker ABD::mCHERRY expressed under the epidermal promoter lin-26 . 0 s time corresponds to the first picture after the laser cut . The yellow line shows the cut region . DOI: http://dx . doi . org/10 . 7554/eLife . 23866 . 00910 . 7554/eLife . 23866 . 010Video 4 . Calcium wave propagation in a wound-healing response after laser ablation . The calcium sensor GCaMP3 was expressed under the epidermal dpy-7 promoter ( which provides strong expression in the dorso-ventral cells ) . 0 s time corresponds to the first picture after the laser cut . The yellow line shows the cut region . This is the same embryo as that shown in Video 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 23866 . 010 To compare the response between different conditions , we detected the cut-opening shape , which we fitted with an ellipse to derive the shape parameters ( see Materials and methods ) . The laser setup we used did not enable us to image the recoil dynamics within the first second after the cut , which other investigators previously used to assess the extent of mechanical stress ( Rauzi and Lenne , 2015; Smutny et al . , 2015; Saha et al . , 2016 ) . To circumvent this issue , we developed a novel analysis method to derive mechanical stress data , based on the equilibrium shape of a thin cut in an infinite elastic isotropic plane , subjected to biaxial loading ( stress applied in two perpendicular directions ) ( Theocaris , 1986 ) . The rationale for approximating the epidermis to such a plane is further outlined in the Appendices 2 and 3 . In these conditions , a thin cut will open to form an elliptical hole at equilibrium ( Figure 2a ) . The opening of the cut reflects mechanical stress in the direction perpendicular to the cut direction . We cut the epidermal actin specifically in the AP and DV directions , which we found to correspond to the stress loading directions ( Figure 2b; Appendix 2 ) . For a cut in the AP direction , the minor axis of the ellipse at equilibrium , bDV , will be proportional to the cut length , l , and to the ratio of stress in the DV direction , σDV , over the Young modulus E of the plane ( Theocaris et al . , 1986 ) ( Figure 2b ) : ( 1 ) bDV= 2σDVEl We will call the ratio bDVl the opening in the DV direction of a cut made along the AP direction ( Figure 2b ) , and similarly bAPl the opening in the AP direction of a cut made in the DV direction . Thus , we used the opening of the hole in a given direction to derive the stress in that direction . We compared the conclusions drawn from this method with methods relying on the recoil dynamics ( Rauzi and Lenne , 2015; Smutny et al . , 2015; Saha et al . , 2016 ) ( Appendix 3 ) . The half-time of the cut border relaxation , which depends on the ratio of viscosity over stiffness , was similar in the AP and DV directions ( Appendix 3 ) , supporting the hypothesis that the seam cell cortex is isotropic . We found an agreement between both methods for the AP versus DV stress ratio , and similar trends for the stress magnitude . To further examine the validity of this method , we performed two tests . First , the theory described above ( Theocaris , 1986 ) predicts that the minor to major axis ratio of the opening ellipse is independent of the initial cut length . We found that it is the case when the cut length varied from 3 μm to 6 μm ( Appendix 4 ) . Second , to prove that the opening observed after laser cutting depends on myosin II activity , we performed cuts in embryos defective for the main myosin II regulator , LET-502/Rho-kinase ( Gally et al . , 2009 ) . As shown in Figure 2c , the opening in the seam cell H1 at the 1 . 5F stage in let-502 ( sb118ts ) embryos changed very little and was significantly smaller than that in WT embryos , consistent with a decrease in mechanical stress . Thus , we feel confident that the method based on the opening shape measures actomyosin-dependent stress and can be used reliably to report on stress differences along the DV and AP directions . We applied the method described above on three seam cells ( head H1 , body V3 , tail V6; Figure 3a ) because myosin II acts mainly in seam cells ( Gally et al . , 2009 ) , and compared the response with embryonic morphological changes . We focused on the anisotropy of stress between the DV and AP directions ( difference of stress along both directions ) in a given cell ( Figure 3b ) . Indeed , in other systems , such as Drosophila embryos ( Rauzi et al . , 2008 ) and C . elegans zygotes ( Mayer et al . , 2010 ) , this parameter is critical . 10 . 7554/eLife . 23866 . 011Figure 3 . Stress anisotropy in seam cells correlates with morphological changes and partially depends on the spectrin cytoskeleton . ( a ) Scheme showing laser ablation experiments in the AP and DV directions for H1 , V3 and V6 seam cells at different stages . A , anterior; P , posterior; D , dorsal; V , ventral . ( b ) Cut opening in H1 , V3 and V6 from the 1 . 3F to the 1 . 7F stages ( see Figure 2b ) . The p-values from two-tailed t-tests are reported . ( c ) Changes in embryo length , head diameter at the level of H1 , body diameter at the level of V3 and tail diameter at the level of V6 , between the 1 . 3F and 1 . 7F stages . N = 10 . ( d ) Scheme showing the measurement of circumferential cell width in the head ( above ) and corresponding section ( below ) . ( e ) The circumferential width of H1 , V3 , head and body DV cells ( averaged for dorsal and ventral cells ) . N = 10 . ( f ) Measures of the cut opening in H1 for WT , spc-1 ( RNAi ) treated and unc-112 ( RNAi ) muscle-defective embryos at a stage equivalent to the 1 . 7F stage . The p-values of two-tailed t-tests are reported . ( g ) Comparison of the stress anisotropy in H1 , defined by DV/AP stress , between WT 1 . 7F stage , spc-1 ( RNAi ) embryos and unc-112 ( RNAi ) embryos at the equivalent 1 . 7F stage . The p-values of Z-tests are reported . The number of embryos used for ablation is given in Supplementary file 1 . For ( b , c , e–g ) , the average ( or calculated ) values and standard errors are reported . DOI: http://dx . doi . org/10 . 7554/eLife . 23866 . 01110 . 7554/eLife . 23866 . 012Figure 3—figure supplement 1 . Elongation curves normalized to the initial embryo length for different genetic backgrounds . ML1540 is the strain carrying the actin-binding domain fused to GFP ( see Materials and methods ) . The average value and standard error are reported . N ≥ 9 for each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 23866 . 012 At the 1 . 3F stage in H1 , there was no significant stress anisotropy; however , as the embryo elongated to the 1 . 5F and 1 . 7F stages , the stress became anisotropic ( Figure 3b ) . In V3 , the anisotropy of stress evolved in the opposite direction , with higher stress anisotropy at the 1 . 3F stage compared to the 1 . 5F stage ( Figure 3b ) . In V6 , the stress was slightly anisotropic at both the 1 . 3F and the 1 . 5F stage ( Figure 3b ) . In all cells , whenever the stress became anisotropic , it was higher in the DV direction . Overall , the opening increased as the embryo elongated from the 1 . 3F to the 1 . 5F stage and from the 1 . 5F to the 1 . 7F stage for H1 . To correlate the stress anisotropy with the morphological changes in the embryo , we used markers labelling cortical actin ( an ABD ) and junctions ( HMR-1/E-cadherin ) . We observed that the head , body and tail diameter ( at the levels of H1 , V3 and V6 , respectively ) decreased at different rates over time ( Figure 3c ) , as also observed by Martin and colleagues ( Martin et al . , 2014 ) . The head diameter did not diminish between the 1 . 3F and 1 . 5F stages when the stress was nearly isotropic , but decreased significantly between the 1 . 5F and 1 . 7F stages as the stress anisotropy increased . Conversely , the body diameter decreased most rapidly between the 1 . 3F and 1 . 5F stages , when the stress was highly anisotropic , then changed at a lower pace beyond the 1 . 5F stage , when the stress became less anisotropic . Finally , the tail diameter decreased nearly linearly between the 1 . 3F and 1 . 7F stages , at a lower rate than the body diameter , coinciding with a smaller anisotropic stress in V6 . Thus , the local morphological changes within the embryo correlate with locally higher stress in the DV compared to the AP direction . To define whether all cells contribute equally to the diameter change , we quantified the circumferential width of the epidermal cells H1 , V3 and their adjacent DV cells ( Figure 3d , e ) . At the level of V3 , the decrease in body diameter came from both seam ( V3 ) and DV cells , whereas in the head , it came mainly from DV cells ( Figure 3e ) . Collectively , our results strongly suggest that stress anisotropy correlates with morphological changes . Furthermore , we found that both seam and DV epidermal cells contribute to the changes in embryo diameter , irrespective of their level of active myosin II . Taking the H1 cell as an example , we considered some cellular factors that could contribute to the stress anisotropy in seam cells: ( i ) actin-anchoring proteins , and ( ii ) muscle-induced tension . To ease comparisons , we defined the anisotropy of stress ( AS ) as ( 2 ) AS=DV stressAP stress= σDVσAP which can be derived from the ratio of the opening along the DV and AP directions , see Equation ( 1 ) . First , we examined the actin-anchoring spectrin cytoskeleton , which is essential for embryonic elongation ( Moorthy et al . 2000; Norman and Moerman , 2002; Praitis et al . , 2005 ) . In spc-1 ( RNAi ) embryos , at a developmental timing equivalent to the 1 . 7F stage in control embryos , we found a smaller opening in both AP and DV directions and a decrease of AS compared to WT ( Figure 3f , g ) . This may account for the slower elongation rate of spc-1 ( RNAi ) embryos and their arrest at the 2F stage ( Figure 3—figure supplement 1 ) . Thus , spectrin partially contributes to the AS at the 1 . 7F stage . Second , we wondered whether muscle contractions , which start after the 1 . 5F stage , could account for AS changes ( Figure 3b ) . Compared to controls , embryos that are depleted in UNC-112/Kindlin , which mediates sarcomere assembly ( Rogalski et al . , 2000 ) , showed a significantly larger opening in both the AP and DV directions at the 1 . 7F stage , but no change in stress anisotropy ( Figure 3f , g ) . This is consistent with their wild-type elongation rate up to the 2F stage ( Figure 3—figure supplement 1 ) . Thus , AS establishment in the H1 cell after the 1 . 5F stage is independent of muscle contractions . To define the possible causal relationship between the AS and embryonic shape changes , we aimed to simplify the shape of the embryo in order to allow the application of classical physical laws such as the Young-Laplace equation , which predicts the relationship between surface tension and the surface curvature . As illustrated in Figure 1 , the embryo has a circular section and a cylindrical or conical shape depending on the stage , in which the epidermis is relatively thin ( 100 nm to 2 µm , depending on areas; www . wormatlas . org ) when compared to the embryo diameter ( 25 µm ) . Within the embryo , the epidermis is subjected to hydrostatic pressure when the section decreases ( Priess and Hirsh , 1986 ) . We can thus model the C . elegans embryo as an isotropic thin-wall ( the epidermis ) vessel with capped ends under hydrostatic pressure , and can determine the relationship between the mechanical stress on the epidermis and the embryo shape . First , we calculated the anisotropy of stress on the wall of such a vessel . For an axisymmetric vessel , the AS on the wall depends on the surface curvature and the radius ( Appendix 5 ) , which for simple geometrical configurations can be written as shown in Figure 4a–c . Typically , the AS factor , or the DV to AP stress ratio , is equal to one for a sphere , equal to two for a cylinder and takes an intermediate value between 1 and 2 for an ellipsoid . We can simplify the geometry of C . elegans embryos between the 1 . 3F and 1 . 5F stages as a curved cylinder ( body ) , attached to a sphere ( head ) ( Figure 4d , e ) . The head evolves into an ellipsoid between the 1 . 5F and 1 . 7F stages ( Figure 4f ) . Thus , the AS of the head can be determined easily . We previously observed that the AP stress among the seam cells at a given stage differs by 20% ( Figure 3b ) . Thus , if we approximate the AP stress as a constant at a given stage , the AS in the body will depend on the ratio of the body to head radius ( Figure 4d , e , Appendix 5 ) . Given the head and body diameter of the embryo ( Figure 3c ) , we can compare the AS predicted by the thin-wall vessel model with those derived experimentally using laser ablation ( Figure 4g ) . These values are nearly identical , showing that AS can be predicted on the basis of embryonic geometry . 10 . 7554/eLife . 23866 . 013Figure 4 . Stress anisotropy induces embryonic morphological changes . ( a , b , c ) Anisotropy of stress ( AS ) for a sphere ( a ) , an ellipsoid ( b ) and a cylinder ( c ) with DV and AP axis defined in the schemes; ( a , b ) show the middle plane; the major and minor axis of the ellipsoid are called a1 and a2 . ( d , e , f ) The embryo is schematized with a spherical ( 1 . 3F and 1 . 5F stages ) or ellipsoidal ( 1 . 7F stage ) head , and a curved cylindrical body . The AS in the head evolves from 1 ( sphere ) to that of an ellipsoid , whereas the body AS depends on the ratio of body to head radius ( R2/R1 ) . ( g ) Comparison of the predicted AS based on embryo diameter measurements ( see Figure 3c ) and the measured AS obtained from laser ablation experiments ( Figure 3b ) . ( h ) Hooke’s law written for an isotropic material like seam cells ( Appendix 6A ) ; ϵAPs and ϵDVs are the relative length changes along the AP and DV directions , respectively . The stress σAPs and σDVs along the AP and DV directions are supposed to be contractile ( so negative ) . E , seam cell Young modulus; A , anterior; P , posterior; D , dorsal; V , ventral . ( i , j ) Dependence of the AP and DV relative length change on the AS for different values of σAPs/E . DOI: http://dx . doi . org/10 . 7554/eLife . 23866 . 013 To examine whether the AS can dictate embryonic morphological changes , we related the deformation of the vessel wall with the forces applied using the Hooke’s law ( Figure 4h , Appendix 6A ) – for instance , Hooke’s law states that the one-dimensional deformation of a spring is equal to the ratio of the applied force to the spring stiffness . Similarly , in a two-dimensional system and for an isotropic material , the deformation is proportional to the mechanical stress ( forces ) and inversely proportional to the Young modulus ( stiffness ) along the different loading directions ( Figure 4h ) . The resulting equations , which assume that seam cells have an isotropic cortex and are subjected to contractile stress , correctly predict that the seam cell dimension increases along the AP axis ( εAP ) with the AS ( Figure 4h–j ) , and decreases along the DV axis ( εDV ) . Indeed , consistent with the equations , the head evolves from a sphere to an ellipsoid between the 1 . 5F to the 1 . 7F stages as the AS becomes greater than 1 ( Figure 3b , c ) . In conclusion , our experimental and modelling data show that the AS induces morphological changes in embryonic seam cells and provide a basis for understanding how the embryo elongates from a mechanical standpoint . As shown in Figure 3e , the head diameter reduction primarily involves changes in the circumferential width in the DV epidermis . As the RhoGAP RGA-2 maintains myosin II activation in these cells at a low level ( Diogon et al . , 2007 ) , actomyosin contractility in DV cells cannot account for such changes . However , in contrast to seam cells , DV epidermal cells have circumferentially oriented actin bundles ( Figure 1b–d ) , which based on recent observation could affect cell stiffness ( Calzado-Martín et al . , 2016; Salker et al . , 2016 ) . We thus hypothesized that the circumferential polarized actin distribution in DV epidermal cells could induce greater stiffness in that direction and thereby influences their deformation . To establish whether this is the case , we investigated both stress and stiffness distribution in the epidermal cells dorsal and ventral to the H1 seam cell using laser nano-ablation ( Figure 5a ) . As these cells are the precursors of the HYP7 syncytium , we will denote them HYP7 . 10 . 7554/eLife . 23866 . 014Figure 5 . The dorso-ventral epidermis behaves differently than the H1 seam cell in ablation experiments . ( a ) Scheme showing laser ablation experiments in the epithelial cell HYP7 dorsal and ventral to H1 , yellow crosses show cut directions . ( b ) Cut opening in the DV and AP directions measured in HYP7 between the 1 . 3F and 1 . 7F stages ( see Figure 2b ) . p-values of two-tailed t-tests are reported . ( c ) Comparison of the DV/AP opening ratio in the seam cell H1 and in the head HYP7 cell . The data were derived from Figures 3b and 5b . To simplify , we will call the cells that will form the future HYP7 syncytium the HYP7 cells . The average ( or computed ) values and standard errors are reported . The numbers of embryos used for ablation are given in Supplementary file 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 23866 . 014 In the HYP7 cell , the opening in the DV direction was larger than that in the AP direction ( Figure 5b; dorsal and ventral cells behaved similarly after laser cutting ) at the 1 . 5F and 1 . 7F stages , as was the case in the H1 cell ( Figure 3b ) . However , the ratio of DV/AP opening in HYP7 was greater than that in H1 ( Figure 5c ) . Assuming that the HYP7 cell cortex has isotropic material properties like those of H1 , our model ( Figure 4; Appendix 5 ) would predict that the DV/AP opening ratio in HYP7 depends only on the head axisymmetric shape and is equal to that of H1 , and would thus contradicts our observations . Hence , this suggests that the HYP7 cell has anisotropic cortical material properties . To model the DV epidermal cell deformation , we examined two classes of anisotropic stiffness materials that have been described previously: orthotropic materials such as bones ( Miller et al . , 2002; Helwig et al . , 2009 ) , and fiber-reinforced materials such as arteries ( Gasser et al . , 2006 ) , articular cartilage ( Federico and Gasser , 2010 ) or fibrous connective tissues ( Ben Amar et al . , 2015 ) . Orthotropic materials have different stiffnesses along orthogonal directions , and thus respond differently to the same stress magnitude along these directions . Fiber-reinforced materials also have different stiffnesses in the directions along and transverse to the fibers; in addition , such materials can respond differently to extensive or compressive stress ( Bert , 1977 ) . To define which model best applies to the DV epidermis , we used continuum linear elastic analysis ( Muskkhelishvili , 1975; Suo , 1990; Theocaris , 1986; Yoffe , 1951 ) ( Appendix 7 ) to interpret the laser-cutting data from the DV epidermis . We discarded the orthotropic model , as it did not adequately describe our data ( Appendix 8 ) , and focused on the fiber-reinforced plane model , which better accounts for the presence of well aligned actin fibers in DV cells . In a fiber-reinforced material that is composed of a matrix superimposed with fibers , the contribution of the fibers to the stiffness of the material depends on their orientation . Along the direction parallel to the fibers , the Young modulus of the composite is much increased due to fiber reinforcement , whereas along the direction perpendicular to the fibers , the contribution of the fibers to the composite stiffness is small . According to our modelling , the Young modulus along the fiber direction increases linearly with a factor K related to the fiber stiffness and density; whereas the stiffness along the direction transverse to the fibers varies as a hyperbolic function of K and reaches a plateau ( Appendix 7 ) . For fiber-reinforcement in the DV direction , the change in Young modulus along the DV and AP directions predicted by modelling is given in Figure 6a , b . Cuts perpendicular to the fibers opened similarly to an isotropic material with the matrix Young modulus , because they locally destroyed the fibers ( Figure 6c; see Equation ( 1 ) above ) . By contrast , cuts along the fibers opened with an equilibrium value that depends on the fiber stiffness and distribution through the factor K defined above ( Figure 6d , Appendix 7 ) . 10 . 7554/eLife . 23866 . 015Figure 6 . Model of cut opening for a fiber-reinforced material . ( a ) Considering a composite material with fiber reinforcement along the DV direction , the ratio of the Young modulus of the composite material along the DV direction , EDV , to the Young modulus of the material without fibers ( matrix ) , E0 , depends linearly on a factor K ( see Appendix 7 ) ; K is related to fiber density and stiffness . ( b ) The ratio of the Young modulus along the AP direction , EAP , to the matrix Young modulus , E0 , varies little with K . ( c ) The opening of the cuts perpendicular to the fibers is similar to an isotropic plane response , and depends on the ratio of DV stress to the matrix Young modulus σDV/E0 . ( d ) The opening of the cuts parallel to the fibers depends on the factor K , the ratio of AP/DV stiffness , EAP/EDV , and the ratio of stress over Young modulus in the AP direction , σAP/EAP . DOI: http://dx . doi . org/10 . 7554/eLife . 23866 . 015 Since the H1 seam and the head HYP7 cells are adjacent along the circumference ( Figure 3d ) , they should be under the same DV stress due to tension continuity across cell-cell junctions . According to Equation ( 1 ) , if the stress in two cells is the same , their opening should vary inversely with their respective Young moduli . As the DV opening of HYP7 was about 1 . 5 times smaller than that of H1 ( Figure 7a ) , we infer that the Young modulus of the HYP7 matrix without fibers was about 1 . 5 times stiffer than that of H1 ( Appendix 9 ) , suggesting that these cells have distinct material properties . Comparing the DV and AP opening for the HYP7 cell , we found that the factor K increased during early elongation ( Figure 7b; Appendix 10 ) . More importantly , the calculated ratio of DV/AP Young moduli also increased , and was greater than the DV/AP stress anisotropy ( Figure 7c ) . 10 . 7554/eLife . 23866 . 016Figure 7 . The anisotropy of stiffness in the HYP7 cell helps the embryo to elongate . ( a ) The cut opening in HYP7 is linearly related to the cut opening in H1 . The slope of the linear regression gives the ratio of the HYP7 matrix without fibers to H1 Young moduli . ( b ) The factor K increases during elongation from the 1 . 3F to 1 . 7F stages . ( c ) The ratio of DV/AP stiffness increases and is greater than the AS during early elongation . The data were derived from Figure 3c and Appendix 10 . ( d ) Hooke’s law written for a fiber-reinforced material such as DV cells ( Appendix 6B ) . ϵAP and ϵDV are the relative length change along the AP and DV directions , respectively . The stresses σAP and σDV along the AP and DV directions , respectively are supposed to be tensile ( so positive ) . EAP and EDV are the Young moduli in the DV cells along the AP and DV direction , respectively . ν1 and ν2 are Poisson’s ratios in DV cells; ω is the EDV/EAP ratio . A , anterior; P , posterior; D , dorsal; V , ventral . ( e ) Dependence of the AP relative length change ϵAP on the ratio of DV/AP stiffness ω for different values of AS and σAP/EAP; the ν1 value is taken to be 1 . ( f ) Dependence of the DV relative length change ϵDV on the ratio of DV/AP stiffness ω for different values of AS and σDV/EDV; the ν2 value is taken to be 1 . For ( a–c ) , the average ( or computed ) values and standard errors are reported . DOI: http://dx . doi . org/10 . 7554/eLife . 23866 . 016 To understand how a change in stiffness affects head HYP7 deformation , we again applied Hooke’s law to these cells ( Appendix 6B; Figure 7d ) . As myosin II activity in DV cells is low , their cortex should be exposed to tensile stress induced by actomyosin contractility in the seam cells . The cell length along the AP direction increased when the stiffness anisotropy ( DV/AP stiffness ratio ) increased ( Figure 7d , e ) , whereas the trend was opposite in the DV direction ( Figure 7d , f ) . Thus , the stiffness anisotropy helps the HYP7 cell to extend along the AP direction and to shrink along the DV direction . Interestingly , the equations predict that increasing stress anisotropy has an opposite effect on HYP7 cell deformation , as it prevents these cells from extending antero-posteriorly ( Figure 7e ) . Altogether , our model strongly suggests that when the DV/AP stiffness anisotropy increases and is greater than the DV/AP stress anisotropy , as observed in the head HYP7 ( Figure 7c ) , elongation along the AP direction is favored . Furthermore , our data demonstrate that the distinct mechanical properties of the cells composing a complex tissue enables its morphogenesis , which does not require all cells to be contractile . Myosin II is not polarized ( Figure 1e , f ) , but to find out whether actin distribution accounts for the stress and stiffness anisotropies , we carried out an analysis of actin filament alignment in the seam cells H1 and V3 , as well as in the head HYP7 cell . We found that the polarization of actin filaments in seam cells correlated with the observed pattern of stress anisotropy . Indeed , in H1 at the 1 . 3F stage , actin filaments had a nearly isotropic angular distribution correlating with the isotropic stress ( Figure 8a , Figure 3b ) , whereas they became increasingly aligned along the DV direction from the 1 . 3F to the 1 . 7F stage ( Figure 8a ) , mirroring the increasing stress anisotropy from 1 . 3F to 1 . 7F ( Figure 3b ) . Likewise , actin alignment decreased along the DV direction in V3 from the 1 . 3F to the 1 . 5F stage , in parallel to the decrease of stress anisotropy between those stages ( Figures 8b and 3b ) . The changes in H1 ( Figure 8c ) , but not in V3 ( Figure 8d ) , were statistically significant . In the HYP7 cell , actin filaments already acquired a preferential DV alignment at the 1 . 3F stage ( Figure 8c ) , but became increasingly organized along the DV direction as the embryo elongated to the 2F stage , with a highly significant difference between the 1 . 5F and 1 . 7F stages ( Figure 8c and f ) . These changes correlated with the increased stiffness anisotropy observed in the HYP7 cell ( Figure 7c ) . 10 . 7554/eLife . 23866 . 017Figure 8 . Actin filament organization correlated with stress and stiffness anisotropy pattern . ( a–c ) Angle distribution of actin filaments in the seam cell H1 ( a ) , seam V3 ( b ) and the HYP7 cell ( c ) , at different elongation stages . D , dorsal; V , ventral; A , anterior; P , posterior . 90° correspond to DV direction . ( d , e , f ) Comparison of the peak values at 90° ± 8° ( DV direction ) of angular distribution showed in ( a , b , c ) , respectively . p-values of two-tailed t-tests are reported . ( g ) ( Left ) The anisotropy of mechanical stress generated by the polarized actomyosin network and medial myosin pulses promote Drosophila germband extension . ( Right ) The interplay of stress anisotropy ( generated in seam cells - red ) and stiffness anisotropy ( DV cells – white ) promote C . elegans embryo elongation . Note that , although myosin II does not display a polarized distribution within individual C . elegans epidermal cells as it does in Drosophila germband epithelial cells , its enrichment in seam cells along the circumference is reminiscent of the localized myosin II enrichment at vertical junctions in Drosophila . A , anterior; P , posterior . For ( a–f ) , the average values and standard errors are reported . DOI: http://dx . doi . org/10 . 7554/eLife . 23866 . 017 We have attempted to test functionally how actin organization could affect stress and stiffness by manipulating actin polymerization through two different strategies to express cofilin during early elongation . However , we could not obtain meaningful results . Altogether , we conclude that the pattern of actin distribution showed a good correlation with the observed stress and stiffness anisotropy . It will remain important to define the mechanisms that mediate changes in actin distribution , and ultimately to determine whether this distribution is a cause or a consequence of anisotropy . Classical experiments in embryology have outlined how the juxtaposition of cells that have different properties is crucial in powering important morphogenetic movements , such as Xenopus gastrulation ( Hardin and Keller , 1988; Keller and Winklbauer , 1992 ) . In this work , we have dissected the mechanical contributions of the different epidermal cells driving C . elegans embryonic morphogenesis at single-cell resolution , highlighting the importance of juxtaposition of cells with different properties . Combining laser nano-ablation and continuum mechanics modelling , we first highlight the importance of stress anisotropy in the seam cells . Second , we emphasize that stiffness anisotropy is equally important for embryonic elongation but matters in another epidermal cell type , the DV cells . Thereby , we reveal the critical role of tissue material properties in morphogenesis . Many studies analysing morphogenetic processes have focused on 2D epithelial sheets such as the Drosophila mesoderm ( Martin et al . , 2009 ) , germband ( Rauzi et al . , 2008 , 2010; Blankenship et al . , 2006; Fernandez-Gonzalez and Zallen , 2011 ) , amnioserosa ( Solon et al . , 2009; Gorfinkiel et al . , 2009 ) , wing and thorax ( Aigouy et al . , 2010; Bosveld et al . , 2012 ) , or the zebrafish enveloping cell layer ( Behrndt et al . , 2012 ) during embryonic development . They have revealed the role of contractile actomyosin pulses and planar polarity in coordinating events over long distances . The C . elegans embryonic elongation is distinct from those situations because it does not involve myosin-II-polarized distribution nor actomyosin pulses . Interestingly , this process still requires stress anisotropy , outlining that stress anisotropy can be produced by different means . We suggest that several factors contribute to establish stress anisotropy in C . elegans . First , the actin network displayed a more polarized dorso-ventral distribution in seam cells when the stress anisotropy was higher , which should increase the stress in that direction . Second , akin to a planar polarized distribution , myosin II activity displays an asymmetric distribution along the embryo circumference in cells with different material properties ( Figure 8g ) . Intriguingly , tissue culture cells can sense the spatial stiffness distribution ( Walcott and Sun , 2010; Fouchard et al . , 2011; Trichet et al . , 2012; Lange and Fabry , 2013 ) , raising the possibility that seam cells sense the higher circumferential stiffness and respond with higher DV-oriented stress through the mechanosensitive adherens junctions ( le Duc et al . , 2010; Yonemura et al . , 2010 ) . Third , we found that the spectrin cytoskeleton has a significant role in establishing normal levels of stress magnitude and anisotropy . Spectrin is known to impinge on actin filament alignment and continuity in DV cells ( Praitis et al . , 2005; Norman and Moerman , 2002 ) and could thus affect DV stiffness anisotropy by reducing the level of actin fiber alignment . Finally , although myosin II activity is low in DV cells ( Diogon et al . , 2007 ) , the remaining activity might create some DV-oriented stress feeding back on seam cells . By modelling the DV cells as a fiber-reinforced material , we reveal how the polarized cytoskeleton in DV cells increases their stiffness to orient the extension in the AP direction , acting like a ‘molecular corset’ . Related ‘molecular corsets’ have been described and proposed to drive axis elongation in other systems ( Wainwright , 1988 ) . In Drosophila , a network of extracellular matrix fibrils was proposed to help elongate developing eggs ( Haigo and Bilder , 2011 ) . In plant cells , the orientation of cellulose microfibrils determines the axis of maximal expansion . In the latter , stiffness anisotropy also helps overcome stress anisotropy ( Green , 1962; Baskin , 2005 ) . Importantly , C . elegans embryos reduce their circumference during elongation , whereas Drosophila eggs and plants increase it . This suggests that to conserve the actin reinforcement properties when the diameter decreases , C . elegans DV epidermal cells should have a mechanism to actively shorten the actin bundles , as observed in a biomimetic in vitro system ( Murrell and Gardel , 2012 ) . Our experimental data were consistent with the predictions from Hooke’s law . They prove that the actomyosin cortex preferentially squeezes the embryo circumferentially , and that the stress anisotropy is tightly linked to the geometry of the embryo . By quantitatively assessing the contribution of stiffness anisotropy in tissue elongation , we have emphasized its importance relative to the more established role of stress anisotropy . The precise relationship between both anisotropies remains to be investigated . Thus , the juxtaposition of cells with different ‘physical phenotypes’ , seam epidermis expressing stress anisotropy and DV epidermal cell showing stiffness anisotropy , powers C . elegans elongation , as previously suggested in chicken limb bud outgrowth ( Damon et al . , 2008 ) or chick intestinal looping ( Savin et al . , 2011 ) . We did not mention other potential stress-bearing components , such as microtubules and the embryonic sheath ( Priess and Hirsh , 1986 ) , as the former mainly serves to enable protein transport ( Quintin et al . , 2016 ) whereas the function of the latter will be the focus of an upcoming work . In conclusion , our work shows that tissue elongation relies on two fundamental physical quantities ( mechanical stress and tissue stiffness ) , and provides the most advanced mesoscopic understanding to date of the mechanics at work during the first steps of C . elegans embryonic elongation . Bristol N2 was used as the wild-type ( WT ) strain and animals were maintained as described in Brenner ( 1974 ) . The strain ML1540: mcIs50[lin-26p::vab-10 ( abd ) ::gfp; myo-2p::gfp] LGI carrying the actin-binding domain ( ABD ) of the protein VAB-10 under the epidermal promoter lin-26 has been described elsewhere ( Gally et al . , 2009 ) . The endogenous NMY-1::GFP reporter strain was built by CRISPR knock-in ( ML2540: nmy-1 ( mc82 ) [nmy-1::gfp] LGX; the NMY-2::GFP reporter strain LP162 nmy-2 ( cp13 ) [nmy-2::gfp + LoxP] LGI was a generous gift from Daniel Dickinson . For parallel calcium and actin imaging during ablation , we used the strain ML2142: mcIs43 [lin26p:: vab-10::mCherry; myo-2p::gfp]; juIs307[dpy-7p::GCaMP3] carrying a calcium sensor under the epidermal promoter dpy-7p and mCherry-labeled VAB-10 ( ABD ) under the lin-26 promoter . The thermosensitive Rho kinase mutation let-502 ( sb118ts ) was crossed with ML1540 to give the strain ML2216: let-502 ( sb118ts ) ; mcIs50[lin-26p::vab-10 ( abd ) ::gfp; myo-2p::gfp] LGI . For determining morphological changes , we used the strain ML2386: mcIs50[lin-26p::vab-10 ( abd ) ::gfp; myo-2p::gfp] I; xnIs97[hmr-1::gfp] III , which expresses both a junctional marker ( HMR-1/E-cadherin ) and an actin marker ( VAB-10 ( ABD ) ) . For actin alignment analysis , we used the ML1966 unc-119 ( ed3 ) mcIs67 [dpy7p::LifeAct::GFP; unc-119 ( + ) ] strain , expressing the actin reporter LIFEACT under the dpy-7 epidermal promoter . RNAi experiments were done using injection of double-stranded RNA synthesized from PCR-amplified genomic fragments using a T3 or T7 mMESSAGE mMACHINE Kit ( Ambion , Austin , TX , USA ) . The embryos were analyzed from 24 hr to 48 hr post-injection . Freshly laid embryos or embryos from dissected hermaphrodites were mounted on 5% agarose pads in M9 buffer and the coverslip was sealed with paraffin oil . DIC time-lapse movies were recorded at 20°C using a Leica DM6000 upright microscope with a 40X oil immersion objective . For each embryo , a Z-stack of 7–8 focal planes with 4 μm step size was acquired . The length of embryos was estimated by tracing the embryo body axis ( through the middle of the embryo ) . Fluorescence time-lapse movies were recorded at 20°C using a spinning-disk Zeiss microscope Axio Observer . Z1 using a 63X oil immersion objective . Other fluorescence images were acquired with the same microscope using a 100X oil immersion objective . To determine the morphological changes in the embryo , sections of the embryo imaged with junctional and actin markers at the level of H1 , V3 and V6 were reconstructed to determine the radius , seam and DV cell width along the circumferential direction . All images were analyzed using the ImageJ ( FiJi ) software ( NIH , Bethesda , Maryland , USA; http://rsb . info . nih . gov/ij/ ) and MATLAB R2014b ( The MathWorks Inc . , Natick , MA ) . Z-stack images of LIFEACT::GFP fluorescence expression in the epidermis were acquired using a confocal Leica SP5 microscope with a 63X oil immersion objective and zoom factor 8 . We used a step size of 0 . 08 μm , a pinhole opening of 0 . 6 Airy Unit and projected 2 μm around the actin cortex . The embryos were rotated on the scan field to have the same antero-posterior orientation . The acquired images were deconvoluted using the Huygens Essential software from Scientific Volume Imaging ( Hilversum , Netherlands ) . We chose a region of interest ( ROI ) of 4 × 4 μm2 within the seam cell H1 or dorso-ventral epidermal cell HYP7 , and of 3 × 3 μm2 within the seam cell V3 to perform Fast Fourier Transform ( FFT ) . We used a high-pass filter to remove the low frequencies then did inverse FFT . We found that the high pass filter removed changes in intensity due to unequal labelling or out of focus signals but retained the actin texture . Finally , we used an ImageJ plugin , ‘Spectral Texture Analysis’ , written by Julien Pontabry to derive the angle distribution of actin texture . This plugin performed the Fast Fourier Transform ( FFT ) of the given ROI and computing coefficients in Fourier space , such as the angle distribution of the given structure , as detailed in Gonzalez and Woods ( 2008 ) . Digital image processing , Nueva Jersey , chapter 11 , section 3 . 3 ( Gonzalez , 2008 ) . For ablations , we compared embryos of the same developmental timing . To do so , we recorded the elongation curve of different genetics background ( Figure 3—figure supplement 1 ) and took embryos at the corresponding developmental time from the beginning of elongation . Thus , unc-112 ( RNAi ) embryos elongating up to 1 . 7F similarly to WT have the same length as WT at 1 . 7F stage . By contrast , spc-1 ( RNAi ) embryos elongated slower than WT ( Figure 3—figure supplement 1 ) and thus at a time corresponding to 1 . 7F in a control embryos were shorter than wild-type embryos at 1 . 7F stage . For comparison between let-502 ( sb118ts ) embryos , measurements were carried out at 25 . 5°C and the embryos were taken when muscles started to twitch ( at around 1 . 5F in control embryos ) . Laser ablation was performed using a Leica TCS SP8 Confocal Laser Scanning microscope , with a femtosecond near-infrared Coherent Chameleon Vision II , Ti:Sapphire 680–1080 nm laser , 80 MHz . To make a line cut , a region of interest with a length varying from 3 μm to 6 μm and a width of 0 . 08 μm ( 1 pixel width ) was drawn . We used a laser wavelength varying from 800 nm to 900 nm , which gave consistent ablation responses . The laser power was tuned before each imaging section to obtain local disruption of the cortex response ( >80% of the cases , visible opening , no actin accumulation around cell borders in the repair process and the ablated embryos developed normally ) . Typically , the power of the laser was 2000 mW , and we used 50% power at 100% gain . Wounding response ( actin accumulation around cell borders in the repair process , embryo died afterwards ) was rarely observed at the power used for local disruption , but more often when the power was increased to 60–65% . The first time point was recorded 1 . 44 s after cutting , which corresponded to the time needed to reset the microscope from a two-photon to a regular imaging configuration . The image scanning time recorded by the software was usually less than 400 ms , so the total exposure time of the chosen ROI to multiphoton laser was less than 1 ms . The cuts were oriented either in the antero-posterior ( AP ) or dorso-ventral ( DV ) directions relative to the global orientation of the embryos . After ablation , the embryos were monitored to see whether they continued to develop normally or whether they expressed the desired phenotype . More precisely , we verified whether embryos ablated at 1 . 3F and 1 . 5F developed past the 2F stage , whether embryos ablated at 1 . 7F developed past the 2 . 5F stage , and whether unc-112 ( RNAi ) embryos and spc-1 ( RNAi ) embryos arrested at 2F stage . The shape of the cut opening was detected using the Active Contour plugin ABsnake ( Boudier , 1997 ) . A starting ROI was drawn around the opening as the initiation ROI for ABsnake . After running the plugin , the results were checked and corrected for detection errors . The detected shape was fitted with an ellipse to derive the minor axis , major axis and the angle formed by the major axis with the initial cut direction . The average opening of the five last time points before the repair process began ( Figure 2—figure supplement 1 , from around 8 to 10 s after cutting ) was taken as the opening at equilibrium . The standard error of the mean is reported . The curve fit was performed on the average value of the cut opening ( defined as the minor axis/initial cut length ) using GraphPad Prism 5 . 00 ( San Diego , California , USA ) and the equation of one-phase association:y=y0+ ( Plateau−y0 ) ( 1−e−γt ) where y0 is the initial width of the cut opening , Plateau is the minor axis of the opening at equilibrium and γ is the relaxation rate . The standard error of the mean given by the software is reported . The two-tailed t-test was performed on the average of the last five time points ( from about 8 s to 10 s ) of the cut opening using MATLAB R2014b ( The MathWorks Inc . , Natick , MA ) . Z-test was performed using QuickCalcs of GraphPad Prism ( San Diego , California , USA ) to compare the anisotropy of stress ( AS ) , the relaxation half-time and the initial recoil speed of the cut opening .	Animals come in all shapes and size , from ants to elephants . In all cases , the tissues and organs in the animal’s body acquire their shape as the animal develops . Cells in developing tissues squeeze themselves or push and pull on one another , and the resulting forces generate the final shape . This process is called morphogenesis and it is often studied in a worm called Caenorhabditis elegans . This worm’s simplicity makes it easy to work with in the laboratory . Yet processes that occur in C . elegans also take place in other animals , including humans , and so the discoveries made using this worm can have far-reaching implications . As they develop , the embryos of C . elegans transform from a bean-shaped cluster of cells into the characteristic long shape of a worm , with the head at one end and the tail at the other . The force required to power this elongation is provided by the outer layer of cells of the embryo , known as the epidermis . In these cells , motor-like proteins called myosins pull against a mesh-like scaffold within the cell called the actin cytoskeleton; this pulling is thought to squeeze the embryo all around and cause it to grow longer . Six strips of cells , running from the head to the tail , make up the epidermis of a C . elegans embryo . Myosin is mostly active in two strips of cells that run along the two sides of the embryo . In the strips above and below these strips ( in other words , those on the upper and lower sides of the worm ) , the myosins are much less active . However , it is not fully understood how this distribution of myosin causes worms to elongate only along the head-to-tail axis . Vuong-Brender et al . have now mapped the forces exerted in the cells of the worm’s epidermis . The experiments show that , in the strips of cells on the sides of the embryo , myosin’s activity causes the epidermis to constrict around the embryo , akin to a boa constrictor tightening around its prey . At the same time , the actin filaments in the other strips form rigid bundles oriented along the circumference that stiffen the cells in these strips . This prevents the constriction from causing the embryo to inflate at the top and bottom strips . As such , the only direction the embryo can expand is along the axis that runs from its head to its tail . Together , these findings suggest that a combination of oriented force and stiffness ensure that the embryo only elongates along the head-to-tail axis . The next step is to understand how this orientation and the coordination between cells are controlled at the molecular level .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2017	The interplay of stiffness and force anisotropies drives embryo elongation
Mouse CA1 pyramidal neurons express apamin-sensitive SK2-containing channels in the post-synaptic membrane , positioned close to NMDA-type ( N-methyl-D-aspartate ) glutamate receptors . Activated by synaptically evoked NMDAR-dependent Ca2+ influx , the synaptic SK2-containing channels modulate excitatory post-synaptic responses and the induction of synaptic plasticity . In addition , their activity- and protein kinase A-dependent trafficking contributes to expression of long-term potentiation ( LTP ) . We have identified a novel synaptic scaffold , MPP2 ( membrane palmitoylated protein 2; p55 ) , a member of the membrane-associated guanylate kinase ( MAGUK ) family that interacts with SK2-containing channels . MPP2 and SK2 co-immunopurified from mouse brain , and co-immunoprecipitated when they were co-expressed in HEK293 cells . MPP2 is highly expressed in the post-synaptic density of dendritic spines on CA1 pyramidal neurons . Knocking down MPP2 expression selectively abolished the SK2-containing channel contribution to synaptic responses and decreased LTP . Thus , MPP2 is a novel synaptic scaffold that is required for proper synaptic localization and function of SK2-containing channels . At most excitatory synapses in the central nervous system , such as the Schaffer collateral to CA1 synapses in the stratum radiatum of the hippocampus , excitatory neurotransmission is largely mediated by ionotropic AMPA-type ( α-Amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid ) and NMDA-type glutamate receptors . Yet , an emerging theme is that several conductances that limit membrane depolarization also make substantial contributions to the integrated excitatory post-synaptic potential ( EPSP ) . For example , synaptically evoked Ca2+ influx into dendritic spines activates apamin-sensitive SK2-containing channels ( small conductance Ca2+-activated K+ channels type 2; KCNN2 ) , and their outward K+ conductance shunts the AMPAR-mediated depolarization , effectively reducing the EPSP ( Ngo-Anh et al . , 2005; Faber et al . , 2005 ) . Kv4 . 2-containing channels are expressed in spines , close to , but not in the PSD ( Kim et al . , 2007 ) . Synaptic activity evokes Ca2+ influx through R-type voltage-gated Ca2+ channels in spines that boosts nearby Kv4 . 2-containing A-type K+ channels to further decrease the AMPA-mediated depolarization ( Wang et al . , 2014 ) . In addition , Ca2+-activated Cl- channels are expressed in the spines and provide further inhibitory contributions ( Huang et al . , 2012 ) . Indeed , the sum of these repolarizing conductances may reduce the depolarizing AMPA-NMDA component by more than 50% . It is likely that each of these components can be regulated by a variety of second messenger pathways , greatly expanding the repertoire of targets to fine-tune synaptic transmission . For example , the Ca2+ sensitivity of SK2 channels is regulated in an activity-dependent manner by co-assembled protein kinase CK2 and protein phosphatase 2A ( Bildl et al . , 2004; Allen et al . , 2007 ) that are engaged by cholinergic signaling ( Giessel and Sabatini , 2010 ) . Moreover , the various contributions to synaptic responses may be dynamic , changing in response to distinct patterns of activity . Synaptic SK2-containing channels undergo protein kinase A ( PKA ) -dependent endocytosis upon the induction of LTP by theta burst pairing . The endocytosis of synaptic SK2-containing channels acts together with the PKA-dependent exocytosis of additional GluA1-containing AMPARs to mediate the expression of LTP ( Lin et al . , 2008 ) . Moreover , after the initial expression of LTP and loss of the SK2-containing channel contribution , homeostatic mechanisms act to re-establish the synaptic SK balance ( Lin et al . , 2010 ) . Similarly , Kv4 . 2-containing channels expressed in spines undergo PKA-dependent endocytosis after the induction of LTP ( Kim et al . , 2007; Hammond et al . , 2008 ) . Therefore , the appropriate localization , spatial distribution , and orchestrated dynamics of these protein complexes provide a powerful regulator of excitatory neurotransmission and plasticity . One class of proteins that plays a major role in synaptic organization and dynamics are the MAGUKs ( Elias and Nicoll , 2007 ) , of which there are 10 subfamilies . These modular , usually multivalent scaffolds bind to synaptic receptors , channels , and signaling molecules to anchor them into their proper locations within the post-synaptic membrane ( Oliva et al . , 2012 ) , creating a spatially and temporally restricted signaling domain ( Hammond et al . , 2008; Colledge et al . , 2000; Dell’Acqua et al . , 2006 ) . Thus , within the post-synaptic density of excitatory synapses PSD-95 binds to NMDARs ( Cousins and Stephenson , 2012 ) , while SAP97 binds to AMPARs ( Howard et al . , 2010; Leonard et al . , 1998 ) , and Shank and Homer may serve as modular organizers of the lattice of synaptic MAGUKs ( Sheng and Kim , 2000; Hayashi et al . , 2009 ) . However , the molecular mechanisms that engender synaptic localization and dynamics to SK2-containing channels are not well understood . There are two major isoforms of SK2 that are expressed in CA1 pyramidal neurons; SK2-L ( long ) has an extended intracellular N-terminal domain compared to SK2-S ( short ) and the two isoforms co-assemble into heteromeric channels ( Strassmaier et al . , 2005 ) . In mice that selectively lack SK2-L expression , the SK2-S channels are expressed in the plasma membrane of dendrites and dendritic spines , yet fail to become incorporated into the post-synaptic membrane . Consequently , the SK2-containing channel contributions to EPSPs and plasticity are absent , and this loss of synaptic SK2-containing channel function enhances hippocampus-dependent learning tasks ( Allen et al . , 2011 ) . To identify proteins that might serve to localize synaptic SK2-containing channels , candidate SK2 interacting proteins were identified . One of them , the MAGUK protein MPP2 ( membrane palmitoylated protein 2 ) , is localized to the PSD and is essential for synaptic SK2-containing channel function . Proteomic analyses ( on high-resolution quantitative mass spectrometry ) of SK2-containing channels immunoaffinity-purified from rodent whole brain membrane preparations suggested that the MAGUK protein , MPP2 might be an interaction partner . To further investigate this interaction , two newly generated antibodies targeting MPP2 were tested . Probing Western blots of proteins prepared from total mouse brain with either MPP2 antibody detected a predominant band at ~55 kDa . Similarly a single band , ~60 kDa , was detected in proteins prepared from HEK293 cells expressing FLAG-tagged MPP2 , but not from mock transfected cells . For both brain and HEK293 , pre-incubating the antibodies with the respective immunizing antigen abolished the bands ( Figure 1A , B ) . 10 . 7554/eLife . 12637 . 003Figure 1 . MPP2 interacts with SK2 . ( A , B ) Western blots were prepared using mouse brain homogenate ( Brain; 100 μg ) or HEK293 cell extracts ( 1% of 10 -cm plate ) , either transfected to express FLAG-MPP2 ( HEK ) or empty plasmid ( HEK mock ) . Duplicate lanes were prepared and one set was probed with the indicated MPP2 antibody ( left panels ) while the second set was probed with the same antibody after pre-absorbing with the immunizing antigen ( right panels ) . Native MPP2 and transfected FLAG-MPP2 were detected only in the left panels . ( C ) Bar graphs illustrating abundance ratios determined for the indicated proteins in APs with two anti-MPP2 antibodies and IgG as a negative control . Horizontal lines denote threshold for specificity of co-purification . ( D ) Western blots of proteins prepared from HEK293 cells transfected with C8-tagged SAP97 , Lin7A , Lin7C , and MPP2 , and probed with anti-C8 antibody ( left ) , anti-MPP2a antibody ( middle ) , and anti-MPP2b antibody ( right ) . The MPP2 antibodies recognized only MPP2 . DOI: http://dx . doi . org/10 . 7554/eLife . 12637 . 00310 . 7554/eLife . 12637 . 004Figure 1—figure supplement 1 . Coverage of the primary sequences of all proteins shown in Figure 1C . Peptides identified by mass spectrometry are in red; those theoretically accessible but not detected in MS-analyses are in black , and peptides not accessible to MS-analyses under the settings used ( peptides with mass values below 738 Da ( [absolute lower mass cutoff ) ] or above 3000 Da ( [practical mass limit of the C18 RP-HPLC separation ) ] are given in grey ) . The sum of amino acids in red is either related to the accessible primary sequence ( relative coverage ) or to the entire primary sequence ( absolute coverage ) ; transmembrane domains are underlined . DOI: http://dx . doi . org/10 . 7554/eLife . 12637 . 00410 . 7554/eLife . 12637 . 005Figure 1—figure supplement 2 . Coverage of the primary sequences of all proteins shown in Figure 1C ( continued ) . See caption above . DOI: http://dx . doi . org/10 . 7554/eLife . 12637 . 005 Therefore , these two antibodies were used for affinity purifications combined with quantitative mass spectrometry ( see Materials and methods ) . The results demonstrated that both antibodies robustly purified MPP2 and co-purified SK2 as well as SK3 . In addition , these experiments identified DLG1 ( SAP97 ) and Lin-homologs 7A and 7C as proteins co-assembling with MPP2 in rodent brain ( Figure 1C; Supplementary file 1; Figure 1—figure supplement 1; Figure 1—figure supplement 2 ) . To rule out off-target effects of the two MPP2 antibodies , we tested their binding of C8-tagged of SAP97 , Lin7A , or Lin7C , C8-tagged versions of each of these proteins after expression in HEK293 cells ( Figure 1D ) . While all 8-tagged proteins were recognized by anti-C8 antibody only C8-MPP2 was recognized by anti-MPP2 antibodies ( Figure 1D ) . MPP2 , a member of the p55 Stardust family of MAGUK proteins , is predicted to be a 552 amino acid protein . Like most MAGUKs , MPP2 is a modular protein comprised of several distinct protein-protein interaction domains . There are two predicted L27 domains , followed by a single predicted PDZ domain , and then the SH3-HOOK-GK domains . To test for direct interaction with SK2 channels , SK2-L and SK2-S , the SK2 isoforms that contribute to synaptic SK2-containing channels ( Allen et al . , 2011 ) , and either C8-MPP2 or PSD-95 were co-expressed in HEK293 cells . Immunoprecipitations were performed using an anti-SK2 antibody raised in guinea pig that is directed to the common C-terminal domain of the two SK2 isoforms , or using IgG as a control . Precipitated proteins were prepared as Western blots . Probing with anti-SK2 antibody that was raised in rabbits and directed against the same C-terminal sequence , demonstrated equivalent SK2 expression for input and after immunoprecipitation in each sample ( Figure 2—figure supplement 1 ) . Probing with anti-C8 antibody detected a band of the appropriate apparent molecular weight for C8-MPP2 in the sample co-expressed with SK2-S plus SK2-L but not in the IgG control sample ( Figure 2A ) . Probing with anti-PSD-95 antibody did not detect co-immunoprecipitation with SK2 ( Figure 2B ) . The results demonstrated that C8-MPP2 but not PSD-95 was specifically co-precipitated with SK2 . To examine the possibility that SAP97 interacts with SK2-S , C8-SAP97 was co-expressed in HEK293 cells together with either myc-SK2-S or GluA1 , a known SAP97 interaction partner ( Leonard et al . , 1998 ) . Immunoprecipitations were performed using anti-SK2 antibody , anti-GluA1 antibody , or IgG as a control . Western blotting with anti-C8 antibody revealed that C8-SAP97 only co-immunoprecipitated when co-expressed with GluA1 , despite equivalent levels of input and immunoprecipitated proteins ( Figure 2—figure supplement 1 ) . To test if MPP2 interacts with the unique N-terminal domain of SK2-L that is required for synaptic localization GST pull-down experiments were performed . While PDZ interactions are important for many MAGUK-partner protein interactions , the unique N-terminal domain of SK2-L does not contain a PDZ ligand motif and using the PDZ domain from MPP2 failed to show an interaction with the N-terminal domain of SK2-L . Therefore , the SH3-HOOK-GK domain of MPP2 was employed . This domain of MPP2 as well the SH3-HOOK-GK domains from SAP97 , a MAGUK scaffold of a distinct subfamily from MPP2 ( Oliva et al . , 2012 ) , or the SH3-HOOK-GK domain from CaCNB4 , a non-canonical MAGUK protein that is a beta subunit for voltage-gated Ca2+ channels ( Van Petegem et al . , 2004 ) were prepared as GST-fusion proteins , bound to glutathione-agarose beads , and used as baits for the prey , the His-tagged N-terminal domain of SK2-L . A GST-fusion protein of the C-terminal domain of Kv1 . 4 , and a His-fusion protein of PSD-93 ( Chapsyn ) were also prepared as positive interaction controls ( Lunn et al . , 2007 ) . Coomassie staining and Western blotting with anti-GST antibody verified equivalent amounts of the baits , either as input or bound to beads ( Figure 2—figure supplement 2 ) . After exposure to the baits , bound prey proteins were eluted and prepared as Western blots . Probing with an anti-His antibody showed that the SH3-HOOK-GK domain from MPP2 , but not from SAP97 or CaCNB4 , specifically pulled down the N-terminal domain of SK2-L . As expected , the C-terminal domain of Kv1 . 4 pulled down PSD-93 ( Figure 2C ) . Taken together these results suggest that MPP2 interacts with the unique N-terminal domain of SK2-L that is required for synaptic localization ( Allen et al . , 2011 ) . 10 . 7554/eLife . 12637 . 006Figure 2 . MPP2 interacts with the N-terminal domain of SK2-L . ( A ) Co-immunoprecipitations . Western blots prepared from HEK cell lysates expressing SK2-S and SK2-L , either alone ( mock ) or together with C8-MPP2 , immunoprecipitated with either anti-SK2 antibody or IgG ( control ) , and probed for C8-MPP2 . Adjacent blot shows input of C8-MPP2 . ( B ) A similar experiment except using PSD-95 instead of C8-MPP2 . PSD-95 is expressed but does not co-IP with SK2 . ( C ) GST-pull-downs . Blot probed with anti-His antibody shows input prey proteins , His-SK2-L N-terminal domain and His-PSD-93 . After exposure to GST-baits representing SH3-HOOK-GK domains of SAP97 , CaCNB4 , or MPP2 , the His-SK2-L N-terminal domain was specifically retained by GST-MPP2 . Positive control shows interaction between GST-C-terminal domain of Kv1 . 4 and His-PSD-93 . Co-immunoprecipitations and GST-pull-downs were performed in triplicate . DOI: http://dx . doi . org/10 . 7554/eLife . 12637 . 00610 . 7554/eLife . 12637 . 007Figure 2—figure supplement 1 . Co-immunoprecipitations . ( A ) Immunoprecipitation and input of SK2-S plus SK2-L . Western blot probed with anti-SK2 antibody raised in rabbit . First three lanes show immunoprecipitates from transfected HEK293 cell lysates co-transfected with empty vector ( mock ) , C8-MPP2 , or PSD-95 ( see Figure 2 ) . Immunoprecipitations were performed using anti-SK2 antibody raised in guinea pig . Last three lanes show input material prior to immunoprecipitation . ( B ) Co-immunoprecipitation of C8-SAP-97 co-expressed in HEK293 cells with GluA1 but not with myc-SK2-S . Western blot using anti-C8 antibody detects input of C8-SAP-97 co-expressed with myc-SK2-S or GluA1 . C8-SAP-97 co-immunoprecipitated with anti-GluA1 antibody but not with anti-SK2 antibody or IgG . ( C ) Western blot using anti-myc antibody detects myc-SK2-S co-expressed with C8-SAP-97 , input and after immunprecipitation with anti-SK2 antibody . Higher MW bands correspond to aggregates of SK2-S . ( D ) Western blot using mouse monoclonal anti-GluA1 antibody detects GluA1 co-expressed with C8-SAP-97 , for GluA1 input and GluA1 immunoprecipitated with rabbit polyclonal anti-GluA1 antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 12637 . 00710 . 7554/eLife . 12637 . 008Figure 2—figure supplement 2 . GST-fusion protein expression . ( A ) Coomassie stained gel showing input bacterial lysates ( left lanes ) for the indicated GST-fusion proteins prior to being bound to glutathione agarose beads , and after binding to beads ( right lanes ) . ( B ) Western blot of the gel in panel A , probed with anti-GST antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 12637 . 008 To examine the anatomical localization of MPP2 , the anti-MPP2 antibodies were used for immunohistochemistry on hippocampal sections . At the light level , MPP2 was expressed throughout the hippocampus , and in area CA1 , MPP2 was predominantly expressed in the dendritic arbors ( Figure 3A , B ) . To determine the subcellular profile of MPP2 expression , pre- and post-embedding immuno-gold electron microscopy ( iEM ) was performed . Using pre-embedding techniques , sections from three animals revealed immunoparticles for MPP2 prominently labeled dendritic spines and dendritic shafts both along the plasma membrane and at intracellular sites . Post-synaptically , most immunoparticles for MPP2 were found in spines ( 84%; 436 particles ) versus in dendrites ( 16%; 82 immunoparticles ) , with most immunoparticles in either compartment localized to the plasma membrane ( 84%; 365 immunoparticles in spines and 65%; 54 immunoparticles in dendrites ) . Next , post-embedding techniques were applied to determine if MPP2 is expressed in the post-synaptic density . Sections from three animals revealed that immunoparticles for MPP2 prominently labeled the post-synaptic densities of dendritic spines , as well as being detected in dendritic shafts ( Figure 3C , D ) . Immunoparticles for MPP2 were not detected in stratum pyramidale , and were also absent pre-synaptically . Therefore , MPP2 is a synaptic MAGUK protein . 10 . 7554/eLife . 12637 . 009Figure 3 . Localization of MPP2 in the hippocampus . ( A , B ) Light microscopic images of anti-MPP2 antibody labelling in hippocampus . MPP2 was present throughout the hippocampus and was prominent in the dendritic arbors of area CA1 . ( C , D ) Electron micrographs of the hippocampus showing immunoparticles for MPP2 in the stratum radiatum of the CA1 region of the hippocampus , as detected using a post-embedding immunogold method . Immunoparticles for MPP2 were detected along the PSD ( arrowheads ) of dendritic spines ( s ) of CA1 pyramidal cells establishing asymmetrical synapses with axon terminal ( at ) , as well as at extrasynaptic sites ( arrows ) of dendritic spines ( s ) . Den , dendritic shaft . Scale bars in C , D: 0 . 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12637 . 009 To test whether MPP2 expression is important for synaptic SK2 channel function , two short hairpin RNAs ( shRNAs ) targeting the 3’ untranslated region ( 3’ UTR ) of Mpp2 mRNA were co-expressed in area CA1 by in utero electroporation ( e14-16 ) of a plasmid that also directed expression of the fluorescent protein , GFP . Four- to five-week-old mice were then used to prepare fresh hippocampal slices . Whole-cell current clamp recordings were made from CA1 pyramidal neurons , either transfected , as reported by GFP expression , or non-transfected control cells . Synaptic stimulations of the Schaffer collateral axons evoked EPSPs . After establishing a stable baseline , apamin ( 100 nM ) was applied to the slices . For control cells , apamin increased EPSPs to 144 . 4 ± 4 . 3% of baseline ( n = 15; p<0 . 0001 ) , consistent with previous studies ( Ngo-Anh et al . , 2005; Lin et al . , 2008; Giessel and Sabatini , 2011 ) . In contrast , apamin did not significantly affect EPSPs from transfected CA1 pyramidal neurons ( 94 . 6 ± 3 . 3%; n = 14 ) ( Figure 4A–C ) . 10 . 7554/eLife . 12637 . 010Figure 4 . MPP2 is required for synaptic SK2-containing channel function . ( A ) Time course of the normalized EPSP amplitude ( mean ± s . e . m . ) for baseline in control ACSF ( Ctrl ) and during wash-in of apamin ( 100 nM ) as indicated above in MPP2 sh-transfected cells ( open red symbols , n = 14 ) and non-fluorescent control cells ( black symbols , n = 15 ) mice . ( B ) Average of 15 EPSPs taken from indicated shaded time points in aCSF ( black ) and 16–20 min after application of apamin ( red ) ; shaded areas are mean ± s . e . m for control non-fluorescent cells ( ctrl , upper traces ) and MPP2 sh-transfected cells ( MPP2 sh , bottom traces ) . ( C ) Scatter plot of relative ESPS peak compared to baseline from the individual slices in panel A non-fluorescent control ( ctrl , black symbols ) and for MPP2 sh transfected ( red symbols ) . Horizontal bar reflects mean response . DOI: http://dx . doi . org/10 . 7554/eLife . 12637 . 010 To further verify the efficacy of MPP2 knock-down , pre-embedding iEM was performed on hippocampal sections from MPP2 shRNA-transfected animals . To identify transfected CA1 pyramidal neurons , mouse anti-GFP antibody was detected using HRP-linked anti-mouse secondary antibody , and rabbit anti-MPP2 antibody was detected using immunogold particles coupled to anti-rabbit secondary antibody . Immunoparticles were then counted in the same number of profiles belonging both to dendritic spines and dendritic shafts on single sections . In spines on neurons expressing GFP , immunoparticles for MPP2 were reduced by 81% compared to spines on non-transfected control neurons ( GFP-negative: 417 particles in 60 spines; GFP-positive: 101 particles in 60 spines; p<0 . 001 ) . A similar reduction ( 78% ) was seen in dendritic shafts ( GFP-negative: 155 particles in 25 dendrites; GFP-positive: 44 particles in 25 dendrites; p<0 . 001 ) ( Figure 5A–D ) . 10 . 7554/eLife . 12637 . 011Figure 5 . Efficient knock-down of MPP2 expression in CA1 pyramidal neurons . Immunoreactivity for MPP2 in the CA1 region of the hippocampus , as revealed using a double-labelling pre-embedding method . ( A–D ) The peroxidase reaction end product ( HRP ) indicating GFP immunoreactivity filled CA1 pyramidal cells , whereas immunoparticles for MPP2 were mainly located along the plasma membrane and at intracellular sites of pyramidal cells . Immunoparticles for MPP2 were distributed in both GFP-positive ( crossed arrows ) and GFP-negative ( arrows ) dendritic spines ( s ) and dendritic shafts ( Den ) of pyramidal cells . However , there was a striking reduction of immunoparticles for MPP2 in GFP-positive profiles compared to GFP-negative profiles ( see text ) . at , axon terminal . Scale bars in A-D: 0 . 2 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 12637 . 011 To demonstrate that the effects of shRNA transfection were specifically due to MPP2 knock-down , the shRNA ( GFP ) plasmid was co-transfected with a plasmid that directed expression of shRNA-immune MPP2 and a plasmid that expressed the red fluorophore , mApple . In doubly transfected neurons , apamin increased EPSPs similar to control , non-transfected neurons ( 148 . 2 ± 4 . 2%; n = 13; p<0 . 0001 , compared to shRNA knock-down ) ( Figure 6A , B ) . Thus , the consequences of the shRNA expression are mediated by knock-down of MPP2 . 10 . 7554/eLife . 12637 . 012Figure 6 . Co-expression of sh-immune MPP2 with MPP2 shRNA rescues synaptic SK2 function . ( A ) Time course of the normalized EPSP amplitude ( mean ± s . e . m . ) for baseline in control ACSF and during wash-in of apamin ( 100 nM ) as indicated above in cells transfected MPP2 sh and MPP2 sh immune ( n = 13 ) . Inset: representative cell showing average of 15 EPSPs taken from indicated shaded time points in ACSF ( black ) and 16–20 min after application of apamin ( red ) ; shaded areas are mean ± s . e . m . ( B ) . Scatter plot of relative ESPS peak compared to baseline from the individual slices for non-fluorescent control , MPP2 sh-transfected cells and MPP2 sh transfected with immune MPP2 ( rescue ) . Horizontal bar reflects mean . DOI: http://dx . doi . org/10 . 7554/eLife . 12637 . 012 There are at least two functionally distinct populations of apamin-sensitive SK2-containing channels in CA1 pyramidal neurons . One population resides in the spines and is activated by synaptically evoked NMDAR-dependent Ca2+ influx ( Ngo-Anh et al . , 2005 ) . Another population resides in the dendrites and may be activated by somatic voltage steps that induce Ca2+ influx through voltage-gated Ca2+ channels ( Stocker et al . , 1999; Gerlach et al . , 2004; Bond et al . , 2004 ) . To determine if MPP2 knock-down specifically affects synaptic SK2-containing channels or also affects SK2-containing channels in the dendrites transfected or control neurons were clamped at −55 mV and stepped to 20 mV; tail currents were elicited upon stepping back to −55 mV ( Figure 6A ) . The apamin-sensitive component of the tail current ( Figure 7A inset ) effectively measures the SK2-containing channel component activated by somatic depolarization and showed that MPP2 knock-down did not significantly affect somatic activation of SK2-containing channels ( ctrl: 41 . 9 ± 8 . 3 pA , n = 13; transfected: 53 . 7 ± 10 . 5 , n = 7; P = 0 . 44 ) ( Figure 7A , B ) . These results show that knocking down MPP2 expression disrupts the synaptic SK2-containing channel component but does not affect the channels in the dendrites that are activated by somatic voltage pulses . 10 . 7554/eLife . 12637 . 013Figure 7 . Dendritic SK channel function is not reduced by MPP2 knock-down . ( A ) Representative traces of voltage-clamp recordings of IAHP after a 200 ms depolarizing pulse to 20 mV in an MPP2 sh-transfected cell . Apamin ( red trace ) blocks a component of the IAHP . Inset: subtraction of the traces before and after apamin application yielded the apamin-sensitive ImAHP . ( B ) Bar graph of apamin-sensitive tail current measured at 100 ms following repolarization to −55 mV . Data presented as mean ± s . e . m . for control non-fluorescent cells ( n = 13 ) and MPP2 sh-transfected cells ( n = 7 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12637 . 013 SK channel activity modulates the induction of synaptic plasticity ( Stackman et al . , 2002 ) and the activity-dependent endocytosis of synaptic SK2-containing channels contributes to the expression of LTP ( Lin et al . , 2008 ) . To determine the consequences of MPP2 knock-down and loss of synaptic SK2-containing channel activity on LTP , a theta burst pairing protocol was delivered in which Schaffer collateral stimulations were paired with back-propagating action potentials evoked by somatic current injections . In non-transfected CA1 pyramidal neurons , this induced LTP of 407 . 9 ± 46 . 3% ( n = 8 ) , while in transfected , MPP2 knock down neurons , LTP was significantly reduced ( 243 . 0 ± 14 . 8%; n = 13; p<0 . 001 ) ( Figure 8 ) . Consistent with the loss of synaptic SK2-containing channels , MPP2 knock-down reduces LTP . 10 . 7554/eLife . 12637 . 014Figure 8 . Loss of MPP2 reduces LTP . ( A ) Time course of the normalized EPSP amplitude ( mean ± s . e . m . ) from control non-fluorescent cells ( ctrl , closed black symbols , n = 8 ) and MPP2 sh-transfected cells ( MPP2 , open red symbols , n = 13 ) . The TBP protocol was delivered at time 0 . Inset: representative cell showing average of 15 EPSPs taken from indicated shaded time points in ACSF ( black ) and 25–30 min after the induction of LTP ( red ) ; shaded areas are mean ± s . e . m . ( B ) . Scatter plot of relative ESPS peak compared to baseline from the individual slices for non-fluorescent control ( ctrl ) and MPP2 sh-transfected cells . Horizontal bar reflects mean response . DOI: http://dx . doi . org/10 . 7554/eLife . 12637 . 014 The results presented here identify the synaptic MAGUK protein , MPP2 ( p55 ) , that is required for synaptic SK2-containing channel function . Synaptically evoked EPSPs in CA1 pyramidal neurons are increased by the SK channel blocker , apamin , but in CA1 pyramidal neurons expressing shRNAs directed against Mpp2 mRNA apamin has no effect . MPP2 knockdown selectively affects synaptic SK2-containing channel function , as the SK2-containing channels expressed in the dendrites that are activated by somatic voltage steps are not altered . Consistent with the effects of MPP2 on synaptic SK2-containing channels , MPP2 knock-down reduces the expression of TBP-induced LTP by ~30% . This is slightly more than the component of LTP attributed to SK2 endocytosis in untransfected CA1 pyramidal neurons , ~17% ( Lin et al . , 2008 ) . This might reflect effects of MPP2 knock-down on other interaction partners ( see below ) . Previous results showed that synaptic SK2-containing channels are heteromeric assemblies that contain two isoforms of SK2 , SK2-S and SK2-L . Compared to SK2-S , SK2-L has an additional 207 amino acids in the intracellular N-terminal domain and SK2-S is otherwise entirely contained in SK2-L ( Strassmaier et al . , 2005 ) . In a transgenic mouse selectively lacking SK2-L , the SK2-S channels are expressed in the dendrites and even in the plasma membrane of dendritic spines , but they are specifically excluded from the PSD , and apamin fails to boost synaptically evoked EPSPs . Re-expressing SK2-L reinstates synaptic function as measured by apamin sensitivity of EPSPs ( Allen et al . , 2011 ) . These results implicated the unique N-terminal domain of SK2-L in directing SK2-containing channel synaptic localization and function , and suggested that the N-terminal domain of SK2-L might interact with a partner protein to engender PSD localization . Indeed , MPP2 binds to the unique N-terminal domain and knocking down MPP2 expression phenocopied the effect of SK2-L deletion on synaptic responses . MAGUK proteins recruit and stabilize AMPA and NMDA receptors in the PSD . Our results suggest that MPP2 similarly serves to stabilize SK2-containing channels in the PSD . MPP2 is a member of the p55 Stardust subfamily of MAGUK scaffold proteins , named after the founding member MPP1 , the major palmitoylated protein in erythrocytes ( Alloisio et al . , 1993 ) . Similar to other MAGUK scaffold proteins , MPP2 is modular , consisting of two L27 domains that may mediate homo- or heterophilic interactions , a single PDZ domain followed by SH3-HOOK-GK domains . Biochemical studies showed that MPP2 is enriched in the PSD fractions of rat brain , and pull-down assays to test for interactions suggested MPP2 may interact with itself as well as a number of other synaptic proteins , among them are PSD-95 , SAP97 , GKAP , CASK , GRIP , neuroligin , and CaMKII . The SH3-HOOK-GK domain of MPP2 was implicated in mediating these interactions ( Jing-Ping et al . , 2005 ) , similar to the interaction between MPP2 and the N-terminal domain of SK2-L . The results presented here , using an unbiased approach also identified SK3 as immunopurifying with MPP2 . SK2 and SK3 can form heteromeric channels in brain ( Strassmaier et al . , 2005 ) , and SK3 is expressed in CA1 pyramidal neurons ( Ballesteros-Merino et al . , 2014 ) . Different from SK2 that has two N-terminal isoforms , there is only one SK3 N-terminal isoform and it is similar to the extended N-terminal domain of SK2-L , harboring several islands of homology that might mediate interactions with MPP2 . The proteomics analyses of MPP2 also identified three additional scaffold proteins , the MAGUK protein , DLG1 ( SAP97 ) , as well as Lin7A and Lin7C as MPP2-interacting proteins . In epithelial cells , MPP7 , a closely related member of the p55 Stardust family , dimerizes with Lin7 proteins , an interaction mediated by the C-terminal L27 domain , and the dimeric complex then associates with DLG1 via the N-terminal L27 domain that is insufficient to mediate DLG1 interactions in the absence of bound Lin7 ( Bohl et al . , 2007 ) . DLG1 also binds the C-terminal PDZ ligand on the GluA1 subunit of AMPA receptors ( Leonard et al . , 1998 ) . Indeed , upon the induction of LTP at Schaffer collateral to CA1 synapses , additional GluA1-containing AMPA receptors undergo exocytosis at a perisynaptic site followed by translocation into the post-synaptic membrane , increasing the AMPA component of EPSPs ( Yang et al . , 2008 ) . This exocytosis is dependent on the PDZ ligand at the C-terminus of the GluA1 subunit ( Lin et al . , 2010; Yang et al . , 2008; Shi et al . , 2001 ) , and exocytosis of GluA1-containing AMPA receptors is prerequisite to SK2-containing channel endocytosis; specifically blocking GluA1-containing AMPA receptor exocytosis prevents the rapid , subsequent endocytosis of SK2-containing channels ( Lin et al . , 2010 ) . Moreover , immunopurification of AMPA receptors from whole brain identified MPP2 as a protein that co-purified with AMPA receptors ( Schwenk et al . , 2012 ) . Synaptic SK2-containing channels reside in very close proximity to synaptic NMDA receptors within the PSD , providing a molecular microdomain that facilitates their functional coupling ( Ngo-Anh et al . , 2005; Lin et al . , 2008 ) . PSD-95 interacts with NMDARs ( Kornau et al . , 1995 ) , is crucial for the proper synaptic localization of ionotropic glutamate receptors ( Schnell et al . , 2002; Elias et al . , 2008 ) , and interacts with SAP97 ( Cai et al . , 2006 ) . It will be interesting to determine whether MPP2 additionally interacts , directly or indirectly with PSD-95 to maintain the spatial synaptic relationship between SK2-containing channels and NMDA receptors . MPP2 contains multiple different protein-protein interaction domains that may bind not only receptors and channels but , additionally , signaling molecules as well as connections to the cytoskeleton . These observations raise the possibility that there is a dynamic protein lattice encompassing SK2-containing channels , AMPARs and NMDARs , and regulatory proteins that is woven together by molecular interactions between scaffold proteins to precisely tune synaptic responses during basal neurotransmission and plasticity . All procedures involving animals were performed in accordance with the guidelines of Oregon Health and Science University ( Portland , OR ) , animal care protocol number: IP00000191; University of Freiburg ( Freiburg , Germany ) , Regierungspräsidium Freiburg , AZ: 35–9185/G-12/47; and University of Castilla-La Mancha ( Albacete , Spain ) . MPP2 shRNAs were designed using an online algorithm ( http://sirna . wi . mit . edu ) . Two sequences targeting the 3’UTR of mouse Mpp2 mRNA ( NM_016695; 1811–1833; 3620–3642 ) were synthesized for shRNA expression and cloned into a vector that drove their expression from the U6 promoter . This same plasmid also directed GFP expression from the ubiquitin promoter . For rescuing MPP2 knock-down , the MPP2 shRNA plasmids ( GFP ) were co-transfected with a plasmid expressing the MPP2 coding sequence , cloned between the CAG promoter and the 3’ UTR from bovine growth hormone . This plasmid was co-transfected with a plasmid directing mApple expression from the ubiquitin promoter . To generate antibodies to MPP2 , rabbits were immunized with synthetic peptides representing amino acids 116–145 and 480–507 of mouse MPP2 ( NP_057904 ) . To test the antibodies using Western blots , mouse whole brains were homogenized in ice-cold homogenization buffer ( 0 . 32 M sucrose , 1 mM ethylenediaminetetraacetic acid , 1 mM ethyleneglycoltetraacetic acid , 10 mM Tris-HCl , pH 7 . 2 , 0 . 4 mM phenylmethylsulfonyl fluoride ) . The resultant homogenate was subjected to centrifugation at 1000 g for 10 min to remove nuclei and debris . The protein concentration was determined with Lowry’s method . Human embryonic kidney 293T ( HEK293T ) cells were transfected with pFLAG-CMV vector ( Clontech , Palo Alto , CA ) encoding mouse FLAG-MPP2 in a 10 cm dish . Cells were harvested in 0 . 5 ml of PBS . Homogenates and suspended cells were mixed with an equal volume of 2 x sodium dodecyl sulfate ( SDS ) sampling buffer ( 63 mM Tris-HCl , pH 6 . 8 , 4% SDS , 20% glycerol , 0 . 002% bromophenol blue ) , and denatured with 50 mM ( ± ) -dithiothreitol at 55°C for 30 min . Proteins ( 100 µg of brain homogenates and 0 . 5 µl of cell lysates ) were separated using 10% SDS-polyacrylamide gel electrophoresis , and electroblotted onto an Immobilon-P Transfer Membrane ( Millipore , Billerica , MA ) . After blocking with 5% skimmed milk for 30 min , blotted membranes were incubated with primary antibodies ( 1 µg/ml ) for 1 hr , then with peroxidase-conjugated secondary antibodies for 1 hr ( Jackson ImmunoResearch , West Grove , PA; 1:10 , 000 ) . Tris-buffered saline ( 10 mM Tris-HCl , pH 7 . 5 , 150 mM NaCl ) containing 0 . 1% Tween-20 was used as the dilution and washing buffer . Immunoreactions were visualized with the ECL chemiluminescence detection system , and captured using an ImageQuant LAS 500 ( GE Healthcare , Buckinghamshire , UK ) . For specificity control , anti-MPP2a and anti-MPP2b ( 1 μg/ml ) were mixed with 20 μg/ml of GST fusion proteins of the respective immunizing antigen . Anti-SK2 antibodies have been previously characterized ( Lin et al . , 2008; Allen et al . , 2011; Ballesteros-Merino et al . , 2012 ) . Mouse monoclonal anti-Myc ( 05–419 ) , mouse monoclonal anti-GST ( 05–311 ) , rabbit polyclonal anti-GluA1 ( AB1504 ) , and mouse monoclonal anti-GluA1 ( MAB2263 ) antibodies were from EMD Millipore ( Billerica , MA ) . Anti-6XHis mouse monoclonal antibody ( 37–2900 ) and anti-PSD-95 mouse monoclonal antibody ( MA1-045 ) were from Thermo Scientific ( Waltham , MA ) . Anti-GFP ( ab92 ) mouse monoclonal antibody was from Abcam ( Cambridge , MA ) . Mouse monoclonal C8 antibody was obtained as supernatant from Chessie 8 cells , LN 10300SP ( National Cell Culture Center , Biovest , Minneapolis , MN ) . Secondary antibodies were from Santa Cruz Biotechnology ( Dallas , TX ) , unless otherwise noted . Proteomic analysis of MPP2 from rat brain was performed as described previously ( Schwenk et al . , 2012; Schwenk et al . , 2010; Bildl et al . , 2012 ) . Briefly , affinity-purifications ( APs ) with anti-MPP2a , anti-MPP2b , and pre-immunization immunoglobulins ( IgG ) were performed on rat brain membrane fractions solubilized with CL-91 ( Logopharm , March , Germany ) and whole eluates were subjected to high-resolution mass spectrometry . Proteins retained in APs and identified by mass spectrometry were quantified by integration of peptide m/z signal intensities over time ( peak volumes , PVs ) that were extracted from FT full scans using MaxQuant ( v . 1 . 4 . 1 . 2; http://www . maxquant . org; with integrated off-line mass calibration ) . Relative abundance of proteins in anti-MPP2 samples versus control ( abundance ratio or rPV ) was determined by the TopCorr method ( Bildl et al . , 2012 ) as the median of either ( i ) six individual peptide PV ratios of the best correlating protein-specific peptides ( as determined by Pearson correlation of their abundance values; TopCorr6 ) , or ( ii ) the 50% best of all individual peptide PV ratios ( Top50 ) . The linear dynamic range of the TopCorr method is about 4 orders of magnitude ( detailed in [Bildl et al . , 2012] ) . The data used for rPV determination ( including peptide sequences , PV and rPV values , as well as the respective medians and selections methods ) are summarized in Supplementary file 1 . The coverage of the primary sequences of all proteins shown in Figure 1C is presented in Figure 1—figure supplement 1 and Figure 1—figure supplement 2 . HEK293 cells were transfected using Lipofectamine 2000 ( Thermo Scientific ) . After 48 hr cells were washed , collected with PBS buffer , and pelleted . Cells were then solubilized with lysis buffer ( 20 mM HEPES , pH 7 . 5 , 150 mM NaCl , 10% glycerol , 2 mM EDTA , 1 mM PMSF , and protease inhibitors ( Hoffman La Roche , Basel , Switzerland ) containing 1% β-D-dodecyl maltoside ( Sigma-Aldrich , St . Louis , MO ) for 30 min at RT . The lysate was centrifuged at 14 , 000 rpm for 20 min at 4°C . The supernatant was incubated with the indicated primary antibody for 30 min then 20 μl of protein A/G agarose beads ( Thermo Scientific ) were added for incubation overnight at 4°C with rotation . The beads were washed three times in washing buffer ( mM ) ( 20 HEPES , pH 7 . 5 , 150 NaCl , 10 KCl , 2 EDTA , 10% glycerol ) . Proteins were eluted with 40 μl of 2X SDS sample buffer at 37°C . Bound and eluted proteins were subsequently separated by SDS-PAGE and transferred to PVDF membrane ( BioRad , Hercules , CA ) . After blocking with 5% skimmed milk for 1 hr , the membrane was probed with the indicated antibodies over night at 4°C . HRP-conjugated secondary antibodies were applied for 30 min at RT . Blots were detected with SuperSignal ECL ( Thermo Scientific ) and developed with GeneMate Blue film ( BioExpress , Kaysville , UT ) . The N-terminal domain of SK2-L ( NP_001299834; 67–273 ) was cloned into pET33b for His tagged expression . Plasmids for expression of His-tagged rat PSD-93 ( P85-T280 ) and GST-Kv1 . 4-ct ( H570-V665 ) were gifts from Dr . Paul Slesinger ( Lunn et al . , 2007 ) . For GST-fusion proteins , the SH3-HOOK-GK domains of mouse MPP2 ( 216–552; NM_016695 . 3 ) , SAP97 ( 577–927; NM_007862 . 2 ) , and CaCNB4 ( 45–371; NM146123 . 2 ) were cloned into pGex4T3 . Expression levels of the GST-baits were determined by Coomassie staining and Western blotting with anti-GST antibody ( Figure 2—figure supplement 2 ) , or for His-prey , with anti-His antibody ( Figure 2 ) . GST pull-downs were performed as previously described ( Allen et al . , 2007 ) , with minor modifications . Glutathione agarose beads ( Sigma-Aldrich ) were resuspended and 50 μl of slurry was used per reaction . Beads were washed one time with pull-down wash buffer ( PWB; 20 mM HEPES , pH 7 . 8 , 10% glycerol , 100 mM KCl , 0 . 1 mM EDTA , 0 . 1 mM dithiothreitol [DTT] , 0 . 1% Igepal CA-630 [Sigma-Aldrich] ) and then incubated at 4°C for 2 hr with bacterial lysate containing the GST-fusion protein . Bead–protein complexes were washed one time with PWB followed by a 5 min wash at 4°C with PWB plus 0 . 1% BSA , and two 1 min washes with PWB . These 'baits' were incubated at 4°C overnight with bacterial cell lysate from cultures expressing the His-tagged 'prey' , washed four times with PWB , added to SDS sample buffer , heated at 95°C for 5 min and resolved by SDS-PAGE . Following transfer the Western blot was probed with anti-His antibody . Immunohistochemical reactions at the electron microscopic level were carried out using immunogold methods as described previously ( Luján et al . , 1996 ) . Timed-pregnant mice were anesthetized with isofluorane , their abdominal cavity cut open , and the uterine horns/sac exposed . Approximately 2 μl of DNA solution ( ~2 mg/ml ) was injected into the lateral ventricle of e14-e16 embryos , using a glass pipet pulled from thin walled capillary glass ( TW150F-4 , World Precision Instruments , Sarasota , FL ) and a Picospritzer III microinjection system ( Parker Hannifin , Hollix , NH ) . The head of each embryo within its uterine sac was positioned between tweezer-type electrodes ( CUY650P10; Sonidel , Dublin , Ireland ) , and 5 square electric pulses ( 35 V; 50 ms; 1-s intervals ) were passed using an electroporator ( CUY21; Sonidel ) . After electroporation , the wall and skin of the abdominal cavity of the pregnant mouse was sutured-closed , and embryos were allowed to develop normally . Hippocampal slices were prepared from 4- to 5-week-old C57BL/6J mice . Animals were anesthetized with isoflurane and decapitated . The cerebral hemispheres were quickly removed and placed into cold artificial CSF ( ACSF ) equilibrated with carbogen ( 95% O2/5% CO2 ) . Hippocampi and cortex were removed , placed onto an agar block , and transferred into a slicing chamber . Transverse hippocampal slices ( 300–350 μm ) were cut with Leica VT1200s ( Leica Biosystems , Buffalo Grove , IL ) and transferred into a holding chamber containing regular ACSF . Slices were incubated at 34°C for 30 min and then at room temperature for ≥1 hr before recordings were performed . The slicing solution consisted of sucrose-ACSF ( in mM ) : 70 sucrose , 80 NaCl , 2 . 5 KCl , 21 . 4 NaHCO3 , 1 . 25 NaH2PO4 , 0 . 5 CaCl2 , 7 MgCl2 , 1 . 3 ascorbic acid , 20 glucose and regular ACSF consisted of ( in mM ) : 125 NaCl , 2 . 5 KCl , 21 . 5 NaHCO3 , 1 . 25 NaH2PO4 , 2 . 0 CaCl2 , 1 . 0 MgCl2 , 15 glucose , equilibrated with carbogen . CA1 pyramidal cells were visualized with IR/DIC optics ( Olympus BX51W1; Olympus Scientific Solutions , Waltham , MA ) and a CCD camera . Whole-cell , patch-clamp recordings were obtained from CA1 pyramidal cells using a Multiclamp 700B ( Molecular Devices , Sunnyvale , CA ) , digitized using an ITC-18 analog-to-digital converter , and transferred to a computer using Patchmaster software ( Heka Instruments , Bellmore , NY ) . Patch pipettes ( open pipette resistance , 2–4 MΩ ) were filled with ( in mM ) 133 K-gluconate , 4 KCl , 4 NaCl , 1 MgCl2 , 10 4- ( 2-hydroxyethyl ) -1-piperazineethanesulfonic acid ( HEPES ) , 4 MgATP , 0 . 3 Na3 guanosine triphosphate ( Na3GTP ) , and 10 K2-phosphocreatine ( pH 7 . 3 ) . Electrophysiological records were filtered at 3 . 3 kHz and sampled at 10 kHz . Series resistance was electronically compensated to greater than 70% . A bias current was applied to maintain the membrane potential in current clamp at −65 mV . The input and residual series resistance in current clamp was determined from a 20 pA hyperpolarizing pulse applied at the end of each sweep . The input resistance in voltage clamp was determined from a 5 mV hyperpolarizing pulse applied at the beginning of each sweep . All recordings were from cells with a resting membrane potential less than −60 mV and a stable input resistance . All electrophysiological recordings were performed at 22–24°C , and data were not corrected for a junction potential of ~15 mV of the internal solution with respect to the bath ACSF . EPSPs were recorded in whole-cell mode . Capillary glass pipettes ( tip diameter , ∼5 μm ) filled with ACSF and connected to Digitimer constant current stimulus isolation unit ( AutoMate Scientific , Berkeley , CA ) were used to stimulate pre-synaptic axons in stratum radiatum as described in Results . SR95531 ( 2–10 μM ) and CGP55845 ( 1 μM ) were present to reduce GABAA and GABAB contributions , respectively . To prevent epileptic discharges in the presence of GABAergic blockers , the CA3 region was microdissected out of slices used for EPSP recordings . LTP was induced by a theta burst pairing protocol , as previously described ( Lin et al . , 2008 ) . Data were analyzed using IGOR ( WaveMetrics , Lake Oswego , OR ) . Data are expressed as mean ± SEM . Paired two-sample t-tests were used to determine significance of data in the same pathway , and non-parametric Wilcoxon Mann-Whitney two-sample rank test was used to determine significance between groups of data; p<0 . 05 was considered significant . D-AP5 , CNQX , CGP55845 , and SR95531 were obtained from Tocris Bioscience ( Ellisville , MO ) . All other chemicals were obtained from Sigma-Aldrich unless specified . All perfusing solutions were modified from regular ACSF unless otherwise noted .	The neurons in the brain communicate with each other by releasing chemical messengers across structures called synapses . This signaling always occurs in the same direction: at a given synapse , one neuron sends signals that bind to receptor proteins on the surface of the receiving neuron . Repeatedly signaling across a synapse strengthens it , making it easier to communicate across , and sometimes such stimulation can cause a persistent strengthening of the synapse: this is known as long-term potentiation . Changes in synaptic strength are important for learning and memory . In the synapses formed between a type of brain cell called CA1 neurons , a protein called SK2 forms part of an ion channel in the membrane of the receiving neuron and is important for synaptic strengthening and long-term potentiation . To work correctly , the SK2 channels must be precisely positioned at the synapse , but the mechanisms responsible for this positioning were not clear . Now , by experimenting with purified proteins taken from the CA1 neurons of mice , Kim et al . show that SK2 physically interacts with a scaffold protein called MPP2 . Further experiments revealed that MPP2 is responsible for positioning SK2 at the synapses , and this allows SK2-containing channels to contribute to long-term potentiation and synaptic strengthening . During synaptic strengthening , it is possible that SK2 disengages from MPP2 , which influences learning . The next step is to understand the processes that dictate this behavior .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Membrane palmitoylated protein 2 is a synaptic scaffold protein required for synaptic SK2-containing channel function
The transposition of bacteriophage Mu serves as a model system for understanding DDE transposases and integrases . All available structures of these enzymes at the end of the transposition reaction , including Mu , exhibit significant bends in the transposition target site DNA . Here we use Mu to investigate the ramifications of target DNA bending on the transposition reaction . Enhancing the flexibility of the target DNA or prebending it increases its affinity for transpososomes by over an order of magnitude and increases the overall reaction rate . This and FRET confirm that flexibility is interrogated early during the interaction between the transposase and a potential target site , which may be how other DNA binding proteins can steer selection of advantageous target sites . We also find that the conformation of the target DNA after strand transfer is involved in preventing accidental catalysis of the reverse reaction , as conditions that destabilize this conformation also trigger reversal . Transposons are mobile DNA elements that move or copy their DNA sequence from one location to another . They have exhibited a remarkable ability to spread , such that sequences derived from transposons are pervasive in the genomes of prokaryotes and eukaryotes alike ( Aziz et al . , 2010 ) . Among transposons whose behavior has been examined in vitro , the transposable Escherichia coli bacteriophage Mu is one of the most active ( Harshey and Bukhari , 1981; Mizuuchi , 1983 ) and well-studied ( reviewed in Harshey , 2012 ) . Its transposase , MuA , belongs to the large DDE family of recombinases ( Baker and Luo , 1994; Rice and Mizuuchi , 1995 ) . This family is named for the amino acid residues in their shared RNase-H-like catalytic domains that bind the divalent metals necessary for catalysis . In addition to the transposases for many common transposons , the DDE family also includes retroviral integrases , which use the same reaction mechanism to integrate viral genomes into host chromatin ( Fujiwara and Mizuuchi , 1988; Li et al . , 2006 ) . To catalyze transposition ( Figure 1A ) , DDE recombinases like MuA bind specific sequences at each end of their element and synapse them together in a complex known as the transpososome ( or , for retroviral integrases , the intasome ) ( Surette et al . , 1987; Wei et al . , 1997 ) . The transpososome then hydrolyzes the phosphate backbone at the boundary between the transposon and flanking host DNA , and catalyzes the attack of the resulting 3’ hydroxyl groups from each transposon end into the ‘target’ destination DNA . This critical second chemical step is referred to as strand transfer . In the case of bacteriophage Mu , the entire prophage acts as a transposon , and strand transfer occurs at positions 5 bp apart in the target DNA ( Grindley and Sherratt , 1979 ) . Rather than catalyzing multiple turnovers like a typical enzyme , MuA remains bound to the branched strand transfer products until forcibly disassembled by the ATP-powered host chaperone ClpX ( Burton et al . , 2001; Mhammedi-Alaoui et al . , 1994 ) , after which transposition can be completed by the host’s own DNA repair and replication machinery ( Mizuuchi , 1984; North and Nakai , 2005 ) . Although MuA is still active after removal by ClpX in vitro , it may be degraded by the ClpXP chaperone-protease complex in vivo ( Levchenko et al . , 1995 ) . 10 . 7554/eLife . 21777 . 003Figure 1 . Transposition by Bacteriophage Mu and Available Structures of DDE family members . ( A ) Diagram of replicative transposition . The transposable element ( here , the phage genome ) is in green with transposase binding sites in red and blue . Transposase subunits ( here , MuA , in light yellow circles ) synapse element ends and catalyze their nicking and subsequent joining to target DNA ( black ) . In vivo , the flanking DNA ( grey ) is double stranded and may represent the entire host chromosome , which may also provide the target site . ( B ) The in vitro Mu transposition system used here utilizes short linear fragments for both the target DNA and phage ends , with 3 nt of single stranded flanking DNA . ( C ) Strand transfer complex structures . DNA is colored as in ( A ) , with protein components partially transparent and colored according to their bound DNA . Generated from PDB IDs 4FCY ( Mu ) , 3OS0 ( PFV ) , 5HOO ( Mos1 ) , and 5EJK ( RSV ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 00310 . 7554/eLife . 21777 . 004Figure 1—figure supplement 1 . Mu end and target DNA fragments . ( A ) The Mu end DNA fragments used for transposition reactions . The bottom strand becomes joined to the target DNA during strand transfer . The bold red 3’ adenine provides the 3’ hydroxyl group that is the nucleophile for this reaction . In fluorescence anisotropy and some FRET experiments ( where indicated in the text ) , this adenine lacked a 3’ hydroxyl group so as to suppress strand transfer . Blue arrows mark the binding sites for MuA; the R2 binding site is replaced with a binding site for the Sin recombinase in the SinMu system . The Mu end DNA is identical to that used to generate the Mu STC crystal structure ( Montaño et al . , 2012 ) . ( B ) Linear target DNA substrates . The G:G mismatch , when used , was placed in the center of the sequence by replacing a C in the bottom strand with a G , indicated in green . Red arrows mark the position of the predominant strand transfer product when the mismatch is present . The 45 bp target DNA was used in the experiments in Figure 5 to discourage melting of the strand transfer products during heating , and is identical in sequence to the 35 bp target used elsewhere except for 5 bp added at each end . Target DNAs were labeled at the 5’ end of the top strand with 32P for all radiographic experiments , and with the Atto565 fluorophore for fluorescence anisotropy and FRET . FRET substrates included an additional Atto647N fluorophore at the 5’ end of the bottom strand . The 35 bp target ( with a mismatched base pair ) is also identical to that used to generate the Mu STC crystal structure ( Montaño et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 004 Transposases and integrases face two challenges when interacting with target DNA sites . First , successful transposition requires that they avoid selecting their own DNA as a target , as intra-self strand transfer can lead to deletion of some or all of the mobile element and a double strand break . To this end , MuA and some other DDE transposases interact with a partner protein whose role is to remain bound only to distant , non-self target sites . For Mu , this is the MuB protein encoded by the phage itself ( Maxwell et al . , 1987; Mizuno et al . , 2013 ) , whereas retroviral integrases bind to host nucleosomes to choose a target ( Pryciak and Varmus , 1992 ) . However , the mechanism ( s ) by which transpososomes resist the high local concentration of self-DNA but are activated to attack DNA bound by their target selection partner are not well understood . After strand transfer , the transposase must also avoid catalysis of the reverse reaction ( termed ‘disintegration’ ) in the time before host machinery completes transposition . Strand transfer merely exchanges of one pair of 3’ hydroxyl groups and phosphodiester bonds for another and so should not provide a net release of chemical bond energy that could drive the reaction towards its products . Nevertheless , the Mu transpososome displays a strong bias towards catalysis of only the forward strand transfer reaction ( Au et al . , 2004; Lemberg et al . , 2007; Mizuuchi et al . , 2007 ) . It seems likely that this bias stems in some way from product binding energy , as transpososomes remain tightly bound to their products . Here we show that target DNA bending contributes to overcoming both of the above challenges . Four crystal structures of transpososomes and intasomes from the DDE family that include the target DNA are available , including that of the Mu transpososome ( Figure 1C ) ( Maertens et al . , 2010; Montaño et al . , 2012; Morris et al . , 2016; Yin et al . , 2016 ) . Beyond the shared catalytic domain fold , the specifics of the protein-protein and protein-DNA contacts in these structures are very different . Nevertheless , they have all converged on a bend in the target DNA following strand transfer . Clues that target DNA bending can play a role in target site selection come from studies of target site preferences . In the absence of target selection partners , Mu and retroviral integrases prefer target sites with more easily deformable sequence steps , suggesting that DNA flexibility is interrogated prior to strand transfer and can guide target site selection ( Haapa-Paananen et al . , 2002; Pryciak and Varmus , 1992; Serrao et al . , 2015 ) . This is supported by a structure of the prototype foamy virus ( PFV ) intasome bound to , but not integrated into , target DNA , which shows the target DNA bent in a conformation very similar to that in the strand transfer complex ( Maertens et al . , 2010 ) . On the other hand , it has also been suggested that target DNA bending could prevent the reversal of strand transfer by driving the products ( the new target 3’ hydroxyl group and transposon-target phosphodiester bond ) out of the active sites to reduce the conformational strain of the bend ( Maertens et al . , 2010; Montaño et al . , 2012 ) . Both ideas have yet to be tested directly in vitro . In this work , we measure how target DNA flexibility and bending affect target DNA binding , strand transfer , and disintegration by the Mu transpososome . We show that increasing DNA flexibility has a dramatic positive effect on the transposition reaction and that bending occurs during binding , implying that bending is required but carries a steep energetic cost . Afterwards , disintegration is very rare and occurs under extreme conditions that coincide with disruption of the bend . Further , a mutant transpososome with compromised target DNA binding and bending can be rescued by pre-bent DNA and is particularly prone to strand transfer reversal . Our results are the first to biochemically link unbending to reversal , and also point to target DNA bending as a key energetic barrier to strand transfer . This barrier would allow DNA deformation generated by other proteins to steer target site selection , and provide a way to channel product binding energy into preventing strand transfer reversal . Mu is an attractive model system because transpososomes can be assembled in vitro from MuA protein and short DNA substrates ( Figure 1B ) ( Savilahti et al . , 1995 ) . The DNAs we use here mimic the products of transposon-host junction cleavage , thus removing that earlier reaction step from our analysis ( compare Figure 1A versus Figure 1B ) . They retain only 3 nt of flanking DNA on the non-transferred strand ( Figure 1—figure supplement 1 ) . The assembled transpososome without target DNA is referred to as the cleaved-donor complex ( CDC ) , and a transpososome that has bound and attacked target DNA as the strand transfer complex ( STC ) . All MuA constructs used here lacked domain Iα ( residues 1–76 ) , a site-specific DNA binding domain which in vivo binds an enhancer sequence within the interior of the Mu genome to aid transpososome assembly ( Mizuuchi and Mizuuchi , 1989 ) . This domain was also omitted in the crystal structure of the Mu STC ( Montaño et al . , 2012 ) , is not required for transposition in vitro , and is in fact slightly inhibitory in the absence of the enhancer ( Yang et al . , 1995 ) . We first sought to determine whether target DNA flexibility has an effect on target DNA binding and the kinetics of the strand transfer reaction . We use two methods to increase the flexibility of duplex DNA: DMSO added to the reaction buffer ( Escara and Hutton , 1980; Herrera and Chaires , 1989 ) , and/or a single G:G base-pairing mismatch incorporated at the center of the target DNA sequence ( Rossetti et al . , 2015 ) . In addition to altering the biophysical properties of DNA , both have been used in previous studies as general enhancers of the transposition activity of MuA in vitro ( Baker and Mizuuchi , 1992; Savilahti et al . , 1995; Yanagihara and Mizuuchi , 2002 ) . Because DMSO also affects earlier transpososome assembly steps , and to eliminate spare Mu ends or MuA subunits that could compete with our intended target DNAs , we first purified the CDC form of the transpososome by gel filtration chromatography . To prevent premature catalysis of strand transfer , CDCs were assembled and purified in a buffer containing EDTA and lacking Mg2+ . A 2:1 mixture of MuA protein to Mu end DNA in this buffer results in a gel filtration peak shifted to a very high molecular weight that we identify as CDCs based on size and near-stoichiometric reactivity with target DNA ( Figure 2 ) . 10 . 7554/eLife . 21777 . 005Figure 2 . Target DNA bending during Mu transposition affects strand transfer kinetics . ( A ) Purification of transpososome tetramers constructed in vitro by gel filtration . Full transpososomes ( black line ) are separable from lower molecular weight complexes Also shown are MuA protein and phage end DNAs alone ( dashed lines ) . The contents of only the tetramer peak were used in our assays here . ( B ) Strand transfer kinetics visualized by denaturing gel electrophoresis and 5’-32P-labeled target DNA . Where indicated , the 35 bp target DNA sequence includes a G:G base-pairing mismatch and/or the reaction buffer was supplemented with 15% DMSO . Strand transfer results in cleavage of the labeled strand . ( C ) Quantification of the strand transfer kinetics experiments described in ( B ) . The vertical axis represents the fraction of total lane signal present in product band ( s ) . The X-axis is broken between 60 and 120 min to enhance the readability of early timepoints . Error bars represent mean ± standard error of the mean ( SEM ) , n = 4 independent CDC preparations , one of which is shown in ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 00510 . 7554/eLife . 21777 . 006Figure 2—figure supplement 1 . Minor strand transfer products viewed at high contrast . Pictured are the same radiographic gel images given in Figure 2B , but with the cropping and contrast modified to highlight minor strand transfer products . Note that the mismatch drives strand transfer to occur predominantly at a single position , which is centered around the mismatch , at the expense of the numerous minor products that result from fully base-paired DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 00610 . 7554/eLife . 21777 . 007Figure 2—figure supplement 2 . Modifying target DNA flexibility or removing domain III do not alter the concerted nature of MuA-catalyzed strand transfer . ( A ) The sequence of the hairpin target DNA fragment used here to detect single ended strand transfer events . This differs from the target DNA used throughout this work ( Figure 1—figure supplement 1 ) only in the addition of the hairpin turn sequence shown in bold . As in that Figure , green indicates the nucleotide position that can be changed to generate a base-pairing mismatch . ( B ) Schematic for the experiment to detect single ended events . Single-ended strand transfer can cleave the 5’-labeled hairpin target DNA to create a novel 58 nt product . ( C ) The same experimental protocol as described for Figure 2 was used with the hairpin target DNA to monitor for single-ended events . A faint 58 nt band for these events appears across all conditions . ( D ) As in ( C ) , except Δ domain III CDCs ( as in the experiments in Figure 4 ) were used . ( E ) Quantification of the images in ( C ) and ( D ) . The vertical axis represents the percent of the total radioactivity contained in all bands that exists in the 58 nt single ended product band . Across all conditions , single ended events accumulate rapidly to a level ≤3% of total target DNA and remain nearly constant after 10–20 min of reaction time . Because levels remain constant , we interpret this to be the result of a tiny population of defective MuA protein that is capable of participating in assembly but not strand transfer . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 007 We monitored the effect of DNA flexibility on the kinetics of strand transfer by sampling reactions containing 100 nM each of purified CDC and 32P-labeled target DNA as a function of time . In these experiments , the target DNA strand is labeled at the 5’ end and becomes cleaved as a result of strand transfer . As has been observed previously ( Yanagihara and Mizuuchi , 2002 ) , a mismatched base pair directs the vast majority of strand transfer events to occur centered around it , hence the single dominant product ( Figure 2B ) . The fully base paired target DNA , which is identical in sequence except for the central nucleotide in one strand , results in two major products ( and a number of minor products , see Figure 2—figure supplement 1 ) . It is likely that the major insertion site remains approximately centered even without the mismatched base pair because our 35 bp target DNA is not much larger than the total target DNA binding surface of the CDC . Increased target DNA flexibility via the G:G mismatch or DMSO significantly enhances the rate of strand transfer ( Figure 2B , C ) , without perturbing the CDCs’ ability to perform concerted strand transfer with both ends ( Figure 2—figure supplement 2 ) . Using the steepest slope between any two timepoints in Figure 2C as an estimate of initial rate , the initial rate for the unmodified reaction was about 0 . 1 µM−1 min−1 , and it required about 40 min to convert 30% of the target DNA into strand transfer product . When the reaction buffer was supplemented with 15% ( v/v ) DMSO , the initial rate was about 0 . 8 µM−1 min−1 and was 30% complete in only 4 min . The mismatch was even more powerful , leading to an initial rate of about 3 . 0 µM−1 min−1and 30% completion in about 1 min . The combined effect of both modifications was additive , albeit only marginally ( shown most clearly in the 2 and 5 min timepoints in Figure 2C ) . To determine whether accelerated strand transfer rates result from tighter binding to the more flexible target DNA substrates , we used fluorescence anisotropy to measure the affinity of the interaction between purified CDCs and target DNA ( Figure 3A ) . To measure these values under conditions where the MuA active site and the DNA ionic environment would be as realistic as possible , we performed these experiments in the presence of Mg2+ . Rather than by withholding divalent metals , we prevented strand transfer by using Mu end DNAs lacking the terminal 3’OH group on the transferred strand , which is the nucleophile for strand transfer ( Figure 3—figure supplement 1 ) . Binding affinities followed a similar pattern to reaction rates: The KD of CDCs for mismatched target DNA ( regardless of DMSO ) was about 15 nM , and for fully base paired target DNA with DMSO , about 29 nM . In the case of the mismatched target , good fits to the data required a cooperative binding model with a Hill coefficient of 3 . 3 ( for the mismatch alone ) and 2 . 6 ( when combined with DMSO ) . We suspect this arises from transient CDC disassembly events that become significant at low nanomolar concentrations . Conversely , we were unable to detect enough binding to normal DNA under normal buffer conditions to confidently fit a binding curve , but can visually estimate that KD might lie between 0 . 5–1 µM . Thus , increasing the flexibility of the target DNA can enhance its affinity for transpososomes by at least 33-fold . 10 . 7554/eLife . 21777 . 008Figure 3 . Enhanced flexibility triggers tight target DNA binding . Fluorescence anisotropy measurements of transpososome binding to Atto565-labeled target DNAs . Dots are experimental data , with error bars representing 95% confidence intervals derived from 15 measurements ( see Materials and methods ) . Solid lines represent fits to the data to obtain the KD values indicated in the legends . ( A ) Wild-type transpososomes , ( B ) Δ domain III transpososomes ( see Figure 4A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 00810 . 7554/eLife . 21777 . 009Figure 3—figure supplement 1 . Strand transfer and/or target DNA nicking are blocked by terminating the transferred strand with dideoxy-adenosine . The results of a 1 hr reaction between 200 nM 5’-32P-labeled target DNA and a pre-formed transpososome mixture using the indicate protein construct ( s ) . The first two lanes are from reactions in which MuA protein and Mu end DNA were omitted . Across all conditions , the reactions in which the transferred strand of the Mu end DNA ended in a 3’ dideoxy-A do not show any evidence of strand transfer or other breaks in the target DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 009 These results indicate that bending or otherwise deforming the target DNA poses a significant energetic barrier to binding target DNA . We also note that the binding and kinetic data do not exactly correspond . For instance , although the mismatch and DMSO produced similar enhancements in target DNA affinity ( with KDs differing by only two-fold ) , the mismatch had a much stronger effect on the strand transfer reaction than DMSO ( with initial rates differing by about four-fold ) , even though the kinetic experiments were carried out at 100 nM , above the KDs for either of these target interactions . This suggests that there may be additional localized target DNA conformational changes , beyond those that are required for initial binding , that are necessary to properly position the target DNA phosphate backbone in the active site , and that are better facilitated by the mismatch than by DMSO . The high binding affinity of CDCs for flexible target DNA suggests that it must be bent in order to make optimal contacts with the transpososome , and the crystal structure of the Mu STC suggests that domain IIIα , which forms a positively charged alpha helix , stabilizes the bent target DNA conformation . The two copies of domain IIIα from the catalytic MuA subunits pair in the middle of the target DNA U-turn ( Figure 4A ) , presumably countering the electrostatic repulsion between the two arms of the target DNA , as well as providing additional protein-DNA contacts . To test the importance of domain III in DNA binding , we repeated the above kinetic and binding experiments using CDCs lacking domain III on the catalytic MuA subunits . Note that domain IIIβ , which comprises the C-terminal ~60 residues of MuA , was not included in the crystal structure because is not required for transposition in vitro , but has been included in the constructs used in this work . Although domain III can be removed by truncating MuA constructs at residue 560 , transpososome assembly requires domain IIIα to be present on the two MuA subunits not involved in catalysis ( Aldaz et al . , 1996 ) . In order to localize truncated subunits only at the catalytic positions , we utilized our ‘SinMu’ system ( Figure 4—figure supplement 1 ) ( see also Ling et al . , 2015 ) . SinMu is a chimeric protein in which the sequence-specific DNA binding domains of MuA have been replaced with that of the unrelated Sin recombinase . It can be placed at the non-catalytic positions in the transpososome by a corresponding substitution in the Mu end DNA sequence ( Figure 1—figure supplement 1 ) . Chimeric ‘SinMu’-containing CDCs that include the full MuA C-terminus at all subunit positions behave indistinguishably from wild type CDCs ( Figure 4—figure supplement 1 ) , and subsequent truncation of the catalytic subunits’ domain III does not hinder assembly and purification ( Figure 4B ) . 10 . 7554/eLife . 21777 . 010Figure 4 . Δ Domain III Transpososomes are rescued by enhanced target DNA bending . ( A ) Domain III from the catalytic MuA subunits is used to make contacts to the bent target DNA . The solid ( non-transparent ) alpha helices indicated with arrows are the portions domain III resolved in the MuA STC crystal structure . Colors as in Figure 1C , with MuA active site residues as orange spheres . ( B ) Gel filtration chromatography of Δ domain III CDCs . Moving to the chimeric SinMu system and truncating the catalytic MuA subunits to remove domain III does not hinder formation and subsequent purification of CDCs . ( C ) Strand transfer kinetics visualized by denaturing gel electrophoresis and 5’-32P-labeled target DNA , as in Figure 1D except using Δ domain III transpososomes . Lower panels are the product band ( s ) from the upper panels at greatly increased contrast . ( D ) Quantification of the strand transfer kinetics experiments described in ( C ) . The vertical axis represents the fraction of total lane signal present in product band ( s ) . Error bars represent mean ± SEM , n = 4 CDC preparations , one of which is shown in ( C ) . ( E ) Rescue of the strand transfer activity of truncated transpososomes by circular DNAs . 32P-labeled Mu ends increase in size as a result of strand transfer . Samples were taken 2 hr after transpososomes were mixed with Mg2+ and indicated target DNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 01010 . 7554/eLife . 21777 . 011Figure 4—figure supplement 1 . The SinMu system does not inherently alter interactions with target DNA . ( A ) A molecular model of the SinMu system . We use this to direct removal of domain III to only the catalytic ( upper ) subunits in the transpososome . Because the DNA binding domains from non-catalytic subunits ( circled ) do not contact any other component of the transpososome , they can be altered to be specified by an alternate DNA binding domain and DNA sequence from the Sin recombinase . Colors are as in Figure 1B , with the modeled Sin substitutions in orange . ( B ) Quantification of strand transfer kinetics experiments , performed as described for Figure 2 . Solid lines represent the activity of wild type MuA transpososomes , and are the same data plotted in Figure 2C . Dashed lines represent SinMu transpososomes in which the catalytic subunits are WT ( not truncated ) , under identical conditions . Error bars represent the mean ± the standard error of the mean ( SEM ) from four independent time-series . ( C ) Fluorescence anisotropy measurements of target DNA binding . The left panel is the same data as in Figure 3A . The results of identical experiments performed with SinMu transpososomes are shown in the right panel . Error bars represent 95% confidence intervals derived from five technical replicates of 3 independent concentration series ( 15 total data points , see Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 01110 . 7554/eLife . 21777 . 012Figure 4—figure supplement 2 . Minicircle target DNA . ( A ) Minicircle DNA was constructed from the two linear DNA fragments shown: Circle A ( C–A ) and Circle B ( C–A ) . Each piece included an approximately centered high affinity binding site for the DNA bending protein IHF . These fragments end in reciprocal complementary four nt overhangs for annealing and ligation . The C-A fragment was designed to optionally include a G:G base-pairing mismatch at the position indicated by the bold green residues . ( B ) Minicircle target DNA . This is the 126 bp circular target DNA used in Figure 4 . The component linear fragments from ( A ) are indicated . The position of the optional G:G mismatched base pair is marked again in bolded green , and the strand transfer attack positions it would produce are marked again with red arrows . The oblong shape is given for illustrative convenience and is not meant to have any correlation the actual physical conformation of the minicircle DNA molecule . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 01210 . 7554/eLife . 21777 . 013Figure 4—figure supplement 3 . Minicircle rescue of Δ domain III transpososome strand transfer activity is not an artifact of the sequences of the DNAs used to form the minicircles . In these experiments , the Mu end DNA used to form transpososomes was labeled with 32P . For each lane , the target DNA substrate , presence of a G:G mismatched base-pair in that substrate , and the presence of DMSO in the reaction buffer are indicated above the images . The left half of the bottom panel is the same image as in Figure 4E . C-A and C-B target DNAs are the linear fragments that were ligated to form the circular target ( given in Figure 4—figure supplement 2 ) ; C-A included the G:G mismatch where indicated . Note that the mismatch-bearing C-A is the only circle component able to rescue Δ domain III strand transfer activity on its own , and only in the presence of DMSO , which is the same behavior as for the primary linear fragment used throughout this work . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 013 We found that CDCs where the catalytic MuA subunits lack domain III ( ‘Δ domain III CDCs’ ) display very little strand transfer activity , except under particular conditions . Rather than just enhancing the reaction , the combination of DMSO and a target base-pairing mismatch were nearly essential for the truncated CDCs to catalyze strand transfer into short linear target DNAs at detectable levels ( Figure 4C , D ) . The reaction rate is otherwise exceedingly slow ( <1% of input after two hours under the next best condition , the mismatch alone ) . Measurements of target DNA binding by the truncated CDCs once again followed a similar pattern to strand transfer kinetics: binding was relatively robust when both modifications are present simultaneously ( KD ≈ 38 nM ) , but undetectable otherwise ( Figure 3B ) . Thus , domain III does indeed provide many of the protein-DNA contacts that are important to capture a target . High target DNA flexibility , however , can make up for the loss of these contacts . This confirms our assertion that the contacts are optimized for bent DNA . To investigate further the connection between binding and bending , we tested whether the strand transfer activity of the truncated CDCs could be rescued by providing pre-bent DNA for use as a target , rather than the linear DNA fragments used thus far . We generated pre-bent DNA by creating 126 bp DNA minicircles ( Figure 4—figure supplement 2 ) . Instead of just being more flexible , minicircles should be naturally and permanently held in a bent state . This should provide an even lower energy barrier to bending . Accordingly , we found that minicircle DNAs can rescue the strand transfer activity of Δ domain III CDCs under a number of conditions ( Figure 4E and Figure 4—figure supplement 3 ) . While Δ domain III CDCs generate a substantial amount of strand transfer products into linear DNA only in the presence of both the mismatched base-pair and DMSO , strand transfer into minicircles requires only one or the other . Thus the loss of transpososome-target DNA interactions can be mitigated by pre-bending the DNA . This is also further evidence that DMSO and the base-pairing mismatch allow the DNA to be more easily bent , and that target DNA binding is linked to bending . The strand transfer reaction does not change the overall number or type of covalent bonds , and is entropically unfavorable in that it links together previously separate DNA segments . The free energy landscape of this reaction might thus be expected to favor its reversal ( hereafter referred to as ‘disintegration’ ) . Nevertheless , our data above show that strand transfer is capable of going nearly to completion ( Figure 2C ) , indicating that the STC is able to suppress disintegration . Given that Mu transpososomes are not true catalysts ( each transpososome catalyzes a single strand transfer reaction , and then remains bound to the reaction products ) , it is likely that MuA suppresses disintegration by using product binding energy to bias the reaction direction toward the final product complex . Our finding that CDCs bind particularly tightly to artificially flexible DNA , and the observation that strand transfer should make any target DNA more flexible by nicking both strands , led us to investigate the relationship between target DNA conformation and disintegration . We first wanted to establish the baseline rate at which the STC catalyzes disintegration . Published reports currently differ regarding this rate . It is difficult to address because it requires a starting pool of pure STCs ( free of unreacted target DNA , which would be indistinguishable from the product of disintegration ) , and any purification protocols used must preserve the activity and structure of the STCs . This has been previously approached , to our knowledge , in two ways: ( 1 ) forming STCs by mixing MuA , Mu end DNAs and target DNA and Mg2+ , as was done for crystallization ( Montaño et al . , 2012 ) , then purifying the resulting STCs by electrophoresis in an agarose gel in TBE buffer ( which chelates Mg2+ ) , and ( 2 ) circumventing the issue of unreacted target DNA by directly assembling transpososomes on branched DNA substrates that mimic the strand transfer products ( Au et al . , 2004; Mizuuchi et al . , 2007 ) . Both methods result in an off-pathway pseudo-disintegration reaction that has been referred to as a ‘foldback’ ( see Figure 5—figure supplement 1 ) . The foldback products imply that some of the STCs in those experiments were either destabilized prior to the assay itself or not properly assembled to being with . To avoid these issues , we devised a new protocol ( Figure 5—figure supplement 2 ) : STCs are formed by reaction of CDCs with labeled target DNA , but are immobilized on neutravidin-coated magnetic beads via biotinylated Mu end DNAs , such that remaining unreacted labeled target DNA can be rapidly washed away at the end of the STC formation step without stripping the Mg2+ ions from the active site . Furthermore , an excess of high-affinity ( mismatch-containing ) cold competitor target DNA is added in the final wash step to trap any complexes that undergo subsequent disintegration . Any residual unreacted labeled target DNA that was not removed in the washing steps can be quantified by sampling the mixture immediately after purification . Monitoring these purified STCs as a function of time shows that they are impressively robust against disintegration under our normal reaction conditions , with , at most , less than 2% of strand transfer products reverting to re-ligated target DNA after 60 min ( Figure 5A ) . This is true whether or not DMSO , a mismatch in the target DNA , or domain III were present . Overall , these data suggest that disintegration events themselves are normally rare , regardless of target binding affinity . 10 . 7554/eLife . 21777 . 014Figure 5 . Strand Transfer Complex Disintegration . ( A ) Disintegration is not detectable under normal reaction conditions . Substrates and potential products were visualized by denaturing gel electrophoresis and 5’-32P-labeled target DNA . STCs were rapidly purified by immobilization on magnetic beads into normal reaction buffer ( see Materials and methods ) , which included DMSO where indicated , and incubated at 30°C . ( B ) Disintegration under modified reaction conditions . The same procedure as in ( A ) , but STCs were purified into a modified buffer ( see text ) and held for 1 hr at the indicated temperatures . Target DNAs in these experiments have 5 bp added to each end to prevent melting during heating . Lanes labeled C are a sample taken immediately after purification , as in the 0 min timepoint in ( A ) . ( C ) Quantification of replicates of the experiment shown and described in ( B ) . The vertical axis represents the fraction of strand transfer product present immediately after purification that became re-ligated after treatment . Error bars represent mean ± SEM , n = 4 independent STC purifications . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 01410 . 7554/eLife . 21777 . 015Figure 5—figure supplement 1 . Pseudo-disintegration of strand transfer products by the ‘foldback’ pathway . ( A ) A schematic of foldback disintegration . Mu STCs react to the melting of the 5 bp overlap in the target DNA by catalyzing cleavage of the Mu end-target DNA junctions using a target DNA 3’OH as the nucleophile . The difference between foldback , shown here , and a true reversal of the strand transfer reaction lies in the choice of which target 3’OH is used in the reaction . Although the Mu end DNA products are the same , the target site becomes split into two hairpin ends . ( B ) The sequence of a SinMu end DNA fragment with 10 nt of flanking DNA past the cleavage site ( compare to Supplemental Figure 1A ) . We find that Mu end DNAs with this extended single stranded flank can trigger foldback disintegration under certain conditions . As in Supplemental Figure 1A , the nucleophile for strand transfer is shown in bold and red , and protein binding sites are indicated by black and blue arrows . ( C ) A target DNA fragment designed for simple detection of foldback products . The products of foldback disintegration from target DNA fragments where strand transfer occurs approximately in the center of the sequence ( as is the case elsewhere in this work ) are the same length ( once denatured ) as the input target DNA fragment , which makes the products of true vs . foldback disintegration difficult to differentiate . The target DNA fragment here uses an off-center mismatched base pair ( bold green ) to direct strand transfer to positions ( red arrows ) where it is possible to differentiate the 5’-labeled intact target DNA ( 45 nt ) from strand transfer products ( 15 and 25 nt ) and from foldback products ( 35 and 55 nt ) . ( D ) Our disintegration assay does not trigger foldback disintegration . Here , the same protocol as in Figure 5 was used , but the target DNA was that from ( C ) and labeled at both 5’ ends , and incubation time was 25 min . The expected strand transfer products are present , but foldback products are not . ( E ) Foldback disintegration is triggered under certain conditions . Here , we used the Mu end DNA described in ( B ) to form reversal-prone ( see Figure 5 ) Δ domain III strand transfer complexes . Purification of these STCs by gel electrophoresis followed by supplementation of MgCl2 and heating for 25 min . using the protocol from ( Au et al . , 2004 ) does trigger foldback disintegration at high temperatures ( note the new 35 and 55 nt products in the 60° lane ) . This demonstrates our ability to detect foldback if it occurs . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 01510 . 7554/eLife . 21777 . 016Figure 5—figure supplement 2 . Method for generating STCs for disintegration experiments . A schematic for the experiments performed in Figure 5 . CDCs are assembled on Mu end DNAs in which the transferred strand is labeled at the 5’ end with biotin ( see also Figure 1—figure supplement 1 ) . Four unpaired thymidine residues separate the biotin label from the Mu end DNA sequence . Magnetic beads coated in Neutravidin ( Nav ) are present during CDC assembly . Excess MuA and Mu end DNA is then washed away , and replaced by buffer containing Mg2+ and the intended 32P-labeled target DNA to permit strand transfer to occur . Excess labeled target DNA is then also washed away , the buffer ( optionally ) exchanged to increase pH and glycerol content , and cold mismatch-bearing target DNA competitor is added . Immediately , a sample is taken to record the baseline level of residual unreacted labeled target DNA , and then samples can be taken to detect disintegration as a function of time and/or incubation temperature . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 016 Previous studies have indicated several factors that could stimulate reversal: increased temperature , increased pH , and higher concentrations of glycerol in the reaction buffer ( Au et al . , 2004; Lemberg et al . , 2007 ) . We note that the first two modifications might weaken binding between MuA and bent target DNA . Indeed , we found that the combination of slightly increased buffer pH ( 7 . 9 instead of 7 . 4 ) , increased glycerol content ( 16% instead of 5% ) , and high temperatures ( 60°C ) triggered disintegration ( Figure 5B ) . Under these conditions , WT STCs converted between 10–15% of strand transfer products back into intact target DNA after 1 hr , ( Figure 5C ) . In our assay , this occurs without producing the ‘foldback’ products that have complicated previous studies ( Figure 5—figure supplement 1D , E ) . Unlike the forward strand transfer reaction , disintegration was not highly affected by including a mismatch in the target DNA . Remarkably , when subjected to the same buffer and heat challenge , Δ domain III STCs catalyze 4-fold more disintegration than WT STCs , reverting about 40% of strand transfer products back into intact target DNA . To observe more directly when the transpososome bends the target DNA and if the bend changes as a result of strand transfer , we measured Förster resonance energy transfer ( FRET ) between Atto565 and Atto647N fluorescent labels positioned at opposite 5’ ends of our 35 bp target DNA . Target DNA bending would bring the labels at each end of the target DNA into closer proximity , reducing the fluorescence intensity and lifetime of the donor Atto565 label . We used time correlated single photon counting ( TCSPC ) to capture this decrease in donor lifetime . By measuring changes in fluorescence lifetimes , we are able to separate out contributions from unbound/unreacted target DNA and approximate the distribution of lifetimes ( and thus the distribution of relative proximities ) present in a sample ( Figure 6 and Figure 6—figure supplement 1 ) . We do this by first observing the behavior of the doubly-labeled target DNA alone , free of any transpososomes ( Figure 6 , first column ) . The 35 bp target DNA should keep the fluorophore pair >116 Å apart , well outside the 69 Å Forster radius ( R0 ) for this dye pair . Indeed , the most reasonable fits to the decay of the DNA-only samples result from fitting a single discrete lifetime ( shown as black bars ) along with a minor ( <10% of total photons ) population of very short lifetimes ( shown in grey ) . We take the former to be the lifetime of the donor fluorophore in unbent DNA ( where FRET is negligible ) , and the latter to be a combination of imperfections in the instrumentation and anomalous fluorophore behavior . Decay measurements taken in the presence of transpososomes were then fit as the sum of this unbound fluorophore lifetime and a gaussian distribution representing any shorter lifetimes resulting from transpososome-induced bending . 10 . 7554/eLife . 21777 . 017Figure 6 . Target DNA bending measured by TCSPC FRET . The measured fluorescence decay for each condition was fit with a model combining a discrete lifetime ( black bars ) and a distribution of lifetimes ( colored gaussian distributions ) . In each plot , the area under each component represents its molar fraction in the fit; the center , full width at half maximum ( FWHM ) and relative abundance is given for each FRET distribution . In the absence of transpososomes ( left-most column ) , little to no FRET can be detected and the vast majority of fluorophores decay with a discrete lifetime . This discrete lifetime was held constant as CDCs lacking the strand transfer nucleophile ( second column ) and active CDCs ( third column ) were added . Δ domain III transpososomes were used for the bottom row . In the fourth column , the STC samples from the bottom two rows were heated to 60°C for 5 min and measured again . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 01710 . 7554/eLife . 21777 . 018Figure 6—source data 1 . Time correlated single photon counting FRET data . TCSPC FRET data are organized into folders by date of collection so that the data can be paired with their proper instrument response function ( IRF ) . For each measurement taken , three files are given: ( 1 ) . ifx: The raw output data from the Vinci fluorometer control software ( 2 ) . txt: A tab delimited text representation of the data , first column is time ( ns ) and the second is photon counts ( 3 ) . pdf: The output of the lifetime fitting done in the Vinci control software and represented in Figure 6 . Filenames are descriptive of the sample and conditions: WT or D3: Wild-type transpososomes , or SinMu transpososomes lacking domain III on the catalytic subunits . DMSO: Where present , indicates 15% ( v/v ) DMSO in the buffer . Mismatch: Where present , indicates that the labeled target DNA contained the central G:G base pairing mistmatch . DO or DA: Target DNA was singly labeled with donor Atto565 fluorophore , or labeled with both donor Atto565 and acceptor Atto647N at opposite ends . 60c: Where present , indicates that the sample and vessel were heated to 60 degrees celsius prior to measurement . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 01810 . 7554/eLife . 21777 . 019Figure 6—figure supplement 1 . Fluorescence lifetime fits across all samples . Each sample was fit as the sum of one gaussian and one discrete lifetime . A donor only fluorophore configuration indicates that the Atto647N FRET acceptor fluorophore was omitted from the target DNA . In the case of DNA only and Donor only samples , no variables were held constant . In samples with transpososomes present , the discrete lifetime from the matching DNA-only sample was held constant ( and thus has no associated error ) in order to account for an unbound/unbent/singly-labeled fraction . FWHM: Full width at half maximum . TCC: target capture complex , CDCs lacking the 3’OH strand transfer nucleophile were used . STC: strand transfer complex , CDCs fully capable of strand transfer were used . α refers to the pre-exponential factor for the decay component . DO , Donor fluorophore only; DA , Donor and Acceptor fluorophores present . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 019 The bend in bound target DNA prior to strand transfer can be measured by utilizing , once again , CDCs lacking the terminal Mu end 3’OH nucleophile for that reaction . Comparing normal conditions to either addition of DMSO or DMSO plus a mismatched base-pair in the target DNA , our analysis indicates that all three show a population of fluorophores in a bent ( reduced fluorescence lifetime ) configuration ( Figure 6 , second column ) . Once the contribution from unbound or unbent DNA is removed from each , the lifetime distributions from all three are centered at about 3 . 3 ns . This indicates that bound DNA is bent to approximately the same degree under all conditions and that enhanced flexibility acts largely to increase the bound/bent fraction . That the normal target DNA/normal buffer condition shows a large unbound/unbent fraction is not surprising , given we have shown it exhibits poor affinity and struggles to utilize target DNA to completion . There is some suggestion in the fitted distributions that more flexible DNA is bent more stably , as judged by the tighter widths ( FWHM ) of the FRET distributions , but this trend may be outside of the resolution of this analysis . If strand transfer is allowed to proceed ( by using unmodified Mu end DNAs; Figure 6 , third column ) , the bend remains nearly unchanged for WT transpososomes across both the normal and highest flexibility conditions . Thus , any conformational changes in the target DNA brought about by strand transfer must be confined mainly to the vicinity of the nicks . FRET also confirmed that Δ domain III CDCs have a reduced ability to bend target DNA . We have shown that they can bind relatively tightly to target DNA when both DMSO and a mismatched base pair are present ( Figure 3B ) . However , FRET data for these same complexes indicates very little bending , despite the experiment being carried out at concentrations where the vast majority of the target DNA would be bound . This is direct evidence that our truncated CDCs are bending deficient . However , it is curious that despite a lack of bending they are still able to carry out strand transfer at a rate comparable to WT . It may be that there is a conformation of the target DNA that is competent for strand transfer ( and still enhanced to a great degree by a base-pair mismatch ) which does not necessarily bring the distant ends of the target DNA closer together , e . g . bends that are not properly phased for a U-turn , or it may be that the mismatch and DMSO allow rapid but very transient sampling of the strand-transfer-competent conformation . Nevertheless , after strand transfer proceeds and opens nicks in the target DNA , we find that the target DNA in Δ domain III STCs becomes significantly more bent and adopts a bend very similar to that seen in WT CDCs and STCs . This suggests that the breaks in the target DNA phosphate backbone allow it to access a bent conformation that is low enough in energy as to be resilient to the loss of some transpososome-DNA contacts . We also note that this convergence on the same bend coincides with the convergence in WT and Δ domain III behavior in resisting disintegration ( at normal temperatures ) . Finally , to determine whether unbending occurs under the high temperatures that trigger disintegration , we used a Peltier heater built into the cuvette holder of our TCSPC instrument to bring the mismatched target + DMSO STC complex samples rapidly to 60°C for 5 min prior to taking a repeated measurement . Heating the samples resulted in sizeable broadening of the FRET distribution , particularly for the disintegration-prone Δ domain III transpososomes , and an increase in the fitted amount of unbent target DNA ( Figure 6 , fourth column ) . This means that disintegration coincides with conditions where the target DNA conformation is less stably controlled ( and a fraction of STCs may have lost the bend in the target DNA entirely ) . In this study , we have shown that the conformation of the target DNA has dramatic consequences for transposition by phage Mu . Our model for this process is shown as a free energy reaction coordinate diagram in Figure 7A . In the absence of outside factors , newly assembled transpososomes bind to potential target DNA with low affinity . We hypothesize that this is because the target DNA must be strongly bent to optimize its contacts with CDCs , but free energy released by making those contacts does not compensate well for the conformational strain required to bend the target DNA . This is supported by our findings that enhancing the flexibility of the target DNA and pre-bending it both greatly enhance its affinity for CDCs , while removing some of the contacts that stabilize the bent form by deleting domain III greatly reduces its affinity . We also found that strand transfer rates correspond relatively well with target binding affinity , which suggests that getting target DNA into the active sites of the CDCs can be rate-limiting for strand transfer and thus provides a control point for determining the choice of target . This aspect of Mu transposition was exploited by Yanagihara and Mizuuchi ( 2002 ) who showed it can be used in vitro to pinpoint a single mismatch in the presence of a vast excess of unperturbed DNA . Furthermore , in vivo transposition into unperturbed target DNA may be even slower than in our assays , as the presence of intact flanking DNA has been shown to further inhibit target capture ( Williams and Baker , 2004 ) . 10 . 7554/eLife . 21777 . 020Figure 7 . Bending the target DNA slows target capture and resists disintegration . ( A ) Cartooned as a simplified free energy diagram , with the unmodified system as the bold black line . Target DNA capture is energetically uphill because the target DNA must be bent . This equilibrium can be altered to favor target capture by factors that promote DNA flexibility ( here , mismatched base-pairs , DMSO buffer , or mini-circularization , dashed red line ) , or to make target capture even more uphill by removing some contacts between the transpososome and the target DNA ( here , deleting domain III , dashed purple line ) . Strand transfer interrupts the target DNA backbone , thus releasing the conformational strain induced by the bend while preserving or solidifying the protein-DNA contacts between the transpososome and target . This leaves the strand transfer products in a deep thermodynamic optimum that prevents disintegration . Our results indicate that domain III stabilizes the target DNA conformation to a greater degree after strand transfer . The dashed grey line represents a hypothetical transposase that binds to target DNA favorably and without inducing any conformational strain . ( B ) The features of the energy landscape outlined in ( A ) have the effect of resisting target DNA capture until an outside factor ( e . g . , MuB ) can compensate for the uphill free energy difference . After strand transfer , the release of bending strain combined with product binding energy ensure that strand transfer is effectively irreversible . DOI: http://dx . doi . org/10 . 7554/eLife . 21777 . 020 Why should transposition be slow by default ? Why scan through potential target sites ? Although Mu and many of its transposon and retrovirus relatives do not hunt for specific target sequences , they do need to avoid inserting into their own genomes even though their own DNA is necessarily present at high local concentration . We propose that transposition is repressed by default except in the presence of DNA bound by target selection proteins . For Mu , target selection is driven by the distribution of MuB protein , whose reported biochemical activities appear at first glance paradoxical ( reviewed in Dramićanin and Ramón-Maiques , 2013 ) . MuB forms filaments along dsDNA in the presence of ATP ( Maxwell et al . , 1987; Mizuno et al . , 2013 ) , and binds the C-terminal region of MuA ( Wu and Chaconas , 1994 ) . Contact with MuA , even the monomeric form present before a transpososome is formed , triggers ATP hydrolysis and DNA release by MuB ( Adzuma and Mizuuchi , 1988; Greene and Mizuuchi , 2002 ) . Thus DNA near the Mu ends is probably cleared of MuB before an active transpososome is formed . However , MuB-bound DNA molecules are strongly preferred as transposition targets . Until recently , it has been difficult to explain how DNA sites coated or encased in MuB can be attractive targets . Our results suggest how transposition targeting mechanisms might activate transposition: their ability to present pre-deformed or pre-bent DNA . The structural data available for MuB and IstB ( the related targeting ATPase from IS21elements ) suggest that they have the ability change the helical conformation of the DNA they coat and/or to form bundles of protein-DNA filaments ( Arias-Palomo and Berger , 2015; Dramićanin et al . , 2015; Mizuno et al . , 2013 ) . Loops could be extruded from these large protein-DNA complexes by transient dissociation events within a filament or as the DNA crosses from one collinear filament to an adjacent one within a bundle . These loops could act like our pre-bent minicircles , triggering tight binding and rapid transposition . This is supported by the observation that Mu has a bias towards target sites immediately adjacent to MuB-coated DNA , rather than inside MuB filaments themselves ( Ge et al . , 2011 ) . For retroviral integration , nucleosomes provide pre-bent DNA by their very nature . Retroviral integration happens directly on the nucleosome , and a high-resolution structure of this event has been obtained by cryo-electron microscopy ( Maskell et al . , 2015 ) . Nucleosome DNA could be a favored target as long as the energy needed to dislodge a DNA loop slightly from the nucleosome surface and into the integrase active sites is less than that required to bend DNA de novo . There is biochemical evidence for this energetic compromise , as integration is known to occur predominantly at positions that are bound least tightly to the histone core ( Maskell et al . , 2015; Serrao et al . , 2015 ) . Much of the conformational strain accrued by the bent target DNA could be released by strand transfer nicking the target DNA backbone . These nicks would allow the adjacent DNA to locally relax to a less strained conformation while maintaining or strengthening the energetically favorable MuA-target contacts . Additionally , the structural consequence of relaxing the strain in the bent target DNA could be to eject one or both of the strand transfer product moeities ( the target 3’OH or new phosphodiester bond ) from the active site , which is the case for the four available strand transfer complex structures ( Maertens et al . , 2010; Montaño et al . , 2012; Morris et al . , 2016; Yin et al . , 2016 ) . The equilibrium of the strand transfer reaction may therefore be strongly biased toward products because the final strand transfer complex is much lower in energy than the initial states ( CDCs both before and after capturing target DNA ) . This thermodynamic minimum may be so deep that thermal energy is insufficient to bring the 3’ OH and scissile phosphate group back into reacting position at any appreciable rate . The existence of a kinetic barrier to disintegration is supported by our finding a lack of disintegration products at 30 degrees , even under conditions where we would have expected any transiently-formed disintegrated target DNA dissociate because its affinity for CDCs is so weak: Δ domain III transpososomes with no DMSO , and fully base paired DNA with WT transpososomes and no DMSO ( and with an excess of higher affinity cold target DNA present to trap any released CDCs ) . This implies that under these conditions , the transpososomes simply do not sample the equilibrium between the target capture and strand transfer complexes . Disintegration could be detected , however , at high temperature , which FRET showed also destabilizes the conformation of the STC target DNA . We suggest that heat weakens the protein-DNA contacts holding the strand transfer product moieties away from the active site and provides the thermal energy necessary to escape the STC energy minimum . Disintegration at 60 degrees was much more efficient for Δ domain III STCs than for WT ones . Because the Δ domain III CDCs have very low affinity for target DNA , this might be explained by their releasing transiently disintegrated DNA more readily . However , WT complexes have a much lower affinity for normal vs . mismatched target DNA , but these produced similar amounts of disintegration product . This suggests instead that domain III – target DNA interactions have an even stronger stabilizing effect on the final strand transfer complex than they do on the intermediate captured target , probably because the binding surface is particularly tailored to a target DNA conformation that is only readily accessible after nicking . Therefore removal of domain III appears to adjust the equilibrium between the target capture complex and the final STC . An alternate way to consider the energetics of this reaction is to consider the free energy diagram if there were no DNA bending , or in the most extreme case , no contacts to any part of the DNA except the scissile phosphate . In that case ( dashed grey line in Figure 7A ) , transpososomes would likely bind to any free DNA immediately as a target and there were be no energetic difference between the target capture complex and the final STC ( the equilibrium while bound to the enzyme would be 1 ) . Regarding the strand transfer step , we propose that removing domain III , heat , and the other buffer modifications that helped to promote reversal , move the system slightly toward this scenario by weakening the contacts between the target DNA and MuA . Heat also by definition helps cross the kinetic barrier between the two states ( in both directions ) . In summary , we find that target DNA is strongly bent upon initial capture by CDCs ( as shown by the FRET data in Figure 6 ) . Because the energy required to induce this bend in the DNA is only partially compensated for by energetically favorable protein-DNA contacts , CDCs have very weak affinity for unperturbed target DNA , which is reflected in very slow rates of strand transfer into those targets . As cartooned in Figure 7A , we found that this barrier can be overcome in several ways: by lowering the free energy of the bent DNA conformation , using DMSO or a mismatch , or by raising the free energy of the initial unbound state by ligating it into a strained pre-bent minicircle . We suggest that in the natural pathway for Mu transposition , MuB protein does the latter , presenting pre-bent DNA to the CDCs . The affinity for target DNA could be manipulated in the opposite direction by removing domain III . That Δ domain III transpososomes are specifically defective in stabilizing DNA bending was supported by our observation that they could be rescued by increasing the flexibility of the target DNA or by pre-bending it . The strand transfer reaction greatly increases the stability of the final protein-DNA complex by releasing strain in the target DNA while maintaining , and probably increasing , the protein-DNA contacts . The stability of this complex shifts the equilibrium of the strand transfer reaction strongly towards products . We therefore propose that target DNA bending therefore serves two purposes ( Figure 7B ) : ( 1 ) it helps enforce proper choice by rendering binding of CDCs to naked target DNA very weak , and ( 2 ) it alters the energetic landscape in a way that drives the strand transfer reaction forward . Proteins were expressed in the Rosetta DE3 Escherichia Coli strain ( Merck KGaA , Darmstadt Germany ) from coding sequences cloned into the pET3c plasmid . Transformed cells were grown at 37°C in LB media supplemented with 100 µg/mL ampicillin to OD600 ≈ 0 . 7 , then protein expression was induced by addition of 0 . 66 mM IPTG and an additional 20 µg/mL ampicillin . At 2 hr after induction , the cells were collected by centrifugation at 8000 rpm , and stored at −80C until lysis . Cells were resuspended in a solution of 25 mM HEPES pH 7 . 5 , 1 mM EDTA , 1 M NaCl , 10% sucrose , 10% glycerol , 5 mM DTT , and Complete protease inhibitor ( Roche Diagnostics , Indianapolis IN ) and lysed by two passes through an LV1 Microfluidizer ( Microfluidics , Westwood MA ) . Cell debris was removed by centrifugation and the resulting supernatant was fractionated by addition of ammonium sulfate to 30% saturation . The precipitate was collected by centrifugation at 18 , 000 rpm in an SS-34 ( Thermo Fisher Scientific , Waltham MA ) rotor for 30 min , and the resulting pellet was resuspended in 20 mM MES pH 5 . 5 , 0 . 2 M NaCl , 0 . 5 mM EDTA , 5% glycerol , and 1 mM DTT ( Buffer A ) . This was passed over a HiPrep Heparin FF 16/10 affinity column ( GE Healthcare , Chicago IL ) and eluted with a gradient from 0 . 2 M ( Buffer A ) to 2 M ( Buffer B ) NaCl . Fractions containing the protein were diluted back into Buffer A and the heparin affinity chromatography was repeated . Fractions containing the protein were concentrated and further purified by gel filtration chromatography using HiLoad 16/600 Superdex 75 prep-grade column ( GE Healthcare ) equilibrated with 25 mM HEPES pH 7 . 5 , 0 . 4 M NaCl , 0 . 5 mM EDTA , 5% glycerol , and 1 mM DTT . Peak fractions were combined , dialyzed into the same buffer supplemented to 20% glycerol , concentrated , and stored at −80°C . The final concentration of proteins was determined by measuring their absorbance at 280 nm . The sequences of the linear DNA substrates used in this work are given in Figure 1—figure supplement 1 . All oligonucleotides , including biotin and fluorophore modifications , were synthesized by Integrated DNA Technologies ( IDT , Coralville IA ) . Oligonucleotides modified with biotin or fluorophores were HPLC purified by IDT , and 32P labeled oligonucleotides were PAGE purified by IDT . Oligonucleotides were resuspended in TE ( 10 mM Tris pH 8 , 1 mM EDTA ) , and those not purified by IDT were desalted using P6 spin columns ( Bio-Rad Laboratories , Hercules CA ) equilibrated in TE . The concentration of all oligonucleotides was verified by measuring their absorbance at 260 nm . DNA substrates were radiolabeled at the 5’ end using γ-32P ATP ( PerkinElmer , Waltham MA ) and T4 polynucleotide kinase ( Thermo Fisher Scientific ) . Oligonucleotides were 3’ modified with dideoxy-adenosine by 4 hr treatment with Terminal Transferase ( New England Biolabs , Ipswich MA ) and a 25-fold molar excess of dideoxy-ATP . Double stranded DNA substrates were created by mixing complementary oligonucleotides of the desired sequence in equimolar amounts in a buffer of 10 mM Tris pH 8 , 100 mM NaCl , and 1 mM EDTA , heating to 80°C , and then slowly cooling to room temperature . CDCs were formed in a buffer of 25 mM HEPES pH 7 . 4 , 200 mM NaCl , 5% glycerol , 0 . 6 mM Zwittergent 3–12 ( Merck KGaA ) , and 0 . 5 mM EDTA . The reaction buffer used during experiments was identical to this , except EDTA was omitted and replaced with 10 mM MgCl2 . Where indicated , these buffers also contained 15% ( v/v ) DMSO . Unless otherwise specified , this reaction buffer was used for all experimental procedures involving transpososomes . Cleaved donor complexes ( CDCs ) were formed by mixing protein and Mu end DNAs in a 2:1 MuA protein:Mu end DNA ratio ( or a 1:1:1 MuA protein:SinMu protein:Mu end DNA ratio ) and incubating at 30°C for ≥1 hr . In cases where gel filtration was used to purify CDCs , DMSO was included in the formation buffer . Purification of CDCs by gel filtration was performed using a Superdex 200 Increase 10/300 column ( GE Healthcare ) equilibrated in formation buffer lacking DMSO . After gel filtration the tetramer peak was collected , spin concentrated , quantified by measuring absorbance at 260 nm , and used immediately . Target binding or strand transfer reactions were carried out at 30°C and initiated by diluting fresh CDCs and the appropriate target DNA into reaction buffer . This buffer contained DMSO where indicated . For electrophoretic analysis , strand transfer reactions were stopped by phenol:chloroform extraction . CDCs were formed as described above , except with DNAs ending in dideoxy-A and in reaction buffer ( with mM MgCl2 ) , which was present during all steps . Purified CDCs were serially diluted ( in reaction buffer ) , mixed with 6 nM Atto565-labeled target DNA , and arrayed into a Corning 3575 black polystyrene 384 well microplate ( Corning , Corning NY ) . Binding reactions were incubated for 1 hr at room temperature . A Victor X5 plate reader ( PerkinElmer ) was used to read the anisotropy of the Atto565 fluorophore , using a 531 nm excitation and 595 nm emission filters and a 1 . 0 s counting time . The raw parallel and perpendicular photon counts were adjusted to account for the instrument response ( G factor and plate positioning artifacts ) , and the resulting anisotropies were fit to an equilibrium binding model , accounting for receptor depletion , using the optimize . curve_fit function from the Python scipy package ( Research Resource Identifier ( RRID:SCR_008058 ) . For each data point , 15 total measurements of anisotropy were made: three separate serial dilutions each measured by the plate reader five times . Minicircle DNAs were constructed by ligating together two linear DNA fragments with complementary sticky ends . These fragments contained a binding site for the Integration Host Factor ( IHF ) protein , which was added prior to ligation so that the DNA fragments would be bent in order to encourage circular ligation . IHF protein was purified using a protocol described previously ( Swinger and Rice , 2007 ) . The sequences of the minicircle DNAs used in this work are given in Figure 4—figure supplement 2 . In detail , the two DNA pieces were mixed in an equimolar ratio along with a twofold molar excess of IHF in T4 Ligase buffer ( New England Biolabs ) for 30 min at room temperature . T4 DNA Ligase ( New England Biolabs ) was then added and the mixture incubated for 12 hr at room temperature . Proteins were removed by phenol:chloroform extraction followed by P6 column ( Bio-Rad Laboratories ) buffer exchange into fresh T4 Ligase buffer . T4 DNA Ligase was added back in and incubated at room temperature for an additional 2 hr to remove any remaining nicks . This reaction mixture was phenol:chloroform extracted and separated by gel electrophoresis on an 8% polyacrylamide TBE gel . The band corresponding to the circular product was identified by UV shadowing , excised , and extracted by shredding the gel slice and soaking in a ~15 fold volume excess of 10 mM Tris pH 8 , 250 mM NaCl , and 2 mM EDTA for 8 hr at room temperature . Gel fragments were removed by filtration . To remove any remaining linear fragments or nicked circles , an equal volume of 2x BAL-31 nuclease buffer and 5 µL BAL-31 nuclease ( New England Biolabs ) were added and incubated for two hours each at 30°C and 40°C . The nuclease was removed by addition of EGTA to 20 mM followed by phenol:chloroform extraction and P6 column buffer exchange into a buffer of 10 mM Tris pH 7 . 5 , 100 mM NaCl , 15% glycerol , 2 mM EDTA , and 2 mM EGTA . Minicircles were stored at −20°C . CDCs were formed as described above using Mu end DNAs modified with biotin on the 5’ end of the transferred strand ( Figure 1—figure supplement 1 ) . After allowing an hour for CDC formation , Sera-Mag SpeedBeads NeutrAvidin particles ( GE Healthcare ) were added to ~0 . 1% ( w/v ) final concentration and incubated for an additional 30 min at 30°C . Using a magnet to immobilize bead-bound CDCs , the mixture was washed to remove unbound protein and DNA and replace the CDC formation buffer with reaction buffer ( including MgCl2 and DMSO ) . An approximately two-fold molar excess of the indicated target DNA was then added for STC formation . This strand transfer reaction proceeded for 2 hr at 30°C , with gentle mixing every 30 min to keep the magnetic beads evenly dispersed . During the final 5 min , heparin was added to a final concentration of 0 . 05 mg mL−1 to encourage dissolution of any incompletely formed MuA-DNA complexes . Immobilized strand transfer complexes were then washed four times into the desired buffer for disintegration ( which always contained MgCl2 ) . A ~5 fold molar excess of unlabeled cold competitor mismatched target DNA was added in the final wash . A sample was always taken immediately after the final wash ( time = 0 min ) to record the initial baseline levels of reacted/unreacted DNA . After this , the reaction was either sampled as a function of time or divided equally among three temperature conditions ( 30°C , 45°C , and 60°C ) for 1 hr . Disintegration was carried out where indicated in either the standard reaction buffer used throughout this work , or reaction buffer modified to pH 7 . 9% and 16% glycerol . Samples of the reaction were taken by dilution into a 10-fold volume excess of 97% formamide +10 mM EDTA that had been pre-heated to 100°C . Samples included 100 nM of labeled target DNA and either 1 . 5 µM ( for the fully base-paired target DNA / no DMSO condition ) or 1 . 0 µM ( all other samples ) purified CDCs . This mixture was equilibrated at 30°C for 1 hr prior to measurements . For each condition , measurements were made of both a donor-only ( Atto565 ) and a donor-acceptor ( Atto565 + Atto647N ) sample . Time-domain lifetimes were measured on a ChronosBH lifetime fluorometer ( ISS , Inc . , Champaign IL ) using Time-Correlated Single Photon Counting ( TCSPC ) methods . The fluorometer contained Becker-Hickl SPC-130 detection electronics and an HPM-100–40 Hybrid PMT detector ( Becker & Hickl GmbH , Berlin Germany ) . Tunable picosecond pulsed excitation at 565 nm was provided by a Fianium SC400–2 supercontinuum laser source with integrated pulse picker and AOTF ( NKT Photonics , Birkerød Denmark ) . Emission wavelengths were selected with a Semrock Brightline FF01-593/40 bandpass filter ( Semrock , Inc . , Rochester NY ) . The Instrument Response Function ( IRF ) was measured in a 1% scattering solution of Ludox LS colloidal silica in matched buffers . Multi-component exponential decay lifetimes and distributions were fit via a forward convolution method in the Vinci control and analysis software ( ISS , Inc . ) . For both donor only and donor-acceptor samples , all fitted parameters are given in Figure 6—figure supplement 1 . Raw data in both the original and a tab-separated plain text format , as well as fitting outputs from the Vinci software in PDF format , are available for download as Figure 6—source data 1 . Plots were prepared using the Python package Matplotlib ( RRID:SCR_008624 ) . Visualization and rendering of macromolecular structures for figures used Pymol ( The PyMOL Molecular Graphics System , Version 1 . 8 Schrödinger , LLC; RRID: SCR_000305 ) .	Pieces of DNA called transposons can move or copy themselves around the genome . Some viruses – such as HIV and Mu ( a virus that infects bacteria ) – act as transposons to hide their DNA by inserting it into their host’s genome . Mu , HIV and many transposons all work in the same , somewhat unusual way . Like many chemical reactions , joining DNAs together needs a source of energy to make it happen , yet these viruses and transposons do not need high energy inputs to work . In addition , they do not look for a specific DNA sequence to insert their DNA into . This gives them the advantage of inserting copies of their DNA anywhere in the host’s genome , but also means that multiple copies might mistakenly insert into each other . Visualizations of the insertion process show that the DNA that the viruses insert their DNA into is always bent like a U-turn . Why does this bending occur ? It may be that the bending helps the virus to choose where in the DNA to insert and acts as a way to power the chemical reaction that joins the DNA . To investigate this possibility , Fuller and Rice performed experiments using purified fragments of DNA and the enzyme from Mu that does the DNA joining chemistry . The results revealed that making the insertion site DNA easier to bend made the insertion much faster . Furthermore , a mutant enzyme that struggled to bend the DNA had trouble keeping the chemistry going , and so the viral DNA would accidentally pop back out after it was joined . Thus the insertion site DNA is like a spring: the enzyme puts a lot of energy into bending it , but once the viral DNA has been inserted that energy is released to power the reaction to completion . Fuller and Rice conclude that if other proteins were to pre-bend or otherwise make the DNA more flexible , this would tell the DNA-joining enzyme where to insert , which helps explain the roles of known targeting proteins for Mu and HIV . Further work is now needed to investigate whether these other targeting proteins exist for other viruses and transposons , and to identify them .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2017	Target DNA bending by the Mu transpososome promotes careful transposition and prevents its reversal
Microphthalmia-associated transcription factor ( MITF ) is the master regulator of the melanocyte lineage . To understand how MITF regulates transcription , we used tandem affinity purification and mass spectrometry to define a comprehensive MITF interactome identifying novel cofactors involved in transcription , DNA replication and repair , and chromatin organisation . We show that MITF interacts with a PBAF chromatin remodelling complex comprising BRG1 and CHD7 . BRG1 is essential for melanoma cell proliferation in vitro and for normal melanocyte development in vivo . MITF and SOX10 actively recruit BRG1 to a set of MITF-associated regulatory elements ( MAREs ) at active enhancers . Combinations of MITF , SOX10 , TFAP2A , and YY1 bind between two BRG1-occupied nucleosomes thus defining both a signature of transcription factors essential for the melanocyte lineage and a specific chromatin organisation of the regulatory elements they occupy . BRG1 also regulates the dynamics of MITF genomic occupancy . MITF-BRG1 interplay thus plays an essential role in transcription regulation in melanoma . Microphthalmia-associated transcription factor ( MITF ) , a basic helix-loop-helix leucine zipper ( bHLH-Zip ) factor , regulates specification , survival , and proliferation of normal melanocytes , and controls proliferation , migration and invasion of melanoma cells ( Goding , 2000; Widlund and Fisher , 2003; Steingrimsson et al . , 2004 ) . The level of functional MITF expression determines many of the proliferation and invasion properties of melanoma cells ( Hoek and Goding , 2010 ) and siRNA-mediated MITF silencing induces senescence in several melanoma lines ( Strub et al . , 2011 ) . We previously reported a genome wide analysis of MITF target genes in 501Mel melanoma cells ( Strub et al . , 2011 ) . Chromatin immunoprecipitation coupled to deep sequencing ( ChIP-seq ) identified MITF binding sites and integration with RNA-seq following siRNA-mediated MITF knockdown showed that MITF directly and positively regulates genes involved in DNA replication and repair and mitosis . In contrast , MITF represses genes involved in melanoma invasion . To better understand the molecular function of MITF , we used tandem affinity purification to isolate MITF from 501Mel melanoma cells and mass spectrometry to identify its interacting partners . We report here the first comprehensive characterisation of the MITF interactome and we show that MITF interacts physically and functionally with a novel form of the PBAF chromatin-remodelling complex specific for neural crest derived cells comprising both BRG1 and CHD7 . We also show that MITF and SOX10 recruit BRG1/PBAF to the nucleosomes flanking critical enhancer elements in the melanocyte lineage . Using 501Mel melanoma cell lines stably expressing an N-terminal FLAG-HA epitope-tagged MITF ( F-H-MITF ) to levels comparable to that of endogenous MITF ( Figure 1A , line J ) , we performed tandem affinity purification of soluble nuclear-and chromatin-associated fractions ( Drané et al . , 2010 ) . Mass spectrometry and silver nitrate staining showed numerous proteins in the tandem immunoprecipitation from F-H-MITF cells , while almost no proteins were detected in immunoprecipitations from untagged 501Mel cells ( Figure 1B and Supplementary file 1 ) . Two independent purifications and mass spectrometry analyses were performed and only proteins identified specifically in the F-H-MITF precipitations with no peptides in the control precipitations are discussed . In addition , contaminating ribo-nucleoprotein particles , spliceosome components , and chaperone proteins have been excluded from the subsequent analysis . 10 . 7554/eLife . 06857 . 003Figure 1 . Purification of MITF-associated complexes . ( A ) Western blot of 501Mel cell lines stably expressing Flag-HA-tagged-MITF ( F-H-MITF ) . ( B ) The immunoprecipitated material from the soluble nuclear extract ( SNE ) was separated by SDS PAGE and stained with silver nitrate . F-H-MITF is indicated along with * that designates a contaminating protein seen in the control immunoprecipitations . Lane M corresponds to a molecular mass marker indicated in kDa . ( C ) Immunoblot detection of HERC2 , BRG1 , USP7 , USP11 , XRCC5 , and XRCC6 in the MITF-associated complexes . ( D ) Summary of proteins and complexes interacting with MITF . Shown are the proteins found specifically in the immunopurifications of F-H-MITF classified according to their function and organisation into known complexes . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 003 Previously described MITF partners such as its heterodimerisation partners TFEB , TFE3 and TFEC ( Steingrimsson et al . , 2002 ) and its cofactor β-catenin ( CTNNB1 ) ( Schepsky et al . , 2006 ) ( Figure 1D and Supplementary file 1 ) were detected . We also identified novel potential MITF partners . The BPTF , SMARCA1 ( SNF2L ) , SMARCA5 ( SNF2H ) , and RBBP4 components of the NURF chromatin-remodelling complex associated with MITF specifically in the chromatin fraction . Components of the DNA damage response machinery including XRCC5 and XRCC6 ( Ku80 and Ku70 ) , DNA-dependent protein kinase ( PRKDC ) , BRCA2 as well as MSH2 and MSH6 were identified along with the HECT domain-containing E3-ligase HERC2 , previously implicated in DNA repair ( Bekker-Jensen et al . , 2010 ) , and UBR5 a second HECT domain-containing E3-ligase with functions in both DNA repair and transcription ( Cojocaru et al . , 2011; Gudjonsson et al . , 2012 ) . NEURL4 , a known HERC2-interacting protein ( Al-Hakim et al . , 2012 ) , was also found along with the de-ubiquitinase enzymes USP7 and USP11 that were preferentially represented in the SNE . In contrast , USP13 shown to regulate MITF stability ( Zhao et al . , 2011 ) was not detected in our experiments . Several interactions were verified by immunoblot as HERC2 , BRG1 , USP7 , USP11 , XRCC5 , and XRCC6 were detected in the F-H-MITF immunoprecipitations from the soluble nuclear fraction , but not in the untagged controls ( Figure 1C ) , with enrichment of USP7 and USP11 in the SNE compared to the CAE and the selective presence of HERC2 in the SNE . We identified several other complexes interacting with MITF . Four subunits of TFIIIC , a RNA polymerase III cofactor were detected along with the SMCA1 , SMC3 , STAG2 , and PDS5 cohesin subunits . TRIM28 ( TIF1β , KAP1 ) , a co-repressor belonging to a sub-family of TRIM proteins comprising TRIM24 ( TIF1α ) and TRIM33 ( TIF1γ ) was found as were the HDAC1 and HDAC2 and the HP1 proteins known to associate with TRIM-co-repressor complexes ( Herquel et al . , 2011 ) . We identified TRRAP , RUVBL1 , RUVBL2 , and BAF53A . These proteins are already known to form a cofactor for MYC ( Park et al . , 2001 , 2002; Murr et al . , 2007 ) . Interestingly , a mutation in TRRAP has recently been associated with human melanoma ( Wei et al . , 2011 ) that together with its interaction with MITF suggests a role in melanomagenesis . In addition to transcription complexes , the RFC1 , RFC2 , RFC4 , and RFC5 subunits of DNA replication factor C associate with MITF along with the MCM3 , MCM5 , and MCM7 subunits of the MCM complex that forms at DNA-replication origins . AKAP8 and AKAP8L were also identified and AKAP8 has been shown to interact with the MCM complex ( Eide et al . , 2003 ) , although these protein kinase A anchoring proteins have additional functions ( Collas et al . , 1999 ) . The kinase PLK1 was found in the chromatin-associated fraction suggesting that it may phosphorylate MITF on chromatin . Finally , we detected multiple subunits of the nuclear pore complex . This may reflect control of MITF sub-cellular localization or a coupling of MITF-driven transcription and RNA-export . Together , these data describe a comprehensive set of MITF-interacting proteins . While BRG1 was reported to interact with MITF ( de la Serna et al . , 2006; Keenen et al . , 2010 ) , the composition of the complex had not been determined . We identified BRG1 ( SMARCA4 ) as well as the PBRM1 ( BAF180 ) , SMARCC2 ( BAF170 ) , SMARCD2 ( BAF60B ) , and ACTL6A ( BAF53A ) subunits along with CHD7 reported to associate with BRG1 in human neural crest cells ( Bajpai et al . , 2010 ) ( Figure 1D ) . The MITF interacting complex most closely resembles the PBAF variant with the presence of PBRM1 ( Trotter and Archer , 2008; Reisman et al . , 2009 ) . To investigate BRG1 complex composition in 501Mel cells , extracts were precipitated with an anti-BRG1 antibody showing co-precipitation of BRG1 with BAF200 , BAF155 , BAF53A , BAF180 , and BAF170 , whereas BAF60B and BAF250B were not co-precipitated ( Figure 2A ) . Peptides for BAF60B were detected in the MITF interactome , but this subunit did not associate with BRG1 in 501Mel cells , while BAF60A co-precipitated with BRG1 , suggesting that MITF interacts with BAF60B independently from the BRG1 complex . The MITF-interacting complex also comprises CHD7 as seen by the reciprocal co-precipitation of BRG1 and CHD7 and their co-precipitation with MITF ( Figure 2A–B ) . Together , these observations show that MITF interacts with a novel form of PBAF complex comprising BRG1 and CHD7 ( Figure 2C ) . 10 . 7554/eLife . 06857 . 004Figure 2 . Composition of BRG1 complexes in 501Mel cells . ( A ) BRG1 associates with CHD7 in 501Mel cells . Following immunoprecipitation of 501Mel cell extracts with anti-BRG1 antibody or HA beads as control the eluted fractions were probed with antibodies for the indicated proteins . ( B ) Following immunoprecipitation of 501Mel cell extracts with anti-CHD7 antibody or HA beads as control the eluted fractions were probed with antibodies for the indicated proteins . ( C ) Table summarising the known subunits of BRG1-containing complexes highlighting catalytic subunits with ATPase activity , common subunits and specific subunits . The composition of the complex interacting with MITF based on mass-spectrometry and immunoblots is schematised . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 004 To address the function of BRG1 in melanoma cells , we performed si/shRNA knockdown . si/shRNA knockdown of BRG1 strongly reduced MITF protein and mRNA levels ( Vachtenheim et al . , 2010 and Figure 3A–B ) . The effects of BRG1 silencing on gene expression were therefore very similar to the loss of MITF itself with reduced expression of MITF target genes involved in cell cycle , pigmentation and signalling and activation of several genes of the senescence-associated secretory phenotype ( SASP ) whose expression is also induced upon MITF-knockdown ( Ohanna et al . , 2011 ) ( Figure 3B ) . BRG1 silencing arrested 501Mel proliferation and cells showed a flattened enlarged morphology with multiple cytoplasmic projections , similar to senescent cells upon MITF knockdown ( Figure 3C [Strub et al . , 2011] ) . Up to 40% of the BRG1 knockdown cells showed senescence-associated β-galactosidase staining ( Figure 3C ) . 10 . 7554/eLife . 06857 . 005Figure 3 . BRG1 is essential in melanoma cells and melanocytes . ( A ) Immunoblots of 501Mel cells transfected with control or anti-BRG1 siRNA or cells infected with lentiviral vectors expressing control or anti-BRG1 shRNA . ( B ) Reverse transcription real time qPCR ( RT-qPCR ) performed on 501Mel cells infected with lentivirus vectors expressing control ( C ) or anti-BRG1 shRNA . The ratio of expression of the indicated genes is shown . ( C ) Phase contrast microscopy of 501Mel cells infected with the indicated shRNA vectors after 5 days of puromycin selection . Magnification X10 . The lower panel shows cells stained for senescence-associated β-galactosidase . Arrowheads indicate representative stained cells . Inserts show enlargements of representative cells . ( D ) Immunoblots of Hermes 3A extracts following infection with lentiviral vectors expressing control , anti-BRG1 , or anti-MITF shRNAs . ( E ) Phase contrast microscopy of Hermes 3A cells infected with the indicated shRNA vectors after 5 days of puromycin selection and stained for senescence-associated β-galactosidase . Arrowheads indicate representative stained cells . Inserts show enlargements of representative cells . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 00510 . 7554/eLife . 06857 . 006Figure 3—figure supplement 1 . Gene expression changes in 501Mel and Hermes 3A cells . ( A ) A Comparative analysis of RNA-seq data from shBRG1 and shMITF 501Mel cells . Venn diagrams indicate the overlap between genes that are up and down-regulated in the shBRG1 and shMITF cells . The results of ontology analysis of the commonly up- and down-regulated genes are shown graphically and classed by p value and the number of genes in each category is indicated . Lists of secreted growth factors and cytokines forming the SASP in the shBRG1 and shMITF cells are indicated . ( B–C ) Comparative analysis of RNA-seq data from shBRG1 and shMITF Hermes 3A cells . Venn diagrams indicate the overlap between genes that are up and down-regulated in the shBRG1 and shMITF cells . The results of ontology analysis of the specifically and commonly up- and down-regulated genes are shown graphically and classed by p value and the number of genes in each category is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 00610 . 7554/eLife . 06857 . 007Figure 3—figure supplement 2 . SOX10 regulates MITF expression in 501Mel cells . ( A ) Western blot analysis of MITF expression in the indicated cell types after siRNA transfections . ( B ) RT-qPCR analysis of gene expression in the indicated cell types after siRNA transfections . ( C ) Western blot analysis of BRG1 and MITF expression in 501Mel CL8 cells following siRNA transfections . Phase contrast microscopy of 501Mel Cl8 cells following siRNA transfections . Magnification X10 . ( D ) RT-qPCR analysis of gene expression in the indicated cell types after siRNA transfections . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 007 RNA-seq following shBRG1 silencing revealed a dramatic effect on gene expression with >4000 genes down-regulated and >5400 genes up-regulated ( p . adj <0 . 05 and log2fold change >1 , Figure 3—figure supplement 1A and Supplementary file 2 ) . ShMITF knockdown resulted in cell cycle arrest and morphological changes associated with senescence ( Figure 3C ) accompanied by the down-regulation of around 600 genes and up-regulation of 747 genes . 60% of genes down-regulated by shMITF showed similar loss of expression upon shBRG1 consistent with the fact that MITF was strongly repressed in the shBRG1 cells ( Figure 3—figure supplement 1A and Supplementary file 2 ) . Commonly down-regulated genes were enriched in ontology terms associated with signalling , cell cycle , and mitosis ( Figure 3—figure supplement 1A ) . Common up-regulated genes were enriched in terms associated with angiogenesis , adhesion , and migration consistent with the dramatic changes in cell morphology . MITF silencing has been shown to induce a SASP comprising CCL2 , CTGF and SERPINE1 ( Strub et al . , 2011 ) . RNA-seq identified a putative SASP in shMITF cells comprising around 20 secreted factors and of these 15 were also induced in the shBRG1 cells , although several key factors such as IL8 and CCL2 were not induced upon BRG1 silencing ( Figure 3—figure supplement 1A ) . Loss of either BRG1 or MITF therefore induced senescence of 501Mel cells . SOX10 , TCF/LEF/CTNNB1 and CREB have been reported to activate MITF expression ( Goding , 2000 ) . We noted that SOX10 expression is strongly repressed in BRG1 knockdown cells , but not in MITF-knockdown cells ( Supplementary file 2 ) . SiSOX10 silencing repressed endogenous MITF expression ( Figure 3—figure supplement 2A–B ) . In 501Mel-Cl8 cells constitutively expressing 3HA-tagged MITF from the CMV promoter ( Strub et al . , 2011 ) , siSOX10 repressed endogenous , but not ectopic MITF . In contrast , siCREB silencing had no effect on MITF expression . SOX10 is therefore a major regulator of MITF expression in 501Mel cells and its diminished expression upon BRG1 knockdown partly explains the concomitant MITF loss . These observations are also consistent with previous reports showing that SOX10 promotes melanoma cell proliferation and that its loss leads to senescence ( Cronin et al . , 2013 ) . To determine whether the shared phenotypes of BRG1 and MITF knockdown cells resulted from the concomitant loss of MITF upon BRG1 silencing or whether BRG1 acts also as an MITF co-factor , we performed shBRG1 silencing in the 501Mel-Cl8 cells . BRG1 knockdown in these cells repressed endogenous MITF expression , but not ectopic 3HA-MITF ( Figure 3—figure supplement 2C ) . Nevertheless , BRG1 silencing elicited a phenotype similar to 501Mel cells characterised by arrested proliferation , and morphological changes . Many MITF target genes were similarly repressed by BRG1 silencing in both 501Mel and Cl8 cells , while SASP components were induced ( Figure 3—figure supplement 2D ) . Together , these data show that BRG1 is essential for MITF expression and that it acts as a cofactor for MITF since ectopic MITF in the Cl8 cells does not activate target genes expression in its absence . We also investigated BRG1 function in untransformed Hermes 3A melanocytes . In contrast to 501Mel cells , shBRG1 silencing had little effect on MITF expression in Hermes 3A cells ( Figure 3D ) , but induced changes in cell morphology with up to 80% of cells showing staining for senescence-associated β-galactosidase ( Figure 3E ) . Within 8 days , the BRG1 silenced cells detached from the plate . ShMITF silencing in Hermes 3A cells also led to growth arrest and a marked changes in morphology , with flattening , enlargement of the cell body and reduced neurite projections ( Figure 3D ) . Despite these changes indicative of senescence , <50% of shMITF-silenced cells showed staining for senescence-associated β-galactosidase . As with shBRG1 , MITF silencing led to cells detaching from the plate within 7 days . Thus , both BRG1 and MITF are essential for melanocyte growth , and in their absence cells undergo growth arrest , senescence , and death . RNA-seq showed that BRG1 silencing in Hermes 3A cells down-regulated 587 genes , and up-regulated 971 genes , many fewer than in 501Mel cells ( Figure 3—figure supplement 1B–C , and Supplementary file 2 ) . Down-regulated genes were involved in pigmentation , cholesterol metabolism as well as intracellular signalling cascades and general cell morphology . Up-regulated genes were involved in cell–cell signalling and adhesion as well as angiogenesis . MITF silencing down-regulated 757 genes and up-regulated 664 genes . Comparisons showed that 38% of genes down-regulated upon BRG1 silencing were diminished by MITF silencing , while 25% of the BRG1 up-regulated genes were also increased upon MITF silencing . Together these results show that BRG1 is critical for Hermes 3A proliferation despite the fact that it regulated a much reduced gene expression programme in these cells . As with the 501Mel cells however , MITF and BRG1 cooperate to regulate a subset of genes involved in several essential cellular processes associated with resistance to apoptosis , cell morphology , and signalling . As BRG1 is essential for proliferation of both melanocytes and melanoma cells in vitro , we asked if BRG1 is also essential for melanocytes in vivo in mice . To bypass the embryonic lethal phenotype of BRG1 germ-line knockout ( Bultman et al . , 2000 ) , we crossed mice with floxed alleles of the Smarca4 gene encoding BRG1 ( Indra et al . , 2005 ) with Tyr-Cre mice to allow selective inactivation of BRG1 in the melanocyte lineage ( Delmas et al . , 2003 ) . We first generated Tyr-Cre::Smarca4lox/+ mice that were crossed to generate the resulting Smarca4mel+/− or Smarca4mel−/− genotypes . The Tyr-Cre::Smarca4mel−/− mice were present in the progeny at the expected ratio indicating no loss of viability ( Figure 4A ) . While the heterozygous Tyr-Cre::Smarca4mel+/− mice were black , homozygous Tyr-Cre::Smarca4mel−/− mice were either completely white or had occasional black spotting presumably arising from rare melanoblasts that had escaped Cre-driven recombination during embryogenesis ( Figure 4B ) . Examination of the hair from two genotypes showed an absence of melanin in Smarca4mel−/− hair . 10 . 7554/eLife . 06857 . 008Figure 4 . BRG1 is essential in mouse melanocytes in vivo . ( A ) Statistics relevant to the phenotype of the mice lacking Brg1 in the melanocyte lineage . Note that as the Tyr-Cre transgene is present on the X chromosome , we detect black mice with the Tyr-Cre::Smarca4lox/lox genotype , but in these animals the Cre-recombinase is subjected to X-inactivation such that recombination does not occur in all melanoblasts . ( B ) Photographs of representative mice of the indicated genotypes . Bright field images of hair from the backs of 8-week-old animals of the two genotypes are also shown . Magnification X40 . ( C ) Labelling of dorsal hair follicle bulbs with antibodies against Sox10 in red and Dct in green along with Hoechst-stained nuclei in blue . Arrows indicate labelled melanocytes in the bulb of Brg1 expressing mice . Scale bars are 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 008 We labelled the hair follicles of the Smarca4mel+/− or Smarca4mel−/− animals with antibodies against Dct and Sox10 . In Smarca4mel+/− mice , Sox10 stained the nucleus of mature bulb melanocytes whose cytoplasm was stained with Dct ( Figure 4C ) . In contrast , there was no staining with these antibodies in Smarca4mel−/− bulbs . There were therefore no identifiable mature melanocytes in the mutant animals . Taken together with the essential role of BRG1 in melanoma and melanocyte cells in vitro , these data indicate that BRG1 is essential for generation of mature melanocytes in vivo . Given the critical function of BRG1 in the melanocyte lineage in vitro and in vivo , we used ChIP-seq to better understand how MITF and BRG1 regulate gene expression in 501Mel cells . We performed BRG1 ChIP-seq on native Mnase-digested chromatin along with a GFP-control ChIP-seq and identified >42 , 400 BRG1-occupied sites located in inter and intragenic regions , but with relative enrichment at the promoter ( Figure 5A ) . Comparison with public ENCODE data showed BRG1-occupied sites often co-localized with or flanking enhancer regions marked with acetylated H3K27 ( ac ) a mark of active enhancers ( Smith and Shilatifard , 2014 ) ( Figure 5—figure supplement 1A–B ) . As melanocyte-specific enhancers are not represented in public ENCODE data , we integrated the BRG1 profile with public H3K27ac data from primary human foreskin melanocytes and with H3K27ac data from a proliferative primary melanoma culture ( Verfaillie et al . , 2015 ) . BRG1 co-localized with enhancers active in melanocytes ( Figure 5—figure supplement 1A–B ) including melanocyte-specific enhancers seen at the MITF and DCT loci ( see below ) . The primary melanocyte and proliferative melanoma H3K27ac profiles are very similar and BRG1 co-localized with around 15 , 000 melanocyte/melanoma H3K27ac marked regions at promoters as well as distal inter and intragenic enhancers ( Figure 5B ) . Analysis of DNA sequence motifs at BRG1 peaks identified binding sites for a large set of transcription factors ( TFs ) , the most abundant of which are NRF1 , KLF5 , SP1-family factors , CTCF , EGR1 , and E-boxes ( Figure 5—figure supplement 1C ) . Thus , BRG1 is recruited by many different TFs to its sites on the 501Mel genome . 10 . 7554/eLife . 06857 . 009Figure 5 . Genome-wide BRG1 occupancy . ( A ) Pie chart showing distribution of BRG1-bound sites with respect to genomic features . ( B ) Clustering of BRG1 occupied sites with primary melanocyte and proliferative melanoma H3K27ac-marked elements and their distribution with respect to genomic features . ( C ) Read density clustering of BRG1 at its occupied sites reveals several profiles of BRG1 occupancy . ( D ) Clusters A–C from panel C were re-clustered and representative meta-profiles are shown to illustrate the distances separating the BRG1 peaks with the large and small arrowheads indicating the stronger and weaker peaks , respectively . ( E ) Re-analysis of RefSeq TSS where BRG1 is present reveals different localizations of BRG1-bound nucleosomes . Several representative meta-profiles of different subclasses are shown . The peak coordinates in each class relative to the TSS are indicated . ( F ) The overall meta-profile of BRG1 TSS occupancy is superimposed on the RNA Pol II meta-profile . The binding profiles of BRG1 and Pol II around the TSS are schematised below the graph . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 00910 . 7554/eLife . 06857 . 010Figure 5—figure supplement 1 . Profiling of BRG1 genome occupancy . ( A–B ) UCSC screenshots illustrating BRG1 occupancy over the ARRDC2 and REL1 loci highlighting co-localization with H3K27ac-marked enhancers either in the Encode data track or in Human Foreskin Melanocytes ( HFM ) . ( C ) Enrichment of TF binding motifs at BRG1-occupied sites . ( D ) Clustering analysis of BRG1 occupancy at RefSeq TSS illustrating the presence of BRG1 at a subset of TSS with different binding profiles . Cluster E showing BRG1 occupancy both upstream and downstream of the TSS was re-clustered to highlight the different profiles as shown in Figure 4E . ( E ) Integrative analysis of BRG1 ChIP-seq data with shBRG1 RNA-seq data . A table shows the number of genes with BRG1-occupied sites ±10 kb or ± 30 kb from the TSS . Venn diagrams indicate the number of these genes that are up- or down-regulated following shBRG1 knockdown . ( F ) Ontology analysis of the BRG1-associated up- and down-regulated genes . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 010 BRG1 occupied sites displayed a variety of profiles . A set of highly occupied sites ( Clusters A–C , Figure 5C ) comprised two peaks separated by various distances . Re-clustering of each of these sets revealed a complex profile comprising in general two more or less sharply defined peaks separated by distances of 450–800 bp ( Figure 5D ) . At other sites however , BRG1 occupied a single sharp peak ( cluster D , Figure 5C ) . BRG1-occupied sites at the TSS also displayed different profiles . BRG1 was preferentially located either upstream or downstream of the TSS ( Figure 5—figure supplement 1D , clusters A–D ) or showed equivalent occupancy at both sites ( cluster E ) , or was absent ( clusters F–G ) . Re-analysis of cluster E revealed BRG1 localized at varying distances between −370 and +440 bp with respect to the TSS ( Figure 5E , clusters A–F ) , although it was present only on the downstream nucleosome in cluster G . BRG1 therefore appears to bind to the last ( −1 ) nucleosome before the nucleosome-depleted region ( NDR ) at the TSS and the first downstream ( +1 ) nucleosome although the distance between these nucleosomes is variable as previously described ( Fenouil et al . , 2012 ) . The metaprofile of all BRG1-occupied TSS showed preferential location of BRG1 at −200 bp and +72 bp relative to the TSS ( Figure 5F ) . Integration of RNA Polymerase II ( Pol II ) ChIP-seq data from 501Mel cells showed the peak of promoter paused Pol II located immediately upstream of the BRG1-bound +1 nucleosome . BRG1 therefore binds a large subset of sites both at the TSS and non-TSS regions as paired BRG1-bound nucleosomes separated by variable distances . Annotation of BRG1-occupied sites identified 7168 potential target genes in a window of ±10 kb around the TSS and >10 , 000 potential target genes in a window of ±30 kb ( Figure 5—figure supplement 1E ) . Integration with RNA-seq data showed that 34% of genes down-regulated upon BRG1 silencing had at least one BRG1-occupied site at ±10 kb , while this figure attained 47% using the ±30 kb window . In contrast , only 12% and 18% of up-regulated genes were associated with BRG1-occupied sites using these windows . BRG1-associated and down-regulated genes were enriched in pigmentation , DNA replication , and glucose and fatty acid metabolism ( Figure 5—figure supplement 1F ) . BRG1-associated up-regulated genes were involved in angiogenesis and transcription regulation , notably expression of a set of ZNF TFs that appear to be co-ordinately regulated by BRG1 ( Supplementary file 3 ) . Together these data show that BRG1 is recruited by many TFs in 501Mel cells to regulate the expression of a large set of target genes . If BRG1 acts as a cofactor for MITF , they should co-localize on the genome . Integration of BRG1 and MITF ChIP-seq data showed that of >16 , 000 MITF occupied sites , 5867 were co-occupied by BRG1 and were located in both inter and intragenic regions , but enriched at promoters ( Figure 6A ) . Co-localization can be observed at the MITF locus itself , where MITF occupied multiple sites , several of which co-localized with BRG1 and melanocyte H3K27ac-marked enhancers throughout this locus ( Figure 6B ) . A similar situation was seen at the DCT locus , where MITF and BRG1 co-localized at H3K27ac marked melanocyte-specific enhancers . 10 . 7554/eLife . 06857 . 011Figure 6 . Sites co-occupied by BRG1 and MITF . ( A ) Read density clustering of BRG1 and MITF at MITF-occupied loci reveals co-occupied sites with different profiles and their distribution with respect to genomic features . ( B ) UCSC screenshots showing MITF and BRG1 occupancy over the MITF and DCT loci highlighting their co-localization along with that of melanocyte-specific H3K27ac-marked enhancers . A GFP ChIP was performed as a negative control . ( C ) Meta-profiles of BRG1 occupancy in clusters A–D of panel A . The lower panel shows a UCSC screenshot of MITF and BRG1 occupancy over the TYR locus highlighting the binding of MITF between two BRG1-occupied nucleosomes . ( D ) Meta-profile showing occupancy by BRG1 , MITF , and Pol II at the TSS . The binding profiles of each factor BRG1 and Pol II around the TSS are schematised below the graph . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 01110 . 7554/eLife . 06857 . 012Figure 6—figure supplement 1 . TF binding motifs at MITF and BRG1-bound sites . ( A ) Ontology analysis of genes associated with BRG1-MITF-co-occupied sites . ( B ) Identification of transcription factor binding motifs at BRG1-MITF-occupied sites . MEME-ChIP identified several motifs enriched at these sites . The most prominent are the E-box and M-box corresponding to MITF binding , SOX10 , CREB , and ETS1 as well as motifs with no known annotation . ( C ) Identification of transcription factor binding motifs at BRG1-occupied sites that are strongly enriched in siMITF senescent cells . ( C–D ) Identification of transcription factor binding motifs at MITF-occupied sites that are associated with down and up-regulated genes after MITF silencing . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 01210 . 7554/eLife . 06857 . 013Figure 6—figure supplement 2 . MITF co-localizes with SOX10 . ( A ) Pie chart illustrating distribution of SOX10-occupied sites with respect to genomic features . ( B ) DNA sequence motifs at SOX10 occupied sites . ( C ) Ontology analysis of genes with associated SOX10 binding sites . ( D ) Clustering at SOX10 occupied sites illustrates co-localization with MITF . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 013 At co-occupied sites , MITF binds between two BRG1 peaks as observed at the TYR locus ( Figure 6C ) . BRG1 peaks were either positioned almost symmetrically with respect to MITF or displayed shoulders of density either upstream or downstream presumably corresponding to additional , but less well-positioned BRG1-occupied nucleosomes ( Figure 6C ) . Genes associated with co-occupied sites are involved in previously defined critical functions of MITF such as lysosome biogenesis ( Ploper et al . , 2015 ) , pigmentation , cell cycle , apoptosis , and DNA damage response ( Figure 6—figure supplement 1A , and Supplementary file 4 ) . Co-occupied sites showed enrichment not only in E and M-boxes for MITF , but also in motifs for SOX10 , CREB , and ETS1 ( Figure 6—figure supplement 1B ) , TFs that may cooperate with MITF and BRG1 to regulate associated target genes . In addition , we also identified a set of 1065 TSS , where BRG1 occupies the nucleosomes flanking the TSS and MITF binds the NDR ( Figure 6D ) . Genes associated with these promoters were involved in lysosome formation , cell cycle , and RNA processing and splicing ( data not shown ) . As the SOX10 motif was so highly represented at MITF-BRG1 sites and that it physically interacts with BRG1 complexes ( Weider et al . , 2012 ) , we performed SOX10 ChIP-seq in 501Mel cells to profile its genomic occupancy . Almost 6000 SOX10-occupied sites were detected , but unlike MITF they were mainly located at distal regulatory elements with only around 80 sites at the proximal promoter . ( Figure 6—figure supplement 2A ) . SOX10 recognition motifs were strongly enriched at SOX10 bound sites some of which were organised as degenerate palindromes ( Figure 6—figure supplement 2B ) . Recognition motifs for TCF/LEF , MITF , and TFAP2A were also enriched at these sites . Genes associated with these sites were enriched in several ontology terms associated with cell motion/adhesion , cell morphogenesis and organisation ( Figure 6—figure supplement 2C ) . We integrated the SOX10 data with that of MITF , BRG1 , and H3K27ac . SOX10 and MITF co-occupied 3674 sites where MITF was located either 3′ or 5′ to SOX10 ( Figure 6—figure supplement 2D ) . SOX10 , MITF , BRG1 , and H3K27ac were found at 1929 sites ( Figure 7A , clusters A–D ) while varying levels of SOX10 , MITF and BRG1 were found at an additional ≈2000 sites ( cluster E ) , whereas SOX10 and BRG1 co-occupied 1159 sites ( cluster F ) and 972 sites displayed essentially only SOX10 ( cluster G ) . The above data define a specific organisation of MITF-associated regulatory elements ( MAREs ) where MITF alone or with SOX10 bind between two BRG1-bound nucleosomes . This analysis also defined 1929 MAREs at active melanocyte/melanoma enhancer elements associated with 1511 genes involved in pigmentation , cell motility and adhesion , actin cytoskeleton organisation , and signalling ( Supplementary file 4 ) . 10 . 7554/eLife . 06857 . 014Figure 7 . MITF , SOX10 , and YY1 co-localize with BRG1 . ( A ) Read density clustering illustrating regions of co-localization of the indicated factors and melanocyte H3K27ac . The meta-profiles for the indicated clusters are shown to the right illustrating the binding of MITF and SOX10 between the two BRG1-bound and H3K27ac-marked nucleosomes . The binding profiles are schematised on the right of the figure . ( B ) Co-localization of MITF and SOX10 on YY1-occupied sites . ( C ) Similar to panel A , read density clustering illustrates MAREs with co-localization of H3K27ac , BRG1 , MITF , and SOX10 at YY1-bound regions . A meta-profile is shown to the right along with a schematic representation of these sites . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 01410 . 7554/eLife . 06857 . 015Figure 7—figure supplement 1 . Identification of YY1-TFAP2A associated MAREs . ( A ) Clustering at YY1 occupied sites illustrates co-localization with BRG1 . ( B ) Clustering at TFAP2A occupied sites illustrates co-localization with BRG1 and H3K27ac . ( C ) Identifcation of TFAP2A sites co-occupied by MITF . The re-clustering shown to the right illustrates that MITF may be localized at various distances up- or down-stream of TFAP2A . ( D ) Clustering identifies MAREs where MITF co-localizes with YY1 , TFAP2A , BRG1 , and H3K27ac . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 015 We also found a large overlap between BRG1 and YY1 genome occupancy ( Figure 7—figure supplement 1A ) . Despite the fact that YY1 ChIP-seq was performed in a distinct melanoma line ( Li et al . , 2012 ) , we identified 2853 sites where YY1 co-localized with MITF and 3060 SOX10-YY1 co-occupied sites ( Figure 7B ) . All three proteins co-localize at 1070 sites together with BRG1 and 530 of these sites are marked with H3K27ac ( Figure 7C ) . These data define a set of MAREs comprising YY1 , BRG1 , MITF , and SOX10 located at active melanocyte/melanoma enhancers . These 530 MARES are associated with genes involved in pigmentation , melanocyte development , cell motion as well as apoptosis ( data not shown ) . TFAP2A is a TF involved in melanoma ( Huang et al . , 1998 ) and normal melanocyte function ( Brewer et al . , 2004; Van Otterloo et al . , 2010 ) and co-localizes with MITF at promoters involved in pigment cell differentiation in primary human melanocytes ( H Seberg , E van Otterloo , and RA Cornell , manuscript in preparation ) . In addition , TFAP2A binding motifs are enriched at BRG1 bound sites ( Figure 5—figure supplement 1B ) . In agreement with this , of the 13 , 693 TFAP2A-bound sites in human melanocytes , 6432 sites co-localized with BRG1 and H3K27ac defining a large set of active TFAP2A-bound regulatory elements ( Figure 7—figure supplement 1B ) . Also , at more than 1000 sites , MITF was located at varying distances 5′ or 3′ to TFAP2A , and YY1 and TFAP2A colocalized at 4819 sites ( Figure 7—figure supplement 1C and data not shown ) . Integrative analysis identified 762 YY1-TFAP2A sites also occupied by BRG1 and MITF and marked by H3K27ac , defining a set of TFAP2A-YY1-containing MAREs ( Figure 7—figure supplement 1D ) . Analysis of the genes associated with these MAREs showed enrichment in those involved in lumen organisation , melanosomes , and transcriptional regulation ( data not shown ) . To determine whether MITF and/or SOX10 actively recruit BRG1 to the genome , we performed BRG1-ChIP-seq following siRNA-mediated silencing of MITF , SOX10 or siLuciferase ( Luc ) as control . At the TYR locus , siMITF silencing led to the specific loss of BRG1 at a promoter proximal site ( * in Figure 8A ) , whereas at the upstream site where MITF co-localized with SOX10 little change was observed ( arrowheads in Figure 8A ) . In contrast , siSOX10 and consequent loss of both SOX10 and MITF ( see above ) , diminished BRG1 at all sites . Interestingly , BRG1 occupancy was observed throughout the region between these sites . siSOX10 silencing led to BRG1 loss across the entire region suggesting that it is recruited to the SOX10/MITF sites and then may spread across the locus . A similar situation was observed at the CIT locus encoding a MITF-regulated gene essential for cell cycle progression , with two downstream regulatory elements comprising MITF and SOX10 sites ( * and arrowheads in Figure 8B ) . At both sites , BRG1 was strongly reduced upon siMITF or siSOX10 silencing . A global analysis identified sites showing a strong and preferential loss of BRG1 following , siSOX10 ( clusters B and G–H , Figure 8C ) , siMITF ( clusters E and F ) or both , ( cluster D ) , defining sites where MITF and SOX10 independently or cooperatively recruit BRG1 ( Figure 8D ) . These data indicate that MITF and SOX10 actively recruit BRG1 either cooperatively or specifically to a large number of genomic loci . 10 . 7554/eLife . 06857 . 016Figure 8 . MITF and SOX10 actively recruit BRG1 . ( A ) UCSC screenshots illustrating occupancy over the TYR locus and highlighting the selective recruitment of BRG1 by MITF and/or SOX10 . * illustrates a region where BRG1 occupancy is diminished upon siMITF silencing , while the arrow indicates a region where BRG1 is lost upon siSOX10 silencing . ( B ) * illustrates a region where BRG1 occupancy is diminished upon siMITF silencing , while the arrow indicates a region where BRG1 is diminished upon both siMITF and siSOX10 silencing . ( C ) Read density clustering illustrating set of sites whose BRG1 occupancy is diminished following siMITF or siSOX10 silencing compared to siLuc . The meta-profiles of several representative clusters are shown to the right of the figure . ( D ) UCSC screenshots showing examples of BRG1 recruitment by SOX10 or MITF . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 01610 . 7554/eLife . 06857 . 017Figure 8—figure supplement 1 . BRG1 and MITF repress gene expression . ( A–B ) UCSC screenshots illustrating that MITF recruits BRG1 to regulatory elements of the SERPINE1 and IL24 loci . BRG1 is diminished at these genes in siMITF-silenced cells where these genes are activated . ( C ) BRG1 re-localizes over the genome in siMITF senescent cells . UCSC screenshot illustrating that BRG1 occupancy is strongly increased over the CCL2 locus in siMITF-silenced cells where this gene is strongly activated . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 017 While many genes associated with MAREs were down-regulated upon MITF silencing , MAREs may also be involved in gene silencing . Expression of SERPINE1 and IL24 , two SASP components , was strongly up-regulated in senescent siMITF cells ( see above ) . At both loci , BRG1 and MITF were present at either the promoter and/or distal regulatory regions and were lost upon siMITF coinciding with activation of their expression ( Figure 8—figure supplement 1A–B ) . These observations suggest that BRG1 recruitment by MITF at these genes acts to silence their expression . Furthermore , BRG1 was re-localized over the genome to new sites during this process for example to CCL2 that was strongly induced upon siMITF silencing ( Figure 8—figure supplement 1C ) . We also noted that different sequence motifs were enriched at MITF-bound sites associated with either up or down regulated genes ( Figure 6—figure supplement 1C–D ) . In particular , we observed that nuclear receptor half sites were enriched at loci associated with both classes . The finding of nuclear receptor half sites associated with MITF is unexpected as little is known about nuclear receptor function in melanoma although a therapeutic role of LXRb agonists has been demonstrated ( Pencheva et al . , 2014 ) . Nuclear receptors can therefore be added to the list of factors that may interact functionally with MITF . We next investigated whether on the other hand MITF occupancy is affected by BRG1 silencing using Cl8 cells where BRG1 loss did not affect expression of ectopic 3HA-tagged MITF ( see above ) . Anti-HA ChIP-qPCR at MITF sites not associated with BRG1 co-occupancy showed only a mild reduction in MITF occupancy ( Figure 9A , group A ) . In contrast , at many co-occupied sites , BRG1 silencing resulted in a marked increase in MITF occupancy . This was observed for example at the GPR110 , UVRAG , and SOX6 loci ( group B ) , where MITF binds to sites located between BRG1-bound nucleosomes ( Figure 9B ) . Increased occupancy is not however seen at all co-occupied sites since at the DCT and SOX10 loci MITF occupancy was not affected ( group C ) . Thus , BRG1 regulates the dynamics of MITF binding to a subset of sites10 . 7554/eLife . 06857 . 018Figure 9 . BRG1 controls dynamics of MITF binding . ( A ) ChIP-qPCR of 3HA-MITF in 501Mel-CL8 cells at the indicated loci following transfection with siLuc or siBRG1 . The protamine 1 locus ( PRM1 ) was used as a negative control . ( B ) UCSC screenshots illustrating binding of MITF between two BRG1-occupied nucleosomes at selected loci assayed by ChIP-qPCR in panel A . sThe GPR110-1 and GPR110-2 sites assayed in Panel A are indicated in panel B . ( C ) A model for regulatory elements in the melanocyte lineage . Melanocyte lineage enhancers comprise combinations of MITF , SOX10 , YY1 , and also TFAP2A and ETS1 ( not represented for simplicity . Note also that Pol II and the PIC are present at active enhancers where enhancer RNAs are made . For simplicity these are also not represented . ) bound to a nucleosome-depleted region . MITF but also these other factors recruit BRG1/PBAF to the nucleosomes flanking the combinations of transcription factors . BRG1/PBAF also occupies the nucleosomes flanking the TSS and a subset of these promoters further comprises a MITF binding site close to the TSS . DOI: http://dx . doi . org/10 . 7554/eLife . 06857 . 018 A comprehensive characterisation of the MITF interactome revealed its association with multiple complexes , including PBAF and NURF , TFIIIC , cohesins and multiple enzymes of the ubiquitin cycle . TFE3 , TFEB , and TFEC also co-purify with MITF raising the question of which sites are bound as MITF homodimers or MITF-TFE heterodimers and whether binding as homo- or heterodimers has functional consequences for transcription regulation . Mouse genetic studies have not revealed functions for the TFE factors in development of the melanocyte lineage ( Steingrimsson et al . , 2002 ) , however their role in melanoma has not been fully evaluated . A striking finding of our study is the interaction of MITF with multiple proteins involved in DNA damage and repair , suggesting that MITF may influence these processes . Nevertheless , we so far found no recruitment of MITF to DNA damage sites ( unpublished observations ) , the significance of these interactions therefore remains to be determined . Similarly , MITF associates with TFIIIC suggesting that it may regulate Pol III transcription . Alternatively , the presence of cohesin subunits in the interactome suggests that MITF may associate with TFIIIC at ‘Extra TFIIIC’ sites that organise chromatin structure in association with the cohesin complex ( Noma et al . , 2006; Kirkland et al . , 2013 ) . MITF also interacts with complexes involved in DNA replication . Assembly of DNA replication origins can be influenced by the presence of transcription regulatory elements and MYC that also interacts with MCM subunits , has been shown to directly influence DNA replication ( Dominguez-Sola et al . , 2007; Méchali , 2010 ) . Perhaps MITF interaction with these replication complexes facilitates replication origin assembly at MITF-occupied sites thereby coupling replication with transcription regulation in a melanocyte-specific manner . MITF interacts with BRG1 , but the related Brahma ( BRM ) protein was not detected , although both proteins are expressed in 501Mel cells ( Keenen et al . , 2010 ) . MITF interacts with a PBAF-like complex comprising CHD7 . We have so far been unable to co-localize BRG1 and CHD7 by ChIP-seq due to the lack of ChIP-seq grade CHD7 antibodies ( our unpublished data ) . In 501Mel cells , BRG1 regulates an extensive gene expression programme including MITF and SOX10 whose expression is lost upon BRG1 silencing and cells undergo growth arrest and senescence characterised by a SASP similar to that seen upon MITF silencing . This requirement for BRG1 to drive a large gene expression programme required for melanoma cell proliferation in vitro contrasts with its tumour suppressor function in human cancers including melanoma ( Hargreaves and Crabtree , 2011; Shain and Pollack , 2013; Bastian , 2014; Wang et al . , 2014 ) . BRG1 is also essential in Hermes-3A cells and the overlapping gene expression changes suggest that BRG1 acts as an MITF cofactor in these cells . Nevertheless , while the impact of BRG1 silencing is less extensive than in 501Mel cells the changes in gene expression lead to senescence and eventual cell death . As BRM is expressed in Hermes-3A cells , there is a potential for redundancy between BRG1 and BRM that could account for the less extensive effect , although it is important to note that this is not the case in 501Mel cells . BRG1 knockout in developing mouse melanocytes in vivo leads to complete loss of pigmentation and immunostaining failed to reveal the presence of Sox10 or Dct expressing cells in the hair follicle of the mutant animals . The absence of expression of these important markers indicates that there are no identifiable melanocytes in the Smarca4mel−/− hair follicles . This together with the arrested proliferation and cell death seen in melanoma cells and melanocytes in vitro , indicate that loss of pigmentation most likely results from an absence of melanocytes in these animals . Together our data identify BRG1 as an essential MITF cofactor in melanoma and melanocyte/melanoblast cells in vitro and in vivo . BRG1 binds extensively over the melanoma cell genome often at active H3K27ac-marked enhancers consistent with previous reports ( Hu et al . , 2011 ) . Nevertheless , we describe here a novel profile where BRG1 often binds as two peaks separated by 250–800 base pairs . These BRG1 sites define two classes of elements . The first is a subset of TSS , where BRG1 occupies the nucleosomes flanking the NDR with the Pol II pause site located immediately 5′ of the +1 nucleosome . This specific positioning of BRG1 at the TSS was not noted in several previous ChIP-seq studies ( De et al . , 2011; Euskirchen et al . , 2011; Ho et al . , 2009; Hu et al . , 2011; Yu et al . , 2013; Morris et al . , 2014 ) , however , Tolstorukov et al . ( 2013 ) reported that BRG1 occupies the nucleosomes flanking the TSS . Tolstorukov et al . also reported that inactivation of BRG1 does not affect the positioning of these two nucleosomes , but rather elicits a strong reduction in their occupancy . Nevertheless , the overall meta-profiles presented by Tolstorukov et al . , analogous to that reported here , overlooked two important features: ( 1 ) that many promoters show BRG1 occupancy of only the −1 or +1 nucleosome , and ( 2 ) that −1 or +1 nucleosome location is variable ( Fenouil et al . , 2012 ) . The second class corresponds to TF binding sites in promoter and enhancer elements . The observed variability in distances separating the BRG1-bound nucleosomes likely reflects the number of TFs bound in the intervening regions . For example , combinations of MITF , SOX10 , YY1 , TFAP2A as well as other TFs such as ETS1 bind between two BRG1-bound nucleosomes . At many of these sites the BRG1-bound nucleosomes are also marked by H3K27ac , thus defining MAREs active in regulating melanocyte lineage gene expression . As SOX10 , YY1 , TFAP2A , and ETS1 all have important roles in melanocytes and/or melanoma , the MAREs identified here define a combinatorial signature of TFs critical for gene regulation in this lineage ( Figure 9 ) . This is further underlined by the recent finding that a combination of MITF , SOX10 , and PAX3 can reprogram fibroblasts into functional melanocytes ( Yang et al . , 2014 ) . Ondrusova et al . reported an MITF-independent pro-survival role for BRG1 in melanoma cells ( Ondrušová et al . , 2013 ) . This observation is in agreement with our findings that BRG1 silencing affects expression of many more genes than MITF and that the binding motifs for a variety of factors other than MITF are enriched at BRG1-occupied sites . BRG1 is likely therefore to act as a cofactor for many other TFs accounting for its MITF-independent functions . Our data are consistent with the idea that MITF and other combinations of TFs bind the DNA between two BRG1-occupied and H3K27ac-marked nucleosomes . Such an organisation has been previously proposed based on extensive ChIP-seq profiling by the Encode consortium showing that TF binding sites are often combinatorial and correspond to GC-rich , DNaseI hypersensitive NDRs flanked by two positioned nucleosomes ( Wang et al . , 2012 ) . Our data support this idea through the identification of combinatorial MAREs and they extend it by showing that the nucleosomes flanking the TF binding sites are often bound by BRG1 and marked with H3K27ac ( Figure 9 ) . These regulatory elements show an analogous organisation to that of the TSS that also comprise a NDR encompassing variable numbers of TF binding sites and the PIC , flanked by BRG1-bound nucleosomes . Although examples of association between BRG1 and TFs have been previously described ( Trotter and Archer , 2008; Reisman et al . , 2009; Euskirchen et al . , 2011 ) , this specific configuration of BRG1 binding to nucleosomes flanking the TF binding sites has not been generally recognised . In yeast , it has however been reported that TF binding sites are rather flanked by nucleosomes bound by ISWI and CHD remodellers ( Zentner et al . , 2013 ) . While there have been many examples of co-localization between BRG1 and TFs , this is the first description of active genome-wide BRG1 recruitment . siRNA-mediated MITF or SOX10 silencing identified sites to which BRG1 was recruited either individually or cooperatively by these TFs . Moreover , we show that MITF associates with PBAF in the soluble nuclear fraction suggesting that they form a complex in the nucleoplasm and are recruited simultaneously to the chromatin to create the NDRs ( Figure 9C ) . Chromatin remodelling has been shown to facilitate TF binding , a good example being that of TAL1 where BRG1 repositions nucleosomes flanking its binding sites ( Hu et al . , 2011 ) . In contrast , we observed that MITF occupancy of many co-occupied sites either shows little change or is increased following siBRG1 silencing suggesting that BRG1 regulates the dynamics of MITF binding . TF binding to chromatin is often extremely dynamic with ChIP capturing only a snapshot of their occupancy ( Voss and Hager , 2014 ) . The increased MITF occupancy seen in ChIP upon BRG1 silencing suggests an increase in its time of residence at occupied sites . Several mechanisms may explain the increased MITF binding upon BRG1 loss . The human genome comprises many more consensus E-box elements than are occupied by MITF . The excess of consensus as well as degenerate E-boxes could potentially act as a sink thus limiting the pool of MITF for binding to functional sites . One function of BRG1-driven dynamics may therefore be to limit MITF sequestration at non-functional sites and ensure a pool of MITF for binding to functional sites . Alternatively , BRG1 may be required to establish the NDRs for MITF binding , for example after mitosis , or at specific stages of the cell cycle . As siBRG1 cells are post-mitotic and senescent , the increased MITF occupancy may reflect a new steady state where MITF remains more stably bound to the NDRs established while BRG1 was still present compared to cycling cells where the NDRs are established and erased in a more dynamic manner . Irrespective of the underlying causes , our results indicate that BRG1 regulates dynamic MITF interactions with chromatin . 501Mel cells cultured in RPMI 1640 medium ( Sigma , St Louis , MO , USA ) supplemented with 10% fetal calf serum ( FCS ) were transfected with a CMV-based vector expressing Flag-HA-tagged MITF and a vector encoding puromycin resistance . Transfected cells were selected with puromycin ( 3 μg/ml ) , and the expression of MITF verified by western blot using the MITF antibody ab-1 ( C5 ) from Neomarkers , the 12CA5 HA antibody ( Roche , Basel , Switzerland ) , or the M2 Flag antibody ( Sigma ) . Details of other cell culture are provided in Supplementary information . Hermes 3A cells were obtained from the University of London St Georges repository . Cell extracts were prepared essentially as previously described and subjected to tandem Flag-HA immunoprecipitation ( Drané et al . , 2010 ) . Cells were lysed in hypotonic buffer ( 10 mM Tris–HCl at pH 7 . 65 , 1 . 5 mM MgCl2 , 10 mM KCl ) and disrupted by Dounce homogenizer . The cytosolic fraction was separated from the pellet by centrifugation at 4°C . The nuclear soluble fraction was obtained by incubation of the pellet in high salt buffer ( final NaCl concentration of 300 mM ) and then separated by centrifugation at 4°C . To obtain the nuclear insoluble fraction ( chromatin fraction ) , the remaining pellet was digested with micrococcal nuclease and sonicated . Tagged proteins were immunoprecipitated with anti-Flag M2-agarose ( Sigma ) , eluted with Flag peptide ( 0 . 5 mg/ml ) , further affinity purified with anti-HA antibody-conjugated agarose ( Sigma ) , and eluted with HA peptide ( 1 mg/ml ) . The HA and Flag peptides were first buffered with 50 mM Tris–HCl ( pH 8 . 5 ) , then diluted to 4 mg/ml in TGEN 150 buffer ( 20 mM Tris at pH 7 . 65 , 150 mM NaCl , 3 mM MgCl2 , 0 . 1 mM EDTA , 10% glycerol , 0 . 01% NP40 ) , and stored at −20°C until use . Between each step , beads were washed in TGEN 150 buffer . Complexes were resolved by SDS-PAGE and stained using the Silver Quest kit ( Invitrogen , La Jolla , CA , USA ) . Mass-spectrometry was performed at the Taplin Biological Mass Spectrometry Facility ( Harvard Medical School , Boston , MA ) . Melanoma cell lines SK-Mel-28 and 501Mel were grown in RPMI 1640 medium ( Sigma ) supplemented with 10% FCS . 293T cells were grown in Dulbecco's modified Eagle's medium supplemented with 10% FCS and penicillin/streptomycin ( 7 . 5 μg/ml ) . Hermes-3A cells were grown in RPMI 1640 medium ( Sigma ) supplemented with 10% FCS , 200 nM TPA , 200 pM cholera toxin , 10 ng/ml human stem cell factor ( Invitrogen ) , 10 nM endothelin-1 ( Bachem , Bubendorf , Switzerland ) , and penicillin/streptomycin ( 7 . 5 μg/ml ) . All lentiviral shRNA vectors were obtained from Sigma ( Mission sh-RNA series ) in the PLK0 vector . The following constructs were used . shBRG1 ( TRCN0000015549 ) and shMITF ( TRCN0000019119 ) . In each case between 5 × 105 and 1 × 106 cells were infected with the indicated shRNA lentivirus vectors and all experiments were performed at least in triplicate . siRNA knockdowns were performed with the corresponding ON-TARGET-plus SMARTpools purchased from Dharmacon Inc . ( Chicago , Il . , USA ) . Control siRNA directed against luciferase was obtained from Eurogentec ( Seraing , Belgium ) . siRNAs were transfected using Lipofectamine RNAiMax ( Invitrogen ) . Transient and stable transfections were performed with 5 μg of expression vectors and using FuGENE 6 reagent ( Roche ) following the manufacturer's instructions . Medium was replaced 24 hr and cells were collected 48 hr after transfection . Cells lysis was performed using LSDB 500 buffer ( 500 mM KCl , 25 mM Tris at pH 7 . 9 , 10% glycerol , 0 . 05% NP-40 , 1 mM DTT , and protease inhibitor cocktail ) . Up to 3 mg of whole cell extracts were diluted in LDSB without KCl to obtain a final concentration of 100 mM KCl and incubated for 12 hr with 5 μg of specific antibody and 50 μl Slurry of protein-G sepharose ( GE Healthcare ) . Beads were washed 3 times in LSDB 300 , twice in LSDB 150 , and boiled in Laemmli buffer before protein separation by SDS–PAGE . For flag immunoprecipitations , extracts were incubated with 50 μl Slurry of Anti-Flag M2-agarose affinity gel ( Sigma ) and washed similarly prior to elution with Flag peptide ( 0 . 5 mg/ml ) . Immunoblots were performed with the following antibodies: MITF ( MS-771-P; Interchim ) , BRG1 ( ab110641; Abcam , Cambridge , UK ) , HERC2 ( 612366; BD Transduction Laboratories , Sparks , MD ) , USP11 ( 3263-1; Epitomics , Burlingame , CA ) , USP7 ( #4833; Cell Signaling , Danvers , MA ) , TRRAP ( 2TRR-2D5; IGBMC ) , NEURL4 ( sc-243603; scbt , Santa Cruz , CA ) , actin ( 2D7; IGBMC ) , XRCC6 ( sc-17789; scbt ) , XRCC5 ( sc-5280; scbt ) , BAF170 ( A301-038A; Bethyl Laboratories , Montgomery , TX ) , BAF155 ( sc-10756; scbt ) , BAF250A ( sc-373784; scbt ) , BAF250B ( sc-32762; scbt ) , BAF200 ( ab56082; Abcam ) , BAF53A ( ab131272; Abcam ) , CHD7 ( ab31824; Abcam ) , BAF180 ( ab137661; Abcam ) , BAF60A ( #611728; BD Transduction Laboratories ) , BAF60B ( ab166622; Abcam ) , SOX10 ( ab155279; Abcam ) , CREB ( #06-863; Upstate Millipore , Molsheim , France ) . The Smarca4lox/lox and Tyr::Cre strains have been described previously ( Indra et al . , 2005; Delmas et al . , 2003 ) . Genotyping of F1 offspring was carried out by PCR analysis of genomic tail DNA with primers detailed in the respective publications . All animals were handled according to institutional and national guidelines and policies . BRG1 ChIP experiments were performed on native Mnase-digested chromatin . 5 × 107 to 5 × 108 freshly harvested 501Mel cells were resuspended in 2 ml ice-cold hypotonic buffer ( 0 . 3M Sucrose , 60 mM KCl , 15 mM NaCl , 5 mM MgCl2 , 0 . 1 mM EDTA , 15 mM Tris–HCl [pH 7 . 5] , 0 . 5 mM DTT , 0 . 1 mM PMSF , protease inhibitor cocktail ) and cytoplasmic fraction was released by incubation with 2 ml of lysis-buffer ( 0 . 3M sucrose , 60 mM KCl , 15 mM NaCl , 5 mM MgCl2 , 0 . 1 mM EDTA , 15 mM Tris–HCl [pH 7 . 5] , 0 . 5 mM DTT , 0 . 1 mM PMSF , PIC , 0 . 5% ( vol/vol ) IGEPAL CA-630 ) for 10 min on ice . The suspension was layered onto a sucrose cushion ( 1 . 2 M sucrose , 60 mM KCl , 15 mM NaCl , 5 mM MgCl2 , 0 . 1 mM EDTA , 15 mM Tris–HCl [pH 7 . 5] , 0 . 5 mM DTT , 0 . 1 mM PMSF , PIC ) and centrifuged for 25 min at 4700 rpm in a swing rotor . The nuclear pellet was resuspended in digestion buffer ( 0 . 32M sucrose , 50 mM Tris–HCl [pH 7 . 5] , 4 mM MgCl2 , 1 mM CaCl2 , 0 . 1 mM PMSF ) and subjected to Micrococcal Nuclease digestion for 5 min at 37°C . The reaction was stopped by addition of EDTA and suspension chilled on ice for 10 min . The suspension was cleared by centrifugation at 10 , 000 rpm ( 4°C ) for 10 min and supernatant ( chromatin ) was used for further purposes . Chromatin was digested to around 80% of mono-nucleosomes as judged by extraction of the DNA and agarose gel electrophoresis . SOX10 and 3HA-MITF ChIP experiments were performed on 0 . 4% PFA-fixed chromatin isolated from 501Mel and Cl8 cells , respectively according to standard protocols as previously described ( Strub et al . , 2011 ) . ChIP-seq libraries were prepared as previously described and sequenced on the Illumina Hi- seq2500 as single-end 50-base reads ( Herquel et al . , 2013 ) . After sequencing , peak detection was performed using the MACS software ( [Zhang et al . , 2008] http://liulab . dfci . harvard . edu/MACS/ ) . Peaks were then annotated with GPAT ( Krebs et al . , 2008 ) using a window of ±10 kb ( or as indicated in the figures ) relative to the transcription start site of RefSeq transcripts . Global clustering analysis and quantitative comparisons were performed using seqMINER ( [Ye et al . , 2011] http://bips . u-strasbg . fr/seqminer/ ) and R ( http://www . r-project . org/ ) . The public human foreskin melanocyte H3K27ac data were taken from the Geo entry GSM958157 . De novo motif discovery on FASTA sequences corresponding to windowed peaks was performed using MEME-ChIP . Motif correlation matrix was calculated with in-house algorithms using JASPAR database . mRNA isolation was performed according to standard procedure ( Qiagen kit , Venlo , Holland ) . qRT-PCR was carried out with SYBR Green I ( Qiagen ) and Multiscribe Reverse Transcriptase ( Invitrogen ) and monitored by a LightCycler 480 ( Roche ) . Detection of Actin gene was used to normalize the results . RNA-seq was performed essentially as previously described ( Herquel et al . , 2013 ) . Gene ontology analyses were performed using the functional annotation clustering function of DAVID ( http://david . abcc . ncifcrf . gov/ ) . Primers for RT-qPCR and ChIP-qPCR were designed using Primer 3 and are listed in Supplementary file 5 . Searching of known TF motifs from the Jaspar 2014 motif database at BRG1-bound sites was made using FIMO ( Grant et al . , 2011 ) within regions of 200 bp around peak summits , FIMO results were then processed by a custom Perl script which computed the frequency of occurrence of each motif . To assess the enrichment of motifs within the regions of interest , the same analysis was done 100 times on randomly selected regions of the same length as the BRG1 bound regions and the results used to compute an expected distribution of motif occurrence . The significance of the motif occurrence at the BRG1-occupied regions was estimated through the computation of a Z-score ( z ) with z = ( x − µ ) /σ , where: − x is the observed value ( number of motif occurrence ) , − µ is the mean of the number of occurrences ( computed on randomly selected data ) , − σ is the standard deviation of the number of occurrences of motifs ( computed on randomly selected data ) . The source code is accessible at https://github . com/slegras/motif-search-significance . git . Biopsies of dorsal skin were isolated and fixed overnight in 4% paraformaldehyde , washed with PBS , dehydrated , paraffin embedded , and sectioned at 5 μm . For antigen retrieval , the sections were incubated with 10 mM sodium citrate buffer , within a closed plastic container placed in a boiling waterbath , for 20 min . Sections were permeabilised with 3 × 5 min 0 . 1% Triton in PBS , blocked for 1 hr in 5% skim milk in PBS , and incubated overnight in 5% milk with primary antibodies . The following antibodies were used: goat anti-Dct at dilution of 1/1000 ( Santa Cruz Biotechnology , sc-10451 ) and rabbit anti-Sox10 , at 1/2000 ( Abcam , ab155279 ) . Sections were washed 3 × 5 min 0 . 1% Triton in PBS , and incubated with secondary antibodies , Alexa 488 donkey-anti-goat , and Alexa 555 donkey-anti-rabbit ( Invitrogen ) for 2 hr . Sections were subsequently incubated with 1/2000 Hoechst nuclear stain for 10 min , washed 3 × 5 min in PBS , dried and mounted with Vectashild . The senescence-associated β-galactosidase staining kit from Cell signaling technology ( Beverly , MA , USA ) was used according to the manufacturer's instructions to histochemically detect β-galactosidase activity at pH 6 .	Melanocytes are pigment-producing cells found primarily in the skin . Many of the genes that help these cells to develop are also thought to affect the development of melanomas: an aggressive form of skin cancer that originates in these cells . One such gene encodes a protein called MITF . This protein binds to DNA and regulates genes that control the development , survival , and spread of melanocytes; it is also linked to the invasive properties of melanomas . The MITF protein works together with partner proteins to control numerous genes , activating some while inhibiting others , by binding to nearby stretches of DNA that act as regulatory elements . Its interactions are therefore widespread and complex . Now , Laurette , Strub et al . have used techniques called tandem affinity purification and mass spectrometry to identify the proteins that interact with MITF . This investigation found many new protein partners for MITF , including proteins involved in DNA damage , repair , and replication . MITF also associates with two proteins—one of which is called BRG1—that are involved in modifying how tightly DNA is packaged inside cells . DNA wrapped around proteins is known as chromatin , and if chromatin is tightly packed , the genes in that stretch of DNA cannot be easily accessed or activated . Removing BRG1 from melanocytes and melanoma cells caused the cells to die or stop growing . When BRG1 was removed from developing mouse embryos , melanocytes failed to form . Further investigation revealed that MITF , together with another protein , localize BRG1 to sites in the melanocyte's DNA to open up the chromatin and regulate nearby genes . Furthermore , Laurette , Strub et al . report that BRG1 binds to many such elements in a characteristic manner , in which two BRG1 proteins flank the stretch of DNA bound by MITF and several other key DNA-binding proteins that together regulate many aspects of melanocyte and melanoma cell physiology . Laurette , Strub et al . have therefore revealed many details about the molecules that activate genes in melanomas and melanocyte cells , as well as the interactions between these molecules . The results could also help researchers to understand how the BRG1 protein organises chromatin packing in other cell types .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "biochemistry", "and", "chemical", "biology" ]	2015	Transcription factor MITF and remodeller BRG1 define chromatin organisation at regulatory elements in melanoma cells
The Drosophila tracheal system is a branched tubular network that forms in the embryo by a post-mitotic program of morphogenesis . In third instar larvae ( L3 ) , cells constituting the second tracheal metamere ( Tr2 ) reenter the cell cycle . Clonal analysis of L3 Tr2 revealed that dividing cells in the dorsal trunk , dorsal branch and transverse connective branches respect lineage restriction boundaries near branch junctions . These boundaries corresponded to domains of gene expression , for example where cells expressing Spalt , Delta and Serrate in the dorsal trunk meet vein–expressing cells in the dorsal branch or transverse connective . Notch signaling was activated to one side of these borders and was required for the identity , specializations and segregation of border cells . These findings suggest that Tr2 is comprised of developmental compartments and that developmental compartments are an organizational feature relevant to branched tubular networks . The importance of temporal- and position-specific gene expression to regional specialization is established for many tissues . It may be universal . In contrast , the roles and contributions of determined cell lineages are not as clear and may not be generally shared . In Drosophila , there are tissues that are dependent on cell lineages - examples include neurons in the developing brain ( reviewed in Spindler and Hartenstein , 2010 ) and the developmental compartments that produce the adult cuticle ( reviewed in Crick and Lawrence , 1975 ) . However , there are also organs such as the salivary gland , gut and tracheal system of the Drosophila embryo that form from groups of non-mitotic cells ( reviewed in Kerman et al . , 2006 ) . The cells of these organs are assigned to their roles after they have exited the cell cycle , and the programs of tubulogenesis and branching that generate them proceed without further cell divisions . Even the complex network of tracheal branches in the embryo forms in this way , and without an apparent role for defined lineages ( Samakovlis et al . , 1996 ) . The tracheal primordia , which have 80–90 cells , have been estimated to derive from approximately ten blastoderm-stage cells ( Campos-Ortega and Hartenstein , 1985 ) . The primordia are first evident in the lateral embryonic ectoderm during stage 10 ( 4–5 hr after egg laying ( AEL ) ) as twenty groups of cells with distinct morphology , and in early in stage 11 , they form discrete pits ( Wilk et al . , 1996 ) . Three post-blastoderm cell divisions have occurred by early stage 11 , but there are no additional mitoses until L3 , when some tracheal cells re-initiate cell cycling ( Guha and Kornberg , 2005; Guha et al . , 2008; Sato and Kornberg , 2002; Weaver and Krasnow , 2008 ) . Beginning at embryo stage 11 , the tracheal pits invaginate , elongate and mold into more complex shapes , following a stereotyped program that generates the major tracheal branches ( Samakovlis et al . , 1996 ) . Although analysis of random clones revealed no consistent relationships between position and kinship that would suggest a role for cell lineage in assigning cells to a particular tracheal branch ( Samakovlis et al . , 1996 ) , there is evidence for region-specific and stage-specific gene expression , and mutant phenotypes suggest that these genes have essential roles in the morphogenetic processes that generate the branches . For example , trachealess ( trh ) is expressed by pit cells and in all tracheal progenitors , and in trh mutants , pits do not form and there is no apparent tracheal development ( Isaac and Andrew , 1996; Wilk et al . , 1996 ) . Spalt ( sal ) is expressed in the dorsal part of the tracheal primordium that will form the dorsal branch ( DB ) and dorsal trunk ( DT ) , and in sal mutants , the dorsal primordium expresses genes such as unplugged , which is normally only expressed ventrally , and its cells do not migrate normally ( Franch-Marro and Casanova , 2002 ) . These and other findings have been interpreted as dorsal to ventral transformations and as evidence that sal has a role in fate specification for particular branches ( Chen et al . , 1998; Kuhnlein and Schuh , 1996 ) . Mutants defective for knirps ( kni ) and for Notch signaling have major branching abnormalities suggestive of general and persistent requirements that begin at the earliest primary branching stages ( Chen et al . , 1998; Ghabrial and Krasnow , 2006; Ikeya and Hayashi , 1999; Llimargas , 1999; Steneberg et al . , 1999 ) . Although these studies support the idea that specialization and branch formation are dependent on region-specific expression of several fate-determining genes , the expression patterns of these genes have not been precisely correlated ( at cellular resolution ) with branching morphologies . There is evidence supporting the presence of a kni-dependent border between DT and DB cells in the embryo ( Chen et al . , 1998 ) , but it is not known whether the different branches , either alone or in combination , exist as distinct regions that develop with unique genetic addresses . Developmental compartments are regions whose constituent cells share a unique genetic address and are clonally isolated . They are polyclones ( Crick and Lawrence , 1975 ) - groups of cells that represent all the descendants of a small group of founder cells that grow but never mix with cells of other compartments or tissues . Compartments of the imaginal discs and abdominal histoblasts generate the epidermal and neuronal cells of the adult cuticle ( Chen and Baker , 1997; Garcia-Bellido et al . , 1973; Kornberg , 1981a; Morata and Lawrence , 1978 , 1979 ) . They do not include associated cells , such as the myoblasts and tracheal cells that develop together and in tight juxtaposition with the epithelial cells of the wing imaginal disc . The common ancestry of cells in the developmental compartments in the epithelia of imaginal discs is essential to the identity and function of these domains , and the adult cuticle does not develop normally if the compartment boundaries do not restrict cells to grow on one side or other ( Blair et al . , 1994; Diaz-Benjumea and Cohen , 1993; Kornberg , 1981a , b; Morata and Lawrence , 1975 ) , or do not properly delimit the expression of certain genes to the cells of a particular compartment ( Dominguez et al . , 1996; Tabata et al . , 1995 ) . The compartments are domains of gene expression and pattern whose geographical positions and limits are precisely defined . In the wing disc , the compartment borders set up and coincide in space with developmental organizers ( Diaz-Benjumea and Cohen , 1993; Tabata et al . , 1995 ) and juxtapose groups of cells with opposite developmental polarity ( Chen and Struhl , 1996; Chuang and Kornberg , 2000; Garcia-Bellido and Santamaria , 1972; Lawrence and Morata , 1976; Lawrence et al . , 2007; Tabata et al . , 1995 ) , but the generality of developmental compartments beyond the epithelial progenitors of the insect cuticle is uncertain . It is not known whether cell lineage domains that have been identified in other tissues , for example in the adult thoracic muscles ( Lawrence , 1982 ) and midgut ( Marianes and Spradling , 2013 ) , share these properties . The work described here investigated cell lineage parameters in the Drosophila tracheal system at the L3 stage . In contrast to the process that forms the tracheal system in the embryo , the pupal and adult tracheal systems develop as their constituent cells proliferate . As noted above , the first post-embryonic cell divisions in Tr2 occur during L3 when its branches begin to reorganize in preparation for metamorphosis . We undertook a classical clonal analysis of the branches of Tr2 and found evidence of coincident lineage and gene expression domains . At the beginning of L3 , the branches of Tr2 are sparsely populated by large cells ( Figure 1A ) . The DT has 16–18 , the DB 8 , and the transverse connective ( TC ) 5–7 ( Guha and Kornberg , 2005; Guha et al . , 2008; Lin , 2009; Weaver and Krasnow , 2008 ) ; there is no air sac primordium ( ASP ) . The other tracheal metameres are similarly constituted , but in early L3 , programs of cell division that are unique to Tr2 repopulate its branches with many cells and grow the ASP ( Figure 1A , B ) ( Guha and Kornberg , 2005; Sato et al . , 2008 ) . At late L3 , the DT has approximately 360 ± 40 cells ( Lin , 2009 ) . The DB also reinitiates cell cycling during L3 ( Weaver and Krasnow , 2008 ) . Despite the difference in cell numbers between Tr2 and the other metameres , the general overall size of the branches in each metamere remains similar as the L3 larva grows because the proliferating Tr2 cells are smaller . To better understand the growth dynamics of the Tr2 tracheoblasts during L3 , we induced random clones of marked cells in the DT and analyzed their size and distribution . 10 . 7554/eLife . 08666 . 003Figure 1 . Cell divisions in the second tracheal metamere during the third instar . ( A ) Drawings of the Tr2 branches before ( early L3 ) and after ( late L3 ) the onset of cell divisions; dorsal trunk ( DT ) , dorsal branch ( DB ) , visceral branch ( VB ) , spiracular branch ( SB ) , transverse connective ( TC ) , lateral trunk ( LT ) , air sac primordium ( ASP ) , region in which btl-Gal4 is expressed ( purple ) . ( right panel ) Nuclei in Tr2 visualized by fluorescence of GFP expressed by the btl-Gal4 driver . ( B ) Bar graph representing clones of indicated sizes in the DT at the indicated times post L2-L3 molt; the numbers of DTs examined , numbers of cells in the clones and color code are listed below . ( C ) Three representative DT clones showing cell proliferation . DOI: http://dx . doi . org/10 . 7554/eLife . 08666 . 003 We induced recombination in late embryos ( tracheoblasts do not divide between embryo stage 11 and >10 hr after the L2 to L3 molt ) and examined 1680 tracheal preparations for marked DT cells at various times during the L3 period . 185 marked patches were identified . At 0–2 hr after the L2-3 molt , 22/24 marked cells appeared to be isolated singles without marked neighbors , as would be expected prior to onset of proliferation and for the observed frequency of recombination ( 11% ) . At 16–18 hr , 80% were still singles , but at 18–20 hr , 69% were clones of two adjacent cells . This suggests that re-entry into the cell cycle was synchronous and that most DT cells divide according to a similar schedule . The clones formed contiguous patches ( Figure 1C ) , indicating that the DT cells do not tend to migrate or intermix with their neighbors; there was no apparent bias to their distribution within the DT . Divisions continued with approximate synchrony and with cycle time ( at 23°C ) of 10–12 hr ( Figure 1B ) . The time interval during L3 that follows the onset of mitotic cycling can therefore accommodate five divisions , and if all of the starting population contributes to growth , the five divisions are sufficient to produce all the cells in the DT of the wandering L3 . The maximum clone size is predicted to be 32 , suggesting that larger patches we identified represent >1 independent clone whose descendants grew together . To identify genes that are expressed in Tr2 branches during L3 , we screened approximately 1300 enhancer trap lines whose expression was uncharacterized , as well as a group of candidate genes that express in domains of the embryo ( Chen et al . , 1998; Franch-Marro and Casanova , 2002; Kuhnlein and Schuh , 1996; Llimargas , 1999; Thomas et al . , 1991 ) or larval ( Furriols and Bray , 2001; Pitsouli and Perrimon , 2010 , 2013 ) trachea for which either antibodies , enhancer trap lines , Gal4 lines , or enhancer reporter lines were available . Figure 2 shows expression patterns for seven genes that are expressed in discrete regions of the L3 Tr2 . These genes encode the transcription factors Sal , Cut and Kni , the Notch ligands Serrate ( Ser ) and Delta , the EGF ligand Vein , and Wingless ( Wg ) . Sal , Ser and Delta expression was detected in DT cells , but not in cells of the DB ( Figure 2A–C ) . There was a distinct expression limit of Ser and Delta-expressing DT cells near the TC junction ( Figure 2B , C ) , but the limit of Sal expression in this region was not as distinct ( Figure 2A ) . Ser was detected in the medial ASP and in the spiracular branch ( Figure 2B ) , and Delta expression was detected in the distal ASP ( Figure 2C ) . We detected Cut expression throughout the spiracular branch , TC and the LT , and in the proximal ASP ( Figure 2D ) , Kni expression in the DB , visceral branch and medial ASP ( Figure 2E ) , Wg expression in the spiracular branch ( Figure 2F ) and Vein expression throughout the visceral branch , ASP , TC and LT and in the proximal DB ( Figure 2G ) . 10 . 7554/eLife . 08666 . 004Figure 2 . Patterns of gene expression in the second tracheal metamere of third instar larvae . ( A-–C ) A Tr2 preparation stained for Sal , Ser and Delta; ( A ) Sal expression was specific to the DT; ( A’ ) anti-Sal antibody stained the DT , but not DB , TC or VB; ( B ) Ser was expressed in the DT , ASP and SB; anti-Ser antibody stained the DT but not DB , TC or VB ( B’ ) , the distal ASP ( B” ) and the SB ( B”’ ) ; ( C ) Delta was expressed in the DT and ASP; ( C’ ) anti-Delta antibody stained the DT specifically and the distal ASP; ( D ) Cut was expressed in the TC , SB , ASP and LT; anti-Cut antibody stained the TC ( D’ ) but not Ser-expressing DT cells ( D” ) , and the SB , LT and proximal ASP ( D”’ , D””; Cut expression in the myoblasts was erased manually in the image file to highlight tracheal expression only ) ; ( E ) Knirps was expressed in the DB , VB and ASP; anti-Knirps antibody stained nuclei in the DB and VB ( E’ ) but not Sal-expressing DT cells ( E” ) and stained the medial region of the ASP ( E”’ ) ; nuclei in ( D”’ ) , ( E’ ) and ( G” ) contained GFP ( btl-Gal4 UAS-nlsGFP ) ; ( F ) wg was expressed in the SB; ( F’ ) wg expression in the SB indicated by GFP fluorescence ( wg-Gal4 UAS-GFP ) but not in the TC ( btl-CD8:Cherry ) ; ( G ) vein was expressed in the proximal DB , TC , VB , ASP and LT; vein expression indicated by staining for vein-lacZ ( red ) in DB adjacent to Ser-expressing DT cells ( G’ ) , in the TC but not SB ( G” ) , and in the VB , TC and ASP ( G”’ , G”” ) . ( H , H’ ) Sal expression ( red , nuclear ) and Ser expression ( green , non-nuclear ) detected in DT cells ( arrows ) in upper level and lower level optical sections; arrowheads indicate 5 cells in the TC domain that express Sal but not Ser; nc ( node cell ) , DAPI-stained nuclei ( blue ) ; ( H” ) projection image showing Sal ( red ) and Cut ( green ) expressing TC cells; 2 cells in the TC domain ( arrowheads ) stained for both Sal and Cut . ( I ) Drawing with the DT , DB and TC expression domains indicated in purple , green and brown , respectively . The clonal analysis has not established whether the VB and SB are expression domains distinct from the TC . DOI: http://dx . doi . org/10 . 7554/eLife . 08666 . 004 Several borders that define domains of expression for these genes appeared to coincide . For example , at the junction of the DT and DB , the DT cells , which expressed Sal , Ser and Delta ( Figure 2A’ , B’ , C’ ) but not Kni , were precisely juxtaposed to Kni-expressing DB cells ( Figure 2E” ) . Although the location of the border appeared to vary between different preparations , we attribute these varied appearances to the way the preparations rotated as they were mounted for viewing and to the fact that proximal/distal position of the border is not precisely the same around the circumference of the branch junction; in no specimen did we detect cells that expressed Kni together with Sal , Ser , or Delta . Another common border appeared to separate the DT and TC . Cells on the trunk side of this border expressed Ser and Delta; cells on the other side expressed Cut and Vein . This border is more complex than the DT/DB border in several respects . First , the border extends a significant distance into the area of the DT and does not align with the DT/TC junction . Cells in this area of the DT are members of the TC lineage domain . Second , cells located in this part of the TC domain varied in number and size , and in contrast to the other Tr2 branches , there were large cells in this region in most wandering stage L3s . The number of large cells varied between 0–3 . This may be a consequence of delayed entry into mitotic cycling by these cells and the possibility that some of the wandering L3s that we analyzed had not yet reached the stage when these cells start to divide; but we have not established the reason for the variability . Third , whereas all cells in the DT lineage domain expressed Sal and Ser at high levels and no cells in the TC lineage domain expressed either gene at high levels , in some samples , low level expression of Sal was detected in 1–2 TC domain cells that were adjacent to the DT domain on the “lower layer” next to the wing disc ( Figure 2H , H’ , H” ) . All cells in the region of the TC domain in the DT expressed Cut and Vein . We note that there are also small regions of gene expression or lineage ambiguities at the anterior/posterior ( A/P ) compartment borders of the wing and eye-antennal imaginal discs ( Blair , 1992; Morata and Lawrence , 1979 ) , and suggest that the TC is also a lineage domain despite the apparent ambiguity for Sal at the DT/TC boundary . The TC lineage domain is defined by expression of Cut and Vein and by lack of expression of Ser and Delta , and it includes a portion that is physically within the DT . Based on this model , we suggest that Tr2 has distinct domains of gene expression that correlate with the DB , DT and TC . No distinct domains of gene expression that delimited the ASP from the TC were observed . Figure 2I depicts these DT , DB and TC domains . To characterize the role of Ser and Delta in the L3 Tr2 , we monitored Notch signal transduction with the NRE-lacZ Notch pathway transcriptional reporter ( Figure 3A ) ( Furriols and Bray , 2001 ) . Previous reports describe Notch signaling and Notch reporter expression in branching morphogenesis and in specifying the number of fusion cells during embryo tracheal development ( Ghabrial and Krasnow , 2006; Ikeya and Hayashi , 1999; Llimargas , 1999; Steneberg et al . , 1999 ) , and Notch signaling has been described to be generally present at tracheal branch junctions of L3 trachea ( Furriols and Bray , 2001 ) and has been characterized in the spiracular branches ( Pitsouli and Perrimon , 2013 ) . Studies of Notch signaling in the Tr2 metamere have not been reported . We examined stages of embryo development subsequent to fusion of the dorsal trunk ( post stage 16 ) , and detected expression of NRE-lacZ at both the DT/DB and DT/TC junctions ( Figure 3B ) . In the L3 Tr2 , NRE-lacZ expression was also detected in the DT/DB and DT/TC junctions , as well as in the ASP , in the TC adjacent to the spiracular ( Figure 3C , D ) and in 1–2 cells of the visceral branch proximal to the TC ( not shown ) . Notch signaling in the ASP is activated by Delta that is expressed in ASP-associated myoblasts ( Huang and Kornberg , 2015 ) ; we did not investigate the function of Delta or Serrate expression by ASP cells or the source of the activating ligand for Notch activation in the TC or visceral branch . 10 . 7554/eLife . 08666 . 005Figure 3 . Discrete regions of Notch activation in the second tracheal metamere . ( A ) Drawing of a late L3 Tr2 metamere with red areas indicating the regions that express the Notch reporter NRE-lacZ; ( B ) Tr2 of a stage 14–16 embryo with tracheal nuclei marked with GFP fluorescence ( btl-Gal4 UAS GFP ) , stained with anti-β-galactosidase antibody to identify cells expressing NRE-lacZ at DT/DB and DT/TC junctions ( arrows ) ; ( C , C’ ) NRE-lacZ expression ( red ) at DT/DB and DT/TC junctions of L3 Tr2; Low ( D ) and high magnification ( D’ D” ) images showing NRE-lacZ expression in the ASP , TC and VB; ( E ) Double stained preparation shows coincidence of NRE-lacZ and Cut expression in the TC; ( F-–F” ) images showing coincidence of NRE-lacZ and Knirps expression in the DB; ( G ) image showing juxtaposition of NRE-lacZ expression in the DB and TC with Delta; ( H ) image showing sal expression juxtaposed with NRE-lacZ expression in the DB and overlapping with NRE-lacZ expression in the TC; ( I–I” ) images showing coincidence of NRE-GFP and vein-lacZ expression at the DT/DB and DT/TC junctions; ( J–J” ) images showing juxtaposition of DB cells that express high levels of NRE-lacZ with cells that express the Notch target Hindsight; ( K ) the Notch target Fzr is expressed in DB cells adjacent to cells that express high levels of NRE-lacZ . DOI: http://dx . doi . org/10 . 7554/eLife . 08666 . 005 At the DT/DB and DT/TC junctions , NRE-lacZ expression coincided precisely with the boundaries that are defined by the expression domains of Cut , Kni , Delta and Sal ( Figure 3E–H ) . All of the NRE-lacZ expressing dorsal branch cells expressed Kni; all of the NRE-lacZ expressing TC cells expressed Cut . These results show that boundaries that define the DB , DT and TC gene expression domains are sites of Notch signaling . In the DB , expression of NRE-lacZ was highest in the cells that abut the Ser/Delta/Spalt expressing DT cells , and it decreased with increasing distance from the boundary . Similarly , expression of NRE-lacZ in the “TC domain” was highest in the cells that are in the DT , and expression decreased with increasing distance from the boundary . Expression of vein and of the Notch targets Hindsight and Fizzy-related appeared to correlate with the level of Notch activation in the proximal DB ( Figure 3I–K ) . vein expression was highest in the cells with the most NRE-GFP expression , but expression of Hindsight and Fizzy-related in the DB was not detected in the cells with highest levels of Notch activation . These results suggest that Notch signaling may pattern the proximal DB . We analyzed cell growth behavior in the DT , DB and TC by inducing marked clones and mapping their distribution . Similar clonal analysis studies of the wing imaginal disc revealed that in different discs , clones occupied varied locations and produced varied shapes in the wing , indicating that the descendants of particular single cells do not generate designated areas ( Bryant , 1970; Garcia-Bellido et al . , 1973 ) . The clone borders were “wiggly” except at compartment borders where they were straight ( reviewed in Lawrence and Struhl , 1996 ) . Although the tracheal branches are tubes , not epithelial sheets , we were able to map clones in the DT , DB and TC . Most of the clones arose in the DT ( as expected because of the greater relative number of founder cells ) , and the number of cells around its circumference was large enough that we were able to evaluate the contours of DT clones . We generated marked clones using eight different regimens that varied clone type ( e . g . , “standard flipout” , dual flipout clones , MARCM , and M MARCM ) as well as time and length of induction ( Table 1 ) . The clone frequency for regimens A and F were low and few specimens had more than one marked patch . Clone frequency for regimens B-E , G , H was greater: most DTs had clones , most DTs with clones had an average of more than one discrete patch of marked cells , and because many marked areas included more than 32 cells ( the period during L3 when the cells divide after clone induction limits the number of mitoses to a maximum of five ) , these areas must be the descendants of more than one founder cell and represent more than one clone . Nevertheless , the clonal patches we observed in the DT , DB and TC appeared to have located randomly in these branches , and clones in the DT appeared to have “wiggly” borders except at the boundaries that delimit the DT , DB and TC expression domains . Among the 361 specimens we analyzed that had marked cells , 202 marked patches were identified that lined a portion of these borders . Ten example specimens are shown in Figure 4 and the others are in the Figure 4—figure supplement 1–7 . The varied size and location of the clones in Figure 4 is evident in these examples that were selected to show representative clones that abut the DT/DB ( Figure 4A–F ) and DT/TC ( Figure 4G–J ) borders . These borders were identified by morphology , by expression of NRE-lacZ ( Figure 4B , G ) , and in Figure 4 ( C , D ) , by the juxtaposition of two clones that had been independently generated and differentially marked , and precisely correlated with Ser expression . 10 . 7554/eLife . 08666 . 006Table 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 08666 . 006 Regimen Marking system # Specimens # Marked DT HS stage HS DT/DB border clones DT/TC border clones DT side DB side DT DB DT side TC side A GFP , NRE lacz 106 40 48-50h AEL 8' 35° 3 0 5 0 B GFP , NRE-lacz 59 24 24-32h AEL 5' 37° 7 1 8 0 C GFP 29 11 24-32h AEL 5' 37° 2 1 1 1 D MARCM , α-Delta 52 43 4-6h AEL 60' 38° 1 3 4 5 E MARCM , α-Cut 48 34 4-6h AEL 60' 38° 6 2 6 5 F M MARCM , α-Ser 251 30 4-6h AEL 30' 38° 3 2 2 3 G GFP & LacZ , α-Ser 119 114 ( GFP ) 79 ( LacZ ) * 24-26h AEL 15' 37° 11 11 6** 8 0 H GFP & LacZ 116 61 ( GFP ) 34 ( LacZ ) * 24-26h AEL 6' 37°9 5 0 6 1 TOTAL 780 361 42 25 6 40 15 Genotype A: NRElacz/hsFLP; actin>y >GAL4 , UAS-GFP/ ; /MKRS B: NRElacz/hsFLP; actin>y >GAL4 , UAS-GFP/ ; /MKRS C: hsFLP/Y; actin>y >GAL4 , UAS-GFP/ ; /MKRS D: hsFLP122 , tubGAL4 , UAS-NLS-GFP/ ; tubGal80 FRT40a/FRT40a; / E: hsFLP122 , tubGAL4 , UAS-NLS-GFP/ ;tubGal80 , FRT40a /FRT40a; / F: hsFLP , tubGAL4 , UAS-GFP-NLS; / ;RpS17 , tubGal80 , FRT80a/FRT80a G: hsFLP/ orY; actin>y >GAL4 , UAS-GFP/ ;actin>stop>lacZ-NLS/MKRS H: hsFLP/ orY; actin>y >GAL4 , UAS-GFP/ ;actin>stop>lacZ-NLS/MKRS * # DTs with clones at or near the DB and TC borders ** specimens with independently marked DT and DB clones10 . 7554/eLife . 08666 . 007Figure 4 . Marked clones that meet but do not cross into or from the dorsal trunk . ( A-–D ) Marked clones in the DT line the DT/DB junction . ( A ) large patch of GFP-expressing cells abuts the DT/DB junction; Cut expression in TC ( red ) ; clone induction by regimen Experiment E ( Table 1 ) . ( B ) several GFP expressing clones in DT ( Expt . A ) , one of which abuts the DT/DB junction defined by NRE-lacZ expression ( red ) . ( C ) several GFP expressing clones in DT ( white arrows ) , one of which abuts the DT/DB junction and is juxtaposed to a LacZ-expressing DB clone; a single cell clone expressing GFP in the TC ( yellow arrow ) abuts the DT/TC junction and a GFP-expressing TC clone ( * ) ; Ser expression ( blue ) ; ( Expt . G ) . ( D ) LacZ-expressing DT clone abuts the DT/DB junction and is juxtaposed to a GFP-expressing DB clone; Ser expression ( blue ) ; ( Expt G ) . ( E , F ) GFP expressing DB clones that abut the DT/DB junction ( Expt . F ) . ( G–J ) GFP expressing clones that abut the DT/TC junction that was also defined by expression of NRE-lacZ ( G; Expt . B ) , Cut ( H; Expt . E ) , Delta ( I; Expt . D ) and Ser ( J; Expt . F ) . ( B , G , H ) DAPI ( blue ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08666 . 00710 . 7554/eLife . 08666 . 008Figure 4—figure supplement 1 . Dorsal trunk clones at the DT/DB border . GFP-expressing clones induced by regimens A-H; arrows point to DT/DB junction . ( A , B ) NRE-LacZ ( red ) ; ( C , H4 , 5 ) DAPI ( blue ) ; ( D ) anti-Delta ( red ) ; ( E ) anti-Cut ( red ) ; ( F ) anti-Ser ( red ) ; the abnormal morphology of the DT/DB junction was characteristic of the Minute genotype; ( F1 ) all M cells in the clone expressed Ser; ( F1’ ) red channel only showing limit of Ser expression at DT/DB junction . ( G , H ) Dual clone regimens produced clones that expressed GFP ( green ) , LacZ ( red ) or both ( yellow ) ; clones appear in the projection images in ( H1 , H2 ) to be juxtaposed at DT/DB junction but are not . DOI: http://dx . doi . org/10 . 7554/eLife . 08666 . 00810 . 7554/eLife . 08666 . 009Figure 4—figure supplement 2 . Dorsal branch clones at the DT/DB border . GFP-expressing clones induced by regimens B-H; arrows point to DT/DB junction . ( B ) NRE-LacZ ( red ) ; ( C ) DAPI ( blue ) ; ( D ) anti-Delta ( red ) ; ( E ) anti-Cut ( red ) ; ( G , H ) Dual clone regimens produced clones that expressed GFP ( green ) , LacZ ( red ) or both ( yellow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08666 . 00910 . 7554/eLife . 08666 . 010Figure 4—figure supplement 3 . Dorsal trunk and dorsal branch clones abutting the same DT/DB border . GFP- and LacZ-expressing clones induced by regimen G; arrows point to DT/DB junction . Regimen G produced clones that expressed GFP ( green ) , LacZ ( red ) or both ( yellow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08666 . 01010 . 7554/eLife . 08666 . 011Figure 4—figure supplement 4 . Dorsal trunk clones at the DT/TC border . GFP-expressing clones induced by regimens A-H; arrows point to DT/TC junction . ( A , B ) NRE-LacZ ( red ) ; ( C ) DAPI ( blue ) ; ( D ) anti-Delta ( red ) ; ( E ) anti-Cut ( red ) ; ( F ) anti-Ser ( red ) ; ( F1 , F1” ) images of upper- and bottom-most optical sections; ( F1’ ) higher magnification image from ( F ) showing coincidence of GFP and Ser expression; the abnormal morphology of the DT/TC junction was unique to this sample; all M cells in the clone expressed Ser; ( G , H ) Dual clone regimens produced clones that expressed GFP ( green ) , LacZ ( red ) or both ( yellow ) . Panel ( D3 ) , which is the same specimen as panel ( D1 , Figure 4—figure supplement 1 ) , is a projection image of sections representing one surface that illustrates the GFP labeled DT clone at the DT/TC border . DOI: http://dx . doi . org/10 . 7554/eLife . 08666 . 01110 . 7554/eLife . 08666 . 012Figure 4—figure supplement 5 . Transverse connective clones abutting the DT/TC border . GFP-expressing clones induced by regimens C , D , E , F , H; arrows point to DT/TC junction . DAPI ( blue ) ; ( D ) anti-Delta ( red ) ; ( E ) anti-Cut ( red ) ; ( F ) anti-Ser ( red ) ; ( H ) Dual clone regimen produced clones that expressed GFP ( green ) and LacZ ( red ) . ( D3 ) is also shown in Figure 4 ( I ) , ( F1 ) is also shown in Figure 4 ( J ) ; both are included here to show separate projection images of upper and lower layers . DOI: http://dx . doi . org/10 . 7554/eLife . 08666 . 01210 . 7554/eLife . 08666 . 013Figure 4—figure supplement 6 . Large marked cells in the TC domain of the DT . Panel labels refer to marking regimens for clones; arrows indicate the one or two large marked cells in the TC domain and ( * ) indicates node cells . ( A ) NRE-LacZ ( red ) ; ( D ) anti-Delta ( red ) ; ( E ) anti-Cut ( red ) ; ( F ) anti-Ser ( red ) ; ( G , H ) dual clone regimens produced clones that expressed GFP ( green ) , LacZ ( red ) or both ( yellow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08666 . 01310 . 7554/eLife . 08666 . 014Figure 4—figure supplement 7 . Marked patches of cells that straddle the DT/DB and DT/TC borders . Panel labels refer to marking regimens for DT/DB clones ( upper panels ) and DT/TC clones ( lower panels ) ; borders indicated by arrows . ( A , B ) NRE-LacZ ( red ) ; ( D ) anti-Delta ( red ) ; ( E ) anti-Cut ( red ) ; ( F ) anti-Ser ( red ) ; ( G , H ) dual clone regimens produced clones that expressed GFP ( green ) , LacZ ( red ) or both ( yellow ) . DT/DB: ( E2 ) Most GFP-expressing cells in this patch are in the DB , 2–4 are in DT . ( G4 ) GFP-expressing patch contains approximately 8 cells in the DT and a larger number in the DB; an independent LacZ-expressing ( red ) clone is in the DT; merging of red and green fluorescence from different optical sections ( yellow ) . ( H1 ) Patch contains 3–4 GFP-expressing DB cells and a larger number in DT , which also has cells expressing both GFP and LacZ ( yellow ) . ( H3 ) Patch contains 3–4 DT and a larger number of DB cells expressing GFP . DT/TC: ( A1 also in Figure 4—figure supplement 1; H2 also H2 upper panel ) . In ( A1 , E1 , E4 , F1 , H1 , H2 , H3 , H4 , H5 ) , the patch of GFP-expressing DT cells included one large cell in TC domain . ( D1 , D1’ ) Projections of optical sections at different planes; a GFP-expressing clone defined the DT/TC border from DT side in ( D1 ) and included a large TC domain cell in ( D1’ ) . ( E2 ) GFP-expressing DT patch included one upper layer and one lower layer large TC domain cell . ( E3 ) GFP-expressing patch in the DT contiguous with 6–8 GFP-expressing TC cells . ( G1 ( also G1 upper panel ) ) One GFP-expressing DT cell next to one GFP-expressing TC cell and one LacZ-expressing cell ( red ) on other surface . ( G2 ) One GFP-expressing DT cell next to one GFP-expressing TC cell . ( G3 ) Two GFP-expressing DT cells next to one GFP-expressing TC cell , and a LacZ-expressing TC clone ( red ) also abutting the border . DOI: http://dx . doi . org/10 . 7554/eLife . 08666 . 014 Of the 79 marked patches that lined the DT/DB border from one side or other , all had many cells ( Figure 4—figure supplement 1–3 ) . Although 26 other specimens had marked cells on both sides of the DT/DB border ( Figure 4—figure supplement 7 ) , these specimens had multiple large patches of marked cells that in combination averaged 34% ( ± 23% , std . dev . ) coverage of the DT surface . It is most likely therefore that the patches that meet on either side of the DT/TC border represent more than one clone and therefore are not inconsistent with the model we propose - that the DT/DB border is also a line of clonal restriction . 87 marked patches were identified that lined the DT/TC border from either side ( Figure 4—figure supplement 4–6 ) . All 40 DT and 15 of the TC clonal patches had multiple cells , but as noted above , most specimens had 1–3 large cells in the TC domain within the DT , and 1–2 of these were marked in 32 of the 87 . 15 specimens were identified that had marked cells on both sides of the DT/TC border ( Figure 4—figure supplement 7 ) ; in 13 of these , the only marked TC cell was a single large one . Because the number of specimens with marked DT patches at the border ( 55 ) was similar to the number of specimens with marked TC cells ( 62 ) , we conclude that the number of cells on either side of the DT/TC border at the time of recombination was approximately the same . And because the number of specimens with marked cells on both sides of the DT/TC border ( 15/102 ) relative to those with clones only on one side ( DT only ( 40 ) TC only ( 47 ) = 87/102 ) is consistent with the expected number of independent clones in the DT and TC domains , we suggest that the DT/TC border is also a line of clonal restriction . To investigate which functions might be served by the genes that are expressed specifically in the DT , DB and TC domains , we analyzed clones that either lacked sal function or ectopically expressed sal . In contrast to marked control clones in the DT , which integrated with their unmarked neighbors and were not morphologically distinct ( Figure 5A ) , clones that lacked sal function appeared to sort out from their neighbors and to bulge abnormally from the plane of the trunk tube ( Figure 5B–F ) . The sal mutant cells expressed Kni and vein ( Figure 5C , D ) but did not express either Delta or Cut ( Figure 5B , E ) . All the mutant cells expressed NRE-lacZ ( Figure 5F ) . We also analyzed clones that ectopically expressed kni , and they also appeared to sort out; and they expressed vein ( Figure 5G ) . This expression signature ( vein , Kni , Sal- , Delta- , Cut- ) is the same as that of DB cells , suggesting that sal function is required for DT identity and that without sal function or with Kni , they transform to DB identity . The sorting out phenotype is consistent with the idea that the presence of mutant cells created an ectopic juxtaposition of cells with different identities . Expression of NRE-lacZ in the mutant cells that lacked sal function ( Figure 5F ) suggests that Notch signaling was activated at the ectopic borders that formed where the two groups of cells abut , just as it does at the normal DT/DB junction . 10 . 7554/eLife . 08666 . 015Figure 5 . Mutant clones transform dorsal branch , dorsal trunk and transverse connective cells . ( A ) Control DT clone of GFP-expressing cells with no effect on morphology or expression of Delta; ( B-–F’ ) Clones of sal mutant DT cells generated bulges , apparently sorting out , and lacked Delta expression ( B , B’ ) , ectopically expressed vein-lacZ ( C ) and Knirps ( D ) , not normally expressed by DT cells , but not Cut ( E , E’ ) and activated the NRE-lacZ reporter ( F , F’ ) . ( G , G’ ) Clone ectopically expressing Knirps in the DT activated vein-lacZ expression . ( H , I ) Clones that ectopically expressed Sal in the DB reduced the diameter of the DB and induced neighboring wild type cells to express vein-lacZ ( H , H’’ ) and ectopically expressed Delta ( I , I’’ ) . ( J , K ) Clones that ectopically expressed Sal in the TC ectopically expressed Delta , which is not normally expressed in the TC ( J , J’ ) , but did not express Cut , which is normally expressed in the TC ( K , K’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08666 . 015 We also analyzed clones that ectopically expressed Sal , and noted three phenotypes associated with clones in the DB . First , cells adjacent to Sal-expressing cells strongly expressed vein ( Figure 5H ) , which is normally expressed at high levels in the DB only by cells near the DB/DT junction that activate Notch signaling ( Figure 5C , F ) . Second , Sal-expressing cells expressed Delta ( Figure 5I ) , which is normally expressed in Sal-expressing DT cells and not in the DB ( Figure 2A , C ) . Third , the diameter of the branch in the region affected by Sal-expressing clones was reduced in the cells that expressed Sal and greater in adjacent regions that expressed vein ( Figure 5H ) . This morphology has features in common with the normal DT/DB junction that has an expanded diameter at the proximal DB , vein-expressing side . Clones in the TC that over-expressed sal also up-regulated Delta ( Figure 5J ) and down-regulated Cut ( Figure 5K ) . The effects of ectopic sal expression in the DB and TC suggest that sal expression transformed their cells to a DT identity . The role of Notch at the DT/DB junction was characterized further by examining the role of Delta expression in the DT and of Notch signaling in the DB . Clones that ectopically expressed Delta in the DB non-autonomously activated vein-lacZ expression and Notch signaling in adjacent cells , and the affected regions had a diameter greater than the normal DB ( Figure 6A , C ) . vein was also expressed in DB clones that ectopically expressed NotchACT , and the morphology of the region with the mutant cells was abnormal and more characteristic of the expanded proximal DB ( Figure 6B ) . In contrast , Delta-expressing clones in the DT , which also activated Notch signaling in adjacent cells , did not alter the morphology of the DT ( Figure 6D ) . Because vein is normally expressed by DB cells in the expanded region at the DT/DB junction , these phenotypes suggest that ectopic over-expression of Delta in the DB transformed adjacent cells to a junction identity and that normally , Delta-activated Notch signaling at the DT/DB junction induces vein expression and morphological specialization . This conclusion is consistent with the effects of Delta over-expression on another morphological feature of the expanded region of the proximal DB . Anti-cadherin staining revealed that the lumen in this region of the wild type DB is branched ( Figure 6E ) . In addition to the lumen that extends the length of the DB , a short segment of lumen angles obliquely in the direction of the DT . A similar structure was detected in the expanded region associated with Delta-expressing clones ( Figure 6F ) . It was present in the cells adjacent to the clone , and its orientation was toward the clone ( mirror image with opposite polarity relative to the normal branch ) . 10 . 7554/eLife . 08666 . 016Figure 6 . Notch signaling is required by dorsal branch cells at the junction with the dorsal trunk . ( A ) Clone of DB cells that ectopically expressed Delta ( green ) increased the diameter of the DB and induced wild type neighbors to express vein-lacZ ( red ) , normally expressed in the DB only by cells at the DT junction . ( B ) Clone of DB cells ( indicated by brackets ) that ectopically expressed NotchACT ( green ) increased the diameter of the DB and activated vein-lacZ . ( C , D ) Clones of DB ( C ) and DT ( D ) cells that ectopically expressed Delta ( green ) activated NRE-lacZ expression in adjacent cells . ( E ) Image of the proximal DB stained with anti-DE-Cadherin antibody; bifurcated lumen indicated by arrow ( left panel ) and isolated and colored red ( right panel ) . ( F ) Delta expressing clone ( green ) in the DB induced a bifurcated luminal structure with orientation opposite to normal . ( F’ ) Higher magnification view of boxed area of ( F ) . ( G ) Expression of Notch-RNAi did not affect DT morphology or Delta expression ( magenta ) but reduced NRE-lacZ expression at both DT/DB and DT/TC junctions and reduced the diameter of the proximal DB ( arrow ) . Number of DT cells was reduced to levels characteristic of normal younger larvae whose proximal DB is expanded ( H ) . ( I-I” ) Clone that ectopically expressed kni , Notch-RNAi and GFP in the DT did not express Sal ( I , blue ) and did not sort out from DT cells ( I , I” ) . DT/DB compartment border ( CB ) indicated by dashed yellow line , borders of Sal expression by solid white line ( I ) and border of clone by dashed white line ( I' ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08666 . 016 We expressed Notch RNAi in trachea ( btl-Gal4 UAS-NotchRNAi ) and observed that the presence of Notch RNAi had no apparent effect on Delta expression in the DT , but it reduced NRE-lacZ expression to undetectable levels at both the DT/DB and DT/TC junctions , and altered the morphology of the DB ( Figure 6G ) . In the absence of Notch signaling , the proximal DB was not expanded and had the same diameter as other more distal regions of the DB . In the NotchRNAi-expressing trachea of wandering L3 larvae , the number and density of cells in the Tr2 branches appeared to be less than normal , and may reflect developmental delay . Comparison of trachea in younger , control L3 larvae ( 40–42 hrs post L2-L3 molt ) that had a similar density of cells in the DT revealed that the proximal DB was normally broadened at this earlier stage ( Figure 6H ) . We conclude that Notch signaling is required at the DB junction for normal morphogenesis . To investigate whether Notch signaling is necessary for lineage segregation at the compartment border , clones were induced that express kni , NotchRNAi and GFP . Figure 6I shows a DT clone in a preparation that was stained with antibodies against Kni and Sal ( which marks nuclei of DT cells ) . The nuclei in the clone lacked Sal staining ( similar to DB cells ) , and in contrast to clones that ectopically over-express kni in an otherwise normal genetic background ( Figure 5G ) , the clones that also down-regulate Notch do not sort out , but appear to integrate with their neighbors without forming a bulge . The clone also crossed into the DB . The discovery of developmental compartments in the Drosophila wing imaginal disc was a major advance that provided a novel conceptual basis to understand how groups of cells are programmed and how genetic programs regulate development independently of overt morphology ( Crick and Lawrence , 1975; Garcia-Bellido et al . , 1973 ) . Although the concepts were instrumental to understanding the function of the genes that segment the early embryo when these genes were later discovered , and although discoveries of developmental compartments in other imaginal discs and in the abdominal histoblast nests followed , their generality has remained an open question because the types of tissues in which they were found represent a limited subset . Studies that map gene expression in developing tissues have identified many examples in which expression is restricted to well-defined domains , and some studies have correlated these domains with cell lineage ( Buchon et al . , 2013; Marianes and Spradling , 2013 ) , but it is not known whether these cases satisfy other criteria that have been ascribed to developmental compartments - such as association with signaling centers and developmental polarity . Although we do not know if all these criteria are important distinctions , we prefer to reserve the term developmental compartments for contexts that have them . Using this strict definition , the studies reported here are the first to identify developmental compartments in an internal tubular organ . We did not initiate these studies expecting to find developmental compartments , but discovered them in the course of a thorough clonal analysis . It is possible that similar studies may also find them in other internal organs , and in particular in the vascular system of vertebrates that are also branched and have regions with distinct identities . This work unless otherwise indicated . btl-Gal4 UAS-nlsGFP ( Guha and Kornberg , 2005 ) NRE-lacZ ( Furriols and Bray , 2001 ) sal-Gal4 UAS-GFP ( Makhijani et al . , 2011 ) yw hsflp;actin>CD2>Gal4;UAS-nlsGFP ( Guha et al . , 2008 ) yw hsflp;actin>y+>Gal4 UAS-GFP;MKRS/TM6B hsFLP tubGAL4 UAS-nlsGFP;+/+;M ( 3 ) i55 tubGal80 FRT80B/TM6B ( Bloomington #42732 ) hsFLP tubGal4 UAS-nlsGFP;tubGal80 FRT40A/FRT40A tubGal80 ( Bloomington #1816 ) FRT40A Df ( 2L ) 32FP-5/CyO ( Df ( 2L ) 32FP-5 uncovers spalt and spalt-related ) ( Organista and De Celis , 2013 ) UAS-Sal;UAS-CD8GFP ( Bloomington #29715 ) UAS-Delta ( Bloomington #26694 ) UAS-knirps ( Chen et al . , 1998 ) UAS-N-RNAi ( NIG-Fly #3936-R2 ) ) UAS-Nact ( Hwang and Rulifson , 2011 ) vein-lacZ ( Bloomington #11749 ) yw hsflp;actin>y+>Gal4 UAS-GFP;NRE-lacZ yw hsflp;actin>y+>Gal4UAS-GFP;vein-lacZ actin>stop>nlslacZ ( Bloomington #6355 ) en GAL4 UAS-myr-mRFP , NRE-EGFP ( Bloomington #30729 ) fzr-lacZ ( Bloomington #12241 ) wg-Gal4 UAS-CD8:GFP/CyO ( Huang and Kornberg , 2015 ) btl-LHG/CyO;lexO Cherry-CAAX/TM6B ( Roy et al . , 2014 ) The enhancer trap lines ( collection generated by the U . Heberlein lab ) have insertions of the pGawB containing a Gal4 enhancer trap construct . For analysis of cell proliferation in the DT , FLP recombinase was induced by heat shock at 37° for 15 minutes during late embryonic stages , and animals were picked at the L2-L3 molt ( 55 hours later ) and aged for defined periods before being sacrificed for analysis . Flpout clones were obtained by heat shocking L1 larvae ( 24-32h AEL ) at 37° for 5 minutes or L2 larvae ( 48-50h AEL ) at 35° for 8 minutes . MARCM clones were induced in embryos with a 60 minute heat shock at 38° . MARCM clones in a Minute background were induced in embryos with a 30 minute heat shock at 38° . Dual clones ( that either express GFP or LacZ-NLS ) were induced by heat shocking L1 larvae ( 24-26hAEL ) at 37° for 6 or 15 minutes . Because the frequency of clones induced by these regimens was high and most specimens had multiple clones , statistical measures could not be used to calculate the probability that marked patches of cells represent the descendants of one or several founder cells . The Supplements to Figure 4 contain images of all specimens with clones at the DT/DB or DT/TC borders . Supplement 6 shows the 32 that have either one or two marked cells in the TC domain of the DT . Supplement 7 shows the 26 that have marked cells on both sides of the DT/DB border and the 15 that have marked cells on both sides of the DT/TC border ( 13 of which have marked cells in the TC domain limited to only one large cell ) . Clones ectopically expressing kni and Notch-RNAi were induced heat shock at 37° for 10 minutes ( genotype: yw hsflp/+;act5C>y+>Gal4 UAS-GFP/UAS-kni;UAS-Notch-RNAi/+ ) . Loss-of-function MARCM clones of salm , salr mutant cells were generated by subjecting L2 larvae ( 48-50h AEL ) to heat shock at 38° for 1hr; wandering L3 larvae were dissected for analysis . Ectopic expression clones that expressed salm , Delta , knirps , knirps and Notch-RNAi , or NotchAct were induced in 2-4 day old animals by heat shock at 37° for 10 minutes; wandering L3 larvae were dissected for analysis after 48 hours . Larvae were fixed in 4% PF followed by washing with 1x PBS with Ca++/Mg++ . These larvae were blocked in blocking buffer ( 1xPBS , 0 . 5% Donkey /Goat serum and 0 . 1% Triton X for one hour ) . Primary antibody staining was performed using above-mentioned buffer for ~8 hrs at RT or ~ 16h at 4 degrees followed by three washes of 15 minutes in blocking buffer at RT . Secondary antibody staining was done in the blocking buffer for 1-2 hours and washed three times with blocking buffer for 15 minutes followed by DAPI staining for 30 minutes and 2 additional washes in blocking buffer . Thereafter samples were stored in 1x PBS at 4 degrees prior to dissection and samples of larval Tr2 trachea or wing discs were mounted in Vectashield . Primary antibodies: mouse anti-Delta ( C594 . 9B , 1:200 ) , mouse anti-β-galactosidase ( 40A1 , 1:100 ) , mouse anti-Cut ( 2B10 , 1:100 ) , guinea pig anti-Knirps ( 1:200 , gift from J . Reinitz ) , rabbit anti-Sal ( 1:50 , gift from R . Schüh ) , rabbit anti-β-galactosidase ( 1:1000 ) , rat anti-Serrate ( 1:1000 ) , rat anti-Cadherin ( DCAD2 , 1:20 ) . Secondary antibodies , at ( 1:500 ) or ( 1:1000 ) : anti-mouse Alexa 488 , anti-mouse Alexa 555 , anti-mouse Alexa 647 , anti-Rat Alexa 555 , anti-Guinea pig Alexa 555 , anti-Rabbit Alexa 555 anti-Rabbit Alexa488 . Leica SPE confocal was used to image the slides . 20x or 40x oil immersion objectives were used . Z-projections of images were compiled with Image J and Adobe Photoshop was used to merge channels . The area covered by the marked patches in the DTs of the 26 projection images in upper panel of Figure 4—figure supplement 7 , were measured using ImageJ .	As a fruit fly develops , its cells may sort themselves into groups according to the type of cell that they will eventually become . Some groups form ‘developmental compartments’ that are separated by boundaries that cells cannot move across . All the descendants of a cell in a compartment will activate the same specific gene ( called a ‘selector’ gene ) that determines their identity and fate . Similar compartments also form in the developing hindbrains of mammals , but it is not clear how general this mechanism of tissue patterning is . Fruit fly larvae undergo a physical transformation called metamorphosis to become adult fruit flies . Here , Rao et al . discover that the cells in the developing airways ( or trachea ) of the larvae at the start of metamorphosis are organised into compartments . At this stage the cells in the trachea start to divide and grow to make the adult tracheal system . The experiments show that these cells do not spread from one main branch of the tracheal system into another . Instead , the cells cluster in locations where the different branches meet on either side of a straight boundary . The cells on each side of these boundaries activate different genes that regulate their identity and development . For example , cells in one branch of the system switch on a selector gene that makes a protein called Spalt . A pathway known as Notch signaling is activated by cells on the other side of a nearby boundary in a different branch of the tracheal system . This separation of Spalt production and Notch activation establishes a cell communication system that keeps the cells of the different compartments apart . Rao et al . ’s findings reveal a role for the Notch protein in regulating the organization of cells into compartments to form branches in fruit fly airways . A future challenge is to find out if Notch plays a similar role in other branched tissues , such as blood vessels .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2015	Developmental compartments in the larval trachea of Drosophila
TP53 is conventionally thought to prevent cancer formation and progression to metastasis , while mutant TP53 has transforming activities . However , in the clinic , TP53 mutation status does not accurately predict cancer progression . Here we report , based on clinical analysis corroborated with experimental data , that the p53 isoform Δ133p53β promotes cancer cell invasion , regardless of TP53 mutation status . Δ133p53β increases risk of cancer recurrence and death in breast cancer patients . Furthermore Δ133p53β is critical to define invasiveness in a panel of breast and colon cell lines , expressing WT or mutant TP53 . Endogenous mutant Δ133p53β depletion prevents invasiveness without affecting mutant full-length p53 protein expression . Mechanistically WT and mutant Δ133p53β induces EMT . Our findings provide explanations to 2 long-lasting and important clinical conundrums: how WT TP53 can promote cancer cell invasion and reciprocally why mutant TP53 gene does not systematically induce cancer progression . Cancer is driven by somatically acquired point mutations and chromosomal rearrangements , that are thought to accumulate gradually over time . Recent whole cancer genome sequencing studies have conclusively established that the tumor suppressor gene , TP53 , is the most frequently mutated gene in a wide range of cancer types . In tumors expressing wild-type ( WT ) TP53 gene , numerous experimental and clinical data have shown that viruses or cellular oncogene proteins target the p53 pathway , promoting abnormal cell proliferation . Altogether these data strongly suggest that defects in TP53 tumor suppressor activity are a compulsory step to cancer formation . In addition , ample data have also demonstrated that TP53 gene , whether WT or mutant , has a paramount biological and clinical role in response to cancer treatment ( Brosh and Rotter , 2009; Do et al . , 2012; Jackson et al . , 2012; Muller et al . , 2009 ) . We previously reported that the human TP53 gene expresses at least twelve p53 isoforms through alternative splicing of intron-2 ( Δ40 ) and intron-9 ( α , β , γ ) , initiation of transcription in intron-4 ( Δ133 ) and alternative initiation of translation at codon 40 ( Δ40 ) and codon 160 ( Δ160 ) . This leads to the expression of p53 ( α , β , γ ) , Δ40p53 ( α , β , γ ) , Δ133p53 ( α , β , γ ) and Δ160p53 ( α , β , γ ) protein isoforms that contain different transactivation domain , oligomerisation domains and regulatory domains ( for review Joruiz and Bourdon , 2016 ) . All animal models ( zebrafish , drosophila and mouse ) of p53 isoforms and experimental data in human cells of diverse tissue origins have consistently shown that p53 isoforms regulate cell cycle progression , programmed cell death , replicative senescence , viral replication , cell differentiation and angiogenesis . Several clinical studies reported that abnormal expressions of p53 isoforms are found in a wide range of human cancers including breast and colon cancers and that p53 isoforms are associated with cancer prognosis ( Joruiz and Bourdon , 2016 ) . However , it is not known whether they are just markers or play an active role in cancer formation and progression . Recently , we reported that Δ133p53β promotes cancer stem cell potential and metastasis formation in a xenograft mouse model ( Arsic et al . , 2015 ) . However , its physiopathological role , its molecular mechanism , its association with cancer progression and the effect of TP53 mutations on its activities have never been investigated . To date , it is currently thought that all the tumor suppressor activities associated with TP53 or the oncogenic activities associated with mutant TP53 gene expression are implemented by the canonical full-length p53 protein ( also named TAp53α or p53α ) . It is well established that TAp53α , is a transcription factor rapidly activated to restore and maintain cell integrity in response to alteration of cell homeostasis , while most TAp53α missense mutants have compromised tumor suppressor activities and have acquired transforming activities . Compelling evidence indicates that WT TAp53α protein , plays a pre-eminent role in preventing human cancer formation and progression to metastasis . In particular , WT TAp53α protein prevents cell migration and invasion which are the main traits of Epithelial to Mesenchymal Transition ( EMT ) , which itself involves global genome reprogramming that drives cancer progression to metastasis ( McDonald et al . , 2011; Suva et al . , 2013 ) . Conversely , most TAp53α missense mutant proteins promote EMT , inducing cancer cells to leave their primary tumor site and disseminate , leading to metastasis ( Bernard et al . , 2013; Gadea et al . , 2007; Muller et al . , 2009; Roger et al . , 2010 ) . However , there remain many enigmas concerning the association of p53 expression with cancer progression and treatment largely due to the natural existence of multiple alternative transcripts from the TP53 gene . ( De Roock et al . , 2009; Brosh and Rotter , 2009; Petitjean et al . , 2007; Soussi , 2007 ) . The main focus of this study is the role of Δ133p53β in cancer progression . We show that Δ133p53β is associated with poor prognosis in breast cancer , particularly in luminal A breast cancer patients who express WT TP53 . This indicates that Δ133p53β , which is a physiological gene product of TP53 , is a predictor of cancer cell invasion and increased risk of death . We corroborate the clinical analysis by demonstrating experimentally that Δ133p53β promotes invasion using a panel of breast cancer cell models expressing WT and mutant TP53 gene . Similar results were obtained in a panel of colon cancer cell lines . Depletion of endogenous mutant Δ133p53β prevents invasiveness despite unaltered strong expression of mutant TAp53α . Reciprocally , introduction of WT Δ133p53β promotes invasion of WT TP53 cells devoid of endogenous Δ133p53β expression . Altogether , our data indicate that cell invasiveness is regulated by Δ133p53β irrespective of whether they harbor a mutant or a WT p53 gene . Challenging the paradigm that WT TP53 always acts as a tumor suppressor protein , our data show that the WT TP53 gene can also promote invasiveness by encoding Δ133p53β . Hence , our data established that the cell decision is not exclusively determined by canonical full- length p53 protein ( TAp53α ) , either mutant or WT , but by the modular protein complex of p53 isoforms , which reprogram epithelial cells to pro-metastatic cells . The association of Δ133p53 ( α , β , γ ) isoforms with prognosis has not previously been investigated in breast cancer . As there is no antibody specific of Δ133p53α , Δ133p53β or Δ133p53γ protein isoform , we developed a nested RT-PCR method that specifically detects and identifies Δ133p53α , Δ133p53β or Δ133p53γ mRNA variants in tumor samples . Briefly , the RT-PCR method is based on the previously described RT-PCR method ( Khoury et al . , 2013 ) but its sensitivity and specificity have been improved ( A fully detailed protocol is provided in the 'Extended Experimental Procedure' section and the primer sequences are described in Table 1 ) . It enables to specifically detect and identify expression of Δ133p53α , Δ133p53β or Δ133p53γ mRNA variants in tumor samples . This nested RT-PCR analysis allows thus to investigate the association of Δ133p53α , Δ133p53β or Δ133p53γ expression with the clinico-pathological markers and/or patient clinical outcome . 10 . 7554/eLife . 14734 . 003Table 1 . Primers for specific amplification of Δ133p53 isoforms mRNAs by nested RT-PCR . The specific region ( exon or intron ) that each of the primers target is indicated ( Ex: exon; Int: intron ) , the sequences corresponding to the exon junction are underlined . ( F ) : Forward , ( R ) : Reverse . PCR fragment sizes ( bp ) corresponding to p53 isoforms are also highlighted . Quality of reverse-transcription is assessed by quantitative RT-PCR amplification ( SybrGreen ) of actin and p53 mRNAs ( primers are included ) DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 003p53 mRNA variantPCRPrimer name and targeted region5’ – 3’ sequenceAll p53 mRNAFor6 ( F ) ( Ex6 ) TTGCGTGTGGAGTATTTGGATRev7 ( R ) ( Ex7 ) TGTAGTGGATGGTGGTACAGTCAGAΔ133p53α1stD133F1 ( F ) ( Int4 ) TAGACGCCAACTCTCTCTAGRev10 ( R ) ( Ex10 ) CTT CCC AGC CTG GGC ATC CTT G2nd ( 670bp ) D133F2 ( F ) ( Int4 ) ACT CTG TCT CCT TCC TCT TCC TAC AGRDNp53 ( R ) ( Ex9/Ex10 ) CTC ACG CCC ACG GAT CTG AΔ133p53β1stD133F1 ( F ) ( Int4 ) TAGACGCCAACTCTCTCTAGRev10 ( R ) ( Ex10 ) CTT CCC AGC CTG GGC ATC CTT G2nd ( 700bp ) D133F2 ( F ) ( Int4 ) ACT CTG TCT CCT TCC TCT TCC TAC AGp53β ( R ) ( Ex9β ) TCA TAG AAC CAT TTT CAT GCT CTC TTΔ133p53γ1stD133F1 ( F ) ( Int4 ) TAGACGCCAACTCTCTCTAGRev10 ( R ) ( Ex10 ) CTT CCC AGC CTG GGC ATC CTT G2nd ( 670bp ) D133F2 ( F ) ( Int4 ) ACT CTG TCT CCT TCC TCT TCC TAC AGp53γ ( R ) ( Ex9/Ex9γ ) TCGTAAGTCAAGTAGCATCTGAAGGactinactin ( F ) ATCTGGCACCACACCTTCTACAATGAGCTGCGactin ( R ) CGTCATACTCCTGCTTGCTGATCCACATCTGC Total RNAs were extracted from 273 randomly selected primary breast tumors ( Figures 1A , B and Table 1 ) . Expression of Δ133p53α mRNA was detected in 97/273 ( 35 . 5% ) , Δ133p53β mRNA in 23/273 ( 8 . 4% ) and Δ133p53γ mRNA in 56/221 ( 25 . 3% ) of the breast tumors analyzed ( Table 2 ) . To verify that detection of Δ133p53β mRNA is associated with Δ133p53β protein expression , protein extracts from 8 large breast tumors were analyzed by SDS-PAGE and western-blot with KJC8 , a beta p53 specific antibody ( Figure 1—figure supplement 2 ) . Samples identified as expressing Δ133p53β mRNA also express Δ133p53β at the protein level , confirming the RT-PCR analysis data and that Δ133p53β protein is expressed in tumor samples . 10 . 7554/eLife . 14734 . 004Figure 1 . Breast cancer patients expressing Δ133p53β have a poor clinical outcome . ( A ) Schematic representation of human TAp53α , Δ133p53α , Δ133p53β and Δ133p53γ protein isoforms . The two transactivation domains [TADI ( in light purple ) and TADII ( in pink ) ] the Proline-rich Domain ( PrD ) , the DNA-Binding Domain ( DBD in orange ) , and the C-terminal domain comprised of the nuclear localization signal ( NLS; in yellow ) , the oligomerization domain ( OD; in blue ) , and the basic region ( BR; in violet ) are represented . The grey boxes correspond to the five highly conserved regions defining the p53 protein family . The amino-acid positions defining the different p53 domains are indicated . The C-terminal domains of p53β ( DQTSFQKENC ) and p53γ ( MLLDLRWCYFLINSS ) are indicated by green and pink respectively . The molecular weight ( kD ) of each p53 isoform protein is indicated . ( B ) Specific nested RT-PCR amplification of Δ133p53α , Δ133p53β and Δ133p53γ . Total RNAs from 273 primary breast tumors were provided by the Tayside tissue bank . The quality of RNA and the quality of reverse-transcription were assessed as described in Materials and methods section . All samples with low RNA quality and low quality of reverse transcription were discarded to minimize the number of false negative . Each different p53 cDNAs were specifically amplified by 2 successive nested RT-PCR ( 35 cycles each ) using 2 sets of the primer specific of each of the following p53 mRNA variants: p53 ( all isoforms ) , Δ133p53α , Δ133p53β and Δ133p53γ . The nested RT-PCR analysis was performed as described in the Materials and methods section and the extended experimental procedures . Primer sequences are provided in Table 1 . A representative subset of the nested RT-PCR analysis is shown . Tumor sample numbers are indicated . C: negative control , M: Molecular Markers . ( C ) ∆133p53β association with TP53 gene mutation status , HER2 and clinical outcome in breast cancer patients . ∆133p53β expression is associated with p53 mutation but is not frequently expressed in HER2 positive tumors as determined by univariate analysis ( upper table ) . ∆133p53β expression is associated with cancer progression and death in breast cancer by univariate analysis . ( Lower table ) ( D–F ) Only ∆133p53β is associated with cancer progression in breast cancer patients . Non-parametric Kaplan-Meier plots of disease-free survival in relation to Δ133p53α ( D ) , Δ133p53β ( E ) , Δ133p53γ ( F ) expression ( n = 273 ) . ‘I’ indicates censored cases on the curves . p-value is based on Kaplan-Meier log-rank analyses . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 00410 . 7554/eLife . 14734 . 005Figure 1—figure supplement 1 . Non-parametric Kaplan-Meier plots analysis of marker association with disease-free survival and overall survival in the entire cohort of primary breast cancers ( related to Figure 1 ) . Overall survival in relation to ∆133p53α ( A ) , ∆133p53β ( B ) and ∆133p53γ ( C ) . Disease-free survival in relation to breast cancer subtypes ( D ) , invaded node ( >0; E ) , TP53 mutation ( F ) , Grade ( G ) , tumor size ( H ) . Association of ∆133p53β expression with disease-free survival of WT TP53breast cancer patients ( no discrimination by subtype ) ( I ) . The Log-rank analysis and p values are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 00510 . 7554/eLife . 14734 . 006Figure 1—figure supplement 2 . Detection of endogenous beta p53 protein isoforms in breast tumors ( related to Figure 1 ) . The expression of Δ133p53β mRNA was determined by nested RT-PCR in the indicated breast tumors ( A ) . The expression of endogenous beta p53 protein isoforms ( 20 ug/well ) was analyzed by western blotting using KJC8 , a β-specific p53 antibody ( B ) . Actin was used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 00610 . 7554/eLife . 14734 . 007Table 2 . characteristics of breast tumors in the Tayside cohort . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 007VariablesPrimary breast tumors273median age follow up61 . 5 ( range 28 . 7–89 . 1 years ) 6 . 85 ( 0 . 29–13 . 7 years ) grade1 2 3 unknown25 85 159 4typeductal others220 53clinico‐pathological subtypetriple‐negatif luminal A luminal B HER2+ luminal/HER2+49 112 43 20 49size<20 mm >20 mm unknown91 180 2invaded lymph nodes ( node>0 ) negatif positif unknown132 140 1patient outcomedisease free recurrence alive death201 72 196 77p53Wild‐type mutant unknown196 64 13∆133p53α‐ +176 97∆133p53β‐ +250 23Δ133p53γ‐ + unknown165 56 52 Δ133p53β or Δ133p53γ are tightly associated with Δ133p53α expression [21/23 ( 91 . 3% ) ; 53/56 ( 94 . 6% ) , respectively] . Almost all tumors expressing Δ133p53β express Δ133p53α . However , only 25% of tumors expressing Δ133p53γ also express Δ133p53β ( 14/56 , Table 3 ) . The strong association of Δ133p53α with either the Δ133p53β or Δ133p53γ isoforms is consistent with the fact that the three mRNAs are initiated from the same promoter in intron-4 of the human TP53 gene ( Bourdon et al . , 2005 ) . 10 . 7554/eLife . 14734 . 008Table 3 . Repartition of Δ133p53 isoforms in the Tayside breast cancer cohort ( 2x2 table ) . ( A ) Δ133p53α X Δ133p53β . ( B ) Δ133p53α X Δ133p53γ . ( C ) Δ133p53β X Δ133p53γ ( − ) not detected , ( + ) amplified . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 008AΔ133p53α−+TotalΔ133p53β− +174 276 21250 23Total17697273BΔ133p53α−+TotalΔ133p53γ − +138 327 53165 56Total14180221CΔ133p53γ−+TotalΔ133p53β− +156 942 14198 23Total16556221 We analyzed the association of Δ133p53α , Δ133p53β , and Δ133p53γ with the current clinico-pathological subtypes of breast cancer: luminal A ( ER+ , PR+ , HER2- ) , luminal B ( ER+ , PR− , HER2- ) , luminal/HER2+ ( ER+ , PR+/- , HER2+ ) , HER2 positive ( ER- , PR- , HER2+ ) , and triple negative ( ER- , PR- , HER2- ) , according to the recommendations of St Gallen International expert consensus ( Goldhirsch et al . , 2013 ) . None of the three isoforms were associated with histological cancer type , number of metastatic lymph nodes , tumor size or grade ( Table 4 ) and neither Δ133p53α nor Δ133p53γ were associated with breast cancer subtypes . However , Δ133p53β was significantly less frequently expressed in HER2 positive and luminal/HER2+ tumors than in other breast cancer subtypes ( 2/69 , Fisher’s exact test , p≤0 . 042 , Figure 1C ) . TP53 mutations were identified in 64 tumors ( 64/260 , 24 . 6% ) . Δ133p53β expression was significantly associated with TP53 mutation status ( χ2 = 5 . 625 , p≤0 . 018 , Figure 1C ) , whereas Δ133p53α and Δ133p53γ expression levels were not ( Table 2 ) . In all tumors , Δ133p53α , Δ133p53β or Δ133p53γ carried the same TP53 gene mutation , indicating that all the corresponding mRNAs derive from cancer cells and not from normal stromal or inflammatory cells . 10 . 7554/eLife . 14734 . 009Table 4 . Univariate analysis of Δ133p53 isoforms expression in relation to clinical pathological markers . Tumor grade , cancer type ( ductal or others ) , tumor size ( >or < 20 mm ) were analyzed by Fisher’s t-test . Association with the number of invaded lymph nodes were analyzed by Mann-Whitney method . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 009gradetypesizenb invaded Lymph nodes1-23p valueductalothersp value<20 mm>20 mmp value∆133p53α− +76 3498 61p<0 . 21140 8036 17p<0 . 5662 29113 67p<0 . 39p<0 . 97∆133p53β− +102 8144 15p<0 . 54201 1949 4p<0 . 881 10167 13p<0 . 30p<0 . 60∆133p53γ− +75 2187 35p<0 . 25133 4632 10p<0 . 859 19105 37p<0 . 79p<0 . 78 Altogether , these data indicate that almost all tumors expressing Δ133p53β and/or Δ133p53γ also express Δ133p53α . However , Δ133p53β is distinct from Δ133p53α and Δ133p53γ , in that it is less frequently detected in breast tumors overexpressing HER2 and more frequently expressed in tumors expressing mutant p53 than in tumors expressing WT p53 . Therefore , Δ133p53α , Δ133p53β , and Δ133p53γ are not randomly expressed . The association of Δ133p53α , Δ133p53β , and Δ133p53γ expression with cancer patient outcome ( cancer progression/disease-free survival and overall survival/death ) was investigated by univariate analysis of our cohort of breast tumors . Δ133p53β was associated with cancer progression ( CP ) and death ( CP: χ2 = 5 . 953 , p≤0 . 015 , death: χ2 = 7 . 126 , p≤0 . 008 ) ( Figure 1C ) , while Δ133p53α and Δ133p53γ were not . This was further confirmed by Kaplan-Meier log-rank analyses ( Kaplan-Meier log-rank test , CP: χ2 = 6 . 221 , 1 df , p≤0 . 013; death: χ2 = 5 . 731 , 1 df , p≤ 0 . 017 , Figures 1D , E , F , Figure 1—figures supplement 1A , B and C ) . Importantly , Δ133p53β expression was also associated with cancer progression in WT TP53 breast cancer patients ( Kaplan-Meier log-rank test , CP: χ2 = 5 . 232 , p≤0 . 022 ) ( Figure 1—figure supplement 1I ) , indicating that the association between ∆133p53β expression and cancer progression is not due to TP53 mutation . ( Of note the Kaplan-Meier log-rank analyses were also performed for Δ133p53α or Δ133p53γ in WT TP53 breast cancer patients . Their respective expression was not found associated with clinical outcome of WT TP53 breast cancer patients . ) In addition to Δ133p53β expression , other well-established markers of cancer prognosis such as the breast cancer subtypes , TP53 mutation , tumor grade , lymph node metastasis ( absence or presence ) and tumor size ( >20 mm ) were , as expected , also associated with cancer patient outcome in our cohort ( Figure 1—figures supplement 1D–H ) . To clarify the univariate analyses and adjust for possible confounding variables , the association of overall survival and disease-free survival with Δ133p53β , the breast cancer subtypes ( luminal-A , luminal-B , HER2+ , triple-negative and luminal/HER2+ ) , TP53 mutation , tumor grade , lymph node metastasis and/or tumor size ( >20 mm ) were investigated using the multivariate Cox’s regression analysis . The cohort was composed of 112 luminal-A , 43 luminal-B , 49 luminal/HER2+ , 20 HER2+ and 49 triple-negative tumors . We observed that the luminal-A subtype was , as expected , the most significant independent predictor of disease-free survival ( HR = 3 . 09 , 95% CI , 1 . 78 to 5 . 38 , p<1 . 10–4 ) and overall survival ( HR = 4 . 15 , 95% CI , 2 to 8 . 62 , p<1 . 6 10–4 , Table 5A ) . Then , we examined by multivariate Cox’s regression analysis the level of inter-dependence between Δ133p53β , TP53 mutation , tumor grade , lymph node metastasis ( absence or presence ) and tumor size ( >20 mm ) in the luminal-A breast cancer population ( n = 112 ) . We determined by an Omnibus test of model coefficient that the Cox-regression model is fitted ( χ2 = 17 . 589 , 1 df , p<2 . 7 10–5 ) , indicating that the number of luminal-A tumors in our cohort is sufficient to support the conclusions . Among all the parameters included in the analysis ( Δ133p53β , TP53 mutation , tumor grade , lymph node metastasis and tumor size [>20 mm] ) , Δ133p53β stood out as the most significant independent predictor of cancer recurrence and death within the luminal-A subgroup regardless of TP53 mutation status or presence of lymph node metastasis ( Hazard ratio [HR] , 7 . 93; 95% CI , 2 . 52 to 25; p<4 . 31 10–4; [HR] , 3 . 29; 95% CI , 1 . 125 to 9 . 63; p<0 . 03 , Table 5B , respectively ) . It implies that at the time of surgery , detection of Δ133p53β in primary luminal-A breast tumors of patients devoid of lymph node would predict cancer progression . Of note , the multivariate Cox-regression analysis was also performed for Δ133p53α and Δ133p53γ but no significant association with the clinical outcome of luminal-A breast cancer patients was found . 10 . 7554/eLife . 14734 . 010Table 5 . Multivariate analysis of predictor . ( A ) Multivariate Cox’s Regression analyses utilizing the forward step-wise elimination method to determine the degree of inter-dependence between the breast cancer subtypes ( triple negative , Luminal A , Luminal/HER2+ , Luminal B , and HER2+ ) , Δ133p53β , TP53 mutation status , lymph node metastasis ( present versus absent ) , tumor size and tumor grade in relation to disease-free survival and overall survival . ( B ) Multivariate Cox’s Regression analyses utilizing the forward step-wise elimination method to determine the degree of inter-dependence between Δ133p53βTP53 mutation status , lymph node metastasis ( present versus absent ) , tumor size and tumor grade in the luminal A breast cancer patient population . The Fitted model was assessed by an Omnibus test . Hazard Ratio ( HR ) , 95% confidence interval ( CI ) , p values , number of iteration ( itr ) are indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 010An = 273omnibus test of model coefficientsitr . predictorχ2dfp-valueHR95% CIp-valueDisease-free survival1Luminal A17 . 38913 . 00E-053 . 091 . 785 . 381 . 00E-04Overall survival1Luminal A17 . 08813 . 50E-054 . 1528 . 621 . 60E-4Bn = 112omnibus test of model coefficientsitr . predictorχ2dfp-valueHR95% CIp-valueRecurrence1Δ133p53β17 . 58912 . 70E-057 . 932 . 5224 , 964 . 31E-04Death1Δ133p53β5 . 3112 . 10E-023 . 291 . 129 . 633 . 00E-02 Thus , Δ133p53β in luminal-A breast cancer , which predominantly expresses WT TP53 , enhances on average by 8 times the risk of recurrence and by 3 times the risk of death in an otherwise excellent prognostic group . This indicates that WT TP53 gene can be associated with cancer recurrence and death if it expresses Δ133p53β . The clinical study suggests that Δ133p53β may confer a more invasive phenotype to breast cancer cells . We then compared the ability of three different cancer cell lines to invade into type1- Collagen . We used: ( 1 ) luminal MCF7 cells ( WT TP53 and ER+ ) , which depend on estrogen and EGF ( Epidermal Growth Factor ) for their growth and are neither locally invasive nor metastatic in mouse models , ( 2 ) the triple-negative MDA-MB-231 ( mutant TP53-R280K , ER- ) , which was derived from a pleural effusion metastasis , and ( 3 ) its ‘D3H2LN’ variant ( mutant TP53-R280K , ER- ) , which was selected for their enhanced tumor growth and widespread metastasis in mice ( Jenkins et al . , 2005 ) . As expected , MCF7 cells were very weakly invasive and MDA-MB-231 D3H2LN cells showed a significantly higher invasiveness as compared to parental MDA-MB-231 cells ( Figure 2A ) . The difference between MCF7 cells and the two MDA-MB-231 cell lines may result from their TP53 status , since mutant TAp53α protein can activate cell migration , invasion and metastasis ( Gadea et al . , 2007 , 2002; Muller et al . , 2009; Roger et al . , 2010 ) . However , the difference in invasion between the two MDA-MB-231 cell types expressing the same mutant TP53 gene suggests the presence of another mechanism activated in the D3H2LN variant cell line . We examined whether this could be due to differential expression of the p53 isoforms . By quantitative real-time-PCR ( RT-qPCR ) using primers and probe ( TaqMan ) specific of the different subclasses of TP53 mRNA variants TAp53 , Δ133 and alpha , beta and gamma ( intron-9 splice variants ) ( Khoury et al . , 2013 ) , we found that whereas the two cell lines expressed similar levels of TAp53 and alpha TP53 mRNA variants , the more invasive MDA-MB-231 D3H2LN expressed higher levels of Δ133p53 , beta and gamma mRNA variants ( Figure 2B ) . We confirmed , by western blotting , the highest expression level of p53 protein isoforms in MDA-MB-231 D3H2LN cells by using a sheep polyclonal anti-p53 antibody ( Sapu ) , which recognizes each p53 protein isoform with high specificity ( Marcel et al . , 2013 ) . By western blotting , eight bands at 28 , 35 , 38 , 40 , 42 , 45 , 47 , and 51 kDa were revealed in mock-transfected cells . ( Figure 2C , mock lanes a and b ) . All these bands correspond to genuine p53 protein isoforms as they disappear after transfection with siRNA siE7 which targets exon-7 that is present in all p53 mRNA variants ( lanes 'siE7' ) ( Aoubala et al . , 2011; Camus et al . , 2012; Terrier et al . , 2012 ) . In agreement with the RT-qPCR data , TAp53α protein expression is similar between MDA-MB-231 and MDA-MB-231 D3H2LN , while all the other p53 protein isoforms , notably ∆133p53β are expressed at higher levels in MDA-MB-231 D3H2LN . This suggests that the enhanced invasiveness of MDA-MB-231 D3H2LN is not due to variation of mutant TAp53α protein expression level ( Figure 2C , all lanes compare b to a ) . 10 . 7554/eLife . 14734 . 011Figure 2 . p53 isoform expression correlates with cell invasiveness in breast cancer cells . ( A ) Invasion of breast cancer cell lines . MCF7 , MDA-MB231 and MDA-MB231 D3H2LN cells were assayed for 24 hr and the changes in invasion were analyzed , as described in Materials and methods ( Smith et al . , 2008 ) . Invading cells were counted as the number of invading cells at 50 μm divided by the number of non-invading cells at 0 µm . The results are expressed as the fold change ratio compared with MDA-MB231 . Each assay was performed in triplicate for each cell line . The values plotted are means ± SEMs of N = 4 independent experiments; p<0 . 05 . ( B ) Comparative expression of p53 mRNA variants in MDA-MB231 D3H2LN versus parental MDA-MB231 cells . The expression level of p53 isoform mRNA is higher in the highly invasive MDA-MB231 D3H2LN cells than in MDA-MB231 cells . Sub-confluent proliferating breast cancer cells were harvested for quantitative RT-qPCR Taqman assays ( see Materials and methods section ) . p53 isoform expression was quantified relative to the control MDA-MB231 cell line . For all RT-qPCR experiments , expression levels of p53 isoforms were normalized to TBP . Results are expressed as the fold change compared to MDA-MB231 cells and represent means ± SEMs of N = 4 independent experiments; p<0 . 05 . ( C ) Differential expression of endogenous p53 protein isoforms in MDA-MB231 and MDA-MB231 D3H2LN cells . The protein expression levels of p53 isoforms were analyzed by western blotting using the sheep polyclonal p53 pantropic antibody ( Sapu ) . To identify p53 protein isoforms , cells were transfected either with p53 siRNA ( siE7 ) targeting exon-7 common to all p53 mRNA variants or with control siRNA ( siNS , non specific ) . Two exposures ( short and long ) are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 011 To address the role of mutant Δ133p53 isoforms in the enhanced invasiveness of MDA-MB-231 D3H2LN cells , we analyzed cells depleted either of all Δ133p53 isoforms ( by using the siRNAs si133-1 or si133-2 , which specifically target the 5’UTR of all Δ133 mRNAs [Aoubala et al . , 2011] ) , or of all β isoforms ( by using siβ , [Camus et al . , 2012; Terrier et al . , 2012] ) , or of all p53 isoforms except Δ133p53 ( α , β , γ ) by using siTAp53 , a siRNA targeting p53 mRNA splice variant containing TP53 exon-2 ( Figure 3A ) . We assessed knockdown efficacy by RT-qPCR and western blotting using the Sapu and KJC8 antibodies ( Figures 3D and Figure 3—figure supplement 1A ) . Importantly , depletion of Δ133 or β p53 variants ( by transfection of si133-1 , si133-2 or siβ ) significantly decreased the invasiveness of mutant TP53 MDA-MB-231 D3H2LN cells without altering the expression of mutant TAp53α protein ( Figure 3D ) . However , depletion of all p53 isoforms except Δ133p53 ( α , β , γ ) by siTA did not change the invasive activity of MDA-MB-231 D3H2LN cells , indicating that the simultaneous depletion of all TAp53 ( including mutant TAp53α ) did not impair MDA-MB-231 D3H2LN cell invasion ( Figures 3B and Figure 3—figures supplement 1A ) . To confirm the role of the Δ133p53β isoform in promoting invasion potential , we re-introduced a si-resistant Δ133p53β-R280K isoform in MDA-MB-231 D3H2LN cells in which all Δ133p53 ( α , β , γ ) isoforms had been knocked down with si133-1 or si133-2 . As expected , expression of the si133-1- and si133-2- resistant Δ133p53β-R280K rescued invasion ( Figure 3C and Figure 3—figure supplement 1B ) . Similarly , Δ133p53β-R280K expression rescued invasion when cells were depleted of all β isoforms by siβ ( Figure 3C and Figure 3—figure supplement 1C ) . We then compared the effects of Δ133p53α-R280K , Δ133p53β-R280K and Δ133p53γ-R280K isoforms on cell invasion . As the siRNAs si133-1 and si133-2 specifically target the 5’UTR of all Δ133 mRNAs , we tested their respective contribution to the invasive phenotype by re-introducing each of them in MDA-MB-231 D3H2LN cells in which all Δ133p53 ( α , β , γ ) isoforms had been knocked down with si133-1 or si133-2 . All three mutant Δ133p53 isoforms rescued invasion in cells depleted of ∆133p53 variants , with Δ133p53β-R280K conferring the highest invasive activity compared to Δ133p53α-R280K or Δ133p53γ-R280K ( Figure 3C and Figure 3—figure supplement 1B ) . These results indicate that mutant Δ133p53 isoforms , notably the Δ133p53β variant , are sufficient to promote invasion in mutant TP53 MDA-MB-231 D3H2LN cells , consistently with an active role for Δ133p53β isoform in cancer progression . 10 . 7554/eLife . 14734 . 012Figure 3 . Δ133p53β isoform promotes invasion in breast cancer cells . ( A ) RNA chart of the different isoforms of TP53 in this study showing the exons ( boxes ) and introns ( horizontal lines , not to scale ) . Alternative promoters are shown as arrows and alternative splices are depicted with the lines above as they connect different exons . Location of the different siRNAs used is indicated below the chart . Below is a list of the p53 isoforms that remain after transfection of the different siRNAs , indicated by a cross . ( B ) Specific inhibition of some p53 isoforms expression decreases invasiveness . MDA-MB231 D3H2LN cells were transfected with si133-1 or si133-2 , two distinct siRNAs specific for the 5’UTR of Δ133p53 mRNAs; or with siTAp53 , a siRNA targeting TP53 exon-2 depleting all p53 isoforms except the Δ133p53 ( α , β , γ ) isoforms; or with siβ , a siRNA targeting the alternatively spliced exon-9β of TP53; or with siNS , a non-specific siRNA used as negative control . ( C ) ' Rescue' experiments . Re-introductions of si-133–resistant mutant Δ133p53α-R280K , Δ133p53β-R280K or Δ133p53γ-R280K restore the invasive activity in MDA-MB231 D3H2LN cells previously depleted of either Δ133p53 ( α , β , γ ) isoforms after transfection with si133-1 or si133-2 , or β p53 isoforms ( p53β , Δ40p53β and Δ133p53β ) ( n = 4 ) . ( D ) Inhibition of endogenous p53 protein isoforms induces expression of epithelial features associated with decreased invasiveness . The expression of endogenous p53 protein isoforms after transfection of MDA-MB231 D3H2LN cells with si133-2 , siβ , siE7 or control siRNA ( siNS ) was analyzed by western blotting using pantropic p53 isoforms antibody Sapu or KJC8 , a β-specific p53 antibody recognizing p53β , Δ40p53β and Δ133p53β . The expression of two EMT markers ( Vimentin and E-Cadherin ) was determined in parallel . Ku80 was used as a loading control . * cross-reaction . ( E ) Quantification of E-Cadherin mRNA in MDA-MB231 D3H2LN cells transfected with siRNA si133-2 , siβ , siE7 or siNS ( control ) used as a negative control . For all RT-qPCR experiments , expression levels were normalized to TBP . Results are expressed relative to TBP mRNA and represent means ± SEMs of N = 4 independent experiments; p<0 . 05; p<0 . 01 ( F ) WT ∆133p53β promotes cell invasion . Weakly invasive MCF7 cells were transfected with Δ133p53β expression vector or the empty expression vector ( Control ) . Cells were challenged for their invasive potential after 48 hr . The values are plotted as means ± SEMs of at least 3 independent experiments; p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 01210 . 7554/eLife . 14734 . 013Figure 3—figure supplement 1 . Δ133p53 ( α , β and γ ) expression in MCF7 breast cancer cells ( related to Figure 3 ) . ( A ) Quantitative RT-qPCR ( TaqMan ) of Δ133p53 ( α , β and γ ) , TAp53 ( α , β and γ ) or β mRNAs in MDA-MB231 D3H2LN cells transfected with non-specific siRNA ( siNS ) , or p53 specific siRNA si133-1 , si133-2 , siTAp53 or siβ , respectively . For all RT-qPCR ( TaqMan ) experiments , expression levels were normalized to TBP . Results are expressed as relative expression to TBP mRNA and represent means ± SEMs of N = 4 independent experiments; *p<0 . 01 . ( B ) Western blot of ectopically expressed Δ133p53α , Δ133p53β and Δ133p53γ in MDA MB231 D3H2LN breast cancer cells using Sapu p53 pantropic or anti-Flag antibodies after Δ133p53 depletion with si∆133–1 ( left panel ) or si∆133–2 ( right panel ) . ( C ) Western blot withSapu p53 pantropic or anti-Flag antibodies of ectopically expressed Δ133p53β in MDA MB231 D3H2LN breast cancer cells previously depleted of all b p53 isoforms with siRNA siβ ( D ) Expression of ∆133p53 ( α , β and γ ) mRNA variants in MCF7 cells and in HCT116 cells . Expression was determined by nested RT-PCR ( E ) Western blot of ectopically expressed Δ133p53β in MCF7 breast cancer cells using Sapu antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 013 EMT is a gatekeeper process involving global cell reprogramming leading to profound phenotypic changes that include enhanced invasiveness along with loss of epithelial cell-cell adhesion proteins such as E-Cadherin and concomitant gain of mesenchymal protein expression including Vimentin and N-Cadherin . EMT is a reversible process , thus mesenchymal cells can reverse their phenotype and re-express epithelial markers ( Thiery et al . , 2009 ) . Interestingly , depletion of mutant Δ133 or β p53 isoforms reversed mesenchymal traits of MDA-MB-231 D3H2LN cells , by inducing a significant induction of E-Cadherin protein expression and a marked decrease in Vimentin expression , particularly in cells transfected with siβ . Knockdown efficacy was assessed by western blotting using the p53 pantropic polyclonal sapu and β-specific KJC8 antibodies ( Figure 3D ) . Thus , inhibition of endogenous mutant Δ133 or β p53 isoforms induced a switch in the expression of these epithelial features which reflects global reversion of EMT cell reprogramming . Interestingly , re-epithelialization was impaired upon depletion of all p53 isoforms ( Figure 3D , lane siE7 ) , suggesting that other p53 splice variants control epithelial features . Differential induction of E-Cadherin expression was also observed at the mRNA level ( Figure 3E ) . We then addressed whether the introduction of WT Δ133p53β in MCF7 cells that endogenously express all p53 isoforms except Δ133p53β and Δ133p53γ ( Figure 3—figure supplement 1D ) promotes invasion . As shown in Figure 3F and Figure 3—figure supplement 1E , WT Δ133p53β triggered MCF7 cell invasion into type1-Collagen . Altogether , the results in MDA-MB-231 D3H2LN and in MCF7 indicate that mutant and WT Δ133p53β promote invasion of WT and mutant TP53 breast cancer cells . In mutant TP53 cells expressing Δ133p53β , depletion of TAp53α does not inhibit cell invasion , while depletion of the Δ133p53 or β p53 isoforms decreases invasion of cells expressing mutant TAp53α . In WT TP53 cells expressing all p53 isoforms except Δ133p53β , introduction of WT Δ133p53β promotes invasion . It suggests an active role for WT and mutant Δ133p53β in cancer progression corroborating the clinical analysis . To determine whether Δ133p53β can promote invasion in other cancer types regardless of TP53 mutation status , we studied a panel of WT and mutant TP53 cell lines derived from colon carcinoma , another relevant model for cancer cell invasion , by comparing their ability to invade into Matrigel . The cell lines used were derived from the primary colorectal carcinoma ( CRC ) : HCT116 ( WT TP53 ) and SW480 ( mutant p53R273H ) , and from metastatic tumors: LoVo ( WT TP53 ) , SW620 ( mutant p53R273H ) and CoLo205 ( mutant p53 , Y103 del27bp ) . Of note , the SW620 cell line was derived from a metastasis of the primary tumor from which the SW480 cell line was derived one year earlier . Compared to cells derived from primary colon tumors , metastasis-derived cell lines expressing WT or mutant TP53 exhibited a higher ability to invade Matrigel ( Figure 4A ) and had no E-Cadherin at cell-cell junctions ( Figure 4B ) . Of note , although the SW620 and SW480 cell lines have a mutated TP53 gene ( R273H ) and come from the same patient , only SW620 are highly invasive , indicating that the expression of mutant TAp53α-R273H protein is not sufficient to explain invasive activities . We thus measured Δ133p53 mRNA expression by RT-qPCR ( Taqman ) and found it significantly higher in the invasive cell lines whether they express WT or mutant TP53 gene , i . e . LoVo , SW620 and CoLo205 , as compared with HCT116 and SW480 cell lines ( Figure 4C ) , in agreement with the results obtained in the breast cancer cell lines . To confirm the role of Δ133p53 isoforms in cell invasion , we knocked-down these isoforms in mutant TP53 SW620 , or WT TP53 LoVo CRC cells by using si133-1 and si133-2 , which led to a significant decrease invasive capacity in both cell lines ( Figure 4D and Figure 4—figures supplement 1A and B ) . As expected , re-introduction of si133-1-resistant WT Δ133p53β rescued invasiveness of LoVo cells depleted of Δ133p53 isoforms by si133-1 ( Figure 4—figures supplement 1C and D ) . This indicates that Δ133p53 isoforms , notably Δ133p53β , regulate invasion , irrespectively of TP53 mutation status . 10 . 7554/eLife . 14734 . 014Figure 4 . Δ133p53 isoforms promote invasion in colorectal cancer cells . ( A ) Invasiveness of different colorectal cancer cells . Cells were assayed for invasiveness through Matrigel over 24 hr , as described in Materials and methods . Invading cells were counted and the results are expressed as the average of 6 different fields and normalized to HCT116 cells . Each assay was performed in triplicate for each cell line . TP53 mutation status is indicated . The values plotted are means ± SEMs of N = 4 independent experiments . ( B ) Immunofluorescence staining for E-Cadherin in a panel of colorectal cell lines . The images show representative E-Cadherin immunostaining for each colorectal cell line plated at low density . Scale bar: 10 µm . ( C ) Quantitative RT-qPCR ( TaqMan ) of different subclasses of p53 isoform mRNA in a panel of colorectal cell lines . Sub-confluent proliferating colorectal cells were harvested for quantitative RT-PCR assays ( see Materials and methods section ) . For each cell line , results are expressed as the fold change compared to HCT116 cells . For all RT-qPCR experiments , expression levels of sub-types of p53 mRNA variants were normalized to TBP mRNA and represent means ± SEMs of N = 3 independent experiments; p<0 . 05 . ( D ) Invasion of SW620 ( left ) or LoVo ( right ) colon cancer cell lines after depletion of Δ133p53 ( α , β , γ ) isoforms . Cells transfected with Δ133 siRNAs ( si133-1 and si133-2 ) or siNS were examined for their invasive potential after 24 hr . Invading cells were counted and the results are expressed as the average of 6 different fields and normalized to siNS . The values are plotted as means ± SEMs of 3 independent experiments; p<0 . 05 . ( E ) Western-blot analysis of endogenous p53 protein isoforms in HCT116 cells transfected with the non-specific siRNA ( siNS ) or with p53 isoform specific siRNA siE7 , siTAp53 , or siβ . Cells were then assessed for their invasive potential as in Figure 4F . ( F ) HCT116 cells transfected with siE7 , siTAp53 , siβ , or siNS were assessed for their invasive potential . Invading cells were counted and the results are expressed as the average of 6 different fields . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 01410 . 7554/eLife . 14734 . 015Figure 4—figure supplement 1 . Δ133p53 ( α , β and γ ) expression and effect on cell invasion using colorectal cancer cells . A and B Quantitative RT-PCR ( Taqman ) of all Δ133p53 ( α , β and γ ) mRNA in SW620 ( A ) or LoVo ( B ) cells transfected with siRNA si133-1 , si133-2 or siNS ( as negative control ) . ( related to Figure 4 ) . For all RT-qPCR experiments , mRNA expression levels were normalized to TBP mRNA . Results are expressed as relative expression to TBP mRNA and represent means ± SEMs of N = 4 independent experiments; p<0 . 01 . ( C ) Re-introduction of si133-resistant WT Δ133p53β isoform restores invasiveness of LoVo colon cancer cells previously depleted of all Δ133p53 ( α , β , γ ) isoforms with si133-1 ( n = 4 ) . ( D ) Western blot of ectopically expressed Δ133p53β in LoVo cells using KJC8 antibody after Δ133p53 depletion with si∆133–1 . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 015 Among WT TP53 colon carcinoma cells , HCT116 are weakly invasive and endogenously express all p53 protein isoforms , including Δ133p53β ( Figure 4E ) . We investigated whether we could modify HCT116 cell invasion by modulating the ratios of p53 isoforms . First , we depleted endogenous expression of different p53 isoforms using specific siRNAs and compared the invasiveness of HCT116 cells depleted of all p53 proteins other than Δ133p53 isoforms ( i . e . transfected with the siTAp53 siRNA ) or depleted of β isoforms but still expressing TAp53α ( i . e . transfected with the siβ siRNA ) ( Aoubala et al . , 2011; Marcel et al . , 2014 ) ( Figure 4F ) . HCT116 cells transfected with siTAp53 are more invasive than cells transfected with siNS or siE7 , while HCT116 cells transfected with siβ , which depleted p53β , ∆40p53β and ∆133p53β are less invasive than cells transfected with siNS ( Figure 4E and F ) . This suggests that the ratio of TAp53 and β isoforms , including ∆133p53β , defines the invasive activity of HCT116 cells . Altogether our data indicate that the invasive capacity of colon cancer cell lines , as in the breast tumor cell lines , correlates with and depends on the expression of Δ133p53 isoforms , regardless of the TP53 mutation status . To determine whether the Δ133p53β-enhanced invasion directly impacts on the mode of cell migration , we transfected HCT116 with GFP- WT Δ133p53β and tracked them by video-microscopy . We observed two populations of cells , cohesive epithelial cells adherent to the glass and rounded cells loosely attached on top of cohesive cells . GFP-tagged Δ133p53β expression elicited a significant increase of rounded cells with dynamic bleb-like structures on their surface ( Figure 5A , Video 1 and Figure 5—figure supplement 1A as control ) . While recording , we noticed that some GFP-Δ133p53β -positive cohesive adherent cells detached progressively from the substratum , became rounded ( white arrowhead , Figure 5A ) and migrated ( white arrow , Figure 5A ) . Propidium iodide staining and FACS analysis showed that blebbing cells were alive and were not undergoing apoptosis ( Figure 5—figure supplement 1B ) . The number of blebbing cells was 25 times greater than the number of adherent cells in HCT116 cells transfected with Δ133p53β ( Figure 5B ) . Similar rounded and detached cells were obtained using myc-tagged isoforms , ruling out any role of GFP in the observed phenotype ( Figure 5B ) . Since loss of adhesive structures and concomitant acquisition of a rounded-blebbing movement are hallmarks of epithelial-amoeboid transition ( EAT ) , a derivative of EMT which produce highly invasive cells , we further confirmed the EAT-like phenotype of Δ133p53β-expressing cells by measuring E-cadherin and β1-integrin levels , as the loss of both is a marker of EAT ( Friedl and Wolf , 2003 ) . As shown in Figure 5C , E-cadherin and β1-integrin , expressed at high levels in adherent cohesive cells , were barely detected in blebbing cells , implying they had experienced EAT . Δ133p53β also enhanced cell migration and invasion , since its expression in HCT116 cells elicited a two-fold increase in migration ( Figure 5D ) in uncoated transwell chambers and conferred a 5 times higher invasive capacity in Matrigel ( Figure 5D ) compared to GFP-only-transfected controls . 10 . 7554/eLife . 14734 . 016Video 1 . ( related to Figure 5A ) : DIC light microscopy of HCT116 cells expressing GFP-Δ133p53β . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 01610 . 7554/eLife . 14734 . 017Figure 5 . HCT116 cells expressing the Δ133p53β isoform display amoeboid-like movements . ( A ) Still time-lapse images of the accompanying video ( Supplementary data , video1 ) of HCT116 cells transfected with the GFP-tagged Δ133p53β isoform . Cells were observed 48 hr after transfection . For the video , images were captured every 4 min during 12 hr . The panel represents 1/10 images i . e . one image every 40 min . Δ133p53β-transfected cells can be distinguished from non-transfected cells through expression of GFP ( green ) . The Δ133p53β-transfected cells are rounded and exhibit blebbing movements on their surface . The arrow shows a cell that detaches from the others and from the dish during the time-lapse . The arrowhead shows a cell that still adheres to the other epithelial cells and to the substratum at the beginning of the experiment and then becomes progressively rounded . ( B ) Quantitative analysis of blebbing versus adherent cell number in the Myc positive cells . FACS analysis of the percentage of non-apoptotic blebbing Myc-positive HCT116 cells compared to total Myc-positive transfected with cells upon transfection of Myc-Δ133p53β , or Myc-empty expression vector ( vector ) . Results were normalized to Myc-empty vector transfected cells . ( C ) Western blot analysis of the expression of E-Cadherin and β1-integrin in HCT116 cells expressing the GFP-tagged Δ133p53β isoform; Control: GFP-tag vector . Adherent: cells still adherent to the substratum; Blebbing: cells detached from the substratum and showing blebbing movements . Loading normalization was performed using an anti-α-tubulin antibody . ( D ) Δ133p53β-transfected HCT116 cells were quantified for their migration ability after 2 hr of migration through the Boyden chamber or for their invasiveness after 24 hr of the invasion through Matrigel , as indicated . The values are plotted as means ± SEMs of at least 3 independent experiments . ( E ) 3-D LoVo cell scattering . The numbers of scattered cells were quantified using Metamorph software ( left ) . Cells were judged as « scattered » when individual cells or clusters of cells had lost contact with the main colony , as visualized ( right ) . Values ( means ± SEMs ) were calculated from 4 independent experiments ( n = 48 ) **p<0 . 001 . ( F ) Wound healing assay in LoVo cells infected with shRNA non relevant ( shNS: shLuciferase ) or sh∆133p53 . Cells were observed 25 hr after infection . Still time-lapse images of the accompanying videos ( Supplementary data , Videos 2 and 3 ) For the videos , images were captured every 1 hr during 25 hr . The arrows show cells detaching from the others and migrating as individual cells . The arrowhead shows a cluster of cells which leaves the cohesive epithelium and which collectively migrate and enter into the gap . ( G ) Schematic representation of the role for ∆133p53β in reprogramming cells toward the invasive process . For WT or mutant TP53 cells devoid of ∆133p53b expression , introduction of ∆133p53β promotes EMT and invasion . Reciprocally WT or mutant TP53 cells expressing ∆133p53β have enhanced invasive activity . Depletion of ∆133p53β reverts EMT and inhibits invasion . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 01710 . 7554/eLife . 14734 . 018Figure 5—figure supplement 1 . Control experiments of Δ133p53β expression and effects in colon cancer cells . ( A ) Image of HCT116 cells transfected with GFP-empty vector as control . Cells were treated as in Figure 5A . ( B ) ( related to Figure 5A ) : Quantitative analysis of blebbing- versus adherent-cell number in the GFP positive cells as in Figure 5A and Figure 5—figure supplement 1A . FACS analysis of the percentage of non-apoptotic blebbing GFP-positive HCT116 cells compared to total GFP-positive transfected cells after transfection of GFP-Δ133p53β or GFP-empty expression vector . Results were normalized to GFP-empty vector transfected cells . ( C ) ( related to Figure 5G ) : Western-blot of p53 protein isoforms with the p53 pantropic antibody Sapu in LoVo cells infected with shRNA non relevant ( shNS: shLuciferase ) , sh∆133p53 and used for 3-D cell scattering . Actin was used as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 01810 . 7554/eLife . 14734 . 019Video 2 . ( related to Figure 5—figure supplement 1F , upper panels ) . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 01910 . 7554/eLife . 14734 . 020Video 3 . ( Related to Figure 5—figure supplement 1F , lower panels ) : DIC light microscopy of Wound healing assay in LoVo cells infected with shRNA non relevant ( shNS: shLuciferase; Video 2 ) or sh∆133p53 ( Video 3 ) . Cells were observed 25 hr after infection . Images were captured every 1 hr during 25 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 14734 . 020 Our data indicate that expression of WT Δ133p53β promotes EAT in WT TP53 HCT116 cells which endogenously express low levels of Δ133p53β . We therefore next investigated whether we could modify EMT/EAT features in WT TP53 CRC cells that endogenously express high levels of Δ133p53β . The WT TP53 colon carcinoma LoVo cells are strongly invasive and endogenously highly express Δ133p53 isoforms at the mRNA and protein levels ( Figure 4A , C ) . Knockdown of Δ133p53 isoforms in LoVo cells remarkably decreased 3-D cell scattering , a process which shares cellular features reminiscent of cells undergoing EMT/EAT ( Figure 5E and Figure 5—figure supplement 1C ) . To investigate the impact of p53 isoforms-mediated cell scattering on the process of migration , we performed wound-healing assay . LoVo transduced with control shNS progressively detached and migrated as individual cells ( arrows , Figure 5F ) . LoVo depleted of Δ133p53 isoforms still remain cohesive and migrate very slowly into the gap . In this case , some rare clusters of cells leave the cohesive epithelium to enter into the gap ( arrowheads , Figure 5F ) . Interestingly , the cluster of cells remains compact and cells move collectively , but with a reduced velocity compared to individual migrating cells with unaffected ∆133p53 expression ( Figure 5F and Videos 2 and 3 ) . Thus endogenous Δ133p53 protein isoforms regulate both the mode of cell motility and the ability to implement adherens junctions , with a subsequent impact on cell velocity . Altogether , our data indicate that Δ133p53β expression elicits hallmarks of EAT/EMT in CRC cells . All the genetic evidence in the clinic and animal models indicate , unequivocally , the pivotal and fundamental role of the TP53 gene in cancer formation , progression and response to treatment . TP53 is ubiquitously expressed and it regulates different cell responses such as cell repair , proliferation , senescence , differentiation , cell migration and cell death in response to any change of tissue/cell homeostasis , thus maintaining tissue/cell integrity upon stress or damage . Consistently , TP53 mutation is the most frequently mutated gene in human cancers and its mutation status is associated with poor clinical outcome . Any cancer treatments change cell homeostasis and therefore trigger TP53-mediated cell responses . It is currently impossible to accurately predict response to cancer treatment in the clinic because of the large variety of TP53-mediated cell responses . In addition , as missense mutations of TP53 do not totally abrogate its tumor suppressor activity and can confer additional biological activities to TP53 gene , it is very difficult to choose the most efficient cancer treatment based on TP53 mutation status ( Kruiswijk et al . , 2015; Lang et al . , 2004; Li et al . , 2012; Meek , 2015 ) . This indicates that major features of the TP53 gene and its pathway have still to be uncovered and understood and this represents a critical bottleneck preventing major breakthroughs in cancer treatment . To date , it is thought that the diverse biological activities associated with TP53 gene expression are carried out by a single protein isoform , p53 ( TAp53α ) . However , like most human genes , TP53 gene does not express one but at least 12 different p53 protein isoforms which have distinct domain and activities . The WT canonical p53 protein ( TAp53α ) prevents cancer invasion by inhibiting EMT ( Gadea et al . , 2007; McDonald et al . , 2011; Muller et al . , 2011; Roger et al . , 2010; Suva et al . , 2013 ) while the mutant canonical p53 protein ( mutant TAp53α ) was shown to promote cell invasion ( Adorno et al . , 2009; Gadea et al . , 2002 , 2004; Muller et al . , 2009; Roger et al . , 2010; Gadea et al . , 2007 ) . However , despite this clear distinction between WT and mutant TP53 , the situation in the clinic is not clear cut , as a mutant TP53 expression in primary tumors does not systematically lead to metastasis formation and WT TP53 is frequently expressed in metastasis ( Dong et al . , 2007; Kalo et al . , 2007; Oren and Rotter , 2010 ) . Little is known about the biological roles and clinical relevance of p53 isoforms in cancer . In this study , we analyzed the expression of Δ133p53 isoforms ( α , β , γ ) in relation to clinical outcome in primary breast cancers . We determined that all tumors expressing Δ133p53β or Δ133p53γ also co-express Δ133p53α . This indicates that Δ133p53 ( α , β , γ ) isoforms are not randomly expressed in breast cancers , confirming that the internal promoter of TP53 and the alternative ( β , γ ) splicing of p53 mRNA are regulated in tumor cells ( Aoubala et al . , 2011; Marcel et al . , 2014; 2010; Tang et al . , 2013 ) . We also determined that Δ133p53β expression is significantly associated with cancer recurrence in breast cancer patients , even in tumor expressing WT TP53 . In particular , Δ133p53β enhances on average by eight times the risk of recurrence and by three times the risk of death in luminal A breast cancer patients , an otherwise excellent prognostic group . Importantly , detection of Δ133p53β at time of surgery in primary luminal-A breast tumors of patients devoid of lymph node would predict cancer progression . Therefore , far from having its function inactivated in Luminal A breast cancer , the WT TP53 gene would promote breast cancer recurrence and increased risk of death when it expresses Δ133p53β . To corroborate the clinical data , we experimentally investigated how Δ133p53β could confer cell invasion and motility to WT and mutant TP53 breast cancer cells . First , we observed that the invasive activity of MCF7 , MDA-MB-231 and MDA-MB-231 D3H2LN cells is positively correlated to the Δ133p53 mRNA expression level . This was also observed in a panel of WT or mutant TP53 colon cancer cell lines . In mutant TP53 MDA-MB-231 D3H2LN breast cancer cells , depletion of endogenous mutant Δ133p53 isoforms reduces cell invasion , despite unaltered and strong expression of mutant full-length p53 ( TAp53α ) protein . Re-introduction of si-133-resistant mutant Δ133p53α-R280K , Δ133p53β-R280K or Δ133p53γ-R280K in MDA-MB-231D3H2LN cells depleted of Δ133p53 isoforms restores their invasive phenotype , with mutant Δ133p53β-R280K being significantly the most potent . Importantly , depletion of the p53 protein isoforms with siTA that targets only the p53 proteins containing the transactivation domains ( i . e TAp53α , TAp53β , TAp53γ , Δ40p53α , Δ40p53β , Δ40p53γ ) without altering expression of Δ133p53 isoform did not change the invasive activity of MDA-MB-231 D3H2LN . This indicates that the simultaneous depletion of all TAp53 ( including mutant TAp53α ) did not impair MDA-MB-231 D3H2LN cell invasion . It has previously been reported that overexpression of mutant TAp53α promotes integrin and epidermal growth factor receptor ( EGFR ) recycling , thus driving invasion ( Muller et al . , 2009 ) . However , expression of the mutant TAp53α is not sufficient to explain why the triple negative breast cancer MDA-MB-231 and its highly metastatic MDA-MB-231 D3H2LN derived clone have different invasive activity while they both express the same mutant TP53-R280K gene . MDA-MB-231 D3H2LN cells showed a significantly higher invasiveness as compared to parental MDA-MB-231 cells . Interestingly , MDA-MB-231 D3H2LN expressed higher levels of Δ133p53 mRNA variants and proteins ( Figure 2B and C ) . Similar results were observed in mutant TP53 colon cancer cell lines . The SW480 and SW620 colon carcinoma cell lines derive respectively from the primary and secondary tumors resected from the same patient ( Hewitt et al . , 2000 ) . Although SW480 and SW620 cells express the same mutant TP53-R273H gene , SW480 cells are far less invasive than the SW620 cells . Interestingly , SW480 cells express much less Δ133p53 mRNAs than the SW620 cells . Importantly , depletion of endogenous Δ133p53 isoforms from SW620 cells reduces cell invasion , despite unaltered and strong expression of mutant TAp53α protein . This indicates that the role of endogenous mutant Δ133p53 isoforms in promoting invasiveness may be extended to other tissue types and that endogenous Δ133p53 isoform expression regulates cancer cell invasion in TP53 mutant colon and breast cancer cells , irrespective of the full-length p53 ( TAp53α ) expression . Importantly , the invasive activity of Δ133p53 isoforms is not associated with TP53 mutation status since depletion of ∆133p53 isoforms in WT TP53 colon carcinoma LoVo cells abolishes their scattering and invasive activities . ∆133p53 isoform expression thus explains the inconsistent clinical association between TP53 mutation and metastasis . Interestingly , the introduction of WT Δ133p53β in the poorly invasive WT TP53 MCF7 breast cancer cells that express all WT p53 isoforms but Δ133p53β , enhances MCF7 invasive activity . In WT TP53 HCT116 colon cancer cells , that express all WT p53 protein isoform including low level of Δ133p53β protein , we demonstrated that HCT116 invasive activity can be controlled ( enhanced or inhibited ) by siRNAs specific of different p53 isoforms to modulate expression of different subsets of p53 protein isoforms . Hence HCT116 invasive activity is inhibited after depletion of Δ133p53 or β isoforms by siRNA si133 or siβ respectively while HCT116 invasive activity is enhanced after depletion of WT full-length p53 proteins ( TAp53α , TAp53β , TAp53γ , Δ40p53α , Δ40p53β and Δ40p53γ ) by siRNA siTA . Furthermore , ectopic expression of WT Δ133p53β in HCT116 cells enhances their invasive activity . Altogether , Δ133p53β expression provides a rationale for invasive cancers expressing WT TP53 and non-invasive cancers expressing missense mutant TP53 . These data suggest that epigenetic mechanisms that stem Δ133p53β expression , i . e . alternative splicing and induction of the TP53 internal promoter favour cancer invasion and dissemination of metastatic cells , regardless of TP53 mutation status . Our data indicate that all three ∆133p53 isoforms ( ∆133p53α , Δ133p53β and Δ133p53β ) promote invasion . This is in accordance with recent reports showing that ∆133p53α stimulate angiogenesis and cell invasion ( Bernard et al . , 2013; Roth et al . , 2016 ) . Here we identified the ∆133p53β isoform as being the most efficient promoter of invasion , corroborating its role as a predictive indicator of cancer relapse and death in the clinic . We investigated then how Δ133p53β regulate cell invasion in cancer cell lines . Since EMT leads to profound phenotypic changes due to genome-scale epigenetic reprogramming and post-transcriptional regulation , we studied the regulation of EMT by Δ133p53β to illustrate a molecular mechanism of Δ133p53β . Our data indicate that expression of Δ133p53β promotes acquisition of a rounded-blebbing movement , which is associated with decrease of E-cadherin and β1-integrin in colon cancer HCT116 cells . These phenotypic changes are hallmarks of epithelial-amoeboid transition ( EAT ) , a derivative of EMT which produces highly invasive cells . Depletion of ∆133p53 isoforms also induces E-Cadherin expression and concomitantly inhibits Vimentin expression in invasive breast cancer MDA-MB-231 D3H2LN cells . Accordingly , knockdown of Δ133p53 isoforms in colon cancer LoVo cells decreased 3-D cell scattering , a process associated with EMT/EAT . Our study supports a critical role for Δ133p53β ( whether WT or mutant ) in the induction of cell invasion and EMT/EAT . In summary , our data lead us to conclude that the biological activities associated with WT or mutant TP53 gene expression are not carried out by the single full-length p53 protein ( TAp53α ) . p53 isoforms underlie the dual role of WT TP53 gene in preventing or promoting cancer cell invasion . The invasive activities of cancer cells expressing WT or mutant TP53 gene can be enhanced or inhibited by respectively increasing or reducing expression of Δ133p53β ( Figure 5G ) . Testing the prognostic discriminatory potential of Δ133p53β could influence clinical practice and may offer appropriate and yet unexplored therapeutic options for mutant or WT TP53 tumors expressing Δ133p53β . Primary , previously untreated and operable breast cancers from 273 Caucasian women , with sufficient tumor tissue surplus to diagnostic requirements and with complete clinical and pathological data , were analyzed . To minimize variations such as different surgical techniques and skills , which would have an effect on the rates of cancer recurrence and death , all tumors were provided by only one surgeon , Prof Alastair Thompson , breast cancer surgeon at Ninewells hospital in Dundee . Moreover , all patients were treated at Ninewells Hospital according to established clinical protocols . Furthermore , to minimize variation in tumor handling , all tumors were collected , stored , processed , extracted and stained by the Tayside Tissue Bank according to validated and standardized protocols at Ninewells Hospital . RNA quality was assessed using the BioAnalyzer 2100 prior to RT-PCR analysis and all samples with a ratio of 28S/18S < 1 . 2 were discarded . In addition , the Tayside Tissue Bank collected and verified the clinico-pathological data established by two anatomo-pathologists . Furthermore Tayside Tissue Bank collected and anonymised all the clinical data ( treatment , patient follow-up ) . The median age at diagnosis of the breast cancer patient cohort was 61 . 5 years ( range 28 . 7 to 89 . 1 years ) and median follow-up period was 6 . 85 years ( range 0 . 29 to 13 . 7 years ) . Tumor tissues were macro-dissected by a specialist breast pathologist and snap frozen in liquid nitrogen prior to storage at −80ºC . The samples were examined following Local Research Ethics Committee approval under delegated authority by the Tayside Tissue Bank ( www . taysidetissuebank . org ) . Immunohistochemical staining was carried out on 4 μm sections of formalin-fixed paraffin-embedded tumors , as previously described ( Purdie et al . , 2010 ) . To generate cDNA from tumor total RNA extracts , two independent reverse transcription of 300 ng of total RNA were performed and pooled ( Khoury et al . , 2013 ) . The quality of the reverse Transcription ( cDNA ) was assessed by quantification of the Actin and p53 cDNAs using quantitative Real-Time PCR ( Sybr Green ) previously described ( Khoury et al . , 2013 ) . All cDNA samples for which Actin or p53 cDNA could not be detected in less than 30 cycles were discarded as it indicates that the reverse-transcription did not work properly . This step reduces the number of false-negative samples that would bias the statistical analysis . The sensitivity and specificity of the RT-PCR analysis were improved by performing 2 successive PCR of 35 cycles each and by using 2 sets of nested primers for each Δ133p53 mRNA variants . Therefore the RT-PCR analysis specifically amplifies Δ133p53α , Δ133p53β or Δ133p53γ mRNA . The primer sequences are provided in Table 1 and a detailed protocol is described in the 'extended experimental procedures' . In the first PCR of 35 cycles , all Δ133p53 mRNA variants were amplified using a forward primer corresponding to the 5’UTR of the Δ133p53 mRNA ( TP53 intron-4 ) and a reverse primer corresponding to the exon-10 . Then , the product of the first PCR ( 1ul ) is re-amplified in the second PCR of 35 cycles using a nested forward primer located in the 5’UTR of the Δ133p53 mRNA ( intron4 ) and a nested reverse primer spanning the exon junction between exon-9 and exon-10 to specifically amplify the DΔ133p53α cDNA . To specifically amplify the Δ133p53β cDNA , the product of the first PCR ( 1ul ) is re-amplified in the second PCR of 35 cycles using the nested forward primer located in the 5’UTR of the Δ133p53 mRNA ( intron4 ) and a nested reverse primer corresponding to the exon9b encoding the β C-terminal amino-acid sequence . To specifically amplify the Δ133p53γ cDNA , the product of the first PCR ( 1ul ) is re-amplified in the second PCR of 35 cycles using the nested forward primer located in the 5’UTR of the Δ133p53 mRNA ( intron4 ) and a nested reverse primer spanning the exon junction between exon-9 and exon-9g that encodes the γ-C-terminal amino-acid sequence . The RT-PCR analysis was tested in p53-null cells ( H1299 ) and on a panel of cell lines expressing TP53 gene ( MCF7 , MDA-MB-231 , HCT116 ) . The products of the second PCR were then validated by sequencing and after electrophoresis on 1% agarose gel . For the RT-PCR analysis of the tumor samples , the products of the second PCRs were validated for all tumors by electrophoresis on 1% agarose gel . The PCR products of at least 2 tumor samples were sequenced to validate identity of the amplified Δ133p53 mRNA variants . The samples giving rise to a PCR product of the expected size were deemed positive for the expression of the corresponding Δ133p53 mRNA variant otherwise they were deemed negative . The two successive PCR of 35 cycles using two nested sets of primer maximize sensitivity and specificity of amplification . It enables to detect and identify the expression of each Δ133p53 variants even if expressed in a small population of tumor cells , taking account thus of tumor cell heterogeneity . It allows thus to compare their expression and respective association with the clinico-pathological markers and/or patient clinical outcome . The expression of p53 isoforms and analysis of p53 mutation were performed as described in the extended experimental procedures . The statistical clinical analysis was performed as described previously ( Bourdon et al . , 2011 ) . All statistical comparisons were made using Wilcoxon test and a p-value < 0 . 05 was considered to be statistically significant . Human p53 isoform constructs ( Bourdon et al . , 2005 ) were sub-cloned into pEGFPC1 ( Clontech ) or pLPCmyc . Antibodies were purchased from BD-transduction laboratories , Santa Cruz and GE-Healthcare . Dilutions are indicated in extended experimental procedures . Cells were cultured at 37°C in the presence of 5% CO2 in McCOY’5A or DMEM media ( Sigma ) for colon cancer and breast cancer cells , respectively . HCT116 ( RRID:CVCL-0291 ) , MDA-MB231 ( RRID:CVCL-0062 ) , MCF7 ( RRID:CVCL-0031 ) , LoVo ( RRID:CVCL-0399 ) , SW480 ( RRID:CVCL-0546 ) , SW620 ( RRID:CVCL-0547 ) , Colo205 ( RRID:CVCL-0218 ) were purchased from ATCC . MDA-MB231 D3H2LN ( RRID:CVCL-D257 ) were purchased from PerkinElmer . All cell lines were tested for mycoplasma contamination after thawing using Lonza mycoalert assay ( Reference number: LT27-236 ) . Transfections of p53 isoforms and siRNA were carried out using JetPEI kit ( Qbiogen ) and Interferin ( Polyplus ) , respectively , according to the manufacturers’ instructions . For adhesion assays , non-adherent and adherent cells were collected 24 hr after transfection and counted using the Countess cell counting system ( Invitrogen ) . Time-lapse DIC microscopy was performed on a Leica DMIRE2 inverted microscope with an automatic shutter and GFP filter sets , as described in the extended experimental procedures . Non-adherent and adherent cell extracts were obtained and analyzed by western blotting . Detailed protocols are described in the extended experimental procedures . Cells were lysed and protein complexes were co-immunoprecipitated using GFP-Trap beads ( Chromotek ) as previously described ( Arsic et al . , 2012 ) . Sources of antibodies are provided in the extended experimental procedures . Colon cancer cell migration and invasion were performed using Boyden chambers ( Vinot et al . , 2008 ) . Breast cancer cell invasion assays were performed as previously described ( Smith et al . , 2008 ) . Total RNA was extracted with the RNeasy Mini Kit ( Qiagen ) and treated with DNase ( Qiagen ) prior to reverse transcription , which was carried out using oligo ( dT ) ( Invitrogen ) and M-MLV reverse transcriptase ( Invitrogen ) . Sub-classes of p53 mRNA isoforms were quantified by Real-Time PCR ( TaqMan ) as previously described ( Aoubala et al . , 2011; Moore et al . , 2010 ) . All measurements were normalized to the expression of the TATA box-binding protein ( TBP ) gene .	Most cancers are caused by a build-up of mutations that are acquired throughout life . One gene in particular , called TP53 , is the most commonly mutated gene in many types of human cancers . This suggests that TP53 mutations play an important role in cancer development . It is widely considered that the TP53 gene normally stops tumors from forming , while mutant forms of the gene somehow promote cancer growth . Evidence from patients with cancer has shown , however , that the relationship between TP53 mutations and cancer is not that simple . Some very aggressive cancers that resist treatment and spread have a normal TP53 gene . Some cancers with a mutated gene do not spread and respond well to cancer treatments . Recent studies have shown that the normal TP53 gene produces many different versions of its protein , and that some of these naturally occurring forms are found more often in tumors that others . However , it was not clear if certain versions of TP53’s proteins contributed to the development of cancer . Now , Gadea , Arsic , Fernandes et al . show that Δ133p53β , one version of the protein produced by the TP53 gene in human cells , helps tumor cells to spread to other organs . Tests of 273 tumors taken from patients with breast cancer revealed that tumors with the Δ133p53β protein were more likely to spread . Patients with these Δ133p53β-containing tumors were also more likely to develop secondary tumors at other sites in the body and to die within five years . Next , a series of experiments showed that removing Δ133p53β from breast cancer cells grown in the laboratory made them less likely to invade , while adding it back had the opposite effect . The same thing happened in colon cancer cells grown in the laboratory . The experiments showed that Δ133p53β causes tumor cells with the normal TP53 gene or a mutated TP53 gene to spread to other organs . Together the new findings help explain why some aggressive cancers develop even with a normal version of the tumor-suppressing TP53 gene . They also help explain why not all cancers with a mutant version of the TP53 gene go on to spread . Future studies will be needed to determine whether drugs that prevent the production of the Δ133p53β protein can help to treat aggressive cancers .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2016	TP53 drives invasion through expression of its Δ133p53β variant
Perisynaptic glial cells respond to neural activity by increasing cytosolic calcium , but the significance of this pathway is unclear . Terminal/perisynaptic Schwann cells ( TPSCs ) are a perisynaptic glial cell at the neuromuscular junction that respond to nerve-derived substances such as acetylcholine and purines . Here , we provide genetic evidence that activity-induced calcium accumulation in neonatal TPSCs is mediated exclusively by one subtype of metabotropic purinergic receptor . In P2ry1 mutant mice lacking these responses , postsynaptic , rather than presynaptic , function was altered in response to nerve stimulation . This impairment was correlated with a greater susceptibility to activity-induced muscle fatigue . Interestingly , fatigue in P2ry1 mutants was more greatly exacerbated by exposure to high potassium than in control mice . High potassium itself increased cytosolic levels of calcium in TPSCs , a response which was also reduced P2ry1 mutants . These results suggest that activity-induced calcium responses in TPSCs regulate postsynaptic function and muscle fatigue by regulating perisynaptic potassium . Muscle fatigue is defined as the decline in muscle performance that occurs in response to continued muscle activation . Muscle fatigue is a clinically important feature of myopathies such as muscular dystrophy , neuromuscular disorders such as Guillain-Barré syndrome , diseases of the central nervous system ( CNS ) such as multiple sclerosis , or diffuse conditions such as chronic fatigue syndrome and cachexia ( Katirji , 2002 ) . A wide variety of mechanisms in the central and peripheral nervous systems contribute to muscle fatigue . In simplified preparations of muscle and peripheral nerve , central sources of input are eliminated , permitting the examination of peripheral sites of fatigue , such as the presynaptic release of the neurotransmitter acetylcholine ( ACh ) at the neuromuscular junction ( NMJ; Nanou et al . , 2016 ) , sensitivity of postsynaptic ACh receptors , propagation of the muscle action potential along the sarcolemma and into the t-tubules , release of calcium ( Ca2+ ) from sarcoplasmic reticulum , and activation of the contractile apparatus ( Boyas and Guével , 2011 ) . Proposed mediators of fatigue at these sites include changes in the concentration of intracellular and extracellular ions , such as calcium ( Ca2+ ) , sodium ( Na+ ) , potassium ( K+ ) , or protons ( H+ ) ; metabolites , such as inorganic phosphate ( Pi ) , or lactate; and reactive oxygen species ( Allen et al . , 2008 ) . For example , high-frequency stimulation ( HFS ) of nerve or muscle raises the level of extracellular K+ or [K+]o , which may mediate fatigue by depolarizing muscle membrane , inactivating Nav1 . 4 voltage-gated sodium channels at the NMJ , and consequently blocking the production of muscle action potentials ( APs; Cairns et al . , 2015 ) This mechanism may also underlie the muscle weakness observed in patients with hyperkalemic periodic paralysis , a neuromuscular disorder caused by dominant mutations in the Scna4 gene encoding the Nav1 . 4 channel and characterized by episodic muscle stiffness and weakness ( Cannon , 2015 ) . In addition to presynaptic nerve terminals and muscle endplates , terminal or perisynaptic Schwann cells ( TPSCs ) reside at the NMJ . TPSCs are a non-myelinating Schwann cell subtype that influence the regeneration of injured peripheral motor axons ( Son et al . , 1996 ) , maintain developing synapses ( Reddy et al . , 2003 ) , and participate in synaptic pruning ( Smith et al . , 2013 ) . TPSCs also respond to neural activity by increasing cytosolic Ca2+ levels ( Jahromi et al . , 1992; Reist and Smith , 1992 ) and are therefore functionally similar to other perisynaptic glial cells , such as astrocytes and enteric glia . In addition to responding to neurotransmitter released during neural activity by mobilizing Ca2+ , astrocytes regulate the concentration of extracellular metabolites produced by activity through the expression of various ion channels and transporters ( Olsen et al . , 2015; Boscia et al . , 2016; Weller et al . , 2016 ) . Therefore , TPSCs , as the perisynaptic glia of the NMJ , likely act to modulate the concentrations of these ions at the NMJ and thereby regulate muscle fatigue . In astrocytes , activity-induced Ca2+ signaling is largely mediated by neurotransmitter-mediated stimulation of Gq G-protein coupled receptors ( GPCRs ) , leading to the release of Ca2+ from the endoplasmic reticulum ( ER ) through the second messenger inositol-1 , 4 , 5-triphosphate ( IP3; Volterra et al . , 2014 ) . Astrocytic Ca2+ signaling in turn modulates synaptic transmission and contributes to functional hyperemia , although each of these effects remains controversial ( Agulhon et al . , 2010; Bonder and McCarthy , 2014 ) . The interpretation of these effects is complicated by the observation that the mechanisms contributing to activity-induced Ca2+ signaling in the fine processes of astrocytes are distinct from those underlying this signal in the cell body ( Bazargani and Attwell , 2016 ) . Additionally , the diversity of astrocyte subtypes in the brain ( Eugenín León et al . , 2016 ) and of neuronal subtypes associated with individual astrocytes ( Perea and Araque , 2005 ) further challenge the precise identification of activity-induced Ca2+ responses in these cells . Finally , the extent to which IP3R-mediated Ca2+ signaling reflects all of the effects of activity-induced Gq GPCR activation remain unclear ( Agulhon et al . , 2013 ) . TPSCs , by contrast , are only associated with the nerve terminals of cholinergic motor neurons ( MNs ) . The NMJ is large and amenable to optical analysis owing to its discrete location at the central endplate band region of muscle . TPSCs do not elaborate extensive processes or make specialized contacts with the microcirculation . Together , these features make the TPSC suitable for the genetic manipulation of their response to and regulation of neural activity . However , the examination of activity-induced Ca2+ responses in TPSCs and the functional effects of these responses have been largely conducted by imaging individual TPSCs injected with fluorescent Ca2+-binding dyes ( Darabid et al . , 2014 ) . Thus , whether the effects of single TPSC manipulation on individual synapses lead to global effects on neuromuscular function cannot be examined . Using mice that express the genetically-encoded calcium indicator GCaMP3 in all Schwann cells including TPSCs , we observed that high-frequency , motor nerve stimulation-induced Ca2+ signaling within TPSCs of the neonatal diaphragm was completely abolished in the absence of the purinergic 2Y1 receptor ( P2Y1R ) . We therefore utilized P2ry1 mutant mice lacking these receptors as a model to investigate the functional effects of activity-induced , Gq GPCR-mediated Ca2+ release in TPSCs . In order to study the Ca2+ response to neural activity in populations of TPSCs , we first evaluated transgene expression in TPSCs of the diaphragm muscle at postnatal day 7 ( P7 ) from Wnt1-Cre , conditional GCaMP3 ( Wnt1-GCaMP3 ) mice . Wnt1-Cre mice drive Cre-dependent transgene expression in neural crest derivatives , which include Schwann cells ( Danielian et al . , 1998 ) . We first assessed Wnt1-Cre mice by crossing them to mice conditionally expressing the fluorescent reporter TdTomato . Robust expression of TdTomato was observed at early ages in all Schwann cells in the diaphragm , including myelinating Schwann cells of the phrenic nerve as well as non-myelinating TPSCs at the NMJ ( Figure 1A ) , visualized using fluorescent α−bungarotoxin ( α-BTX ) . We then crossed Wnt1-Cre mice to mice conditionally expressing GCaMP3 , which encodes a green fluorescent protein ( GFP ) -conjugated calmodulin that fluoresces upon Ca2+ binding ( Zariwala et al . , 2012 ) . In whole-mounts of P7 diaphragm muscle incubated with GFP antibodies to label GCaMP3 , S100 antibodies to detect Schwann cells , and α-BTX to visualize NMJs , we observed expression of GCaMP3 in all TPSCs ( Figure 1B ) . Together , these results show that Wnt1-Cre drives robust expression of GCaMP3 in TPSCs of the early postnatal diaphragm . We next determined if GCaMP3 expression in TPSCs exhibited activity-induced Ca2+ responses , similar to previous studies ( Jahromi et al . , 1992; Reist and Smith , 1992 ) . Imaging these responses before and after nerve stimulation at low magnification ( 20X ) , we observed large populations of TPSCs that responded to 45 s of 40 Hz tonic phrenic nerve stimulation ( Figure 2A; Figure 2—video 1 ) . Higher magnification images ( 60X ) showed that each individual TPSC , identified by labeling with fluorescent α-BTX ( data not shown ) , responded to HFS ( Figure 2B ) . We used stat maps of the standard deviation of fluorescence intensity ( SD map ) to spatially represent the distribution of Ca2+ transients within individual TPSCs from high-magnification videos and traces of intensity to examine their temporal characteristics ( Figure 2C ) . The onset of these transients occurred 1 . 8 ± 0 . 74 s ( n = 5 , c = 54 ) after the initiation of HFS , reached peak intensity at 2 . 86 ± 0 . 49 s after the onset of the transient , and shortly thereafter declined in amplitude ( time to 50% decay = 13 . 35 ± 4 . 22 s ) . Most transients lasted the entire duration of the nerve stimulation period , although the intensity at the end of the stimulation period was usually less than 10% of that at the initial peak . GCaMP3-expressing TPSCs responded to multiple stimulations , providing a useful tool to measure the effects of different stimuli on the same TPSCs . Similar to a previous study ( Darabid et al . , 2013 ) , the peak intensity of Ca2+ transients observed after a subsequent 45 s bout of 40 Hz stimulation was lower than that of the first ( 17 . 3 ± 2 . 5 vs . 13 . 3 ± 1 . 1 dB , first vs . second stim , n = 5; p<0 . 05 ) . Interestingly , analyses of transients in TPSCs across large regions of the diaphragm from low-magnification videos showed substantial variability in the onset after nerve stimulation ( from 0 . 8 s to 4 . 3 s ) . This pattern could be achieved by differential transmitter release , differential transmitter breakdown in the perisynaptic space , or differential signal transduction in TPSCs . Because individual muscle fibers across the entire diaphragm exhibit shortening at nearly the same onset after nerve stimulation ( Figure 2—video 1 ) , this result does not appear to result from differential transmitter release . Although the diaphragm is activated by short bursts of tonic stimulation of the phrenic nerve during several behaviors ( e . g . , expulsive maneuvers such as wretching; Hodges and Gandevia , 2000 ) , the more native physiological pattern of stimulation that occurs during respiration is phasic with a duty cycle between 25% and 50% , in which each period of activation is 100 ms , at frequencies between 30–70 Hz ( Kong and Berger , 1986; Zhan et al . , 1998; van Lunteren and Moyer , 2003; Sieck et al . , 2012 ) . The duration of TPSC Ca2+ transients was longer in response to 45 s of 40 Hz phasic vs . tonic stimulation ( Figure 2E , F ) . The off-period of each duty cycle could clearly be discerned as a dropoff in transient intensity ( Figure 2E ) , showing the dynamic nature of these responses ( Todd et al . , 2010; Darabid et al . , 2013 ) . At lower frequencies of stimulation , we observed similar differences of Ca2+ signals in response to phasic vs . tonic patterns . Interestingly , in response to 10 Hz stimulation , the lowest phasic rate that produced a measurable response , Ca2+ transients were much slower in onset after nerve stimulation than after 40 Hz stimulation ( Figure 2F ) . We next examined whether lower frequencies were capable of inducing Ca2+ transients in TPSCs , similar to astrocytes ( Sun et al . , 2014 ) . Whereas spontaneous ACh release , 1 Hz and 2 Hz evoked stimulation failed to produce visible Ca2+ transients in TPSCs , 5 Hz stimulation elicited transients in a small number of TPSCs , and 10–40 Hz stimulation produced responses in all TPSCs ( data not shown ) . In response to different durations of 40 Hz stimulation , only several TPSCs responded to 0 . 1 s of nerve stimulation ( i . e . , four pulses ) , whereas all TPSCs responded to 1–30 s of 40 Hz stimulation ( Figure 2—video 2 ) . Finally , in contrast to TPSCs , myelinating Schwann cells along phrenic nerve trunks and branches failed to exhibit Ca2+ responses , similar to previous reports ( Jahromi et al . , 1992 ) . However , bath application of adenosine triphosphate ( ATP ) or muscarine elicited a response in Schwann cells along distal phrenic nerve branches as well as TPSCs ( Figure 2—video 3 ) . Despite their expression of GCaMP3 ( determined by immunohistochemistry; data not shown ) , myelinating Schwann cells along larger phrenic nerve branches failed to respond to either nerve stimulation or bath application of ATP or muscarine . Together , these results show that TPSCs dynamically respond to different patterns of neuronal activity . Previous studies of individual or small cohorts of TPSCs show that a variety of neurotransmitter-derived substances trigger cytosolic Ca2+ accumulation , including ACh through muscarinic AChRs , adenine nucleotides such as ATP/ADP through purinergic P2Y receptors ( P2YR ) , and adenosine , derived from synaptic ectonucleotidase-mediated degradation of adenine nucleotides ( Cunha et al . , 1996 ) , through P1R ( Darabid et al . , 2014 ) . A recent study provided evidence that TPSCs also respond to nerve-derived ACh through nicotinic AChRs ( Petrov et al . , 2014 ) . In contrast to these responses in adult TPSCs , a recent report demonstarted that TPSC Ca2+ signals in neonatal mouse soleus are not mediated by muscarinic or nicotinic ACh receptor ( mAChR ) or by adenosine receptor activation , but rather by P2YRs ( Darabid et al . , 2013 ) . In agreement with this finding , we found that whereas the pan-muscarinic antagonist atropine and the pan-nicotinic antagonist curare failed to block activity-mediated Ca2+ transients in TPSCs of the P7 diaphragm ( n = 8 ) , the wide spectrum P2 antagonist suramin completely eliminated them ( n = 6; Figure 3C ) . We tested whether this response was mediated by P2Y1 receptors ( P2Y1Rs ) , as TPSCs reportedly express this protein ( Darabid et al . , 2013 ) and astrocytic Ca2+ signaling is mediated in part by the Gq GPCR-coupled pathway that is activated by P2Y1Rs ( Fam et al . , 2000 ) . Treatment with the selective P2Y1R antagonist MRS2500 ( 1 μM ) completely blocked activity-induced Ca2+ responses in all TPSCs of the diaphragm ( n = 5; Figure 3C ) . In order to determine whether activity-dependent , P2Y1R-mediated Ca2+ responses were dependent on release from intracellular stores , we examined them in the presence of the sarco-/endoplasmic reticulum Ca2+-ATPase ( SERCA ) inhibitor cyclopiazonic acid ( CPA ) . 15 min after treatment with CPA , these responses were completely abolished ( n = 3; Figure 3C ) . Finally , to test whether the effect of MRS2500 was indeed mediated by P2Y1Rs , we crossed mice expressing a constitutive null mutation of the P2ry1 gene , which encodes P2Y1Rs ( Fabre et al . , 1999 ) , to Wnt1-GCaMP3 mice . Nerve stimulation of P2ry1 mutants completely failed to elicit Ca2+ responses in TPSCs ( n = 7; Figure 3A , C; Figure 3—video 1 ) . Mutant TPSCs exhibited a robust response to muscarine ( Figure 3A , C ) , indicating that the failure of activity to induce these responses was not caused by non-specific effects of the P2ry1 mutation . We were intrigued by the inability of atropine to block activity-mediated Ca2+ responses in TPSCs of P2ry1 WT mice , since ( a ) bath application of muscarine evoked a robust Ca2+ signal . ; ( b ) atropine blocked this effect of bath-applied muscarine ( data not shown ) ; ( c ) ACh is released upon nerve stimulation . On the one hand , this may result from the absence of clustered mAChRs in P7 TPSCs ( Darabid et al . , 2013 ) . Alternatively , the lateral diffusion of nerve-derived ACh to perisynaptic TPSC-derived mAChRs may be limited by the activity of the cholinesterases acetylcholinesterase ( AChE ) or butyrylcholinesterase ( BChE ) in the synaptic cleft . Nerve stimulation of P2ry1 mutants in the presence of the pan-cholinesterase inhibitor neostigmine resulted in a robust Ca2+ response in all TPSCs , suggesting that cholinesterase activity normally prevents this effect ( n = 3; Figure 3B , C ) . Neostigmine also increased the intensity of activity-induced Ca2+ transients in P2ry1 WT mice , demonstrating the additive nature of the effects of purinergic and muscarinic stimulation on this response ( Figure 3D ) . We investigated which nerve-derived , P2Y1R-stimulating ligands were capable of evoking TPSC Ca2+ responses . While bath application of either ATP or ADP induced these responses , neither ADP-ribose nor β-nicotinamide adenine dinucleotide ( βNAD ) did ( n = 4; data not shown; Mutafova-Yambolieva and Durnin , 2014 ) . Bath application of the P1R-activating ligand adenosine also evoked robust Ca2+ transients in TPSCs ( n = 3; data not shown ) , similar to previous studies ( Robitaille , 1995; Castonguay and Robitaille , 2002 ) . However , the onset of this response was markedly delayed , compared to that triggered by purines or ACh mimetics ( ATP = 3 . 2 ± 1 . 6 s , adenosine = 15 . 6 ± 3 . 4 s; p<0 . 001 ) . Together , these pharmacological and genetic studies suggest that activity-induced Ca2+ signaling in neonatal diaphragm TPSCs is mediated by adenine nucleotide-mediated stimulation of P2Y1R . This result thus permits the evaluation of the functional role of this signal . In order to test whether P2Y1R deletion itself or activity-induced P2Y1R-mediated Ca2+ signaling exerted gross effects on synaptogenesis of the NMJ , we first assessed the structure of NMJs by immunohistochemical and ultrastructural techniques . We examined the tripartite NMJ by staining whole-mounts of P7 diaphragm with antibodies against synaptophysin ( nerve terminals ) , GFP ( GCaMP3-expressing perisynaptic Schwann cells ) and α-BTX ( postsynaptic AChR clusters ) . We found no difference in the total number of NMJs , the size of NMJs , the percentage of innervated NMJs , or the apposition of perisynaptic Schwann cells in P2ry1 mutant or WT mice ( Figure 3—figure supplement 1; data not shown ) . Next , we examined NMJs for AChE immunoreactivity , as previous reports of adult NMJs in P2ry1 mutants showed a reduction in the level of expression of this cholinesterase ( Xu et al . , 2015 ) . Using a highly specific antibody that fails to detect expression of this enzyme in AChE mutant mice , we were unable to observe any difference in its expression or synaptic localization ( Figure 3—figure supplement 1 ) . Finally , we examined NMJs by electron microscopy , and found no obvious structural abnormalities ( Figure 3—figure supplement 2 ) . Together , these results suggest that the gross morphological development of the NMJ , at least until P7 in the diaphragm , is unaffected in mice lacking P2Y1R and activity-induced Ca2+ signaling . We next evaluated the effect of eliminating activity-induced Ca2+ signaling on the presynaptic release of neurotransmitter , based on results obtained in previous studies ( Robitaille , 1998; Castonguay and Robitaille , 2001 ) . The resting membrane potential ( RMP ) was unchanged between genotypes ( −68 . 5 ± 5 . 2 vs . −68 . 4 ± 3 . 3 mV; WT vs . mutant; p=0 . 46; WT vs . mutant; n = 3 , c = 13 ) , and was not significantly different after vs . before a nerve stimulation bout between genotypes ( data not shown ) . The frequency , amplitude , rise to peak , and time to 50% decay of miniature EPPs ( mEPPs ) were also unchanged ( resting frequency: 0 . 38 ± 0 . 15 vs . 0 . 46 ± 0 . 14 events/s; p=0 . 11; post-stimulation frequency: 1 . 57 ± 0 . 8 vs . 2 . 01 ± 0 . 8 events/s; p=0 . 15; amplitude: 2 . 52 ± 0 . 5 vs . 2 . 49 ± 0 . 6 mV; p=0 . 44; rise to peak: 3 . 76 ± 1 . 8 vs . 3 . 69 ± 2 . 7 ms; p=0 . 46; time to 50% decay: 5 . 83 ± 2 . 4 vs . 4 . 81 ± 1 . 8 ms; p=0 . 08; WT vs . mutant; n = 3 , c = 13–19 ) . Individual nerve-evoked EPPs , recorded in the presence of μ-conotoxin , were also similar between P2ry1 WT and mutant mice ( Figure 4A ) . In response to HFS , EPP amplitudes at the end of the period were also similar in each genotype . These results demonstrate that basal and HFS-induced ACh release are not affected in the absence of activity-induced Ca2+ signaling in TPSCs ( Figure 4A , D ) . These results also corroborate the finding that AChE expression at the NMJ was unaffected in P2ry1 mutants , as the durations of mEPPs and EPPs were unaffected , whereas they are longer in the absence of this enzyme ( Adler et al . , 2011 ) . In order to assess postsynaptic function , we took advantage of BHC , a drug which blocks contraction of skeletal muscle without affecting neurotransmission and thus allows the electrophysiological and optical evaluation of muscle APs ( Heredia et al . , 2016 ) . We first assessed individual nerve-evoked muscle APs and observed no differences between P2ry1 WT and mutant mice ( Figure 4B ) . In order to determine the effects of HFS on muscle APs , we initially examined neural transmission failure , or the failure to transmit a successful EPP into a muscle AP , by identifying the time at which less than half of the nerve stimuli were transduced into successful muscle APs . Similar to the adult diaphragm ( Heredia et al . , 2016 ) , P7 diaphragm exhibited multiple muscle AP profiles in response to HFS , characterized by the occurrence of failed or subthreshold APs at different timepoints after stimulation , likely reflecting differential fatiguability . We were unable to detect differences in the time to neural transmission failure in any subtype ( Figure 4D ) , consistent with the failure to detect differences in EPP or mEPP amplitude , which reflect the presynaptic release of and the postsynaptic response to ACh , respectively . However , when we examined the features of successfully transmitted muscle APs at different stages of HFS , we found that the amplitudes were smaller and durations longer in P2ry1 mutant relative to WT mice ( Figure 4E ) . These results suggest that the muscle AP itself , rather than the transmission of the nerve impulse to the muscle , is affected in the absence of activity-induced , P2Y1R-mediated Ca2+ responses in TPSCs . Because previous studies reported that impaired muscle APs are correlated with muscle fatigue ( Juel , 1988 ) , we evaluated muscle force in the P7 diaphragm of P2ry1 mutant and WT mice . We used an optical measure of fiber shortening in whole diaphragm to measure muscle peak force and muscle fatigue ( Heredia et al . , 2016 ) . When we examined the effect of 45 s of 40 Hz phrenic nerve stimulation , we detected no difference in the magnitude of peak contraction ( Figure 5A , B ) . However , peak contraction was maintained for longer durations in P2ry1 WT than mutant mice ( Figure 5B ) . These results were also obtained in WT mice treated with the P2Y1R antagonist MRS2500 , demonstrating that the acute inactivation of P2Y1R function is sufficient to enhance fatigue ( Figure 5C ) . When we subjected the diaphragm to multiple bouts of HFS , each separated by a recovery period of 15 min , we found that the initial peak contraction , as well as ability to maintain peak contraction , were significantly reduced in P2ry1 mutants ( Figure 5D ) . Finally , in order to test whether nerve stimulation-induced muscle fatigue was enhanced as a result of impaired neuromuscular synapse transmission , we stimulated muscle rather than nerve . We failed to observe differences in peak contraction or fatigue in response to 40 s of 45 Hz electrical field between P2ry1 WT and mutant mice ( data not shown ) . Collectively , these data demonstrate that the perisynaptic region of muscle fibers of the P7 diaphragm is more sensitive to fatigue induced by HFS in P2ry1 mutant than WT mice . We next examined muscle fiber subtype in the diaphragm , since muscle AP failure profiles with different fatiguability were observed in response to HFS , and since the development of these subtypes reflects the endogenous pattern of nerve stimulation . Although earlier studies indicate that the development of these fiber subtypes first occurs at around P25 in rodent diaphragm ( Zhan et al . , 1998 ) , we observed that both P2ry1 WT and mutant mice contained all four basic fiber subtypes at P7 , as assessed by immunostaining with myosin heavy chain ( MHC ) antibodies that selectively recognize each fiber subtype ( Bloemberg and Quadrilatero , 2012 ) . However , the relative percentage of each of these subtypes was indistinguishable between genotypes , suggesting that the enhanced fatigue in P2ry1 mutants is not caused by a relative increase in fast-fatiguing subtypes of muscle fibers ( Figure 5E ) . A recent study demonstrated that neonatal TPSCs respond to distinct levels of nerve activity during the period of polyneuronal synapse elimination by modulating the magnitude of their Ca2+ response ( Darabid et al . , 2013 ) . Together with the finding that TPSC processes separate competing nerve terminals from each other and from the postsynaptic muscle fiber during this period ( Smith et al . , 2013 ) , these data suggest that TPSC Ca2+ responses may regulate this phenomenon . Further support for this idea comes from the finding that synapse elimination in the CNS is impaired in P2ry1 mutant mice as well as in mice lacking activity-induced Ca2+ responses in astrocytes ( Yang et al . , 2016 ) . In order to evaluate this hypothesis , we confirmed that Ca2+ responses were eliminated in the absence of P2Y1R signaling at P15 , the age at which synapse elimination is largely complete . Similar to those at P7 , TPSC Ca2+ responses at P15 were completely dependent on P2Y1R signaling , both in P2ry1 WT mice treated with MRS2500 or in P2ry1 mutant mice ( Figure 6A , B ) . Fatigue was similarly enhanced , using both optical and tension measurements ( Figure 6C ) . However , when we examined NMJs by neurofilament immunohistochemistry to detect the numbers of innervating axons at individual NMJs at several ages between P7 and P15 , we were unable to observe any differences ( Figure 6D , E ) . Together , these results suggest that activity-induced Ca2+ responses in TPSCs are not required for polyneuronal synapse elimination in the developing diaphragm . A variety of mechanisms underlie muscle fatigue . In order to assess which of these might be affected by TPSC Ca2+ accumulation , we initially examined intracellular Ca2+ release within muscle cells . In order to investigate activity-induced Ca2+ signaling in whole populations of diaphragm muscle cells , we crossed Myf5-Cre to conditional GCaMP3 mice . In unparalyzed muscle under epifluorescence , we measured both fiber length changes and Ca2+ fluorescence intensities in response to HFS . Similar to the results obtained with brightfield recordings obtained above , MRS2500 enhanced the fatigue of a second bout of HFS relative to a first , compared to no treatment ( data not shown ) . Peak intensities and time to 50% decay of Ca2+ transients were significantly affected in response to HFS in the presence of MRS2500 ( Figure 5—figure supplement 1 ) , suggesting that events upstream or concurrent with Ca2+ release mediate muscle fatigue caused by the absence of TPSC Ca2+ signaling . We also used Myf5-GCaMP3 mice to test whether nerve-derived purines were capable of eliciting Ca2+ responses in muscle cells , similar to previous results ( Choi et al . , 2001 ) . In response to 100 μM ATP , muscle cells of the diaphragm of P7 Myf5-GCaMP3 mice failed to exhibit such a response ( Figure 5—figure supplement 1 ) . Direct electrical stimulation of skeletal muscle produces fatigue , similar to indirect , nerve-mediated excitation . Interestingly , in response to high levels of extracellular potassium or high [K+]o , this activity-induced fatigue is enhanced , and muscle APs exhibit lower amplitudes and longer durations ( Cairns et al . , 2015 ) , similar to the response of P2ry1 mutants to nerve activity shown above . These results raise the possibility that activity-induced Ca2+ signaling in TPSCs protects against muscle fatigue by regulating K+ uptake by these cells and therefore persiynaptic [K+o] . similar to reports of other perisynaptic glia ( Wang et al . , 2012a , Wang et al . , 2012b ) . In order to test this idea , we first challenged diaphragms to [K+]o greater or less than normal levels ( 5 mM ) . These challenge experiments were modeled on those used to characterize the effects of hypo- and hyperkalemia on skeletal muscle function ( Wu et al . , 2011 ) . We found that HFS-induced fatigue was disproportionately enhanced in P2ry1 mutant , relative to WT , mice in response to 10 mM [K+]o ( Figure 7A ) . The enhanced fatigue in these mutants was further revealed by multiple bouts of HFS; P2ry1 mutant diaphragm was almost completely unable to contract after the second period of HFS in high [K+]o , in marked contrast to P2ry1 WT diaphragm ( Figure 7A; Figure 7—video 1 ) . We next examined whether the effect of high [K+]o was caused by depolarization of the postsynaptic muscle membrane . After stimulation with several bouts of HFS in 5 mM [K+]o , muscle cells were impaled and recorded before and after changing the [K+]o to 10 mM . While this caused a mild depolarization of the RMP in P2ry1 WT mice , this effect was enhanced in P2ry1 mutants ( Figure 7B ) . Conversely , HFS-induced fatigue was modestly but not significantly ameliorated in response to low [K+]o in P2ry1 mutants ( Figure 7C ) . Collectively , these results demonstrate that P7 diaphragm muscle cells lacking activity-induced , P2Y1R-mediated Ca2+ signaling in TPSCs are more sensitive to high [K+]o , suggesting that Ca2+ signaling modulates the response to [K+]o in these perisynaptic glia . To test if TPSCs respond to and/or regulate K+ at the NMJ , we examined the response of these cells to changes of [K+]o . Interestingly , we found that raising [K+]o to 10 mM resulted in a robust TPSC Ca2+ response ( Figure 7D ) . This was not mediated by indirect depolarizing effects of [K+]o on the phrenic nerve , leading to an activity-induced , P2Y1R-mediated Ca2+ response , because it was still observed in the presence of doses of tetrodotoxin that blocked neurotransmission , and because it was also observed in P2ry1 mutants completely lacking activity-induced Ca2+ responses . In other words , these responses were observed in the absence of nerve stimulation . Interestingly , Ca2+ responses induced by high [K+]o were not caused by influx of extracellular Ca2+ into TPSCs , because they were still observed when external Ca2+ was removed , but rather by release from intracellular stores , because they were abrogated after treatment with CPA ( n = 3; data not shown ) . In order to determine if TPSCs directly respond to manipulations of [K+]o , we performed whole cell voltage recordings of these cells , identified by GCaMP3 expression in P7 Wnt1-GCaMP3 mice and co-localization with α-BTX-labeled NMJs . We observed that the RMP of TPSCs depolarized in response to treatment with 10 mM KCl by an amount that was close to that predicted by the Hodgkin-Goldman-Katz equation ( data not shown ) . Therefore , these data demonstrate that TPSCs are capable of taking up K+ and that elevations of [K+]o depolarize TPSCs , leading to a release of Ca2+ from intracellular stores , similar to depolarization-induced intracellular Ca2+ release reported in neurons ( Ryglewski et al . , 2007 ) . If K+ uptake is affected in TPSCs of P2ry1 mutants lacking activity-induced Ca2+ signaling , then this [K+]o-induced Ca2+ response in these cells might in turn be affected . In order to test this idea , we subjected diaphragms to several bouts of HFS , separated by 15 min , to mimic the effects on muscle fatigue described above , then assessed the effects of high [K+]o on Ca2+ signaling . In contrast to P2ry1 WT mice , mutants or WT mice treated with MRS2500 showed a markedly reduced Ca2+ response to 10 mM [K+]o ( Figure 7D , E ) , suggesting that K+ uptake is impaired in P2ry1 mutants . This failure of TPSCs to regulate perisynaptic [K+]o may contribute to the enhanced muscle fatigue that occurs in P2ry1 mutants lacking activity-induced TPSC Ca2+ responses . Additionally , these results suggest that TPSC Ca2+ responses , and consequently K+ uptake , are positively regulated by nerve activity through feedforward ( i . e . , neurotransmitter-mediated stimulation ) and feedback mechanisms ( i . e . , by suprathreshold [K+]o itself . Our results using Wnt1-GCaMP3 mice demonstrate that activity-induced Ca2+ signaling in neonatal TPSCs of the diaphragm is mediated by P2Y1R activation by nerve-derived adenine nucleotides . The absence of Ca2+ signaling within TPSCs does not appear to affect the structural and molecular development of the NMJ , nor does it alter the presynaptic release of neurotransmitter , but rather affects the postsynaptic AP during sustained HFS , where longer , smaller APs were correlated with a failure to maintain peak muscle force . Because previous studies observed that administration of high K+ induced similar effects on muscle APs subjected to fatiguing muscle stimulation ( Cairns et al . , 2015 ) , we examined muscle fatigue in response to this treatment and found that it was enhanced to a greater degree in P2ry1 mutants lacking activity-induced Ca2+ signaling . This heightened susceptibility to high [K+]o may be caused by impaired K+ uptake by P2ry1 mutant TPSCs , as these cells exhibited a markedly reduced release of Ca2+ from intracellular stores in response to high [K+]o treatment . Collectively , these results suggest that activity-induced , P2Y1R-mediated Ca2+ signaling in TPSCs influences muscle fatigue by regulating perisynaptic [K+]o ( Figure 8 ) . The current study represents the first evaluation of Ca2+ responses using genetically encoded calcium indicators in perisynaptic glia at the NMJ . The onset after nerve stimulation and time to peak intensity of Ca2+- transients using this method are similar to those of published studies using Ca2+- sensitive dyes in TPSCs loaded with Fluo-3 as well as astrocytes expressing GCaMP3 ( Jahromi et al . , 1992; Reist and Smith , 1992; Darabid et al . , 2013; Akerboom et al . , 2013 ) , demonstrating that this genetic technique is a valid tool to measure these responses in large populations of TPSCs . The most striking finding from this study is the complete dependence of TPSC Ca2+- responses on a single GPCR , P2Y1R . Therefore , despite the fact that bath administration of a multiple substances induces widespread Ca2+ signaling in neonatal TPSCs , activity-induced responses are only mediated by adenine nucleotides . In contrast to these results , studies of adult TPSCs support a role for ACh and other factors in mediating these responses ( Darabid et al . , 2014 ) . Therefore , the early exclusive dependence on P2Y1R-activating adenine nucleotides may broaden over time . Alternatively , TPSCs at the NMJs of the diaphragm may continue to depend exclusively on P2Y1R signaling . At the oldest ages at which we were able to examine population responses ( P15-P20 ) , activity-induced Ca2+ signals were completely dependent on the P2Y1R pathway , supporting this latter idea . Moreover , the prevention of ACh diffusion to perisynaptic mAChRs by cholinesterase is unlikely to represent a developmentally transient response . Indeed , a recent report described a functional role for TPSC-derived BChE at the adult NMJ ( Petrov et al . , 2014 ) . Our results fail to support the idea that TPSC Ca2+ responses affect the presynaptic release of ACh , in contrast to previous studies ( Robitaille , 1998; Castonguay and Robitaille , 2001 ) . NMJs at the diaphragm do exhibit dynamic changes of ACh release such as facilitation and depression ( Vautrin et al . , 1993 ) , arguing against the absence of plasticity at diaphragm NMJs as an explanation of these differences . Rather , they may be attributable to different species , muscle , age or technique . Interestingly , using a similar genetic approach to study the effect of eliminating activity-induced Ca2+ responses in astrocytes neurotransmitter release was unaffected ( Agulhon et al . , 2010 ) . On the other hand , postsynaptic function was affected at the NMJ in response to HFS in P2ry1 mutants . Whereas the number of successful muscle APs , reflecting the presynaptic release of and response to ACh , was not different in response to HFS , the characteristics of these APs changed significantly in response to prolonged nerve stimulation . These results suggest that muscle APs may not be transduced as efficiently in P2ry1 mutant mice . The reduced intensity of muscle Ca2+ transients in response to pharmacological blockade of P2Y1Rs supports this contention , as does the enhancement of muscle fatigue in response to pharmacological or genetic inhibition of this pathway . These effects may result from the absence of P2Y1R function in muscle , as the mice used in this study were constitutive mutants . For example , Choi et al . ( 2001 ) found that stimulation of chick muscle fibers with 100 μM ATP triggered Ca2+ release from intracellular stores , a response blocked by the pan-P2 blocker suramin . Together with the expression of P2Y1R by muscle , these findings suggest that nerve-derived ATP may modulate intracellular muscle Ca2+ levels and thus muscle fatigue . However , we failed to observe Ca2+ release in response to ATP in the muscle cells of Myf5-GCaMP3 mice . Moreover , Ca2+ signals in TPSCs were observed in response to treatment with lower doses of ADP/ATP ( 10–20 μM ) than those used to evoke these signals in chick muscle cells . Finally , Ca2+ signals in TPSCs were blocked after P2Y1R was blocked by specific pharmacological and genetic tools , rather than the pan-P2R blocker suramin ( Choi et al . , 2001 ) . Thus , we favor the idea that the reduction of muscle cell Ca2+ mobilization in P2ry1 mutants is caused by the absence of this protein in TPSCs rather than in muscle cells . The effects of HFS on muscle APs suggested the possibility that perisynaptic [K+]o was dysregulated in P2ry1 mutants . Supporting this idea , treatment with high [K+]o enhanced muscle fatigue to a greater extent in P2ry1 mutant than in WT mice . Together with a report that intense exercise increases [K+]o in muscle to 10–14 mM ( Mohr et al . , 2004 ) , a level sufficient to cause muscle fatigue ( Sjøgaard , 1990 ) ; but see Shushakov et al . , 2007 ) , these data indicate that muscle fatigue is enhanced in P2ry1 mutants as a result of elevated perisynaptic [K+]o , caused by a failure of TPSCs to spatially buffer or take up K+ ( Kofuji and Newman , 2004 ) . In order to test this idea , we examined the effects of treatment with high [K+]o , reasoning that if K+ uptake mechanisms in TPSCs were impaired in P2ry1 mutants lacking activity-induced Ca2+ signaling , these effects would be diminished . Indeed , we found that Ca2+ responses to 10 mM [K+]o were markedly reduced in these mutants , providing indirect evidence that K+ uptake was impaired . Interestingly , elevation of [K+]o to 20 mM also induced a robust increase of intracellular Ca2+ in cultured astrocytes ( Duffy and MacVicar , 1994 ) . However , in contrast to the current study , this response was mediated entirely by external influx through voltage-gated Ca2+ channels ( VGCC ) . Therefore , K+-induced Ca2+ responses in TPSCs do not appear to result from depolarization-mediated ingress through VGCC , but rather by the release from intracellular stores . Together , these data suggest that activity stimulates perisynaptic K+ uptake in TPSCs by both feedforward Ca2+ responses initiated by neurotransmitter and feedback signals initiated by high [K+]o itself . The intracellular uptake of K+ has been demonstrated in Müller glia in the retina ( Newman et al . , 1984 ) as well as in other glial subtypes and may be mediated by several mechanisms , including inwardly-rectifying potassium channels ( Kir ) , Na+ , K+ ATPases , and Na+ , K+ Cl- cotransporters . The importance of perisynaptic K+ regulation by glial cells has been demonstrated by several genetic studies . For example , mice lacking Kir4 . 1 in astrocytes exhibit impaired K+ and neurotransmitter uptake , leading to seizures , ataxia and early lethality ( Djukic et al . , 2007 ) . In contrast , Wang et al . ( 2012b ) observed that activity-induced Ca2+ responses in astrocytes are required for K+ uptake through an ouabain-sensitive Na+ , K+ ATPase activity . Future studies will determine which if any inward K+ conductance is expressed in and stimulated by activity within TPSCs , as well as the mechanisms by which increases of intracellular Ca2+ lead to enhanced K+ uptake . Of note , it has been established that nonmyelinating Schwann cells in sympathetic nerves possess Kir currents that are sensitive to neural activity ( Konishi , 1994 ) . In addition to [K+]o , the extracellular concentrations of other ions are dysregulated during muscle fatigue , including [Na+]o and [H+]o ( Allen et al . , 2008 ) . Because astrocytes express an abundance of transporters and ion channels that modulate the levels of these ions in response to neuronal activity , TPSCs may similarly regulate these ions in response to stimulation of muscle , in addition to [K+]o . On the other hand , muscle itself is equipped for this role , expressing an abundance of ion channels and transporters along the extensive t-tubule system that regulate muscle membrane potential and excitability during activity ( Fraser et al . , 2011 ) . However , the importance of ionic homeostasis at the NMJ may depend on additional mechanisms , such as those proposed here . For example , the sensitivity of Nav1 . 4 , which is expressed at high levels in a restricted region in the depths of the postsynaptic junctional folds ( Stocksley et al . , 2005 ) , to the inactivating effects of depolarization ( Cannon , 2015 ) , may require enriched expression of [K+]o buffering proteins by TPSCs ( Figure 8 ) . Consistent with this idea , P2ry1 mutants did not exhibit enhanced fatigue in response to direct muscle stimulation . In summary , we have utilized the diaphragm of neonatal P2ry1 mutant mice as a model to explore the functional significance of activity-induced Ca2+ signals in perisynaptic glia . We found that in the absence of purinergic signaling , postsynaptic rather than presynaptic function was altered , leading to enhanced muscle fatigue in response to HFS . These effects were correlated with elevated [K+]o and reduced responsivity to [K+]o , suggesting that activity-induced Ca2+ responses in TPSCs regulate perisynaptic [K+]o . Future studies will determine the mechanisms underlying K+ uptake and [K+]o-mediated Ca2+ accumulation in TPSCs . Such mechanisms may represent important translational targets in diseases with altered [K+]o . For example , in patients with hyperkalemic periodic paralysis , a genetic disorder caused by Scna4 mutations and characterized by elevated [K+]o , stimulation of Ca2+ signaling and subsequently [K+]o buffering within TPSCs may enhance neuromuscular function . P2ry1 mutant , GCaMP3 or GCaMP6f conditional knockin , and Wnt1-Cre and Myf5-Cre transgenic mice were all purchased from Jax . P2ry1 null mutant mice were backcrossed into the C57/Bl6 strain several times before crossing to other strains , each of which is maintained in the C57/Bl6 strain . We could find no difference in any experiment between P2ry1+/+ and P2ry1+/- mice , so we pooled these samples and denoted them all in the text as ‘WT . ’ Similarly , we found no difference between male and female P2ry1+/-mice , so we pooled these samples . In order to generate P2ry1 mutants expressing GCaMP3 in Schwann cells , we generated Wnt1-Cre; P2ry1+/- and Rosa26-GCaMP3flox/flox; P2ry1+/- mice and crossed them , such that all P2ry1 mutant and heterozygote mice expressed only one copy each of Cre and GCaMP3 . We used a slightly modified common 3’ WT primer to genotype P2ry1 mutant mice ( ATT TTT AGA CTC ACG ACT TTC ) and the recommended primers by Jax for all other alleles . All studies were performed with animals aged postnatal day 7 and 15 ( P7 , P15 ) . To verify knockouts , we performed RT-PCR on muscle-derived RNA with primers against a 300 bp fragment of P2ry1 ( 5’: CTG TGT CTT ATA TCC CTT TCC , 3’: CTC CAT TCT GCT TGA ACT C ) . Animal husbandry and experiments were performed in accordance with the National Institutes of Health Guide for the Care and Use of Laboratory Animals and the IACUC at the University of Nevada . The following reagents were used at the following concentrations: P2Y1R agonists ATP and ADP ( Sigma; 10 or 20 μM ) ; P2Y1R antagonist MRS2500 ( Tocris; 1 μm ) ; P1R agonist adenosine ( Sigma; 100 μM ) ; pan-P2 antagonist suramin ( Sigma; 100 μM ) ; pan-muscarinic agonist muscarine ( Sigma; 10 μM ) ; pan-muscarinic blocker atropine ( Sigma; 10 μM ) ; pan-nicotinic agonist nicotine ( Sigma; 50 μM ) ; pan-nicotinic antagonist curare ( Sigma; 200 μM ) ; pan-cholinesterase inhibitor neostigmine ( Sigma; 1 μM ) ; sarco-/endoplasmic reticulum Ca2+-ATPase ( SERCA ) inhibitor , cyclopiazonic acid ( Sigma; CPA; 10 μM ) ; potassium chloride ( Sigma; 2–10 mM ) ; GIIIb μ-conotoxin ( Peptides International; 2 . 3 μM ) ; skeletal muscle myosin-blocker 3- ( N-butylethanimidoyl ) −4-hydroxy-2H-chromen-2-one ( BHC; Hit2lead; 100 μM ) ; 488- , 594- , 633-conjugated-α-bungarotoxin ( α-BTX; Biotium; 1 μg/mL ) . The diaphragm of Wnt1-GCaMP3 mice was illuminated with a Spectra X light engine ( Lumencor ) . In order to quantify maximal fluorescence ( Fmax ) exhibited by GCaMP3 in Schwann cells , 30 μM CPA was added to deplete sarcoplasmic reticular Ca2+ stores ( Heredia et al . , 2016 ) . Image sequences were captured using an Andor Neo sCMOS camera and a Windows-based PC using Nikon NIS Elements 4 . 1 . Image sequences were recorded at 25 frames per second , and were exported as 8-bit TIFF files into custom-written software ( Volumetry G8d; logic underlying methods in Source code file 1 ) . A Gauss filter ( 3 × 3 pixel , sd = 1 . 0 ) was applied to reduce camera noise , and motion-correction routines were used to stabilize neural and Schwann cell elements in the movie ( see Hennig et al . , 2015 ) . Changes in background fluorescence were stabilized by subtracting the average intensity near the main phrenic nerve branch . An average intensity image was generated before stimulus application ( ‘Pre-stim’ ~1 s ) to quantify basal Ca2+ levels in Schwann cells . These images are presented using a blue->green color lookup table ( CLUT ) . This image was subtracted from the entire movie , thereby filtering out static fluorescent structures and displaying only objects that changed their intensity , i . e . , Ca2+ transients . A number of statistical maps ( stat maps ) were used to portray and analyze the pattern of activity-induced Ca2+ transients in TPSCs . The main stat map type used to portray the amplitude of Ca2+ transients in TPSCs was the standard deviation ( SD ) map . This map was calculated in similar fashion to the average intensity image , except the standard deviation of 16-bit fluorescence intensity units ( SD iu16 ) at every pixel prior to the application of the stimulus ( 0 . 5–1 . 0 s ) extending to 60 s was calculated . Intensity SD projections ( SD iu16 ) were used to portray changes in Ca2+-induced fluorescence as they are more descriptive of the overall changes in intensity during the recording; average intensity projections ( iu16 ) can be somewhat misleading depending on how much of the ‘non-active’ time before and after an event is included in the average . Intensity averages also don’t describe how much the signal fluctuates . Per pixel maximum or maximum-minimum calculations can also be misleading as they include the maximum noise amplitude and/or artifacts ( i . e . , shot noise ) . SD remedies these issues , as it describes intensity fluctuations ( noise or event ) in standardized units and it is less sensitive to the time periods chosen to perform the projection/s . Overall , this approach isolates actively fluorescing structures more clearly than other types of projections . SD maps were color coded using a ‘Fire’ CLUT . Traces of fluorescence intensity were generated from movies and presented as changes in fluorescence with respect to initial fluorescence ( ΔF/Favg ( prestim ) or ΔF/Fo ) . Peak intensities of Ca2+ transients were calculated as a signal-to-noise ratio ( SNR ) in which peak standard deviation values were divided by the prestim standard deviation value . The log10 of this ratio was generated and multiplied by 20 to standardize the SNR as decibels ( dB ) . Decibels ( signal-to-noise ) are commonly used in signal transduction to describe the strength of a signal in terms of its ‘resolvability’ above background . This approach allows one to definitively characterize signals in terms of the ‘noise’ in the surrounding environment . We have begun using imaging decibels to more accurately portray Ca2+ signals in relation to the surrounding environment - which essentially relates the peak amplitude of the Ca2+ event to the standard deviation of the signal before the event ( on a log scale ) . For more information on the logic flow underlying these routines , see source code file entitled , ‘Image Analysis . ’ Drugs were either bath applied in proximity to the motor endplate or pressure injected ( PDES-O2DX; NPI Electronic ) . Drugs dissolved in DMSO were either perfused in or diluted in 1 mL of Kreb’s-Ringer’s before bath application , as bath application of small volumes of DMSO ( ~8 μl into 8 ml chamber ) caused fluorescence within Wnt1-GCaMP3-expressing TPSCs . For experiments with altered potassium , a stock solution of 3M [KCl] was added to the bath to change the concentration from 5 mM to 10 mM immediately prior to imaging . Diaphragms were dissected and pinned on a 6 cm Sylgard-coated dish containing oxygenated Krebs-Ringer’s solution at RT according to standard procedure . Stimulation and recording of intracellular potentials were performed as described ( Heredia et al . , 2016 ) . Briefly muscle APs were recorded after treatment with 100 μM BHC for 30 min , followed by 30 min of washing . Endplate potentials ( EPPs ) were recorded after treatment with the Nav1 . 4 antagonist μ-conotoxin GIIIb ( μ-CTX; 2 . 3 μM ) . Signals were amplified , digitized , recorded and analyzed as described ( Heredia et al . , 2016 ) . The temporal dynamics of EPPs ( rise to peak; time to 50% decay ) were recorded as gross approximations of endplate current ( EPC ) kinetics changes , as has been done previously in the presence or absence of cholinesterase inhibitors ( Fatt and Katz , 1951; Beránek and Vyskocil , 1968; Kuba and Tomita , 1971 ) . Only muscle fibers with resting membrane potentials between −60 and −75 mV were included for analysis . Stimulation episodes of the phrenic nerve over 10 Hz were separated by 30 min rest periods to allow recovery . In order to calculate percent failure ( APs ) or percent transmitter release rundown ( EPPs ) , the average of three potentials at a particular timepoint ( e . g . , the time at which fewer than half of nerve stimuli produced a successful muscle AP ) was taken and expressed as a percent of the average of the first three potentials . For experiments with altered KCl , NaCl was adjusted accordingly to maintain the same Cl- concentration before being perfused into the dish . For recording of whole cell membrane potential , diaphragms from Wnt1-GCaMP3 mice were treated with 10 mg/ml collagenase for 30 min at 37°C . TPSCs near the surface were identified by green fluorescent GCaMP3 signal in somatic cytosol , located adjacent to red fluorescent α-BTX-stained endplates . Whole cell recording configuration was achieved using 2 μm borosilicate pipettes pulled to 5–7 mΩ tip resistance . Whole cell access was 8–20 mΩ , and leak current was <-50 pA when voltage-clamped at −60 mV command potential . Pipette internal solution contained ( in mM ) : 97 . 5 K-gluconate , 32 . 5 KCl , 40 HEPES , 12 Na-phosphocreatine , 2 MgATP , 0 . 5 GTP , and 0 . 5 EGTA . Voltage recordings were made using current-clamp mode ( I = 0 ) on a HEKA EPC 10 amplifier controlled by Patchmaster software . Current injection steps of 100 ms duration were given in increments between −200 to +700 pA . [K+]o . was then altered by superfusion of bath solution from 5 to 10 mM KCl to determine if a potassium conductance existed in these cells at resting membrane potential . Tension recording of muscle force in diaphragm strips ( P15 ) or video recording of muscle shortening in hemidiaphragms ( P7 , P15 ) in response to nerve or muscle stimulation was performed as described ( Heredia et al . , 2016 ) . For experiments with altered KCl , NaCl was adjusted accordingly to maintain the same Cl- concentration before being perfused in . Antibodies against GFP ( Rockland ) , S100 ( Dako ) , synaptophysin ( Santa Cruz ) , neurofilament ( Millipore ) and acetylcholinesterase ( kindly provided by P . Taylor , UCSD ) were used at 1/1000 in PBS containing 1% triton-X and 10% fetal bovine serum to detect proteins in fixed , whole-mount diaphragms . Fluorescently-conjugated α-BTX and fasciculin-2 were added with secondary antibodies . Tissues were confocally imaged with an Olympus Fluoview 1000 . For myosin heavy chain staining , muscles were fresh-frozen , cut at 16 μm , and immediately incubated without fixation in PBS with primary antibodies as described ( Heredia et al . , 2016 ) . P7 mice were transcardially perfused in 1 . 5% glutaraldehyde , 2% paraformaldehyde in 0 . 1M sodium cacodylate . The costal diaphragm was dissected and incubated in fixative at 4°C overnight and then in rinse for several hours at 4°C . Samples were post-fixed in 2% osmium tetroxide , dehydrated , incubated in propylene oxide , embedded in Spurr’s resin and polymerized at 60°C overnight . Ultrathin sections were cut at 90 μm and stained with uranyl acetate followed by lead citrate . Sections were photographed or digitized using a Phillips CM10 transmission electron microscope equipped with a Gatan BioScan digital imaging system . Power analyses were performed using G*power 3 . 010 to determine the numbers of P2ry1 wild-type ( WT ) and mutant mice required . For example , to determine the number of mice ( n ) and cells ( c ) to analyze for electrophysiological recordings , a power of 0 . 8 , significance or alpha of 0 . 05 and effect size or Pearson’s r of 3 . 6 was used . Thus , for these experiments , data was generated from c = 3 per animal or more , and from n = 3 or more . In this case , each c and each n are biological replicates . Differences between means were assessed by unpaired Student t-tests , in some cases with the Bonferonni correction for multiple comparisons , assuming equal variance , or evaluated using analysis of variance ( ANOVA ) with Tukey post-hoc tests . As mentioned above , a p value < 0 . 05 was considered significant . Student t-tests were tested for significance with two tails if the direction of the outcome was not predicted ( e . g . , initial comparisons of morphology , MHC isoform , synapse elimination , electrophysiology , calcium signals and shortening/fatigue between P2ry1 WT and mutants ) and with one tail if an outcome was predicted ( e . g . , subsequent calcium and shortening/fatigue experiments between P2ry1 WT and mutant after treatment with MRS2500 or high/low potassium ) . In source data files , all reported statistical tests from text or figure legends are italicized and in red font .	A muscle that contracts over and over again will become tired . This can sometimes occur after vigorous exercise , but abnormal muscle fatigue is also a feature of various clinical disorders . These include conditions that affect muscles directly , such as muscular dystrophy , as well as disorders of the motor nerves that control muscles , such as Guillain-Barré syndrome . Nerves make contact with muscles at specialized sites called neuromuscular junctions . Failing to send the correct signals to the muscles at these junctions can lead to muscle fatigue . Studies to date have focused on the role of nerve cells and muscle cells in these communication failures . But there is also a third cell type present at the neuromuscular junction , known as the terminal/perisynaptic Schwann cell ( TPSC ) . Stimulating motor nerves in a way that produces muscle fatigue also activates TPSCs . To investigate whether TPSCs contribute to or counteract muscle fatigue , Heredia et al . studied the responses of these cells at the neuromuscular junctions of young mice . Stimulating motor nerves caused TPSCs to release calcium ions from their internal calcium stores . However , this did not occur in mice that lacked a protein called the P2Y1 receptor . In normal mice , activating the P2Y1 receptor directly also made the TPSCs release calcium . This calcium release in turn prompted the TPSCs to take up potassium ions . Nerve and muscle cells release potassium during intense activity , and removal of potassium by TPSCs helped to prevent muscle fatigue . Therapeutic strategies that make TPSCs release more of their internal calcium stores – and thus increase their potassium uptake – could help ease muscle fatigue . A valuable first step would be to use drugs and genetic techniques to show this effect in mice . The results could then guide the development of corresponding strategies in patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Activity-induced Ca2+ signaling in perisynaptic Schwann cells of the early postnatal mouse is mediated by P2Y1 receptors and regulates muscle fatigue
The insect central complex ( CX ) is a conserved brain region containing 60 + neuronal subtypes , several of which contribute to navigation . It is not known how CX neuronal diversity is generated or how developmental origin of subtypes relates to function . We mapped the developmental origin of four key CX subtypes and found that neurons with similar origin have similar axon/dendrite targeting . Moreover , we found that the temporal transcription factor ( TTF ) Eyeless/Pax6 regulates the development of two recurrently-connected CX subtypes: Eyeless loss simultaneously produces ectopic P-EN neurons with normal axon/dendrite projections , and reduces the number of E-PG neurons . Furthermore , transient loss of Eyeless during development impairs adult flies’ capacity to perform celestial navigation . We conclude that neurons with similar developmental origin have similar connectivity , that Eyeless maintains equal E-PG and P-EN neuron number , and that Eyeless is required for the development of circuits that control adult navigation . Work over the past two decades has revealed two important developmental mechanisms that generate neuronal diversity from flies to mice . First , spatial patterning cues produce different pools of neural progenitors ( called neuroblasts in insects ) ; second , neuronal progenitors/neuroblasts sequentially express a series of transcription factors that generate additional neuronal diversity ( Kohwi and Doe , 2013 ) . These so-called ‘temporal transcription factors’ or TTFs are expressed transiently in progenitors , are inherited by neurons born during the expression window , and specify progenitor-specific neuronal identity ( Rossi et al . , 2017; Doe , 2017 ) . For example , the Hunchback ( Hb ) TTF is present in Drosophila embryonic neuroblasts as they produce their first progeny; loss of Hb leads to absence of first-born neurons , whereas prolonging Hb expression generates ectopic first-born neurons ( Isshiki et al . , 2001 ) . While TTFs are clearly important for generating molecularly distinct neuronal subtypes , their role in establishing neuronal morphology , connectivity , and behavior remains relatively poorly understood . Recent work has shown that there are only four bilateral ‘type II’ neuroblasts that generate the intrinsic neurons of the central complex ( CX ) projecting to the protocerebral bridge ( PB ) . These four neuroblasts are named DM1-DM4 ( Yang et al . , 2013; Andrade et al . , 2019 ) or DM1-DM3 and DM6 ( Riebli et al . , 2013 ) ; here we use the DM1-DM4 nomenclature ( Figure 1A ) . Type II neuroblasts have a complex lineage . They repeatedly divide every 1 . 6 hr to generate a series of molecularly distinct intermediate neural progenitors ( INPs ) , which in turn divide every 2–3 hr to produce 4–6 molecularly distinct ganglion mother cells ( GMCs ) that each yield a pair of sibling neurons ( Figure 1B ) ( Bello et al . , 2008; Boone and Doe , 2008; Bowman et al . , 2008; Homem et al . , 2013 ) . Several laboratories have identified candidate temporal transcription factors ( TTFs ) that are expressed in type II neuroblasts , such as the Ecdysone Receptor ( EcR ) ( Figure 1B , horizontal axis; Syed et al . , 2017 ) or in INPs , such as Dichaete and Eyeless ( Figure 1B , vertical axis; Bayraktar and Doe , 2013 ) . Each of these TTFs is required to specify the identity of neurons born during its neuroblast or INP expression window ( Bayraktar and Doe , 2013; Ren et al . , 2017; Syed et al . , 2017 ) . In this study we address how larval brain TTFs contribute to the development and function of the adult insect central complex ( CX ) . The CX is a highly conserved brain region in insects that is thought to play a crucial role in navigation and motor control ( Pfeiffer and Homberg , 2014; Green et al . , 2017; Heinze , 2017; Kim et al . , 2017;Stone et al . , 2017; Franconville et al . , 2018;Giraldo et al . , 2018; Green et al . , 2018 ) . The CX is characterized by four distinct neuropil regions: the Ellipsoid Body ( EB ) , Fan-shaped Body ( FB ) , Protocerebral Bridge ( PB ) , and Noduli ( NO ) ; the CX is also connected to lateral neuropils termed the Gall and the Round body ( ROB ) ( Wolff et al . , 2015 ) . Columnar neurons , which innervate single glomeruli that tile the entire EB and PB neuropil , have been shown to play a key role in navigation ( Pfeiffer and Homberg , 2014; Green et al . , 2017; Heinze , 2017; Kim et al . , 2017; Turner-Evans et al . , 2017; Franconville et al . , 2018; Giraldo et al . , 2018; Green et al . , 2018 ) . There are at least four columnar neuron subtypes ( Figure 1C ) . The E-PG neurons have spiny dendritic arbors in the EB ( hence the E at the front of their name ) and provide outputs to the PB and Gall ( hence the PG at the end of their name ) ; conversely , P-EN neurons have spiny dendritic arbors in the PB and provide outputs to the EB and Noduli . Recently it has been proposed that the E-PG/P-EN neurons form a recurrent circuit that tracks the fly’s orientation in space ( Lin et al . , 2013; Green et al . , 2017; Turner-Evans et al . , 2017; Green et al . , 2018 ) . Two additional columnar neuron classes are PF-R neurons that have dendritic spines in the PB and FB and project axons to the ROB , and the P-FN neurons which have dendritic spines in the PB and project axons to the FB and Noduli ( Figure 1A ) ( Wolff et al . , 2015; Wolff and Rubin , 2018 ) ; both are proposed to have a role in navigation based on anatomical connectivity ( Heinze , 2017; Stone et al . , 2017; Wolff and Rubin , 2018 ) , but their function has not been experimentally determined . Here we map the developmental origin of these four CX neuronal subtypes postulated to have a critical role in navigation . We find that each is derived from a specific temporal window during the INP cell lineage , and that neurons with similar developmental origins have similar axon/dendrite neuropil targets . We confirm that Eyeless , previously shown to be a INP TTF ( Bayraktar and Doe , 2013 ) , is expressed in the latter half of INP lineages; we go on to show that Eyeless is required to promote the identity of the two CX neuron subtypes born late in INP lineages ( E-PG , PF-R ) as well as to repress the identity of the two CX neuron subtypes born during early INP lineages ( P-EN , P-FN ) . In this way , the Eyeless TTF regulates the relative proportion of each neuronal subtype: loss of Eyeless generates fewer E-PG neurons and more P-EN neurons . Importantly , the ectopic P-EN neurons have normal anatomical connectivity . Finally , we show that loss of Eyeless specifically during the larval stages when E-PG neurons are born results in a highly specific defect in adult flight navigation , consistent with the proposed role of E-PGs in maintaining an arbitrary heading to a sun stimulus . Our findings are the first to identify the developmental origin of functionally important adult flight navigation neurons . Moreover , they set the stage for manipulating developmental genetic programs to alter the number and function of each class of adult CX neurons . We used intersectional genetics to map the developmental origin of four CX columnar types ( Figure 1—figure supplement 1 ) . Our strategy was to use the FLP enzyme to permanently open a lexAop-FRT-stop-FRT-GFP reporter in specific populations of INPs and then use adult columnar neuron LexA transgenes to determine the number of each adult columnar neuron type made by each of these INP populations . This approach allowed us to map the developmental origin of neurons labeled by LexA reporters only at pupal or adult stages . We opened the lexAop-FRT-stop-FRT-GFP reporter in all INPs of the type II neuroblast lineages and confirmed that all four types of adult CX columnar neurons are generated by type II neuroblasts ( Figure 1—figure supplement 1A ) . Indeed , we found that type II neuroblasts make all 30 PF-R neurons , all 40 E-PG neurons , all 40 P-EN neurons , and all 50 P-FN neurons across both hemispheres of the adult brain ( Figure 1D ) . We conclude that the four types of CX columnar neurons are all derived from type II neuroblast lineages . The challenge in birth-dating CX neurons from type II neuroblast lineages is that they are generated across two temporal axes , NB and INP . To address this , we systematically dissected one axis at a time . Larval type II neuroblasts produce neurons over five days ( 0–120 hr after larval hatching; ALH ) , with each lineage generating roughly between 40–50 INPs , totaling around 400 neurons and additional glia from each distinct lineage ( Homem et al . , 2013 ) . We used intersectional genetics to determine when each columnar neuron subtype was born during the type II neuroblast lineage . We transiently expressed the FLP recombinase in INPs to permanently open the lexAop reporter at different times during type II neuroblast lineages and assayed for the number of PF-R , E-PG , P-EN , or P-FN adult neurons made at each time-point ( method summarized in Figure 1—figure supplement 1B ) . We found that PF-R neurons were made first in larval type II neuroblast lineages , followed by E-PG neurons , and then by P-EN and P-FN neurons which share overlapping birthdates ( Figure 2A ) . The relatively broad distribution of columnar neuron birthdates is likely due to DM1-DM4 individual lineages generating neuron subtypes asynchronously , but could also represent natural developmental variation or stochasticity in the time of columnar neuron birthdates; it is most consistent with each pool of 30–50 columnar neurons being generated within a 12 hr temporal window in the type II neuroblast lineage ( Figure 2B ) . We next defined columnar neuron birthdates along the INP temporal axis ( see Figure 1—figure supplement 1C ) . Young INPs express Sox family transcription factor Dichaete ( D ) , whereas old INPs express the Pax6 family transcription factor Eyeless ( Bayraktar and Doe , 2013; Eroglu et al . , 2014; Farnsworth et al . , 2015 ) . Here we test whether columnar neuron subtypes arise from a young D+ or old Ey+ temporal window . As expected , all columnar neuron subtypes are labeled when the lexAop reporter is ‘opened’ in all INPs ( Figure 3A–D ) . In contrast , when the lexAop reporter is ‘opened’ only in old INPs , we detect all 40 E-PG and all 30 PF-R adult neurons but no P-EN or P-FN neurons ( Figure 3E–H ) . We conclude that all P-EN and P-FN neurons are born from young INP lineages , whereas all E-PG and PF-R neurons are born from old INP lineages ( summarized in Figure 3I ) . Interestingly , the P-EN and P-FN columnar neurons have a highly similar developmental origin and project to similar CX neuropils ( dendrites to PB , axons to Noduli; Figure 3I ) , whereas E-PG and PF-R columnar neurons have distinct developmental origins and share no similarities in neuropil targets , suggesting that developmental origin may be tightly linked to neuronal morphology and anatomical connectivity ( see Discussion ) . Our birthdating results indicated that INP age might be a major determinant of CX columnar neuron morphology and connectivity . We next tested whether the TTF Eyeless , which is expressed by INPs during the last half their lineage , specifies the identity of PF-R and E-PG neurons , which are born from Ey+ INPs . To knock down Eyeless expression in INPs , we used an eyeless enhancer-Gal4 line ( R16B06-Gal4 ) that is expressed in old INPs ( Farnsworth et al . , 2015 ) to drive a UAS-EyRNAi transgene that we previously showed eliminates all detectable Eyeless protein ( Bayraktar and Doe , 2013 ) . In wild type adults , there are ~40 E-PG neurons and ~30 PF-R neurons ( Figure 4A , B; quantified in G , H ) . In adults where EyRNAi is expressed in old INPs , we found nearly complete loss of PF-R and E-PG neurons ( Figure 4D , E; quantified in G , H ) ; we suggest that these neurons are converted into an early-born INP progeny identity ( for which we have no markers ) , but we can’t rule out that they undergo apoptosis . In addition , we performed an antibody screen for neuronal markers of CX neuronal subtypes , and identified Toy as specifically marking all of the old INP-derived PF-R and E-PG neurons but none of the young INP-derived P-EN and P-FN neurons ( Figure 5—figure supplement 1 ) . Here we show that Toy+ neurons generated by old INPs are also significantly reduced following EyRNAi in old INPs ( Figure 4C , F; quantified in I ) . We conclude that the Eyeless temporal transcription factor is required for the specification of PF-R and E-PG columnar neurons . The P-EN and P-FN columnar neurons derive from early INP progeny , prior to the expression of Eyeless in later-born INPs , raising the question of whether Eyeless expression triggers a switch from early-born P-EN/P-FN production to late-born E-PG/PF-R production . To determine if Eyeless terminates production of early-born P-EN and P-FN columnar neurons , we expressed EyRNAi in old INPs , and assayed for ectopic P-EN or P-FN neurons . In wild type adults , there are ~40 P-EN neurons and ~50 P-FN neurons ( Figure 5A , B; quantified in G , H ) . In adults where EyRNAi was expressed in old INPs , we found an over two-fold increase in the number of P-EN and P-FN neurons ( Figure 5D , E; quantified in G , H ) . In addition , the antibody screen described above identified the transcription factor Runt as specifically marking all early-born P-EN and P-FN neurons but none of the late-born E-PG and PF-R neurons ( Figure 5—figure supplement 1 ) . In wild type , there are ~220 Runt+ adult neurons made by INP progeny , but EyRNAi led to a significant increase to ~580 Runt+ adult neurons ( Figure 5C , F; quantified in I ) , consistent with a role for Eyeless in terminating production of young INP-derived neurons . We conclude that Eyeless maintains equal pools of E-PG and P-EN neurons by triggering a switch from early-born P-EN/P-FN neurons to late-born E-PG/PF-R neurons . Loss of Eyeless extends the production of P-EN neurons into an older stage of INP lineages , creating a mismatch between their molecular temporal identity ( early ) and their time of differentiation ( late ) . We tested whether the ectopic P-EN neurons have a neuronal morphology and anatomical connectivity characteristic of the endogenous early-born neurons , or whether their later birthdate results in different morphology or connectivity . We designed a genetic method for specifically labeling the ectopic late-born P-EN neurons – but not the endogenous early-born P-EN neurons – to trace their morphology and anatomical connectivity ( Figure 1—figure supplement 1D ) . As expected , control RNAi did not result in any ectopic P-EN neurons , although there were a few neurons labeled outside the central brain and a small pattern of fan-shaped body neurons ( Figure 6A–A’’’ ) . In contrast , EyelessRNAi specifically in old INP progeny resulted in the formation of sparse populations of ‘late-born’ ectopic P-EN neurons with projections into the PB , EB , and Noduli ( Figure 6B–B’’’ ) . These are the same neuropils targeted by wild type early-born P-EN neurons . We conclude that ectopic late-born P-EN neurons have morphology indistinguishable from the normal early-born P-EN neurons ( Videos 1–2 ) . To determine if the ectopic P-EN neurons have the same anatomical connectivity as the endogenous P-EN neurons , we expressed the pre-synaptic active zone marker Bruchpilot ( Brp ) specifically in the ectopic P-EN neurons . We found that ectopic P-EN neurons localized Brp to the EB and Noduli , but not to the PB . This is the same as in wild type P-EN neurons ( Figure 6C–C’ , summarized in Figure 6E ) . Furthermore , the ectopic P-EN neurons assembled into proper columns between glomeruli in the PB and tiles in the EB , precisely matching the morphology of endogenous P-EN neurons ( Figure 6D; compare to Figure 8D1 in Wolff et al . , 2015 ) . Thus , ectopic P-EN neurons match the normal early-born P-EN neurons in molecular identity ( R12D09-LexA+ ) , morphology ( PB , EB , Noduli projections ) , and anatomical connectivity ( Brp puncta in EB and Noduli ) . Finally , we assayed the morphology of the ectopic P-FN neurons following EyelessRNAi . We found that the expanded pool of P-FNs innervated the FB and NO , identical to endogenous P-FN neurons , resulting in an enlarged FB and NO ( Figure 6—figure supplement 1 ) . We conclude that reducing expression of the TTF Eyeless leads to a doubling of P-EN and P-FN neurons in the CX , which all have proper neuropil targeting . This shows that neuronal birth-date can be uncoupled from neuronal morphology , because we see P-EN and P-FN neurons born later than normal in the INP lineage , yet they establish morphology that mimics that of the endogenous , early-born P-EN and P-FN neurons ( Figure 6—figure supplement 2 ) . Our finding that the temporal transcription factor Eyeless contributes to the development of CX columnar neurons raises the question of how Eyeless influences CX function . Recent work has shown that silencing adult E-PG neurons impairs flies’ capacity to maintain an arbitrary heading to a bright spot resembling the sun ( Giraldo et al . , 2018; Green et al . , 2018 ) , a finding that we independently confirmed ( Figure 7—figure supplement 1A–D ) . Based on these results , we hypothesized that Eyeless function during development may be required for adult E-PG function in sun navigation . To reduce Eyeless expression , we drove EyelessRNAi in old INPs using R16B06-Gal4 . Temporal control over EyelessRNAi was achieved with the temperature-sensitive Gal4 inhibitor Gal80 . We raised animals at the Gal80 permissive temperature ( 18°C ) to prevent EyelessRNAi expression and shifted to the non-permissive temperature ( 29°C ) for 24 hr at the time E-PG neurons are born and differentiate ( Figure 7A ) . Both control and EyelessRNAi animals exposed to this regime had no major morphological defects in the central complex ( EB shown in Figure 7B , C ) , indicating that E-PG neuron number is likely normal ( see Discussion ) . We then examined how the transient reduction of Eyeless in larval INPs affected the ability of adult flies to maintain an arbitrary flight heading to a fictive sun ( Figure 7D ) . We compared the sun headings of EyelessRNAi flies that received the 29°C heat pulse with two control groups . One control group had an identical genotype but received no heat pulse ( Figure 7E ) . A second control group received the heat pulse but EyelessRNAi was replaced with mCherryRNAi ( Figure 7F ) . In both control groups , we found that flies maintained arbitrary headings , as expected , with a slight bias towards headings where the sun was behind the fly ( Figure 7E , F , I ) . In contrast , flies with transient EyelessRNAi during E-PG development exhibited a marked frontal bias in their heading distribution , which was significantly more frontal than the control distributions ( Figure 7G , I; p<0 . 01 , permutation test ) . The control distributions were not significantly different from each other ( p=0 . 49 ) . Notably , although the heading distributions were distinct , the degree of stimulus stabilization – quantified by calculating the overall vector strength of each flight – was equivalent in the EyelessRNAi genotype and controls ( Figure 7H ) . Moreover , the EyelessRNAi genotype and controls showed equivalent performance orienting to a dark vertical stripe ( Figure 7—figure supplement 1E–H ) , similar to the effect of silencing adult E-PG neurons ( Giraldo et al . , 2018 ) . This suggests that E-PG silencing and EyelessRNAi induce similar , relatively specific navigation deficits rather than a more general deficiency in visual-motor flight control . Taken together , our results indicate that a transient loss of Eyeless specifically in old INPs causes specific deficits in adult flight navigation to that of silencing E-PG neurons . Our findings therefore demonstrate the importance of Eyeless for CX function . The TTF Eyeless is required to specify E-PG neuronal identity , but Eyeless does not persist in adult E-PG neurons , raising the question: What Eyeless target genes regulate E-PG connectivity and function ? We focused on Twin of eyeless ( Toy ) which encodes a transcription factor whose expression is induced by Eyeless in old INPs ( Bayraktar and Doe , 2013 ) and is maintained in their adult post-mitotic neuronal progeny . We used two previously characterized Gal4 drivers ( Kim et al . , 2017; Lovick et al . , 2017 ) to express UAS-toyRNAi specifically in post-mitotic E-PG neurons at different stages in development and confirmed that it removes all detectable Toy protein ( Figure 8 inset A , B ) . We next determined if depleting Toy in post-mitotic larval E-PG neurons using R19G02-Gal4 UAS-toyRNAi altered E-PG survival or morphology . Loss of Toy had no effect on E-PG neuronal number ( n = 5 , p=0 . 92 ) or on connectivity to the EB and PB ( data not shown ) . In contrast , we observed greatly diminished E-PG axonal connectivity to the Gall , where in some cases the E-PG projections appeared nearly absent ( n = 12 , Figure 8A–C ) . We next removed Toy later , beginning ~24 hr after pupal formation using ss00096-Gal4 UAS-toyRNAi , and observed no effect on E-PG neuronal number ( n = 5 , p=0 . 48 ) or projections to the EB , PB , or Gall ( n = 6 , Figure 8D–F ) . Surprisingly , however , loss of Toy produced a significant reduction in the levels of the pre-synaptic active zone marker Bruchpilot ( Brp ) in the Gall ( Figure 8G–I ) . We conclude that Toy is required during larval stages for E-PG connectivity to the Gall , and is required in pupal stages for establishing or maintaining Brp levels at the E-PG axonal terminals in the Gall . To determine how the loss of Toy during pupal stages affects CX function , we tested whether reduction of Toy in the E-PGs affected sun navigation . We observed no significant change in flies’ heading distribution in relation to the sun stimulus , or in the degree to which they stabilized the sun stimulus ( Figure 8—figure supplement 1 ) . Therefore , the loss of Toy in pupal E-PG neurons and the associated reduction of Brp at E-PG axon terminals has no discernible effect on sun navigation . We have shown that distinct classes of CX columnar neurons have unique developmental origins within type II neuroblast lineages . We find that CX columnar neurons map to four bilateral type II neuroblast lineages ( DM1-DM4 ) , confirming previous work ( Wang et al . , 2014 ) . Thus , per brain there are eight parental neuroblasts that generate 30–50 neurons of each subtype , or 4–6 neurons per neuroblast . These 4–6 neurons could arise from 2 to 3 GMCs in a single INP lineage , or as 1 neuron from six different INPs; twinspot MARCM would be needed to determine their precise cell lineage . Our birth-dating results indicate that CX columnar neurons originate from distinct INPs born ~12 hr apart during larval life , except for P-EN and P-FN neurons whose similar birthdates suggest they may arise from the same INPs . Twin-spot MARCM analysis ( Lee and Luo , 1999 ) would be necessary to determine whether P-EN and P-FN neurons arise from the same or different INPs . Interestingly , the two CX columnar neurons born at the same time ( P-EN and P-FN ) have axon projections intrinsic to the CX and target the same neuropils ( PB and Noduli ) . In contrast , the two CX columnar neuron types born at different times ( E-PG and P-FR ) have axon projections extrinsic to the CX and target different neuropils ( Gall and ROB ) . This raises the possibility that neuroblast temporal identity determines whether columnar neuron axon projections are intrinsic or extrinsic to the CX . More generally , the results suggest that neurons with similar temporal identity have matching connectivity . We have mapped the birthdates of only four CX columnar neuron subtypes out of the 60 distinct neuronal subtypes innervating the CX ( Young and Armstrong , 2010 ) . Mapping these other neurons to their type II neuroblast and INP lineages is an important task for the future , which will help identify developmental correlates of neuronal morphology , connectivity , and function . Additionally , significant neuronal diversity may arise from GMCs dividing to make NotchON/NotchOFF sibling neurons , which often have distinct morphology ( Truman et al . , 2010; Lacin et al . , 2014; Wang et al . , 2014; Harris et al . , 2015 ) . The role of Notch signaling in generating hemilineages within type II neuroblast progeny remains unexplored . By mapping the developmental origins of four classes of columnar neurons innervating the central complex , we find that each class derives from a relatively tight window during the neuroblast lineage , and from either young or old INPs ( Figure 3I ) . The fact that all of the four subtypes are restricted to early or late in the INP lineage suggests that the early/late lineage distinction is developmentally important , consistent with our finding that early/late INPs express different TTFs ( Dichaete/Eyeless , respectively ) . Furthermore , mapping the lineage of each neuronal class allowed us to identify a correlation with developmental origin and neuronal morphology ( neurons with similar birth-dates have similar morphology ) . Many other developmental windows have yet to be characterized , for example the neurons derived from young INPs prior to PF-R/E-PG production are unknown , and would be expected to be expanded in the absence of Eyeless; similarly , the neurons derived from the old INPs following production of the P-EN/P-FN neurons are unknown , and would be expected to be missing in the absence of Eyeless . We tested Dichaete and Grainy head for a role in specification of early INP-derived P-EN and P-FN neurons , but observed no phenotype ( data not shown ) ; this is unsurprising for Grainy head , because it is not expressed in the DM1 lineage ( Bayraktar and Doe , 2013 ) which generates P-EN and P-FN neurons . In the future , our intersectional genetic approaches can be used to map the developmental origin of any neuronal subtype for which there exists an adult LexA driver line . For example , we have recently mapped the CX dorsal fan-shaped body ‘sleep neurons’ ( Donlea et al . , 2011; Ueno et al . , 2012; Dubowy and Sehgal , 2017; Donlea et al . , 2018 ) to an old neuroblast developmental window ( M . Syed , LS , and CQD , unpublished ) . We have shown that Eyeless maintains a balance of early-born P-EN/P-FN neurons and late-born E-PG/PF-R neurons by triggering a switch from early-born to late-born neuronal identity . Loss of Eyeless generates fewer E-PG neurons and more P-EN neurons ( Figures 4 and 5 ) . We document the loss of late-born E-PGs here , but many other uncharacterized neurons are also likely to be lost , except during our heat pulse experiments where we tried to specifically target E-PG neurons ( Figure 7 ) . Similarly , we document the production of ectopic P-EN neurons in the absence of Eyeless , but many other early-born neuron populations are likely to be expanded . We considered performing clonal analysis to identify the neurons sharing an INP lineage with our four neural subtypes , but decided against it because INPs make morphologically different neurons at each division ( Wang et al . , 2014 ) ; we would not be able to map these neurons to early or late in the INP lineage , nor would we have molecular or genetic markers for these neurons . Determining the identity and birth-order of neurons within each INP lineage will be a difficult task for the future . Developing markers for the remainder of the 60+ different CX neuronal subtypes will be needed understand the breadth of Eyeless function in generating CX neuronal subtypes . Additional neuronal subtype markers will also be important to test the role of type II neuroblast candidate TTFs ( Ren et al . , 2017; Syed et al . , 2017 ) . We predict that at least some of these candidate TTFs will be required to specify the identity of the four columnar neuron classes described here . We were interested in whether misexpression of Eyeless in young INPs was sufficient to induce ectopic late-born PF-R and E-PG neurons . We could not simply use R9D11-Gal4 to misexpress Eyeless in young INPs , because we previously showed that in this genotype Ey translation is repressed in young INPs ( Farnsworth et al . , 2015 ) . Thus , we permanently expressed Eyeless in INPs and their progeny ( R9D11-FLP , actin-FRT-stop-FRT-Gal4 UAS-eyeless ) but observed loss of all four neuronal subtypes ( data not shown ) . Our interpretation is that permanent high level expression of Eyeless in INPs and their progeny leads to neuronal death , although we cannot rule out that Ey transforms all INP progeny into a late-born cell type that we lack markers to detect . We have shown that the ectopic P-EN neurons formed due to reduced Eyeless levels have morphology and anatomical connectivity that matches the endogenous P-EN neurons ( i . e . Brp+ neurites to the EB and NO , and Brp--neurites to the PB ) ( Figure 6 ) . It is unknown , however , whether these ectopic P-ENs are functionally connected to the normal P-EN circuit partners . This could be resolved through functional imaging experiments testing whether ectopic P-ENs receive the innervation from E-PG or delta7 neurons like endogenous P-ENs ( Franconville et al . , 2018 ) or whether they form functional inputs to known E-PG downstream neurons ( Lin et al . , 2013; Green et al . , 2017; Turner-Evans et al . , 2017 ) . Furthermore , we demonstrate that the Eyeless target gene Toy is required for E-PG axonal connectivity to the Gall . Future work could elucidate the target genes of Toy through RNA-seq that are required for assembling this connectivity , such as downstream cell surface molecules , thus linking INP temporal identity to a direct mechanism for neuronal connectivity in a highly conserved adult brain region . We found that reducing Eyeless expression during early development ( 24–48 hr after larval hatching ) causes a profound shift in how flies orient their flight relative to a fictive sun stimulus . Whereas control populations adopt a broad set of headings , with a slight bias for orientations where the sun is behind ( Figure 7E , F , I ) , EyelessRNAi flies choose flight directions where the sun is in front ( Figure 7G , I ) . A similar shift to a more frontal heading distribution occurs when E-PG neurons are silenced , either following expression of the Kir2 . 1 inward rectifying channel ( Figure 7—figure supplement 1; Giraldo et al . , 2018 ) or with a synaptic transmission blocker in walking flies ( Green et al . , 2018 ) . The consistent shift to a frontal heading after both E-PG silencing and EyelessRNAi suggests that EyelessRNAi affects navigation behavior via perturbation of E-PG neurons , although we cannot rule out an effect on unknown late-born neurons . EyelessRNAi causes no gross deformities in the CX , suggesting E-PGs were not eliminated by EyelessRNAi using this regime , as loss of all E-PG neurons produces severe EB defects ( Xie et al . , 2017 ) . The developmental defects in E-PG neurons could be misexpression of ion channels or other functionally important molecules , rather than apoptosis . In contrast , genetic silencing likely affects all E-PG neurons ( Giraldo et al . , 2018 ) . The fact that similar behavioral effects are induced by our more subtle Eyeless manipulation and E-PG silencing suggest that sun navigation is highly dependent on E-PG neuron activity . One difference between the behavioral effects of EyelessRNAi and E-PG silencing is the degree to which flies stabilize the sun stimulus . Whereas silencing E-PG neurons significantly reduces the overall vector strength , a measure of the heading consistency within a flight ( Figure 7—figure supplement 1; Giraldo et al . , 2018; Green et al . , 2018 ) , there is no such reduction in vector strength in EyelessRNAi flies ( Figure 7H ) . This difference could be due to the more limited scope of the Eyeless manipulation or it could reflect some capacity of the adult CX to compensate for the larval developmental defect . Taken together , our findings demonstrate that a specific navigation behavior – arbitrary orientation to a sun stimulus – depends on the precise expression and function of the Eyeless TTF during larval development . These results raise the question of how other types of navigation depend on the development and function of CX neuronal subtypes . All larvae were grown at 25°C unless noted , and all hours after larval hatching are standardized to grow wild type at 25°C based on published conversions: 18°C is 2 . 25x slower than 25°C , and 29°C is 1 . 03x faster than 25°C ( Powsner , 1935 ) . Primary and secondary antibodies , see Key Resources Table , above . Adult brain dissections were conducted at room temperature with 2–5 day old adult females . Adult brains were dissected in 2% formaldehyde solution in Phosphate-Buffered Saline with . 5% Triton-X ( PBST ) and incubated for 55 min before applying an overnight block solution ( 5% Goat/Donkey serum , Vector Laboratories ) at 4°C . Brains were then washed in PBST for one hour before applying an overnight primary mix at 4°C . Then , brains were washed for one hour at room temperature in PBST , before applying an overnight secondary mix at 4°C . Finally , brains were mounted in 90% glycerol , and imaged immediately . Fluorescent images were acquired on a Zeiss LSM 700 . Adult brain cell counting was performed using the Fiji cell counter plug in , and statistical analysis ( Student’s T test ) was done in Excel . Figures were assembled in Illustrator ( Adobe ) . Relative Brp-density was quantified in Fiji; maximum intensity projections were made , a rectangular ROI selected around the Gall , and a histogram plot of pixel intensity was generated . Background for image was calculated in neighboring ROIs and subtracted from each individual histogram plot-value . Intensity values were then summed together to calculate total intensity , and this was divided by Gall total area , calculated manually in Fiji using polygon selection tool . Qualitative measurements of Gall defects were made by observing whether the total area of the Gall had been reduced , or entirely eliminated , through visual observations in FIJI . We used 3–4 day old females for behavioral experiments . We tethered flies under cold anesthesia , gluing a tungsten wire to the anterior notum with UV-cured glue ( Bondic ) . The head was immobilized relative to the body with a small amount of glue between the head and thorax . Flies recovered for at least 20 min prior to behavioral testing . We coupled the angular velocity of a visual stimulus that was presented via LED panels to the continuously measured difference in wing stroke amplitude . Stroke amplitude was tracked at 60 Hz via Kinefly , a previously described video tracking system ( Suver et al . , 2016 ) . A digital camera equipped with macro lens ( Computar MLM3x-MP ) and IR filter ( Hoya ) captured wing images from a 45° mirror positioned beneath the fly . Backlit illumination of wings was provided by a collimated infrared LED above fly ( Thorlabs #M850L3 ) . We displayed visual stimuli using a circular arena of 2 rows of 12 LED panels ( 24 panels total ) . Each panel had 64 pixels ( Betlux #BL-M12A881PG-11 , λ = 525 nm ) and was controlled using hardware and firmware ( IORodeo . com ) as previously described ( Giraldo et al . , 2018 ) . The gain between stimulus angular velocity and wing stroke amplitude difference was 4 . 75°/s per degree of wing stroke difference . The sun stimulus was a single LED pixel which is ~2 . 4° on fly retina ( Giraldo et al . , 2018 ) , ~30 deg above fly . The stripe was four pixels wide and 16 pixels high ( 15° by 60° ) . Flight experiments were controlled in the ROS environment . Incoming video was collected at 60 Hz and stimulus position data ( i . e . the flight heading ) at 200 Hz . In each experiment , flies navigated in closed loop to the sun stimulus in two distinct 5 min trials , which were separated by a 5 min rest period , during which we gave flies a small piece of paper to manipulate with their legs . Following the second sun flight , flies flew for 5 min in closed loop to the stripe stimulus . We discarded flights in which a fly stopped flying more than once during a sun or stripe presentation; furthermore , we discarded flights from flies that did not complete the two 5 min sun flights . All data analysis was conducted using custom scripts in Python , ( Warren , 2019; archived at https://github . com/elifesciences-publications/elife_2019 ) . The circular mean heading of a flight was computed as the angle of resultant vector obtained via vector summation , treating each angular heading measurement as a unit vector . To determine the vector strength , we normalized the length of the resultant vector by the number of individual headings . Data represent mean ± standard deviation . Two-tailed Student’s t-tests were used to assess statistical significance of anatomical data , with p<0 . 05; p<0 . 01; **p<0 . 001 . To determine the significance of differences in the mean of the vector strength and heading distribution between groups , we used Fisher's exact test with 10 , 000 permutations ( Fisher , 1937 ) . To avoid pseudoreplication , we permuted across flies rather than flights . We computed a 95% confidence interval of the circular mean of each heading distribution by bootstrapping from the observed data . For each experimental condition , we resampled with replacement from the observed flight data ( resampling across flies not flights ) to create 10 , 000 distributions of matched size to the observed data set . Confidence intervals were computed from the circular means of these 10 , 000 distributions . For analysis of the heading distributions and confidence intervals , we considered flights with a vector strength above a minimum threshold of 0 . 2 .	Every task that an animal performs , even a simple one , typically requires numerous signals to pass across complex networks of cells called neurons . These networks develop early in an animal’s life , beginning when progenitor cells called neural stem cells divide over and over to produce new cells . Specific molecular signals then induce these new cells to become different types of neurons . However , in many animals , it is poorly understood what these critical molecular signals are and how they work . Fruit flies , for example , have a network of neurons that control how they navigate when flying . The same type of progenitor cell gives rise to at least four types of neurons in this network; these progenitor cells make an increasing amount of a protein called Eyeless as they age . Sullivan et al . have now specifically disrupted production of the Eyeless protein in the progenitor cells , and found that this altered the relative numbers of navigation neurons . The fruit flies had too many of some types of navigation neurons and too few of others . Fruit flies normally navigate in a variety of directions relative to the sun , which may allow them to disperse and find food . This was not the case in experiments where the production of Eyeless was briefly disrupted when the flies were larvae . In these experiments , the adult flies tended to head towards a bright light ( that represented the sun ) much more often than normal , which would presumably keep them from dispersing effectively . This was true even if the disruption of Eyeless was not long enough to change the numbers of neuron types , showing the protein is important in determining both how these navigation neurons form networks , and whether they are born at all . A better understanding of the complexities of how healthy networks of neurons develop may give scientists more insight into what goes wrong during human developmental disorders that affect the brain . In theory , it may also someday lead to tools that can help to repair the brain if it is damaged .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2019	Temporal identity establishes columnar neuron morphology, connectivity, and function in a Drosophila navigation circuit
Metastasis-initiating cells dynamically adapt to the distinct microenvironments of different organs , but these early adaptations are poorly understood due to the limited sensitivity of in situ transcriptomics . We developed fluorouracil-labeled RNA sequencing ( Flura-seq ) for in situ analysis with high sensitivity . Flura-seq utilizes cytosine deaminase ( CD ) to convert fluorocytosine to fluorouracil , metabolically labeling nascent RNA in rare cell populations in situ for purification and sequencing . Flura-seq revealed hundreds of unique , dynamic organ-specific gene signatures depending on the microenvironment in mouse xenograft breast cancer micrometastases . Specifically , the mitochondrial electron transport Complex I , oxidative stress and counteracting antioxidant programs were induced in pulmonary micrometastases , compared to mammary tumors or brain micrometastases . We confirmed lung metastasis-specific increase in oxidative stress and upregulation of antioxidants in clinical samples , thus validating Flura-seq’s utility in identifying clinically actionable microenvironmental adaptations in early metastasis . The sensitivity , robustness and economy of Flura-seq are broadly applicable beyond cancer research . Metastasis is a multi-step process that begins with migration of cancer cells from the primary tumor into the circulation to reach lymph nodes and the parenchyma of distant organs ( Massagué and Obenauf , 2016; Lambert et al . , 2017 ) . In host organs , disseminated cancer cells interact with a tissue microenvironment that includes organ-specific resident cells , immune cells , perivascular niches , extracellular matrix , cytokines , metabolites , and an oxygen concentration range . This environment eliminates the majority of cancer cells that infiltrate the parenchyma from the circulation , and selects for cells that can adapt , survive as latent entities , and form micrometastases that may eventually grow into clinically manifest metastases . The progression from micro- to macrometastasis is thought to entail a dynamic interaction between disseminated cancer cells and the host microenvironment , which determines an organ-specific pattern of metastatic relapse characteristic of each type of cancer ( Obenauf and Massagué , 2015; Celià-Terrassa and Kang , 2018 ) . Overt metastasis is associated with high morbidity and mortality , and is a major clinical concern . Large metastatic lesions accumulate genetic and epigenetic alterations and stably express specific transcriptional signatures ( Easwaran et al . , 2014; Roe et al . , 2017 ) . In recent years , analysis of these signatures in cells derived from human tumors and xenografts has uncovered numerous factors whose expression mediates organ-specific metastasis in animal models and is associated with organ-specific metastasis in patients ( Ell and Kang , 2013; Kang et al . , 2003; Minn et al . , 2005; Bos et al . , 2009; Boire et al . , 2017; Tavazoie et al . , 2008; Valiente et al . , 2014; Chen et al . , 2016; Shibue et al . , 2012; Bragado et al . , 2013; Gao et al . , 2016 ) . Some of these mediators serve as targets of therapeutic intervention against metastatic cancer ( Celià-Terrassa and Kang , 2018; Sleeman and Steeg , 2010 ) . By contrast , cancer cells in the early stages of metastatic colonization may dynamically alter their gene expression profiles in response to specific stresses experienced in distant organs as they adapt to the host tissue microenvironment and form long-lasting metastatic seeds . These early disseminated cells represent a crucial transition state and may be particularly vulnerable to therapy since they can sometimes be eliminated using adjuvant therapy after surgical resection of primary tumors , unlike established macrometastases . Thus , it is critical to understand the vulnerabilities , dynamic as they may be , of early micrometastases . However , insight into the dynamic early micrometastatic state has been limited by the lack of sensitive techniques for in situ transcriptomic analysis of minute numbers of disseminated cancer cells within large host organs . Current techniques to study cell-type-specific transcriptomes have limitations that preclude their effective application in studying metastasis-initiating cancer cell populations . Single-cell RNA sequencing ( scRNA-seq ) , with or without an intervening fluorescence activated cell sorting ( FACS ) step , allows identification of the transcriptomes of underrepresented cell populations at a single-cell level , but it requires extensive physical and enzymatic processing of the tissue , which disrupts the effects of the host microenvironment while exerting stress on these cells , thus compromising the ability to discern the impact of the host stroma from the transcriptome of the isolated cells . Furthermore , only about 10–20% of the transcripts are captured during the library preparation in scRNA-seq which severely limits the coverage of transcriptome of cells of interest ( Hwang et al . , 2018 ) . In addition , scRNA-seq is challenging to apply in tissues and cell types that are difficult to dissociate into single cells . In situ transcriptomic profiling obviate these problems but lack the necessary sensitivity for disseminated cancer cells that represent less than 1% of the tissue cell population . For example , translating ribosome affinity purification and mRNA sequencing ( TRAP-Seq ) ( Heiman et al . , 2008 ) is not suitable to analyze cells that constitute less than 1% of the total population ( Bertin et al . , 2015; Obenauf et al . , 2015 ) . Direct-enzyme-based metabolic tagging of RNA with thiouracil ( TU ) and ethynyl cytosine ( EC ) in the cells of interest are limited in sensitivity and specificity due to collateral tagging and purification of tagged RNA in cells lacking the enzymes , and requires additional in vitro biotinylation steps ( Cleary et al . , 2005; Gay et al . , 2014; Gay et al . , 2013; Miller et al . , 2009; Hida et al . , 2017 ) . TU tagging has a sensitivity limit of 5% ( Gay et al . , 2013 ) . Thiol ( SH ) -linked alkylation of the metabolic labeling of RNA in tissue ( SLAM-ITseq ) eliminates the noise associated with the purification of RNAs that are not thiol tagged in TU-tagging method ( Matsushima et al . , 2018 ) , but undesired TU tagging through endogenous enzymes in cells lacking UPRT expression remains a limitation . Other methods such as laser capture microdissection/RNA-seq are useful in preserving the spatial information ( Nichterwitz et al . , 2016 ) , however , require sophisticated tools and are challenging to use in rare cell populations that are sparsely distributed in the tissue . Here , we describe the development of a CD-based method for in situ transcriptomic profiling of rare cell populations with high sensitivity ( less than 0 . 01% of an organ ) , and the application of this method to the analysis of organ-specific micrometastatic adaptation . Using this approach , we define microenvironment-dependent transcriptional programs in micrometastatic pulmonary and brain metastases from breast cancer , identify oxidative stress as a lung-specific liability of disseminated cancer cells , and demonstrate that NRF2 activation and upregulation of distinct antioxidant genes are adaptive responses to this stress in lung micrometastases . This oxidative stress and adaptive transcriptional events are reversible upon removal of metastatic cells from the tissue microenvironment , and disappear when metastasis-derived cells are placed in culture . We validate our findings in metastatic tumors from different organ sites from patients with breast cancer . Thus , Flura-seq identifies both a dynamically induced organ-specific stress program activated by metastasis-initiating cancer cells in the pulmonary microenvironment , as well as an adaptive transcriptional program that ensures cancer cell survival , which could be targeted to therapeutic advantage . Cytosine deaminase ( CD ) is a key enzyme of the pyrimidine salvage pathway in fungi and prokaryotes , but is absent in mammalian cells , which instead use cytidine deaminase for the same purpose ( Mullen et al . , 1992 ) . In addition to converting cytosine to uracil , CD can also convert 5-fluorocytosine ( 5-FC ) , a non-natural pyrimidine , to 5-fluorouracil ( 5-FU ) . 5-FU is endogenously converted to fluorouridine triphosphate ( F-UTP ) , which is incorporated into RNA ( Figure 1A , B ) . An antibody-based purification step that specifically captures the 5-FU-tagged RNA would yield a sample suitable for sequencing . Although 5-FU or the combination of CD expression and 5-FC treatment are cytotoxic in some cells ( Austin and Huber , 1993; Kievit et al . , 1999; Longley et al . , 2003 ) , such toxicity requires more than 7 days of treatment ( Hamstra et al . , 2004; Kaliberov et al . , 2006 ) . We hypothesized that short-term 5-FC treatment in CD-expressing cells may avert such toxic effects and minimize transcriptional distortion , thus allowing in situ transcriptomic profiling of rare cell populations . We expressed S . cerevisiae CD in human embryonic kidney 293 T cells ( 293 T-CD cells ) , and treated the cells with 5-FC to yield intracellular 5-FU , which is incorporated into newly synthesized RNA . Antibodies against bromodeoxyuridine ( BrdU ) crossreact with other halogenated uridines incorporated into nucleic acids ( Aten et al . , 1992 ) . Accordingly , untransfected control cells incubated with 5-FU showed positive anti-BrdU immunofluorescence , whereas cells incubated with 5-FC did not ( Figure 1—figure supplement 1A ) . The anti-BrdU antibody also stained 293 T-CD cells when treated with 5-FC , demonstrating that the antibody binds to exogenous or CD-generated 5-FU derivatives but not 5-FC derivatives ( Figure 1—figure supplement 1A ) . To test the specificity and efficiency of RNA isolation , we immunoprecipitated messenger RNA ( mRNA ) from 5-FU-labeled cells with the anti-BrdU antibody and determined the mRNA levels of representative high expression genes ( glyceraldehyde 3-phosphate dehydrogenase , GAPDH; tubulin beta chain , TUBB ) and low expression genes ( chemokine CX3C motif ligand 1 , CXC3L1 ) by reverse transcriptase-polymerase chain reaction ( RT-PCR ) . In 293 T-CD cells , these mRNAs were detectable after 2 hr of treatment with 5-FC , and the levels continued to increase for up to 24 hr ( Figure 1C ) . The relative enrichment of the RNAs was two to three orders of magnitude higher in 293 T-CD cells compared to control 293 T cells ( Figure 1C ) . These results demonstrate that 5-FU tagging allows specific labeling and purification of newly synthesized transcripts . 5-FU can be transported across cell membranes based on its concentration gradient ( Ojugo et al . , 1998; Wohlhueter et al . , 1980 ) . Therefore , we determined whether 5-FU labeling of RNAs using this method would be restricted to CD-expressing cells or collaterally affect neighboring cells . We generated CD-expressing derivatives of MDA-MB-231 ( MDA231 ) cells , a cell line derived from the pleural fluid of a patient with highly metastatic , triple hormone receptor-negative breast cancer ( Cailleau et al . , 1974 ) . The CD-expressing derivative cells , MDA231-CD , were co-cultured with unmodified MDA231 cells ( Figure 1D ) , incubated with 5-FC , and the 5-FU labeling of individual cells was determined based on anti-BrdU immunofluorescence . The co-cultures showed 5-FU-labeling not only in MDA231-CD cells but also in unmodified MDA231 cells ( Figure 1E ) . To limit the diffusion of 5-FU from CD-expressing cells , we implemented a dual strategy . First , we engineered MDA231 cells to co-express CD and uracil phosphoribosyl transferase ( UPRT ) . UPRT directly converts 5-FU to 5-fluorouridine monophosphate ( F-UMP ) , which does not diffuse across cell membranes , bypassing the generation of 5-fluorouridine ( Figure 1—figure supplement 1B ) . We developed a polycistronic vector that allows doxycycline ( Dox ) -inducible co-expression of UPRT , CD and red fluorescence protein ( RFP ) ( Figure 1D ) , and transduced this vector into the cells ( MDA231-CD/UPRT cells ) . Second , since thymine can competitively inhibit cellular uptake of 5-FU ( Yuasa et al . , 1996 ) , we included thymine in the medium as a competitive inhibitor of 5-FU transport . This dual strategy restricted the anti-BrdU immunostaining to cells expressing CD ( Figure 1E ) . Thymine was used in all subsequent in vitro and in vivo experiments . Next , we determined whether this 5-FU-tagging method , ‘Flura-tagging’ , could be used to isolate RNA specifically from cells of interest that were admixed with a large proportion of unlabeled cells . MDA231-CD/UPRT cells were co-cultured with 4T1 mouse breast cancer cells at ratios of 10−3 to 10−4 ( 100 to 1000 MDA231-CD/UPRT cells to 106 4T1 cells ) . After 12 hr of incubation with 5-FC , 5-FU-labeled mRNAs were immunoprecipitated with anti-BrdU antibody , and the proportion of human and mouse mRNA for representative housekeeping genes was determined by qRT-PCR . Notably , human mRNAs were enriched by more than 10-fold relative to mouse mRNAs , despite human cells comprising 0 . 01–0 . 1% of the total cell population ( Figure 1F ) . These results demonstrated the efficacy and specificity of the technique in measuring newly synthesized RNAs from small cell populations of interest in a heterogeneous mixture of cells . To identify potential transcriptional alterations caused by Flura-tagging , we compared the transcriptome of MDA231-CD/UPRT cells treated with two different concentrations of 5-FC ( 50 μM and 250 μM ) , with that of untreated cells that do not express CD/UPRT , using global RNA sequencing analysis ( RNA-seq ) . Over 99% of ~20 , 000 analyzed genes showed statistically similar expression with 50 µM or 250 µM 5-FC ( Supplementary file 1 ) indicating that Flura-tagging introduces minimal alteration in the basal transcriptomes of cells in our experimental conditions . To determine whether Flura-tagging could be used to analyze the transcriptional response to extrinsic regulatory signals , we examined the transcriptional response to TGF-β , a pleiotropic cytokine that regulates the expression of many genes involved in diverse cellular processes ( David and Massagué , 2018 ) . We used the TGF-β response of MDA231 cells ( Padua et al . , 2008 ) as an indicator of the sensitivity and fidelity of our method . MDA231-CD/UPRT cells were treated with 5-FC and either TGF-β or the TGF-β receptor kinase inhibitor SB-505124 ( SB ) . We subjected total RNA from MDA231 cells and immunoprecipitated 5-FU-tagged RNA from MDA231-CD/UPRT cells to RNA-seq analysis . In MDA231 cells , 176 genes showed either an increase or decrease of more than two-fold in transcript levels upon TGF-β treatment ( Supplementary file 2 ) . RNA-Seq analysis of Flura-tagged RNA samples ( ‘Flura-seq’ ) captured the TGF-β transcriptional response of MDA231 cells with high accuracy and fidelity , compared to the RNA-seq control ( Figure 2A , B; Supplementary file 2 ) . It is also noteworthy that Flura-seq showed an enhancement in the fold change of the majority of TGF-β induced genes compared to the control ( Figure 2B–D ) . This is possibly because Flura-seq only detects newly synthesized transcripts , whereas RNA-seq accounts for the total transcripts and thus dilutes the transcriptional response to an acute TGF-β stimulus . On the other hand , Flura-seq identified 575 genes differentially expressed upon TGF-β treatment ( Supplementary file 2 ) . Comparison of the genes uniquely identified by Flura-seq ( 2 . 5 hr post TGF-β treatment ) to the differential gene expression data sets in MDA231 cells 6 hr post TGF-β treatment ( Tufegdzic Vidakovic et al . , 2015 ) showed that 83 of the genes identified only by Flura-seq were induced by TGF-β as detected by RNA-seq at later time points , suggesting that Flura-seq captures early signal-induced gene expression that is missed by RNA-seq due to dilution by the preexisting basal mRNA pool . Collectively , these results show that Flura-seq can accurately capture global changes in gene expression in response to stimuli . Next , we determined whether Flura-seq could be used to characterize transcriptomics in situ from a small number of cancer cells disseminated in an intact organ that would be challenging to achieve using existing technologies . MDA231 cells expressing a GFP-luciferase fusion protein for imaging and bioluminescence analysis and Dox-inducible CD/UPRT for Flura-seq analysis , were inoculated into the tail vein of Foxn1nu immunodeficient mice to allow colonization of the pulmonary parenchyma ( Figure 3A ) . A small proportion of the injected cells survive in the lungs and initiate metastatic outgrowth ( Minn et al . , 2005 ) . At day 31 after inoculation , the cancer cell population was present as micrometastatic colonies throughout the pulmonary parenchyma ( Figure 3—figure supplement 1A , B ) . In tissue sections , the size distribution of these colonies ranged from 112 to 877 cells per cluster , with a mean value of 333 cells ( Figure 3—figure supplement 1C ) . CD/UPRT expression was induced by doxycycline treatment on day 28 , and mice were administered 5-FC ( 250 mg/kg ) and thymine ( 125 mg/kg ) on day 31 for 4 hr to 12 hr before harvesting the lungs for immunoprecipitation of 5-FU-tagged RNAs ( Figure 3A ) . The 5-FC dose was selected based on the non-toxic dose of the structurally related thiouracil in mice ( 250 mg/kg ) that has been used for RNA tagging with thiouracil ( Gay et al . , 2014 ) . We determined the relative fold enrichment of 5-FU tagging in vivo by measuring the relative capture of representative housekeeping human and mouse transcripts . The human mRNAs were enriched more than a 10 , 000-fold compared to the corresponding mouse mRNAs ( Figure 3B ) , indicating that 5-FU tagging occurs primarily in the human cells of interest and that tagged RNAs can be purified efficiently from intact mouse lung tissue . We also compared the relative fold enrichment of 5-FU tagging with TU tagging , an analogous covalent RNA labeling technique ( Gay et al . , 2013; Miller et al . , 2009 ) . To this end , mice harboring lung micrometastases were treated in parallel with TU for 12 hr according to previous studies ( Miller et al . , 2009 ) . Analysis of tested human mRNAs relative to the mouse mRNAs showed approximately 10-fold enrichment with TU tagging compared to over 10 , 000-fold enrichment with 5-FU tagging ( Figure 3B ) . In parallel , we determined the percentage of human cells present in the mouse lungs in these experiments . Approximately 0 . 003% to 0 . 08% of the total cell population comprised of human cells , as determined by RFP expression from the polycistronic UPRT/CD/RFP vector ( Figure 3—figure supplement 1D ) . Since one mouse lung contains approximately 150 million cells ( Perrone et al . , 2010 ) , we estimate that RNA from as few as approximately 5000 human cancer cells per mouse lung could be analyzed by 5-FU tagging ( Figure 3—figure supplement 1E ) . To determine whether 5-FU tagged mRNA from micrometastatic lesions could be used to characterize the in situ transcriptome of cancer cells , mice were treated with 5-FC for 4 hr or 12 hr , and tagged RNAs were immunopurified and sequenced . The sequenced reads were aligned to a hybrid genome containing both human and mouse genomes , so that reads coming from human or mouse cells could be distinctly identified . In mice treated with 5-FC for 4 hr , approximately 53% of the aligned reads were mapped to human genome , whereas 74% of the aligned reads were mapped to human genome when the mice were treated with 5-FC for 12 hr ( Figure 3C ) . Fewer than 1% of the mapped reads in the non-immunopurified input samples were aligned to the human genome while 99% of the reads aligned to the mouse genome ( Figure 3C ) . To further distinguish transcripts derived from the cells of interest ( human cells ) versus other cells ( mouse cells ) , we focused on transcripts that were enriched more than 2-fold relative to input . After applying this enrichment cut-off , the reads were aligned to 7487 human genes and 231 mouse genes ( Figure 3D ) . When the cutoffs were increased to 4 , 8 and 16-fold , the number of human genes identified remained the same , whereas the mouse genes were completely eliminated ( Figure 3D ) . These results demonstrate the sensitivity and specificity of Flura-seq in identifying in situ transcriptomes of cells of interest in vivo ( Figure 3E ) . Next , we applied Flura-seq to define the in situ transcriptomes of breast cancer cells during early stages of metastatic colonization in distinct microenvironments of the brain and lungs . MDA231-CD/UPRT cells were injected intracardially into the arterial circulation of female mice to allow infiltration of multiple organs ( Figure 4A ) . In the lungs and brain , the cells developed micrometastases within 31 days of injection ( Figure 4—figure supplement 1A ) . The cancer cells were also injected into the mammary fat pad ( MFP ) to generate orthotopic mammary tumors ( Figure 4A ) . To identify the genes that are expressed in response to the organ-specific microenvironment , we harvested the brain , lungs , and mammary tumors , and subjected samples to Flura-seq analysis . In parallel , an aliquot of these tissue samples was dissociated into single cells and cultured in selective media to isolate the labeled MDA231 cells as previously described ( Minn et al . , 2005 ) . Following selection and in vitro expansion for 1–2 weeks ( passage 2 ) , these cultures were subjected to RNA-seq analysis ( Figure 4A ) . Principal component analysis ( PCA ) revealed that the in situ transcriptomes of MDA231 cells in different tissues were highly divergent from one other ( Figure 4B ) . In contrast , in vitro culture of the mammary tumor and metastasis-derived cells diminished their transcriptomic differences ( Figure 4B ) . Flura-seq identified several thousand genes that were differentially expressed in different tissues whereas the same cells showed differential expression of only a few hundred genes when cultured in vitro ( Figure 4C , Supplementary file 3 ) . The majority of organ-specific gene expression changes were not preserved when the cells were isolated from the host tissues and expanded in culture . These results suggested that micrometastases have considerable transcriptional plasticity and dynamically regulate gene expression in response to microenvironmental cues . In situ transcriptomic analysis is therefore critical to capture the phenotypic state of micrometastatic cells in the biologically relevant intact tissue context . Analysis of in situ organ-specific transcriptomes unexpectedly revealed that lung micrometastases had the highest content of unique transcriptional activity relative to brain micrometastases and mammary tumors , suggesting that distinct requirements exist for successful metastasis initiation in the lung microenvironment ( Figure 4D , Figure 5—figure supplement 1A ) . Gene Ontology ( GO ) analysis of the differentially expressed cancer cell genes in the different tissues revealed that genes encoding components of the mitochondrial electron transport chain , particularly genes encoding Complex I subunits , were significantly upregulated in lung metastases relative to both brain metastases and orthotopic mammary tumors ( Figure 5A ) . Gene set enrichment analysis ( GSEA ) further confirmed the upregulation of Complex I-encoding genes in lung micrometastases ( Figure 5B ) . The enrichment of these genes was not observed when the cancer cells were isolated from each organ and cultured in vitro under similar conditions ( Figure 5—figure supplement 1B ) , suggesting that the lung microenvironment drives Complex I expression in metastatic cells . In fact , Complex I genes were underexpressed in lung metastasis-derived cells in culture relative to cells derived from brain metastases or mammary tumors , possibly due to re-adaptation of the cells when removed from the lung microenvironment . Complex I activity is a source of reactive oxygen species ( ROS ) ( Balaban et al . , 2005; Murphy , 2009 ) , which at high concentrations cause oxidative stress owing to chemical alteration of proteins and nucleic acids in the cell ( Liou and Storz , 2010; Liou and Storz , 2015 ) . 4-Hydroxynonenal ( 4-HNE ) , a product derived from lipid peroxidation in cells , is a marker of oxidative stress ( Liou and Storz , 2015 ) . A higher level of 4-HNE was present in lung micrometastases compared to the brain micrometastases , as determined by anti-4-HNE immunohistochemistry ( Figure 5C ) , indicating higher oxidative stress in the lung micrometastases . Cells counteract the cytotoxic effect of oxidative stress by upregulating genes that have antioxidant activity ( Espinosa-Diez et al . , 2015 ) . Indeed , analysis of the expression of 63 genes that include all the antioxidant enzymes and the proteins that directly detoxify ROS ( Gelain , 2009 ) revealed that a set of antioxidant genes were specifically upregulated in the lung micrometastases ( Figure 5D ) . To confirm that the transcriptional changes identified reflect changes in protein levels , we performed immunohistochemistry for one of these gene products , glutathione peroxidase 1 ( GPX1 ) , which functions in the detoxification of hydrogen peroxide . Anti-GPX1 immunohistochemistry analysis confirmed high expression GXP1 in lung micrometastases compared to brain micrometastases ( Figure 5E ) . We also tested whether the organ-specific oxidative stress and antioxidant programs are specific to triple negative breast cancer by analyzing lung and brain micrometastases formed by HCC1954 cell line , a HER2+ human breast cancer cell line . The higher oxidative stress and increased expression of antioxidants were also detected in lung micrometastases relative to brain micrometastases in HCC1954 xenograft model ( Figure 5—figure supplement 2A–C ) , indicating that higher oxidative stress and elevated antioxidant program are more general phenomena of early stage lung metastasis in breast cancer . During oxidative stress , the transcription factor nuclear factor erythroid 2-related factor 2 ( NRF2 ) is stabilized , enabling transcription of an antioxidant transcriptional program ( Ma , 2013 ) . Lung micrometastases contained high levels of NRF2 compared to brain micrometastases , based on anti-NRF2 immunohistochemistry ( Figure 5F ) . To determine whether NRF2 transcriptional activity is increased in lung micrometastases , we created a list of 24 NRF2 target genes based on NRF2 chromatin immunoprecipitation-sequencing data curated by Cistrome database ( ENCODE Project Consortium , 2012 ) ( Supplementary File 4 ) , and performed GSEA analysis on our cancer cell transcriptomes . Indeed , the NRF2 signature was enriched in lung micrometastases compared to brain micrometastases and mammary tumors ( Figure 5G ) . Like the Complex I genes , the NRF2 responsive genes were underexpressed in lung metastasis-derived cells placed in culture ( Figure 5—figure supplement 1C ) . Collectively , these results show a specific upregulation of Complex I associated with oxidative stress and a strong NRF2 response in breast cancer cells that survive as lung micrometastases . We investigated Complex I gene expression , and the associated oxidative stress and antioxidant responses in breast cancer patients with metastasis . We analyzed RNA-seq data from breast primary tumors and matched lung metastases from 11 patients ( Siegel et al . , 2018 ) . The lung metastases showed significantly higher expression of Complex I genes compared to mammary tumors ( Figure 6A ) . Matched pair comparison showed that 73% ( 8/11 ) patients had higher expression of Complex I genes in lung metastases than in their matched primary tumors ( Figure 6B ) . 100% ( 8/8 ) of the patients with higher Complex I genes had higher expression of lung-specific antioxidant genes identified by Flura-seq ( Figure 6B ) , and 88% ( 7/8 ) of the patients had higher NRF2 gene signature expression ( Figure 6B ) . A closer examination of differentially expressed genes ( >2 fold ) in lung metastases compared to their corresponding primary tumors revealed that 45% ( 5/11 ) patients overexpressed 26–39 out of 43 nuclear encoded Complex I genes ( Supplementary File 5 ) . We divided these patients into two groups: a high Complex I group of five patients with upregulation of more than 25 Complex I genes , and a low Complex I group of remaining six patients . Complex I high patients were specifically associated with higher expression of lung antioxidant genes and NRF2 signature genes ( Figure 6C , D ) , supporting the conclusion that the high expression of Complex I in lung metastasis is associated with the expression of compensatory antioxidant programs . Moreover , eight antioxidant genes that were upregulated together with Complex I genes in patients’ lung metastases ( Figure 6D ) were also upregulated in Flura-seq transcriptomes from experimental lung micrometastases ( refer to Figure 5D ) . Finally , we sought to determine whether the differences in oxidative stress and antioxidant responses in lung vs . brain metastases were conserved in clinical samples from breast cancer patients . We performed immunohistochemistry for 4-HNE and NRF2 on a tissue microarray ( TMA ) containing lung metastases and brain metastases from more than 40 breast cancer patients . Consistent with the Flura-seq findings , 93% ( 42/45 ) of the lung metastases scored high for 4-HNE immunostaining , whereas only 16% ( 9/55 ) of brain metastases did ( Figure 6E ) . Likewise , 78% ( 32/41 ) of the lung metastases scored high for NRF2 immunostaining versus only 30% ( 14/48 ) in the brain metastases ( Figure 6F ) . There was a strong association between oxidative stress ( 4-HNE ) and NRF2 protein level in majority of the patients ( Figure 6G ) . Collectively , these results demonstrate higher oxidative stress and elevated protective antioxidant program in lung metastases compared to brain metastases in breast cancer patients . To test if the NRF2 signature genes overexpression in breast cancer tumors correlate with organ-specific metastasis prognosis outcomes , we calculated the Hazard ratio for NRF2 signature genes for lung , brain and bone metastasis in breast cancer patients . We found that the Hazard ratio was significantly different for lung metastasis but not for brain and bone metastasis ( Figure 6H ) , indicating that NRF2 overexpression is advantageous for the survival of breast cancer cells in the lungs compared to brain or bone . Previous studies have identified stable , organ-specific transcriptomic programs in cancer cells that were selected on the basis of their ability to form macrometastases and then isolated from these lesions by FACS or in vitro culture prior to transcriptomic analysis ( Roe et al . , 2017; Kang et al . , 2003; Minn et al . , 2005; Bos et al . , 2009; Boire et al . , 2017; Chen and Massagué , 2012; Malladi et al . , 2016; Bruns et al . , 1999; Ikeda et al . , 1990; Ambrogio et al . , 2014 ) . Although these methods successfully identify heritable transcriptional alterations of clinical relevance , these approaches overlook the dynamic transcriptional states that are dependent on tissue-specific microenvironmental cues . Flura-seq now enables the highly sensitive capture of these dynamic transcriptional states , thus shedding light on crucial adaptive processes underway in micrometastases that could not previously be identified . In this study , we applied Flura-seq to identify the in situ transcriptomic programs that are differentially active in cancer cells at early stages of metastatic colonization in the lungs and brain . We identified metabolic gene signatures that were specific to the colonized organ and lost upon removing cancer cells from the tissue microenvironment and placing them in culture . Specifically , we identified mitochondrial Complex I as the top upregulated transcriptional alteration in lung metastases that was dynamic and dependent on an intact tissue microenvironment . Elevated expression of Complex I genes correlated with increased oxidative stress and activation of counteracting antioxidant programs including the upregulation of a distinct set of NRF2-driven antioxidant genes in metastatic cells that seed the lungs . Antioxidant and NRF2 activity were also increased in association with high Complex I expression in lung metastases from breast cancer patients , suggesting a role of these pathways in mitigating the cytotoxic effects of oxidative stress on lung metastatic cells . Lung tissue is exposed to higher concentration of oxygen compared to other organs ( Jagannathan et al . , 2016 ) , and high oxygen concentration can cause oxidative stress ( Halliwell , 2014 ) . It is therefore possible that higher oxygen concentration in the lung micrometastases drives the observed changes . However , we cannot rule out other lung specific microenvironmental cues such as metabolites , cytokines , physical stress , or immune surveillance as sources of the observed changes . These results demonstrate that metastatic tumor cells arising from a single source adopt unique transcriptional profiles depending on their site of colonization . Despite increasing appreciation that metastatic outgrowths frequently exhibit altered metabolic gene expression compared to their primary tumor counterparts ( LeBleu et al . , 2014; Dupuy et al . , 2015; Chen et al . , 2007 ) , whether these metabolic transitions result from the outgrowth of a selected subpopulation predisposed to thrive in a particular location or from the dynamic adaptation of cancer cells to a changing microenvironment remains an open question . Our results support a model wherein tumor cells dynamically adapt to local conditions and suggest that a major determinant of the metabolism of metastatic cells is the site of colonization . These metabolic rearrangements are likely an early event in the establishment of metastatic seeding and may represent a targetable bottleneck against the growth of metastatic lesions . Oxidative stress has been implicated in metastasis , however , the precise role of the stress in metastasis has remained controversial . On one hand , oxidative stress has been observed in cancer cells soon after detachment from epithelia ( Schafer et al . , 2009 ) , and it persists during circulation ( LeBleu et al . , 2014 ) and upon colonization of metastatic sites in model systems ( Piskounova et al . , 2015; Gill et al . , 2016 ) . The lung has been proposed to have pro-oxidant environment due to high oxygen and toxins exposure ( Schild et al . , 2018 ) , and anti-oxidative mediators such as NRF2 ( Wang et al . , 2016; DeNicola et al . , 2015; Menegon et al . , 2016 ) , peroxiredoxin 2 ( Stresing et al . , 2013 ) and thioredoxin-like 2 ( Qu et al . , 2011 ) stimulate the progression of lung cancer and lung metastasis . On the other hand , ROS has also been reported to promote metastasis , and antioxidants have been shown to inhibit metastasis ( Ferraro et al . , 2006; Ishikawa et al . , 2008; Porporato et al . , 2014 ) . The oxidative state and the role of oxidative stress soon after the metastatic cancer cells seed the distant organs before they form macrometastases remain unknown . Our findings demonstrate high oxidative stress in the lung micrometastases of breast cancer , supporting the idea that antioxidant programs promote the progression of lung metastasis and highlighting a critical role for antioxidant mediators in the transition of micrometastases to overt metastases . Surprisingly , however , our data suggest that elevated antioxidant defenses are not a universal hallmark of metastatic lesions . We found that breast cancer brain metastases experience a low level of oxidative stress and antioxidative response . Given that metastatic cells can exhibit reversible metabolic alterations ( Piskounova et al . , 2015 ) , these results raise the possibility that tumor cells undergo multiple metabolic transitions in order to adapt to the changing microenvironments encountered during the metastatic cascade . Indeed , recent evidence suggests that cancer cells from disparate origins may converge to adopt metabolic phenotypes in a given organ ( Schild et al . , 2018; Mashimo et al . , 2014 ) . Techniques such as Flura-seq that enable in situ interrogation of tumor cell phenotypes can reveal to what extent these various metabolic transitions are driven by adaptation to the specific microenvironment versus selection of cancer cells with preexisting traits . Given increasing evidence that cell lineage is a critical determinant of cancer cell metabolism ( Mayers et al . , 2016; Yuneva et al . , 2012 ) it will be interesting for future studies to determine whether lineage-specific metabolic predispositions contribute to the metastatic organ tropisms of different tumor types . More broadly , these studies will help to shed light on the precise factors in the tissue microenvironment that contribute to organ-specific metabolic profiles . Preservation of the intact tissue microenvironment is critical to accurately elucidate the transcriptional state of a cell in vivo . Flura-seq can define in situ transcriptomes from a very rare cell population representing a small fraction ( >0 . 003% ) of an organ . The superior sensitivity of Flura-seq compared to related TU-tagging and EC-tagging may be due to the elimination of a biotinylation step and RNA purification system that distinguishes between cytosine derivatives and uracil derivatives . Flura-seq can be easily applied to any cell type that constitutes a rare subpopulation within the host tissue , such as stem cells and specific subtypes of immune and neuronal cells , in addition to residual cancer cells populations during early stages of metastasis or following the shrinking of a tumor with current therapies . Another feature of Flura-seq is that it only identifies newly synthesized transcripts , which is an advantage in the study of transcriptional responses to cytokines , metabolites , pharmacologic agents , stress signals , and other factors that act by rapidly changing the transcriptomic state of target cells . Further , since Flura-seq involves covalently labeling RNA , it can complement other techniques such as scRNA-seq to combine in situ transcriptomic analysis with profiling of the dissociated cell population with single-cell resolution . Recent advances in scRNA-seq have significantly expanded the application of this technology to the analysis of underrepresented cell types in tissues; however , the method requires extensive physical and enzymatic processing that destroys the tissue microenvironment , and thus microenvironment-dependent gene expression features cannot be accurately captured by scRNA-seq . The higher coverage and applicability of Flura-seq to any tissues and cell types is the principal benefit of Flura-seq over scRNA-seq . Flura-seq involves the expression of exogenous enzymes , CD and UPRT , and treatment of cells or mice with 5-FC . These treatments may alter the levels of certain transcripts , and is therefore important to validate findings made by Flura-seq with alternative methods such as immunostaining , as shown here . This limitation notwithstanding , Flura-seq provides a sensitive , robust and economical alternative to existing in situ transcriptomics techniques . Thus , the power of Flura-seq in studying rare cell populations can be harnessed to address challenging questions of high biological and clinical significance . Human embryonic kidney cells transformed with T-cell antigen ( 293T ) and human breast cancer MDA-MB-231 ( MDA231 ) cells were cultured in DMEM High Glucose medium ( Wheaton ) supplemented with 10% fetal bovine serum and 2 mM L-glutamine . All cell lines have been regularly tested for mycoplasma contamination , and the identity of the cell lines have been authenticated by STR profiling . For the induction of CD or CD/UPRT , cells were treated with 1 μg/ml doxycycline for 24 hr . For 5-FU tagging , cells were treated with 250 μM 5-FC or 5-FU unless indicated . Where indicated , 125 μM thymine was added together with 5-FC . For the induction of TGF-β target genes , cells were treated with 200 pM TGF-β or 2 . 5 μM SB-505124 for 150 min . For 5-FU-tagging during TGF-β treatment , cells were treated with 5-FC and thymine for 30 min before adding TGF-β or SB-505124 . Mouse experiments were performed following the protocols approved by the MSKCC Institutional Animal Care and Use Committee ( IACUC ) . Five- to six-week-old female mice ( Mus musculus ) Hsd:Athymic-Foxn1nu were used in all the experiments . For lung colonization experiments , 50 , 000 MDA231 cells suspended in 100 μl PBS were injected into the tail vein . For organ-specific metastasis experiments , 50 , 000 MDA231 cells or 100 , 000 HCC1954 cells suspended in 100 μl PBS were injected intracardially . For mammary fat pad injection , 50 , 000 MDA231 cells in 50 μl PBS were mixed with 50 μl matrigel and the mixture was injected in the fat pad of mammary gland #4 . Proliferation of injected cancer cells was quantified using bioluminescence imaging following retro-orbital injection of D-luciferin ( Gold Biotechnology ) . CD/UPRT expression was induced by feeding mice doxycycline diet for 2–3 days . For Flura-tagging , mice were injected with 250 mg/kg ( 500 μl ) 5-FC intraperitoneally together with 125 mg/kg ( 500 μl ) thymine subcutaneously . For thiouracil-tagging , mice were injected intraperitoneally with 250 mg/kg ( 500 μl ) of 4-thiouracil . The mice were euthanized 4–12 hr post injection , lungs and brain were harvested and processed for downstream experiments . For RNA analysis , lungs were dissociated using the PRO 200 grinder from PRO Scientific Inc . in RNA extraction lysis buffer . The lung lysates were either used immediately for mRNA extraction or stored at −80°C for later use . For IF , cells were fixed with 4% paraformaldehyde for 10 min , permeabilized with 0 . 2% TritonX-100 for 10 min , blocked with 5% BSA for 1 hr at room temperature , prior to incubation with primary antibodies at 4°C overnight , and secondary antibodies incubated for 1 hr at room temperature . Mouse lung and brain were fixed in 4% paraformaldehyde 24–48 hr at 4°C , embedded in paraffin and sectioned at 5 μm . Paraffin-embedded sections or tissue microarrays were rehydrated using Histo-Clear ( National Diagnostics ) followed by 100-95–70% ethanol and water . Antigen retrieval was performed in a steamer for 30 min in citrate antigen retrieval solution . Tissue sections were blocked with 5% normal goat serum for 1 hr , and incubated with primary antibodies overnight . Secondary antibodies conjugated with fluorophores were used for detection . IHC were performed on BOND RX ( Leica Biosystems ) using standard Epitope Retrieval Solution 2 ( Leica Biosystems ) for 30 min followed by primary antibody incubation for 30 min and BOND polymer refine detection kit-DAB . Automated image analysis was performed using the FIJI software package . Human histopathological sections were obtained under a biospecimen protocol approved by the MSK Institutional Review Board . All human pathology analyses were performed under the supervision of an experienced breast pathologist ( E . B . ) . Cells or tissues were lysed in lysis buffer ( 20 mM Tris-HCl pH 7 . 5 , 500 mM LiCl , 1% LiDS , 1 mM EDTA , 5 mM DTT ) , and mRNAs were extracted using Oligo ( dT ) 25 magnetic beads following the manufacturer’s protocol . The isolated mRNAs were immunoprecipitated using anti-BrdU antibody ( 1–5 μg/sample ) conjugated with Protein G Dynabeads by overnight incubation at 4°C . The mRNAs were incubated with the antibody bead complex in 0 . 8X Binding buffer ( 0 . 5X Sodium Chloride-Sodium Phosphate-EDTA ( SSPE ) with 0 . 025% Tween 20 ) at room temperature for 1–2 hr in a rotator . Subsequently , beads were washed twice with Binding buffer , twice with Wash buffer B ( 1X SSPE with 0 . 05% Tween 20 ) , once with Wash buffer C ( TE with 0 . 05% Tween 20 ) , and once with TE buffer . The bound mRNAs were eluted in 200 μl of 100 μg/mL BrdU for 45 min in a shaker at room temperature . The eluted RNAs were purified using the RNeasy MinElute Cleanup kit following the manufacturer’s protocol . The RNA was eluted in 100 μl RNAase free water . The Flura-tagged RNA elute were re-precipitated as described above , and eluted in 12 . 5 μl final volume . The RNA was either reverse-transcribed using cDNA kit-First Strand Transcriptor following the manufacturer’s protocol , or used for Flura-Seq . TU-tagged mRNAs were purified as described in Miller et al . ( 2009 ) . Brain , lung or mammary tumors were cut into small fragments ( around 1 mm3 ) and transferred to a tissue digestion C-tube . The tumor pieces were incubated with mouse Tumor Dissociation Kit and further dissociated mechanically on a gentleMACS Dissociator as per manufacturer’s protocol . The digestion reaction was stopped with albumin-rich buffer ( RPMI-1640 medium containing 0 . 5% bovine serum albumin ( BSA ) ) . A single-cell suspension was obtained by filtering through a 70 μm cell strainer . The cells were then cultured in DMEM High Glucose media containing 10% FBS , 2 mM L-Glutamine , 200 μg/mL Hygromycin and 8 μg/mL Blasticidin to select MDA231 cells . Harvested lungs were chopped into small pieces ( around 1 mm3 ) , which were then incubated at 37°C in 30 mL digestion buffer ( 5% Fetal Bovine Serum ( FBS ) 1 mM L-glutamine 0 . 35 mg/mL Worthington Type III collagenase , 6 . 25 × 10−3 U/mL dispase , 100 U/mL penicillin , 100 μg/mL streptomycin , 6 . 25 ng/mL amphotericin B ) containing 10 mL trypsin and 30 μl DNAse for 1 hr . The cells were filtered through a 70 μM filter , and were collected by centrifugation . The cell pellets were then resuspended in PBS containing 0 . 1% FBS and 100 μg/ml DAPI , and analyzed using a BD FACS Aria IIU Flow cytometer . CD or CD/UPRT expressing stable cell lines were treated with 1 μg/mL doxycycline for 24 hr , trypsinized , filtered and sorted for RFP positive cells using a BD LSRFortessa Flow cytometer . RNA-seq library preparation . Total RNA was purified using Qiagen RNeasy Mini Kit . Quality and quantity of the RNA were checked using an Agilent BioAnalyzer 2000 . 10 ng of RNA per sample was used for library construction with Sample Prep Kit v2 according to manufacturer’s instructions . Libraries were multiplexed on a Hiseq2500 platform , and more than 25 million raw paired-end reads were generated for each sample . Flura-seq library preparation . RNA was amplified by SMARTer PCR kit with the number of PCR cycles determined empirically based on the amount of purified 5-FU-tagged RNA . The Nextera XT kit was used to prepare sequencing libraries following the manufacturer’s protocol . In our in vivo experiments , 20–24 cycles of PCR were used . In all relevant experiments , mice were randomized prior to different treatments . Comparisons between samples were done in the gene expression analysis , and each group had 2–3 biological replicates that are indicated in the figure legends for each experiment . In the in vitro experiments , biological replicates are samples derived from cells that were plated and processed separately . In the in vivo experiments , the biological replicates represent individual mouse . N described in the Figure legends represents independent biological replicates . The technical replicates are originated from the same sample but divided into different groups . Sample size for each experiment was determined empirically . Reads were quality checked using FastQC v0 . 11 . 5 and mapped to a human ( hg19 ) or hybrid human-mouse ( hg19-mm10 ) genome with STAR2 . 5 . 2b ( Dobin et al . , 2013 ) using standard settings for paired reads . Uniquely mapped reads were assigned to annotated genes with HTSeq v0 . 6 . 1p1 ( Anders and Huber , 2010 ) with default settings . Read counts were normalized by library size , and differential gene expression analysis based on a negative binomial distribution was performed using DESeq2 v3 . 4 ( Love et al . , 2014 ) . In general , thresholds for differential expression were set as follows: adjusted p value<0 . 05 , fold change >2 . 0 or<0 . 5 , and average normalized read count >10 . Genes were considered detectable in the immunoprecipitation samples with a normalized read count >100 . Gene set enrichment analysis was performed using GSVA v3 . 4 ( Hänzelmann et al . , 2013 ) and previously curated gene sets ( Subramanian et al . , 2005 ) . GSEA mountain plots were generated by ‘liger’ R package ( V0 . 1 ) . Primers used for cloning the constructs described in the manuscript are described in Supplementary file 6 . CD ( Addgene 35102 ) , and UPRT ( Addgene 47110 ) were used as template for PCR for subcloning . RFP and IRES were amplified using pTRIPZ ( Dharmacon ) as template . The PCR products were either ligated using DNA Ligase after restriction enzyme digestion and/or by Gibson Assembly .	Cancer cells may not limit themselves to the tissue or organ where they first formed . In some cases , the cells can spread to form tumors in new parts of the body . This process is known as metastasis , and because it is difficult to treat it causes the majority of cancer deaths . To develop new treatments , researchers are trying to learn more about the different steps involved in metastasis . As cancer cells travel through the body they must adapt to the changing environments they encounter , and avoid detection and destruction by the immune system . To do so , they turn different genes on or off . When the cells reach their final destination tissue , they divide to form microscopic clusters , or ‘micrometastases’ , that can grow into new tumors . Micrometastases can sometimes be eliminated by chemotherapy or radiation . Examining which genes are active in the micrometastases may help researchers to find other ways to kill these cancer cells before they can grow into larger tumors that are harder to treat . Basnet et al . have developed a new tool called Flura-seq that documents which genes are active in small clusters of cells in the tissues of living animals . The tool was used to study how breast cancer cells form new tumors in the lungs and brains of mice . The results of the study reveal that lung and brain micrometastases have different patterns of gene activity . In particular , the cancer cells in the lungs turn on antioxidant genes . If they did not , they were killed by a condition known as oxidative stress . This suggests that hindering the activity of the antioxidant genes could help to stop tumors forming in the lungs . Further studies that use the new Flura-seq technique could help researchers to learn more about the early stages of cancer and cancer metastasis . The technique could also be used to study gene activity in other small groups of cells as tissues develop and regenerate .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2019	Flura-seq identifies organ-specific metabolic adaptations during early metastatic colonization
RUNX1 encodes a RUNX family transcription factor ( TF ) and was recently identified as a novel mutated gene in human luminal breast cancers . We found that Runx1 is expressed in all subpopulations of murine mammary epithelial cells ( MECs ) except the secretory alveolar luminal cells . Conditional knockout of Runx1 in MECs by MMTV-Cre led to a decrease in luminal MECs , largely due to a profound reduction in the estrogen receptor ( ER ) -positive mature luminal subpopulation , a phenotype that could be rescued by the loss of either Trp53 or Rb1 . Mechanistically RUNX1 represses Elf5 , a master regulatory TF gene for alveolar cells , and regulates mature luminal TF/co-factor genes ( e . g . , Foxa1 and Cited1 ) involved in the ER program . Collectively , our data identified a key regulator of the ER+ luminal lineage whose disruption may contribute to the development of ER+ luminal breast cancer when under the background of either TP53 or RB1 loss . RUNX1 , RUNX2 , and RUNX3 , and their common non-DNA-binding partner protein CBFβ , form a small family of heterodimeric transcription factors ( TFs ) referred to as Core-Binding Factors ( CBFs ) ( Speck and Gilliland , 2002 ) . They are best known as master regulators of cell fate determination in blood , bone , and neuron , respectively ( Chuang et al . , 2013 ) . RUNX1 is a master regulator of hematopoietic stem cells and multiple mature blood lineages . Translocations and mutations involving both RUNX1 and CBFB are frequently found in human leukemias ( Speck and Gilliland , 2002 ) . Recently , key roles of this family of TFs in epithelial cells and solid tumors also started to emerge ( Taniuchi et al . , 2012; Chuang et al . , 2013; Scheitz and Tumbar , 2013 ) . In particular , in breast cancer , recent whole-genome and whole-exome sequencing studies have consistently identified point mutations and deletions of RUNX1 in human luminal breast cancers ( Banerji et al . , 2012; Cancer Genome Atlas Network , 2012; Ellis et al . , 2012 ) . In addition , mutations in CBFB were also identified in luminal breast cancers from these studies . Its gene product CBFβ is critical for enhancing DNA-binding by RUNX TFs through allosteric regulation ( Bravo et al . , 2001; Tahirov et al . , 2001 ) . Thus , we hypothesized that RUNX1 , together with CBFβ , might play a key role in mammary epithelial cell ( MEC ) lineage determination as a master regulatory TF and that the loss of this normal function might contribute to breast cancer development . There are two major epithelial cell lineages in the mammary gland ( MG ) , luminal lineage ( including ductal and alveolar luminal cells ) , and basal lineage ( the mature cell type in the basal lineage is myoepithelial cell ) ( Figure 1A ) . These two types of MECs are produced by multipotent mammary stem cells ( MaSCs , which are basal cells ) during embryonic development or upon MEC transplantation to cleared mammary fat pads ( Shackleton et al . , 2006; Stingl et al . , 2006; Spike et al . , 2012 ) . In adult MGs , they appear to be maintained by both lineage-specific unipotent stem cells and multipotent basal MaSCs , based on lineage tracing studies ( Van Keymeulen et al . , 2011; van Amerongen et al . , 2012; Rios et al . , 2014; Tao et al . , 2014; Wang et al . , 2014 ) . The gene regulatory network that must be in place to orchestrate lineage specification and differentiation of stem cells into mature MEC types remains largely elusive , although a number of key TFs have been identified in recent years , for example , GATA3 has been shown as a master regulator for both ductal and alveolar luminal cells ( Kouros-Mehr et al . , 2006; Asselin-Labat et al . , 2007 ) ; ELF5 was identified as a master regulator of alveolar cells ( Oakes et al . , 2008; Choi et al . , 2009 ) ; SLUG ( SNAIL2 ) was shown as a master regulator of MaSCs , and it could reprogram differentiated MECs to transplantable MaSCs , together with another TF , SOX9 ( Guo et al . , 2012 ) . In this work , we asked whether RUNX1 is an integral part of this transcription network and how its mutations contribute to breast tumorigenesis . By using genetic , cellular , and molecular approaches , we found that RUNX1 is a key regulator of estrogen receptor ( ER ) -positive mature ductal luminal cells , and that the loss of RUNX1 may contribute to the development of ER+ luminal breast cancer when under the background of either TP53 or RB1 loss . 10 . 7554/eLife . 03881 . 003Figure 1 . Expression pattern of Runx1 in murine MGs . ( A ) Schematic diagram of a simplified version of the MEC hierarchy . MECs can be separated into the luminal and basal lineages . Major MEC subpopulations , their names and name abbreviations , as well as their marker expression patterns are shown . Note: ‘luminal progenitor ( LP ) ’ has been used to refer to progenitor cells for the luminal lineage defined based on either CD61 ( Asselin-Labat et al . , 2007 ) , or CD14 and c-Kit ( Asselin-Labat et al . , 2011 ) , or CD49b ( Li et al . , 2009; Shehata et al . , 2012 ) , and is therefore a mixture of overlapping progenitor cell populations and may include common or separate progenitors for ductal and alveolar luminal cells . ( B ) qRT-PCR analysis of Runx1 , Runx2 , Runx3 , and Cbfb transcripts isolated from luminal and basal cells of adult virgin female mice . ( C–H ) IHC staining for RUNX1 on sections of MGs at different developmental stages: ( C ) adult virgin , ( D–E ) mid-gestation ( the region highlighted in D is shown in E ) , ( F–G ) lactation ( the region highlighted in F is shown in G ) , and ( H ) after involution . Arrows and arrowheads indicate RUNX1-expressing luminal and basal cells , respectively; * indicates lumen . Scale bars = 20 μm . ( I ) Relative expression values of indicated genes determined by microarray analysis of the indicated MEC subpopulations isolated from the MGs of adult virgin female mice . ALs were isolated as YFP+ cells from Wap-Cre;R26Y females ( i . e . , MECs genetically marked by the Wap-Cre transgene ) during mid-gestation . Affymetrix probes used to estimate expression of each indicated gene are 1419555_at , 1422864_at , 1448886_at , 1435663_at , 1449031_at , and 1418496_at for Elf5 , Runx1 , Gata3 , Esr1 , Cited1 , and Foxa1 , respectively . ( J ) Runx1 expression levels were confirmed in sorted LPs , MLs , and ALs ( as in I ) by qRT-PCR . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 00310 . 7554/eLife . 03881 . 004Figure 1—figure supplement 1 . Expression analysis of Runx1 and other select luminal transcription factor ( TF ) genes based on microarray . Relative expression values of indicated genes determined by microarray analysis of the indicated mammary epithelial cell ( MEC ) subpopulations isolated from the mammary glands ( MGs ) of adult female mice from multiple published datasets . ( A ) Based on GEO database accession # GSE40875 ( only showing the nulliparous subsets ) . Basal_CD49hi and Basal_CD49flo are two subpopulations in the lin−CD24loSca1− basal lineage based on higher or lower CD49f expression , respectively . ( B ) Based on GEO database accession # GSE19446 . ( C ) Based on GEO database accession # GSE20402 . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 004 We first measured expression levels of all three Runx genes and their common co-factor gene Cbfb in freshly sorted basal epithelial cells ( Lin−CD24+CD29hi ) and luminal epithelial cells ( Lin−CD24+CD29lo ) ( Figure 1A ) from adult virgin female mice by quantitative RT-PCR ( qRT-PCR ) . Results showed that Runx1 is the predominantly expressed Runx gene in both luminal and basal cells ( Figure 1B ) . Immunohistochemical ( IHC ) staining further confirmed the expression of RUNX1 protein in these two major MEC types in adult virgin MGs ( Figure 1C ) . However , RUNX1 expression is largely absent in alveolar luminal cells ( ALs ) that start to emerge during pregnancy ( Figure 1D–E ) . In the lactating gland , the only MEC type that still expresses RUNX1 is the myoepithelial cell ( Figure 1F–G ) . Upon involution , RUNX1 expression is restored to a pattern resembling that of the virgin gland ( Figure 1H ) . Additionally , we performed microarray expression profiling of sorted subsets of MECs , including basal cells ( Lin−CD24+CD29hi ) , luminal progenitors ( LPs , Lin−CD24+CD29loCD61+ ) , mature luminal cells ( MLs , Lin−CD24+CD29loCD61− , mainly represent ductal luminal cells in virgin MGs ) , and alveolar luminal cells ( ALs , i . e . , MECs genetically marked by Wap-Cre at mid-gestation; Wap-Cre is a transgenic mouse line with Cre expression under the control of the Whey acidic protein [Wap] promoter , a milk protein promoter [Wagner et al . , 1997] ) . Estimation of Runx1 levels based on this microarray dataset confirmed its expression in all MEC subsets except in ALs ( Figure 1I ) . We examined Runx1 expression levels in different subsets of MECs in several additional published microarray datasets ( Asselin-Labat et al . , 2010; Lim et al . , 2010; Meier-Abt et al . , 2013 ) and further confirmed this expression pattern ( Figure 1—figure supplement 1A–C ) ; in particular , in the pregnant MGs , Runx1 was also found expressed in basal MECs but not in luminal MECs ( mainly ALs ) ( Figure 1—figure supplement 1C ) . Lastly , by qRT-PCR , we verified that Runx1 was indeed expressed in sorted LPs and MLs but not in Wap-Cre-marked ALs ( Figure 1J ) . The RUNX1 mutations identified from the recent sequencing studies of human breast cancers include point mutations , frame-shift mutations , and deletions ( Banerji et al . , 2012; Cancer Genome Atlas Network , 2012; Ellis et al . , 2012 ) . We analyzed the breast cancer-associated missense mutations of RUNX1 to determine whether they lead to loss-of-function of RUNX1 ( Figure 2A ) . Based on a previous alanine-scanning site-directed mutagenesis study ( Li et al . , 2003 ) , we found that these missense mutations either affect amino acid residues in RUNX1 that directly contact DNA , or disrupt the overall fold of its DNA-binding RUNT domain or abolish its binding to CBFβ , both of which would also perturb its DNA-binding ( Figure 2B ) . Thus , similar to RUNX1 deletions , the point mutations also lead to loss-of-function of RUNX1 , due to disrupted DNA-binding ability . Therefore , we asked whether and how the loss of Runx1 could affect the development of normal MECs . 10 . 7554/eLife . 03881 . 005Figure 2 . Analysis of RUNX1 mutations . RUNX1 somatic missense mutations identified in human breast cancers disrupt its DNA-binding either directly ( disrupting direct DNA contact ) or indirectly ( disrupting the overall protein fold of its DNA-binding RUNT domain or disrupting CBFβ binding ) . ( A ) RUNX1 full-length protein sequence; RUNT domain is highlighted in blue . The three amino acid residues affected by point mutations in luminal breast cancers ( based on Ellis et al . ( 2012 ) ) are shown in red . Several additional missense mutations ( based on Cancer Genome Atlas Network ( 2012 ) ; Taniuchi et al . ( 2012 ) ) are also highlighted with red font . ( B ) How these missense mutations affect RUNX1 DNA-binding is predicted based on a previous structural and biochemical analysis of the RUNT domain ( Li et al . , 2003 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 005 Runx1−/− mice died during mid-gestation mainly due to hemorrhages in the central nervous system and are thus not suitable to determine the effect of Runx1-loss on MG development ( Okuda et al . , 1996; Wang et al . , 1996 ) . We therefore used a conditional knockout allele of Runx1 ( Runx1L/L ) ( Li et al . , 2006b ) . To facilitate characterization of Runx1-null MECs , we bred in a conditional Cre-reporter , Rosa26-Stop-YFP ( R26Y ) . Cross of the floxed Runx1 mice with the R26Y reporter mice and MMTV-Cre transgenic mice allowed us to simultaneously disrupt Runx1 in MECs and mark the targeted cells by Yellow Fluorescent Protein ( YFP ) ( Figure 3A ) . Lineage analysis revealed that in virgin MGs , MMTV-Cre mainly targeted MECs in the luminal lineage , but it could also lead to Cre-mediated recombination in a portion of basal MECs ( Figure 3B ) . By fluorescence-activated cell sorting ( FACS ) , we isolated YFP+ MECs from MMTV-Cre;Runx1L/L;R26Y females and MMTV-Cre;Runx1L/+ and +/+;R26Y control females and by qRT-PCR , we confirmed the loss of Runx1 expression in YFP+ MECs from MMTV-Cre;Runx1L/L;R26Y females ( Figure 3C ) . Whole-mount analysis of MGs from MMTV-Cre;Runx1L/L;R26Y virgin females or dams on lactation day-0 did not reveal any obvious gross morphological abnormalities , although a portion ( 3 out of 7 ) of MMTV-Cre;Runx1L/L;R26Y females exhibited a slight delay in expansion of their ductal trees during pubertal growth ( Figure 3—figure supplement 1A ) . Surprisingly , however , none of the MMTV-Cre;Runx1L/L;R26Y dams were able to successfully nurse their pups ( Figure 3—figure supplement 1B ) . Most of their pups died within 24 hr postpartum and no milk spots were observed in them compared to pups from MMTV-Cre;Runx1+/+ ( or L/+ ) ;R26Y dams . A closer examination of MGs of lactating MMTV-Cre;Runx1L/L;R26Y females revealed milk stasis and an increasing number of cytoplasmic lipid droplets ( Figure 3—figure supplement 1C ) . Similar phenotypes have also been observed in Runx1 conditional knockout mice with Krt14-Cre ( i . e . , Cre-expressing transgenic mice under the control of the Keratin 14 promoter ) ( Krt14-Cre;Runx1L/L , data not shown ) and in a number of genetically engineered mice with defects in myoepithelial cell contraction and milk ejection ( Li et al . , 2006a; Plante et al . , 2010; Haaksma et al . , 2011; Weymouth et al . , 2012 ) . Since Runx1 is only expressed in myoepithelial cells at this stage ( Figure 1D–G , I , and Figure 1—figure supplement 1C ) , we reasoned that the nursing defects observed are most likely due to a disrupted function of RUNX1 in myoepithelial cells . Additional studies are required to determine this . 10 . 7554/eLife . 03881 . 006Figure 3 . Runx1-loss leads to a reduction in the luminal MEC population . ( A ) Schematic representation of the Runx1 conditional knockout allele in which its exon 4 is flanked by loxP sites , as well as the R26Y conditional Cre-reporter . STOP: transcriptional stopper cassette . Subsequent breeding with MMTV-Cre resulted in mice in which selected subsets of MECs express YFP and lack expression of functional RUNX1 . ( B ) FACS gating strategy for detecting lin−YFP+ ( lin: lineage markers ) MECs , as well as YFP+ lin−CD24+CD29lo luminal ( Lu ) , and lin−CD24+CD29hi basal ( Ba ) MECs in MMTV-Cre;R26Y females . Str: stromal cells . ( C ) qRT-PCR analysis confirming the loss of Runx1 expression in YFP+ MECs sorted from MMTV-Cre;Runx1L/L;R26Y females ( L/L ) . ( D ) FACS analysis showing the reduced lin−YFP+ MEC population ( left plots ) , as well as the reduced lin−YFP+ luminal population ( right plots ) , in MMTV-Cre;Runx1L/L;R26Y female compared to those in MMTV-Cre;Runx1+/+;R26Y control female . ( E–F ) The percentages of lin−YFP+ MEC population ( E ) , as well as the ratios of luminal/basal subpopulations among the lin−YFP+ gate ( F ) , are significantly reduced in MMTV-Cre;Runx1L/L;R26Y females ( n = 9 ) ( L/L ) compared to those in MMTV-Cre;Runx1+/+;R26Y control females ( n = 10 ) ( +/+ ) . p values: : p ≤ 0 . 05; error bars represent mean ± S . E . M . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 00610 . 7554/eLife . 03881 . 007Figure 3—figure supplement 1 . Conditional knockout study of Runx1 in murine MGs . ( A ) Whole-mount carmine staining of inguinal MGs of pubertal ( top ) , adult virgin mice ( middle ) , and lactation day-0 dams ( bottom ) . Scale bars indicate 3 mm . Black arrow indicates the main duct coming from the nipple; indicates the lymph node . Blue arrow indicates the distance between the front of the expanding ductal tree and the lymph node ( used as a reference point ) . ( B ) Weaning record showing MMTV-Cre;Runx1L/L females failed to nurse their pups . ( C ) Hematoxylin and eosin ( H&E ) staining of sections of MGs from MMTV-Cre;Runx1+/+ ( left ) and MMTV-Cre;Runx1L/L ( right ) dams on lactation day-0 . Note the increasing number of cytoplasmic lipid droplets and milk in the lumen of the MMTV-Cre;Runx1L/L dam . Arrowheads point to lipid droplets . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 007 In MMTV-Cre;Runx1L/L;R26Y virgin females , we found that the percentages of the YFP-marked MEC population ( representing Runx1-null MECs ) were significantly reduced when compared to those of the MMTV-Cre;Runx1+/+;R26Y control females ( Figure 3D–E ) . Furthermore , the ratios of the YFP-marked luminal to basal subsets were also significantly reduced in MMTV-Cre;Runx1L/L;R26Y females ( Figure 3D , F ) ; this could be due to an expansion of the YFP-marked basal population or a reduction in the YFP-marked luminal population , or both . However , since the overall population of YFP+ MECs in MMTV-Cre;Runx1L/L;R26Y females was reduced ( Figure 3D–E ) , the reduction in the YFP+ luminal/basal ratio was most likely due to a reduction in the YFP-marked Runx1-null luminal population . Recent studies suggest that most breast cancers , including both basal-like and luminal subtypes , may originate from luminal cells , rather than from basal MaSCs ( Lim et al . , 2009; Molyneux et al . , 2010; Proia et al . , 2011; Keller et al . , 2012 ) . Furthermore , RUNX1 and CBFB mutations have only been found in the luminal subtype of human breast cancers ( Banerji et al . , 2012; Cancer Genome Atlas Network , 2012; Ellis et al . , 2012 ) and our data so far showed that the loss of Runx1 appeared to lead to a reduction in the luminal population ( Figure 3D–F ) , we therefore examined the role of RUNX1 in luminal MECs ( from which luminal breast cancers may originate ) . To determine the overall defects of Runx1-null luminal MECs , we first profiled the transcriptomes of YFP+ Runx1-null luminal cells ( sorted from MMTV-Cre;Runx1L/L;R26Y females ) and control YFP+ Runx1-wild-type ( WT ) luminal cells ( sorted from MMTV-Cre;Runx1+/+;R26Y ) by microarray . By gene set enrichment analysis ( GSEA [Subramanian et al . , 2005] ) , we observed significant enrichment of a previously generated LP signature and downregulation of a ML signature in Runx1-null luminal cells ( Figure 4A ) . These LP and ML signatures were generated previously based on subset-specific genes conserved in the corresponding human and mouse MEC subpopulations ( Lim et al . , 2010 ) . Furthermore , we also observed significant enrichment of multiple gene sets related to p53 signaling in Runx1-null luminal cells in relation to Runx1-WT luminal cells ( Figure 4—figure supplement 1A ) , possibly indicating a stress response in these mutant luminal cells in vivo . Lastly , we examined the expression levels of a number of TF/co-factor genes known to be part of the transcription network that regulates specification and maintenance of luminal MECs . In our microarray data , we found that Elf5 , a TF gene critically required in the alveolar cell lineage and a LP marker ( Oakes et al . , 2008; Choi et al . , 2009; Lim et al . , 2010 ) , was upregulated in Runx1-null luminal cells , whereas several ductal luminal TF/co-factor genes ( e . g . , Gata3 , Foxa1 , Esr1 , Cited1 ) were downregulated ( Figure 4—figure supplement 1B ) . Among these luminal TF/co-factor genes , Foxa1 encodes a pioneer factor that is a key determinant of ERα ( encoded by Esr1 ) function ( Bernardo et al . , 2010; Hurtado et al . , 2011 ) ; Cited1 encodes a selective co-activator for estrogen-dependent transcription , which potentially regulates the sensitivity of luminal cells to estrogen ( Yahata et al . , 2001; Howlin et al . , 2006 ) . When validated by qRT-PCR , we found that although the expression of Gata3 did not seem to be significantly affected in luminal cells upon Runx1-loss , expression levels of Foxa1 , Esr1 , and Cited1 were downregulated in Runx1-null luminal cells , whereas Elf5 expression was upregulated ( Figure 4B ) . 10 . 7554/eLife . 03881 . 008Figure 4 . Runx1 disruption leads to a profound reduction in ER+ MLs . ( A ) GSEA enrichment plots showing correlation of the expression profiles of Runx1-null or WT luminal MECs with previously published conserved human and mouse signatures of LPs ( left ) or MLs ( right ) ( Lim et al . , 2010 ) . ( B ) qRT-PCR validation of TF/co-factor genes known to play roles in luminal lineage specification and maintenance . RNA was isolated from YFP+ Runx1-null and WT primary luminal MECs . ( C ) FACS plots of expression of CD14 and c-Kit , two LP markers ( Asselin-Labat et al . , 2011 ) , in the gated YFP+ luminal MECs ( Lin−CD24+CD29lo ) of adult MMTV-Cre;Runx1L/L;R26Y virgin females and MMTV-Cre;Runx1+/+;R26Y control females . Note the CD14−c-Kit− mature luminal ( ML ) subpopulation was largely lacking in the lower right plot . ( D ) Quantification of the percentages of the ML and LP subpopulations as indicated in C , showing significant reduction in the ML subpopulation in MMTV-Cre;Runx1L/L;R26Y females ( n = 13 ) compared to that in MMTV-Cre;Runx1+/+;R26Y control females ( n = 10 ) . ( E ) qRT-PCR analysis showing significantly reduced Runx1 expression in the LP subpopulation but not in the ML subpopulation in MMTV-Cre;Runx1L/L;R26Y females . ( F ) FACS plots of expression of CD49b , a LP marker , and Sca1 , an ER+ ML marker ( Shehata et al . , 2012 ) in the gated YFP+ luminal MEC population . Note the CD49b−Sca1+ ER+ ML subpopulation was dramatically reduced , whereas the CD49b+Sca1− ER− LP subpopulation was increased in MMTV-Cre;Runx1L/L;R26Y females . ( G ) Quantification of the percentages of the ER+ ML , ER+ LP , and ER− LP subpopulations as indicated in F , showing significant reduction in the ER+ ML subpopulation in MMTV-Cre;Runx1L/L;R26Y females ( n = 4 ) compared to those in MMTV-Cre;Runx1+/+;R26Y control females ( n = 4 ) . p values: : p ≤ 0 . 05; #: p ≤ 0 . 005; ^: p ≤ 0 . 0005; NS = not significant; error bars represent mean ± S . E . M . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 00810 . 7554/eLife . 03881 . 009Figure 4—source data 1 . ( A ) Gene sets from the MSigDB database C2-CGP ( chemical and genetic perturbations , v3 . 1 ) enriched in Runx1-null luminal cells . ( B ) Gene sets from the MSigDB database C2-CGP ( chemical and genetic perturbations , v3 . 1 ) enriched in Runx1-WT luminal cells . ( C ) Gene sets from the MSigDB database C2-CP:KEGG ( KEGG gene sets , v3 . 1 ) enriched in Runx1-null luminal cells . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 00910 . 7554/eLife . 03881 . 010Figure 4—figure supplement 1 . Analysis of the luminal phenotype in MMTV-Cre;Runx1L/L;R26Y females . ( A ) GSEA analysis of Runx1-null and WT luminal MECs showing enrichment of several gene sets related to the p53 signaling pathway in Runx1-null luminal MECs compared to WT luminal MECs . ( B ) Relative expression values of the indicated genes determined by microarray analysis of Runx1-null ( Runx1 L/L ) and WT ( Runx1 +/+ ) luminal MECs . Affymetrix probes used to estimate expression of each indicated gene are 1448886_at , 1419555_at , 1418496_at , 1449031_at , and 1460591_at for Gata3 , Elf5 , Foxa1 , Cited1 , and Esr1 , respectively . ( C ) FACS plots of CD14 and c-Kit expression in the lin− luminal ( upper plots ) and lin−YFP+ luminal ( bottom plots ) MECs of 5-week old MMTV-Cre;Runx1+/+;R26Y and MMTV-Cre;Runx1L/L;R26Y females showing reduced CD14−c-Kit− ML subpopulation within the lin−YFP+ luminal population in MMTV-Cre;Runx1L/L;R26Y female ( bottom right plot ) . ( D ) Quantifications ( for C ) of the percentages of ML or LP subpopulations within the indicated gates showing significantly reduced ML and increased LP subpopulations within the lin−YFP+ luminal gate in 5-week old MMTV-Cre;Runx1L/L;R26Y females ( n = 4 ) ( L/L ) compared to those in 5-week old MMTV-Cre;Runx1+/+;R26Y control females ( n = 4 ) ( +/+ ) . ( E ) FACS plots showing the reduced YFP-marked lin−CD29loCD61− ML subpopulation in adult MMTV-Cre;Runx1L/L;R26Y virgin females ( n = 2 ) compared to MMTV-Cre;Runx1+/+;R26Y control females ( n = 3 ) . ( F ) Quantifications for the basal , LP and ML subpopulations based on CD61 and CD29 staining in E . ( G ) qRT-PCR analysis showing significantly reduced Runx1 expression in the ER− LP subpopulation , partial Runx1 reduction in the ER+ LP subpopulation , and no reduction in the ER+ ML subpopulation from MMTV-Cre;Runx1L/L;R26Y ( Runx1 L/L ) females , based on CD49b and Sca1 staining , compared to those from MMTV-Cre;Runx1+/+;R26Y ( Runx1 +/+ ) control females . p values: ^: p ≤ 0 . 0005; NS = not significant; error bars represent mean ± S . E . M . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 010 Our microarray data for the entire luminal population suggested that Runx1-loss in luminal MECs might lead to either a global block in luminal differentiation or loss of a mature luminal MEC subpopulation . To determine this , we performed FACS analysis of the Lin−CD24+CD29lo luminal lineage . Intriguingly , we found that in both pubertal and adult virgin MMTV-Cre;Runx1L/L;R26Y females , the YFP+ Runx1-null ML subpopulation defined based on CD14 and c-Kit staining ( Lin−CD24+CD29loCD14−c-Kit− , referred to as CD14−c-Kit− MLs hereafter ) ( Asselin-Labat et al . , 2011 ) was significantly reduced , whereas the YFP+ Runx1-null LP subpopulation ( Lin−CD24+CD29loCD14+c-Kit+ , referred to as CD14+c-Kit+ LPs hereafter ) was increased ( Figure 4C–D , Figure 4—figure supplement 1C–D ) . The reduction in the ML subpopulation was further confirmed in adult virgin MMTV-Cre;Runx1L/L;R26Y females based on CD61 staining ( MLs: Lin−CD29loCD61− ) ( Asselin-Labat et al . , 2007 ) ( Figure 4—figure supplement 1E–F ) . The residual YFP+ MECs in the ML gate in MMTV-Cre;Runx1L/L;R26Y virgin females could either represent a truly Runx1-null ML subpopulation ( but reduced in percentage ) or represent YFP-marked MLs that have escaped Cre-mediated disruption of the Runx1L allele ( thus not truly Runx1-null ) . To determine this , we sorted YFP+ CD14−c-Kit− MLs from MMTV-Cre;Runx1L/L;R26Y virgin females and MMTV-Cre;Runx1+/+;R26Y control females; as an internal control , we also sorted YFP+ CD14+c-Kit+ LPs from the same animals . By qRT-PCR , we found that while Runx1 expression in YFP+ LPs from MMTV-Cre;Runx1L/L;R26Y females was significantly reduced , its expression in YFP-marked MLs was only slightly reduced ( Figure 4E ) . This data suggested that many YFP+ MLs in MMTV-Cre;Runx1L/L;R26Y females might have escaped Cre-mediated excision in at least one copy of their Runx1L alleles ( thus they were either Runx1+/+ or Runx1+/− ) . The data thus also suggests that RUNX1 is essential for the emergence or maintenance of the ML lineage . A recent study demonstrated that the luminal cell compartment of the mouse MG could be further resolved into non-clonogenic ER+ MLs , as well as clonogenic ER+ LPs and ER− LPs based on CD49b and Sca1 staining; the ER+ LPs may represent progenitors for ER+ MLs whereas the ER− LPs probably represent alveolar progenitors ( Shehata et al . , 2012 ) . We examined these luminal subpopulations in the YFP-gated luminal population in MMTV-Cre;Runx1L/L;R26Y virgin females . We found that compared to their corresponding subpopulations in MMTV-Cre;Runx1+/+;R26Y control females , the CD49b−Sca1+ ER+ ML subpopulation was significantly reduced in MMTV-Cre;Runx1L/L;R26Y females , whereas the CD49b+Sca1− ER− LP subpopulation was significantly increased; the CD49b+Sca1+ ER+ LP subpopulation was not significantly altered ( Figure 4F–G ) . Of note , since the overall YFP+ luminal population was significantly reduced in MMTV-Cre;Runx1L/L;R26Y females ( Figure 3D–F ) , the increase in the ER− LP subpopulation might be mainly due to a reduction in the ER+ ML subpopulation ( thus proportionally increased the percentage of the ER− LP subset ) , rather than a significant expansion of ER− LPs per se; similarly , although the percentage of the ER+ LP subpopulation was not significantly changed , the absolute number of YFP+ ER+ LPs could still be reduced ( due to an overall reduction in YFP+ luminal MECs ) . In support of this , we measured Runx1 expression in these three luminal MEC subpopulations . We found that whereas the YFP-marked ER− LP subset had a profound reduction in Runx1 expression , the YFP+ ER+ LP subset exhibited a partial reduction in Runx1 transcripts , and the YFP+ ER+ ML subpopulation had almost no loss of Runx1 expression ( Figure 4—figure supplement 1G ) , suggesting RUNX1 is required for both ER+ LPs and ER+ MLs . Collectively , our data suggest that RUNX1 is required for the development or maintenance of the ER+ luminal lineage , and it is particularly essential for the ER+ MLs . From recent whole-genome/exome sequencing studies , RUNX1 and CBFB mutations were only identified in the luminal subtype of human breast cancers ( Banerji et al . , 2012; Cancer Genome Atlas Network , 2012; Ellis et al . , 2012 ) , which are typically ER+ . Paradoxically our data so far in the murine model suggest that loss-of-function of Runx1 leads to a reduction in ER+ luminal MECs in vivo . Furthermore , we have followed MMTV-Cre;Runx1L/L;R26Y females for at least 18 months and have not observed any mammary tumor development in them . This can be explained by a possibility in which RUNX1-mutant breast cancer originates from ER+ luminal MECs and Runx1 disruption alone actually leads to the loss of the cell-of-origin of such cancer . We hypothesized that additional genetic events might be needed to cooperate with RUNX1-loss to promote the development of luminal breast cancer from ER+ luminal MECs . Interestingly , one recent sequencing study unveiled that pathway signatures of RB1 mutation , TP53 mutation , and RUNX1 mutation are co-associated with human luminal B breast tumors ( Ellis et al . , 2012 ) . Furthermore , by carefully examining luminal breast cancer cases with RUNX1 mutations , we noticed that >50% of them are accompanied by mutations or deletions in either TP53 or RB1 genes ( Cancer Genome Atlas Network , 2012 ) . Lastly , our microarray data for luminal MECs suggested that loss of Runx1 might lead to activation of the p53 pathway in luminal cells in general ( Figure 4—figure supplement 1A ) . Based on these observations , we hypothesized that loss of Runx1 in luminal MECs perturbs the fate of ER+ MLs , possibly leading to a stress response and subsequently upregulation of the p53 pathway , which then triggers cell cycle arrest ( or apoptosis ) ; this would cause the Runx1-null ( YFP+ ) MLs to be outcompeted by Runx1-WT ( YFP− ) MLs . If this is the case , then either disruption of the p53 pathway or activation of cell cycle by Rb1-loss might rescue the phenotype of ER+ ML cell loss upon Runx1 disruption . To test this , we bred MMTV-Cre;Runx1L/L;R26Y mice to Trp53 or Rb1 conditional knockout mice ( Trp53L/L or Rb1L/L ) . In the resulting compound mice , we were only able to follow MMTV-Cre;Runx1L/L;Trp53L/L;R26Y or MMTV-Cre;Runx1L/L;Rb1L/L;R26Y females for ∼4–5 months or ∼9–10 months , respectively , due to lethality possibly caused by hematopoietic malignancies ( as MMTV-Cre has leaky expression in bone marrow hematopoietic cells ) . Nevertheless , we were able to analyze MEC subpopulations in their MGs . Upon MMTV-Cre-induced Trp53 or Rb1 loss alone , the percentages of YFP-marked MECs increased dramatically so that the majority of MECs in their MGs became YFP+ ( Figure 5A , increased from ∼20–30% to ∼70–90% ) , suggesting a growth advantage for Trp53-null or Rb1-null MECs ( in relation to their Trp53-WT or Rb1-WT YFP− neighbors ) . However , the percentages of the YFP-marked luminal population were both reduced ( Figure 5A , middle and bottom left plots compared to upper left plot , green circles ) . Interestingly , disruption of Runx1 either together with Trp53 or with Rb1 significantly increased the percentages of the YFP+ luminal population ( Figure 5A–C , increased from ∼4% [Trp53-loss alone] to ∼11% [Runx1/Trp53-loss] [Figure 5B] and from ∼12% [Rb1-loss alone] to ∼23% [Runx1/Rb1-loss] [Figure 5C] , respectively ) . Of particular note , the percentage of the YFP-marked ML subpopulation , which was dramatically reduced upon Runx1-loss alone ( Figure 5A , upper right plot , red circle ) , was reverted back to almost the normal level upon simultaneous loss of Runx1 together with either Trp53 or Rb1 ( Figure 5A , middle right plot for Trp53 [5B for quantification] , bottom right plot for Rb1 [5C for quantification] , red circles ) . To verify the presence of ER+ MECs in the MGs of these compound female mice , we performed IHC staining for ERα and could indeed detect abundant ERα+ luminal MECs in both MMTV-Cre;Runx1L/L;Trp53L/L;R26Y and MMTV-Cre;Runx1L/L;Rb1L/L;R26Y compound females ( Figure 5—figure supplement 1A–B , since the majority of MECs in their MGs were YFP+ , most of these ERα+ MECs should represent MECs with simultaneous loss of Runx1 and Trp53 or Rb1 ) . As the residual YFP+ MECs in the ML gate from MMTV-Cre;Runx1L/L;R26Y females ( Runx1-loss alone ) appear to have escaped Cre-mediated excision in at least one Runx1L allele ( Figure 4E ) , we wanted to determine whether YFP+ MLs in these compound mice had undergone ( or escaped ) Cre-mediated excision of their Runx1L alleles . By qRT-PCR analysis , we observed more than 50% reduction in the Runx1 expression level in the YFP-marked ML subpopulation sorted from MMTV-Cre;Runx1L/L;Rb1L/L;R26Y females ( Figure 5D ) , suggesting a significant portion of these YFP+ MLs should have undergone biallelic excision of their Runx1L alleles . 10 . 7554/eLife . 03881 . 011Figure 5 . Reduction in ER+ MLs upon Runx1 disruption can be rescued by Trp53 or Rb1 loss . ( A ) FACS analysis showing total lin−YFP+ MEC population , lin−YFP+ luminal population , and lin−YFP+ ML and LP subpopulations ( from left to right for each genotype , an example of the gating strategy is indicated in the bottom left plots ) in female mice with the indicated genotypes . Those highlighted in green show increased lin−YFP+ luminal populations upon the loss of both Runx1 and Trp53 or Rb1 ( middle and bottom right plots , respectively ) compared to those of Trp53 or Rb1 loss alone ( middle and bottom left plots , respectively ) ; those highlighted in red show increased lin−YFP+ ML subpopulations upon the loss of both Runx1 and Trp53 or Rb1 ( middle and bottom right plots , respectively ) compared to that of Runx1-loss alone ( upper right plot ) . Lu: luminal; Ba: basal; LP: luminal progenitor; ML: mature luminal cell . ( B–C ) Quantifications for the percentages of each indicated subpopulation in A under either the Trp53-loss ( B ) or Rb1-loss background ( C ) ; in ( B ) Trp53-loss alone ( n = 5 ) , Trp53/Runx1-loss ( n = 3 ) ; in ( C ) Rb1-loss alone ( n = 5 ) , Rb1/Runx1-loss ( n = 5 ) . ( D ) qRT-PCR analysis showing significantly reduced Runx1 expression in both the YFP-marked ML and LP subpopulations in MMTV-Cre;Runx1L/L;Rb1L/L;R26Y females . p values: : p ≤ 0 . 05; ^: p ≤ 0 . 0005; NS = not significant; error bars represent mean ± S . E . M . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 01110 . 7554/eLife . 03881 . 012Figure 5—figure supplement 1 . Abundant ER+ MECs are present in both Runx1/Trp53-null and Runx1/Rb1-null MGs . ( A–B ) IHC staining for ERα showing abundant ER+ MECs ( brown cells ) in MMTV-Cre;Runx1L/L females upon simultaneous loss of either Trp53 ( A ) or Rb1 ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 012 Since Runx1-loss leads to a reduction in ER+ MLs and the residual MECs present in the CD14−c-Kit− or CD49b−Sca1+ ML gate appear to have escaped Cre-mediated disruption of Runx1 ( Figure 4 , Figure 4—figure supplement 1 ) , it is technically challenging to study how RUNX1 controls the fate of ER+ luminal cells at the molecular level in this mouse model directly . Therefore , we first performed molecular studies in human breast cancer cell lines MCF7 and T47D . Although both cell lines are ER+ luminal breast cancer cell lines , a key difference between them at the molecular level is that MCF7 cells express WT p53 , whereas T47D cells carry a TP53 missense mutation ( nonfunctional p53 ) ( Schafer et al . , 2000 ) . Interestingly , despite multiple attempts , we were only able to obtain RUNX1 knockdown ( kd ) stable lines from TP53-mutant T47D cells but not from TP53-WT MCF7 cells . This observation suggests that a similar genetic interaction between RUNX1-loss and TP53-loss may also operate in human ER+ luminal breast cells . We therefore used T47D cells as our cell line model to study how RUNX1 controls the fate of ER+ luminal breast epithelial cells . By Western blot , we found that upon RUNX1 kd , the protein level of ELF5 was increased , whereas the protein levels of both ERα and FOXA1 were reduced , and CITED1 protein level appeared unchanged ( Figure 6A ) . 10 . 7554/eLife . 03881 . 013Figure 6 . RUNX1 controls transcription of select target genes in human ER+ breast cancer cells . ( A ) Western blot showing upregulation of ELF5 and downregulation of ERα and FOXA1 upon RUNX1 knockdown ( kd ) in T47D luminal breast cancer cells . ( B ) ChIP analysis showing significant binding of RUNX1 to multiple ECRs ( evolutionarily conserved regions ) with RUNX-binding sites in the ELF5 locus in T47D cells . ( C ) ChIP analysis showing significant binding of RUNX1 to the −1 . 6 kb and −1 . 9 kb regions of FOXA1 in T47D cells . RUNX1-binding to the −1 . 4 kb region is marginally significant ( p = 0 . 08 ) . In ( B–C ) , RUNX1-binding motifs ( highlighted in red ) and their flanking sequences are shown; note RUNX1-binding motifs in ECR-1 and ECR-3 of ELF5 ( B ) are in the reverse strand . p values: : p ≤ 0 . 05; #: p ≤ 0 . 005; NS = not significant; error bars represent mean ± S . E . M . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 01310 . 7554/eLife . 03881 . 015Figure 6—figure supplement 1 . Opposite expression patterns of RUNX1 and ELF5 proteins upon alveolar differentiation of HC11 cells . Western blot showing downregulation of RUNX1 and upregulation of ELF5 protein levels upon induced alveolar differentiation in HC11 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 015 ELF5 is a master regulator of alveolar cells , a cell type in which Runx1 is not expressed ( Figure 1D–J , Figure 1—figure supplement 1C ) . Interestingly , it was shown previously that RUNX1 is a direct target of ELF5 and is repressed by it , based on chromatin immunoprecipitation ( ChIP ) analysis ( Kalyuga et al . , 2012 ) . In our microarray data for sorted MEC subsets , as well as those publicly available microarray datasets we analyzed , we could always observe a largely opposite expression pattern of Elf5 and Runx1 ( Figure 1I , Figure 1—figure supplement 1 ) . In both basal cells and MLs in which Runx1 is highly expressed , Elf5 is not; Elf5 expression is greatly elevated in ALs whereas Runx1 expression is repressed . This negative correlation in their expression levels could be further confirmed in the HC11 cell line model . While both Runx1 and Elf5 were expressed in uninduced HC11 cells , upon induction of alveolar differentiation , the ELF5 protein level was increased , whereas the RUNX1 protein level was reduced ( Figure 6—figure supplement 1 ) . To determine whether ELF5 is also a direct target of RUNX1 , we performed ChIP analysis on T47D cells and identified significant binding of RUNX1 to multiple evolutionary conserved RUNX-binding sites in the ELF5 locus ( Figure 6B ) . The RUNX1-binding was particularly profound in an enhancer region ∼17 kb upstream of the ELF5 transcription start site ( ECR-1 , Figure 6B ) . Since RUNX1 kd in T47D cells led to downregulation of ERα and FOXA1 ( Figure 6A ) , we asked whether ESR1 ( encoding ERα ) and FOXA1 are direct targets of RUNX1 . We identified a RUNX-binding motif in the ESR1 control region ∼1 . 4 kb upstream of its transcription start site , as well as several RUNX-binding motifs in the FOXA1 control region ∼1 . 4–1 . 9 kb upstream of its transcription start site . By ChIP assay , we confirmed significant binding of RUNX1 to the −1 . 6 kb and −1 . 9 kb motifs in the FOXA1 locus ( Figure 6C ) . Collectively , these data suggest that FOXA1 and ELF5 genes may be direct targets of RUNX1 positively and negatively regulated by it , respectively . To determine whether RUNX1 regulates the expression of these transcription regulators in primary cells in vivo , we took advantage of the rescue of Runx1-null ER+ luminal MECs by Trp53 or Rb1-loss ( Figure 5 ) and measured expression of these genes in FACS-sorted YFP+ luminal MEC subsets . We used MMTV-Cre;Runx1L/L;Rb1L/L;R26Y double mutant and MMTV-Cre;Rb1L/L;R26Y single mutant females for this analysis , as MMTV-Cre;Runx1L/L;Trp53L/L;R26Y double mutant females often exhibit early lethality . When comparing double mutants ( with Rb1/Runx1-loss ) to single mutants ( with Rb1-loss alone ) , we found that both Elf5 and Esr1 appeared upregulated in ER+ LPs and ER+ MLs ( based on CD49 and Sca1 staining ) from double mutants ( Figure 7A–B , left plots ) , and Foxa1 and Cited1 were downregulated in the rescued double mutant ER+ MLs ( Figure 7B , left plot ) . Since we cannot rule out a possibility in which Rb1-loss in MECs also affects expression of these genes , we compared their expression in double mutants to matched WT females as well . From this comparison , we found that both Elf5 and Esr1 were also upregulated and Foxa1 and Cited1 were slightly downregulated in ER+ LPs and ER+ MLs from double mutants ( Figure 7A–B , right plots ) . Furthermore , as we showed above , in MMTV-Cre;Runx1L/L;R26Y females , although the YFP-marked ER+ MLs appeared to have escaped Cre-mediated disruption of the Runx1L allele , the YFP-marked ER+ LP subset did exhibit a partial reduction in Runx1 expression ( Figure 4—figure supplement 1G ) . We therefore asked whether there is any correlation of reduced Runx1 expression to changes in expression of other genes in this MEC subset . Interestingly , we found that in Runx1-mutant ER+ LPs both Elf5 and Esr1 were upregulated and Foxa1 and Cited1 also appeared slightly upregulated ( Figure 7—figure supplement 1A ) . 10 . 7554/eLife . 03881 . 016Figure 7 . Target genes of RUNX1 in ER+ luminal MECs revealed by in vivo expression analysis . ( A–B ) qRT-PCR analysis showing changes in expression of the indicated genes in sorted YFP+ ER+ LPs ( A ) and ER+ MLs ( B ) ( based on CD49b and Sca1 expression ) from 4- to 5-month old MMTV-Cre;Rb1L/L;Runx1L/L;R26Y double mutant females , compared to those from either MMTV-Cre;Rb1L/L;R26Y single mutant females ( left plots ) or MMTV-Cre;R26Y WT females ( right plots ) . ( C–D ) qRT-PCR analysis comparing expression of the indicated genes in CD14−c-Kit− MLs ( C ) and CD14+c-Kit+ LPs ( D ) from 7-month to 2-month old MMTV-Cre;Rb1L/L;Runx1L/L;R26Y double mutant and MMTV-Cre;Rb1L/L;R26Y single mutant ( control ) females . Expression was normalized to those of the corresponding 2-month old double or single mutant females , respectively . Error bars: mean ± S . E . M . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 01610 . 7554/eLife . 03881 . 017Figure 7—figure supplement 1 . Loss of Runx1 in vivo leads to changes in expression of ER program-related genes . ( A ) Changes in expression of the indicated genes in the ER+ LP subpopulation from MMTV-Cre;Runx1L/L;R26Y females ( Runx1 L/L ) compared to MMTV-Cre;Runx1+/+;R26Y WT females ( Runx1 +/+ ) . ( B ) Elf5 expression is de-repressed upon Runx1-loss in all ER− MEC subsets examined , including CD49b+Sca1− ER− LPs , CD14+c-Kit+ LPs enriched for ER− cells ( both sorted from MMTV-Cre;R26Y females ) and ER− basal MECs ( sorted from Krt14-Cre;R26Y females ) . ( C ) Compared to ER+ LPs shown in A , Esr1 expression is not increased in CD49b+Sca1− ER− LPs , CD14+c-Kit+ LPs ( enriched for ER− cells ) , and ER− basal MECs with Runx1-loss . Note Esr1 expression in CD14+c-Kit+ LPs with Runx1-loss is slightly elevated , possibly due to the fact that a small portion of CD14+c-Kit+ LPs are ER+ LPs ( Shehata et al . , 2012 ) , which exhibit ostensibly higher Esr1 expression , as shown in A . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 01710 . 7554/eLife . 03881 . 018Figure 7—figure supplement 2 . RUNX1 reduction leads to hyperproliferation of abnormal ER+ luminal cells in a context-dependent manner . ( A ) Quantification of the percentages of the ER− LP , ER+ LP , and ER+ ML subpopulations sorted from MMTV-Cre;Rb1L/L;Runx1L/L;R26Y females ( n = 4 ) compared to those from MMTV-Cre;Rb1L/L;R26Y control females ( n = 4 ) showing slight upregulation of the ER+ LP and ER+ ML subpopulations from MMTV-Cre;Rb1L/L;Runx1L/L;R26Y females . ( B ) Knockdown ( kd ) of RUNX1 in T47D luminal breast cancer cells , which carry a TP53 mutation , increased their proliferation . ( C–D ) Quantification of IHC staining for ERα showing significant increase in the numbers of ER+ luminal MECs in MMTV-Cre;Rb1L/L;Runx1L/L;R26Y ( C ) and MMTV-Cre;Trp53L/L;Runx1L/L;R26Y ( D ) double mutant females compared to their corresponding single mutant control females . ( E ) qRT-PCR analysis showing expression of indicated genes in the ER− LP , ER+ LP , and ER+ ML subpopulations from MMTV-Cre;R26Y WT females . Expression was normalized to that of the ER+ LP subpopulation ( =1 ) . p values: : p ≤ 0 . 05; #: p ≤ 0 . 005; ^: p ≤ 0 . 0005; error bars represent mean ± S . E . M . ( error bars represent mean ± S . D . in B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 018 Thus , from both cell line and in vivo expression analyses , the gene that exhibits the most consistent change upon Runx1-loss is Elf5 , which appears to be a target gene of RUNX1 repressed by it in ER+ luminal MECs . Intriguingly , we found that upregulation of Elf5 upon Runx1 disruption is not restricted to ER+ luminal cells and/or the Rb1-loss genetic background . In LPs defined based on either CD14+c-Kit+ or CD49b+Sca1− where Elf5 is abundantly expressed , loss of Runx1 further increased their Elf5 expression ( Figure 7—figure supplement 1B ) . Strikingly , in basal MECs where Elf5 is normally not expressed ( Figure 1I , Figure 1—figure supplement 1 ) , loss of Runx1 led to its profound upregulation ( Figure 7—figure supplement 1B ) . These data suggest that in normal MGs , RUNX1 represses expression of Elf5 in almost all MEC subsets in which Runx1 is expressed . Our in vivo data showed that Esr1 is upregulated rather than downregulated ( based on the in vitro data in T47D cells , Figure 6A ) in ER+ luminal MECs upon Runx1-loss . This is most likely due to hyperproliferation of Runx1-null ER+ luminal MECs under the Rb1 ( or Trp53 ) -null background , rather than de-repression of Esr1 expression caused by Runx1-loss . Several lines of evidence support this notion . First , in Runx1/Rb1-double mutant females , we not only observed a slight increase in the percentages of total YFP+ MECs ( Figure 5A ) but also an increase in both the YFP+ luminal subset and , in particular , the YFP-marked ER+ LP and ER+ ML subpopulations ( Figure 5A–C , Figure 7—figure supplement 2A ) . In TP53-mutant T47D cells , we found that kd of RUNX1 leads to a significant increase in their proliferation ( Figure 7—figure supplement 2B ) . Furthermore , we quantified ERα+ luminal MECs in MGs with either Runx1/Rb1-loss or Runx1/Trp53-loss and found that both double mutants contained significantly more ERα+ MECs than single Rb1-loss or Trp53-loss mutants ( Figure 7—figure supplement 2C–D ) . Since our expression analysis was based on FACS-sorted MEC subsets ( e . g . , LP , ML ) and each subset may represent a mixture of both ER+ and ER− MECs ( with different proportions ) , a change in this proportion , due to overrepresentation of the rescued ER+ luminal cells in the FACS-sorted MEC subpopulations from Rb1/Runx1-null double mutants , may contribute to the ostensibly higher Esr1 transcript signals in double mutant ER+ LPs and ER+ MLs ( Figure 7A–B ) . Lastly , we found that Esr1 upregulation in vivo appears restricted to the Runx1-null ER+ luminal MECs; in ER− LPs and ER− basal MECs , we did not observe upregulation of Esr1 expression upon Runx1 disruption ( Figure 7—figure supplement 1C ) . This is apparently different from negative regulation of Elf5 by RUNX1 , in which loss of Runx1 leads to expression of Elf5 even in basal cells in which Elf5 is normally not expressed ( Figure 7—figure supplement 1B ) . This is also different from a recent finding of repression of Esr1 by ID4 , as loss of Id4 leads to widespread upregulation of Esr1 expression in both luminal and basal MECs ( Best et al . , 2014 ) . Collectively , these data suggest that Esr1 is not a direct target repressed by RUNX1 in vivo; the downregulation of ERα in vitro in T47D cells upon RUNX1 kd is likely to be indirect ( e . g . , due to RUNX1 loss-induced upregulation of ELF5 , as overexpression of ELF5 in T47D cells can also suppress ERα expression [Kalyuga et al . , 2012] ) . The higher Esr1 signal makes it challenging to accurately quantify any potential changes in expression levels of ER-related mature luminal genes upon Runx1 disruption , by simply comparing their expression in double to single mutants or to WT controls ( as the matched MEC subsets based on FACS sorting may have different cell compositions , if Runx1-null ER+ luminal MECs become over-populated ) . Interestingly , we noticed that in YFP+ MLs from MMTV-Cre;Runx1L/L;Rb1L/L;R26Y females , Runx1 reduction became more profound in older females . We therefore similarly monitored changes in expression of other TF/co-factor genes over time in animals with the same genotype . This strategy may allow us to control for gene expression changes introduced by differences in cell populations or genetic backgrounds . By using this strategy , we found that in Rb1-null single mutants ( controls ) , Esr1 , Foxa1 , and Cited1 were upregulated and Elf5 was downregulated in their YFP+ CD14−c-Kit− MLs when they aged; however , in YFP+ MLs from double mutants , following Runx1 reduction , although Esr1 was upregulated to a similar level ( to that of single mutants ) , Foxa1 and Cited1 were not , and Elf5 was even further upregulated ( Figure 7C ) . We also observed a similar trend of changes for Foxa1 , Cited1 , and Elf5 in YFP+ CD14+c-Kit+ LPs from the same animals ( Figure 7D ) . Of note , Esr1 expression in these LPs appeared further upregulated in older females , possibly due to hyperproliferation of the rescued ER+ luminal cells within this largely ER− LP subpopulation ( the LP subset defined based on CD14+c-Kit+ contains a small number of ER+ cells [Shehata et al . , 2012] ) . Overall , the data from this time course study further supports that in ER+ luminal cells , RUNX1 negatively and positively regulates the expression of Elf5 and mature luminal TF/co-factor genes ( e . g . , Foxa1 and Cited1 ) , respectively . As shown above , in YFP+ ER+ LPs from MMTV-Cre;Runx1L/L;R26Y females with partial Runx1 reduction , we observed an abnormal expression pattern of these TF/co-factor genes ( i . e . , upregulation of both Elf5 and Esr1 and slight upregulation of Foxa1 and Cited1 , Figure 7—figure supplement 1A ) . This may be explained by a possibility in which a portion of them are committed for differentiation to ER+ MLs by upregulating Esr1; however due to Runx1-loss , Elf5 is not repressed and Foxa1 is not sufficiently upregulated in them , potentially leading to an abnormal population of Elf5+Esr1highFoxa1lowCited1low ML-like cells retained in the ER+ LP FACS gate . A small number of such abnormal Runx1-null ER+ ML-like cells may also be present in the ML FACS gate . These abnormal luminal MECs may be the target cells rescued for proliferation under the Rb1-loss background , and their hyperproliferation may contribute to the unusual Elf5+Esr1highFoxa1lowCited1low expression pattern ( Figure 7C–D ) . As a further support to this notion , we measured expression levels of these TF/co-factor genes in the ER− LP , ER+ LP , and ER+ ML subsets sorted from WT animals and found that Elf5 expression trends down , whereas expression of Esr1 , Foxa1 , and Cited1 similarly trends up , from ER− LPs to ER+ LPs and then to ER+ MLs , and that Runx1 expression is also elevated from ER+ LPs to ER+ MLs ( Figure 7—figure supplement 2E ) . This expression pattern suggests that differentiation of ER+ luminal MECs requires coordinated expression of these factors and Runx1-loss may disrupt their coordinated expression . Collectively , our in vitro and in vivo expression analyses coupled with ChIP analysis suggest that Elf5 is a key target gene of RUNX1 repressed by it in MECs . RUNX1 also positively regulates the expression of mature luminal TF/co-factor genes involved in the ER program and among them , Foxa1 is a direct target of RUNX1 , and RUNX1 does not appear to regulate transcription of Esr1 directly . Among TFs that control cell fates of the two subpopulations of luminal MECs , GATA3 has been shown as a common master regulator for both ER+ ductal luminal cells and ER− alveolar luminal cells ( Kouros-Mehr et al . , 2006; Asselin-Labat et al . , 2007 ) , whereas ELF5 has been identified as a key regulatory TF specific for the alveolar luminal subset ( Oakes et al . , 2008; Choi et al . , 2009 ) . However , what is the TF that specifically controls the fate of the ER+ ductal luminal subset remained largely elusive . In this study , we identified RUNX1 as a key regulator of ER+ luminal MECs . RUNX1 controls the in vivo fate of this luminal subpopulation by repressing the program for an alternative cell fate choice ( i . e . , repressing the key TF gene for alveolar cells , Elf5 ) and by optimizing activation of the ML gene expression program ( i . e . , regulating key mature luminal TF/co-factor genes such as Foxa1 ) ( Figure 8A ) . Loss of Runx1 impairs the fate of ER+ luminal cells , leading to a profound reduction in this luminal subpopulation . However , the loss of either Trp53 or Rb1 can rescue this defect , leading to hyperproliferation of Runx1-mutant ER+ luminal cells , which may eventually progress to ER+ luminal breast cancer , upon acquisition of additional mutations ( Figure 8B ) . Our study thus provides a direct link between a somatically mutated lineage-specific TF , impaired cell fate , and development of luminal breast cancer . 10 . 7554/eLife . 03881 . 019Figure 8 . Model for the role of RUNX1 in ER+ mammary luminal cells and luminal breast cancer . ( A ) Relative expression levels of key TFs in different subsets of MECs are indicated ( ‘+++’ , ‘++’ , ‘+’ , ‘±’ , ‘−’ indicate highest to low to no expression , based on [Figure 1I] and our single cell profiling data for sorted MECs [MPAvB and ZL , unpublished data] ) . RUNX1 or ELF5 controls the ductal or alveolar luminal cell fate , respectively , by antagonizing each other . RUNX1 further controls the fate of ER+ ductal luminal MECs by regulating the ER program via modulating FOXA1 expression . ( B ) Genetic interaction between the loss of RUNX1 and the loss of either TP53 or RB1 plays a key role in the development of RUNX1-mutant ER+ luminal breast cancer . DOI: http://dx . doi . org/10 . 7554/eLife . 03881 . 019 Among RUNX1 target genes , the repressed Elf5 is of particular interest , as it encodes a master regulatory TF for the alternative cell fate of the milk-secreting alveolar lineage in which Runx1 is not expressed ( Figure 1D–G , I–J ) . We showed that ELF5 is a direct target of RUNX1 and is repressed by it ( Figures 6A–B and 7 ) . Thus , combined with the previous observation in which RUNX1 was reciprocally shown as a direct target repressed by ELF5 ( Kalyuga et al . , 2012 ) , these data suggest that RUNX1 and ELF5 are two master regulators for mutually exclusive cell fate choices ( i . e . , ductal vs alveolar fates ) by antagonizing each other's transcription program ( e . g . , RUNX1 promotes the ER program [this study] , whereas ELF5 suppresses it [Kalyuga et al . , 2012] ) ( Figure 8A ) in a way similar to the GATA1-PU . 1 paradigm for regulating the choice between erythroid and myeloid fates ( Huang et al . , 2007 ) . Intriguingly , RUNX1 not only represses Elf5 expression in ER+ luminal cells but also in all other MEC subsets in which Runx1 is expressed ( Figure 7—figure supplement 1B ) . The de-repression of Elf5 in basal MECs may also be of clinical relevance . Recently it was found that RUNX1 protein expression correlates with poor prognosis in ER− breast cancer and more specifically in triple-negative breast cancer ( TNBC ) ( Ferrari et al . , 2014 ) . Furthermore , RUNX1 was also found associated with super-enhancers in an ER− breast cancer cell line ( Hnisz et al . , 2013 ) . As super-enhancers often associate with key oncogenes in cancer cells ( Loven et al . , 2013 ) , these recent findings suggest that RUNX1 may also play an oncogenic role in ER− breast cancers . The link between RUNX1 and ELF5 in basal MECs may explain a potential oncogenic role of RUNX1 in ER− breast cancer/TNBC , as it was shown previously that SNAI2 ( encodes SLUG ) is a target of ELF5 repressed by it ( Chakrabarti et al . , 2012 ) . Thus , it is possible that RUNX1 expression in ER− breast cancer cells may repress ELF5 expression , leading to de-repression ( thus upregulation ) of SNAI2 expression , which then promotes epithelial-mesenchymal transition ( EMT ) and aggressiveness of breast cancer cells . Interestingly , it was shown recently that Snai2-null mice exhibit a nursing defect , due to failed milk ejection caused by defects in basal/myoepithelial cell differentiation ( Phillips et al . , 2014 ) . In Runx1-null mice , upregulation of Elf5 in basal MECs may lead to repression of Snai2 , which may provide an explanation for the similar nursing defect we have observed in our Runx1 conditional knockout mice ( Figure 3—figure supplement 1B–C ) . In luminal breast cancer , our study provides strong evidence to support that RUNX1 plays a key role in this breast cancer subtype as a tumor suppressor in ER+ ductal luminal cells , which may be their cells of origin . All three RUNX TFs have been shown to play context-dependent roles in breast cancer development as either tumor suppressors or oncogenes ( Chimge and Frenkel , 2013 ) . Among them , RUNX3 is also a tumor suppressor as it is often inactivated in human breast cancers and loss of one copy of Runx3 led to spontaneous mammary tumor development in a portion of aged female mice ( Huang et al . , 2012 ) . The tumor suppressor role of RUNX3 in breast cancer is explained by its ability to inhibit ERα-dependent transactivation by reducing the stability of ERα ( Huang et al . , 2012 ) . In contrast , RUNX2 mainly exhibits oncogenic roles in breast cancer by promoting invasiveness and metastasis via its target , SNAI2 ( Chimge et al . , 2011 ) ; however it may also play a tumor suppressor role in breast cancer by antagonizing ERα ( thus , similar to RUNX3 ) ( Chimge et al . , 2012 ) . In this study , we showed that RUNX1 , the most abundantly expressed RUNX TF in MECs , controls the fate of ER+ luminal cells in part by upregulating FOXA1 and repressing ELF5 . Furthermore , RUNX1 has also been shown as a novel tethering factor for recruiting ERα to its genomic sites for ER-mediating transcriptional activation ( Stender et al . , 2010 ) . Estrogen signaling has dual roles in MECs and breast cancer cells; on one hand it has an oncogenic role by promoting proliferation of ER+ luminal breast cancer cells , on the other hand it also has a tumor suppressor role by promoting MEC differentiation and inhibiting metastasis of breast cancer cells ( Chimge and Frenkel , 2013 ) . The tumor suppressor role of RUNX2 and RUNX3 mainly relates to the antagonism between RUNX2/3 and the cancer-promoting program of ER signaling , whereas the tumor suppressor role of RUNX1 largely correlates to its ability to positively regulate the tumor-suppression program of ER signaling . The tumor suppressor role of RUNX1 is also consistent with a previous observation in which RUNX1 was found among a 17-gene signature associated with metastasis as a gene downregulated in metastasis-prone solid tumors , including breast cancer ( Ramaswamy et al . , 2003 ) . Lastly , our study also provides an explanation for the paradox in which RUNX1 is a positive regulator of the ER program , yet its loss-of-function mutations and deletions are only present in ER+ human luminal breast cancers ( often accompanied by mutations or copy number losses in TP53 or RB1 genes ) ( Cancer Genome Atlas Network , 2012; Ellis et al . , 2012 ) . We show that the loss of Runx1 does not appear to affect transcription of Esr1 directly ( thus , the affected luminal cells remain phenotypically ER+ ) but may lead to a crippled ER program , in part due to de-repression of Elf5 and insufficient upregulation of Foxa1 , which may reduce the sensitivity and output of ER signaling , respectively ( Hurtado et al . , 2011; Kalyuga et al . , 2012 ) . The impaired ER program in Runx1-mutant ER+ luminal cells may cause cellular stress , leading to activation of the p53 pathway and subsequently cell cycle arrest and/or apoptosis; as a result , abnormally differentiated Runx1-mutant ER+ luminal cells are outcompeted by their WT neighbors in vivo . However , the loss of Trp53 or Rb1 can relieve the cell cycle arrest or positively activate cell cycle , respectively , and/or rescue apoptosis in them , leading to rescue of the Runx1-mutant ER+ luminal cells . In humans , upon acquisition of additional mutations , the RUNX1/TP53-mutant or RUNX1/RB1-mutant ER+ premalignant luminal cells may progress to ER+ luminal breast cancer , upon acquisition of additional oncogenic events ( Figure 8B ) . Of note , germline mutations of RUNX1 that result in haploinsufficiency of RUNX1 can lead to an autosomal dominant disorder referred to as familial platelet disorder with a propensity to acute myeloid leukemia ( FPD/AML ) ( Song et al . , 1999 ) . Interestingly , in one study that characterized three FPD/AML pedigrees , it was found that one female patient with FPD/AML also developed a breast cancer 2 years after AML was diagnosed , and no other tumors were observed in all three pedigrees ( Preudhomme et al . , 2009 ) . Although the sample size for this study was too small , it certainly raises an intriguing question as to whether germline mutations of RUNX1 predispose FPD/AML patients to luminal breast cancer , but only under a background of either TP53 or RB1 loss . In summary , we identified RUNX1 as a key regulator of the ER+ luminal lineage . Loss of RUNX1 may contribute to the development of ER+ luminal breast cancer under a background of either TP53 or RB1 loss and upon cooperation with other additional oncogenic events . Mice carrying the floxed Runx1 allele ( Runx1L/L ) ( Li et al . , 2006b ) were bred with mice carrying a conditional Cre-reporter , R26Y . Subsequently , these mice were bred with mice that drive expression of Cre recombinase under the control of the mouse mammary tumor virus ( MMTV ) promoter ( MMTV-Cre ) and with mice carrying the floxed Trp53 allele ( Trp53L/L ) or floxed Rb1 allele ( Rb1L/L ) . For studying Runx1 disruption in basal MECs , Cre transgenic mice under the control of the Keratin 14 promoter ( Krt14-Cre ) were also used . Mice were obtained from JAX ( R26Y: 006148; MMTV-Cre: 003553 ) or the MMHCC repository ( Krt14-Cre: 01XF1; Wap-Cre: 01XA8 ) or were a generous gift from Dr Stuart Orkin ( Trp53L/L and Rb1L/L [Walkley et al . , 2008] ) . All animal experiments and procedures were approved by our Institutional Animal Care and Use Committee ( IACUC ) . Whole-mounts of MGs of pubertal , adult virgin , or lactation day-0 mice were fixed and processed as previously described ( Jones et al . , 1996 ) . For histology and immunohistochemical staining , MGs were fixed in 10% formalin and embedded in paraffin . For RUNX1 or ERα detection , antigen retrieval ( Citrate buffer pH 6 . 0 , 20 min boil in microwave oven ) was performed prior to incubation with an anti-RUNX1 antibody ( 2593-1 , Epitomics , Burlingame , CA ) or an anti-ERα antibody ( SC-542 , Santa Cruz Biotechnology , Dallas , TX ) . Signal was detected using the impress reagent kit and DAB substrate ( MP-7401 and SK-4100 , Vector Laboratories , Burlingame , CA ) . Thoracic and inguinal mammary glands were dissected from pubertal or adult virgin female mice and cell suspensions were prepared as previously described ( Shackleton et al . , 2006 ) . Flow cytometric analysis was performed with a DXP11 analyzer ( Cytek , Fremont , CA ) or an Accuri C6 analyzer ( BD Biosciences , San Jose , CA ) . FACS sorting was performed with a FACSAria sorter ( BD Biosciences ) . Data were analyzed with FlowJo ( Tree Star , Ashland , OR ) or CFlow ( BD Biosciences ) . Antibodies used for FACS were purchased from eBiosciences ( San Diego , Ca ) and included CD24-eFluor450 , CD24-eFluor605 , CD29-APC , CD61-PE , c-Kit-PE-CY7 , CD14-PE , CD49b-PE , Sca1-APC and biotinylated CD31 , CD45 , and TER119 ( i . e . , lineage [Lin] markers ) , as well as Streptavidin-PerCP-CY5 . 5 . We also used a Sca1-APC-CY7 antibody purchased from BD biosciences ( San Jose , CA ) . Total RNA from sorted subsets of MECs was prepared by the RNeasy kit ( Qiagen , Valencia , CA ) and amplified with the Ovation RNA Amplification System V2 ( Nugen , San Carlos , CA ) . YFP-marked luminal cells were sorted from adult virgin MMTV-Cre;Runx1L/L;R26Y or MMTV-Cre;Runx1+/+;R26Y littermates . Normal MEC subsets , including MaSCs , LPs , and MLs , were sorted from WT C57/B6 adult virgin females; alveolar luminal cells ( ALs ) were sorted as YFP+ cells from Wap-Cre;R26Y females at mid-gestation . Mouse Genome 430 2 . 0 Array ( Affymetrix , Santa Clara , CA ) was used to generate the expression profiles . All arrays were normalized by dCHIP and analyzed by GSEA as described ( Subramanian et al . , 2005 ) , using MSigDB database v3 . 1 ( http://www . broadinstitute . org/gsea/msigdb/index . jsp ) . For qRT-PCR , cDNA was generated with Omniscript ( Qiagen ) according to the manufacture's protocol and real-time PCR was performed using FastStart SYBR Green Master ( Roche , Indianapolis , IN ) . ΔΔCt method was used for normalization to the control group and to the endogenous control ( Hprt ) . Primers are listed in Supplementary file 1 . Cells were cross-linked with 1% formaldehyde at room temperature for 10 min , quenched with 0 . 125 M glycine for 5 min and washed with PBS , harvested by scraping and lysed in cell lysis buffer ( 0 . 1% SDS; 0 . 5% NP40; 1 mM EDTA; 10 mM Tris–HCl , pH 7 . 4; 0 . 5% NaDOC ) . 200–1000 bp DNA fragments were obtained after sonication . After 10 min centrifugation at max speed at 4°C , supernatant was used for IP overnight at 4°C . 30 μl Dynabeads Protein G beads ( Invitrogen , Carlsbad , CA ) and 1 μg antibody were used for each IP . One tenth of lysate was saved as input . The following antibodies were used: rabbit anti-RUNX1 ( ab92336 , Abcam , Cambridge , MA ) , rabbit IgG ( sc-2027 , Santa Cruz Biotechnology ) . The beads were washed twice with the following buffers , 3 min each: low-salt buffer ( 0 . 1% SDS; 1% Triton X-100; 1 mM EDTA; 10 mM Tris–HCl , pH 7 . 4; 300 mM NaCl; 0 . 1% NaDOC ) , high-salt buffer ( 0 . 1% SDS; 1% Triton X-100; 1 mM EDTA; 10 mM Tris–HCl , pH 7 . 4; 500 mM NaCl; 0 . 1% NaDOC ) , LiCl buffer ( 10 mM Tris–HCl , pH 8; 0 . 25M LiCl; 1 mM EDTA , pH 8; 1% NP-40; 1% NaDOC ) , and TE . Precipitated materials were eluted with 300 μl elution buffer ( 1% SDS; 0 . 1 M NaHCO3; 50 mM Tris–HCl , pH 8; and 10 mM EDTA ) . Chromatin was reverse-cross-linked by adding 12 μl of 5 M NaCl and incubated overnight at 65°C . DNA was obtained after RNaseA treatment , protease K treatment , phenol/chloroform extraction , and ethanol precipitation . DNA was analyzed by qPCR , normalized to the input DNA . Primers are listed in Supplementary file 1 . shRNAs for RUNX1 were purchased from Open Biosystems ( Huntsville , AL; shRNA sequences are listed in Supplementary file 1 , data from a pool of TRCN0000013659-D1 and TRCN0000013662-D4 were shown ) . After lentiviral infection and puromycin selection , stable shRNA-expressing cell lines were generated . For Western blotting , whole-cell extracts were prepared by boiling cells for 10 min at 95°C in SDS sample buffer ( 50 mM Tris [pH 6 . 8]; 100 mM DTT; 2% SDS; 0 . 1% bromophenol blue; 10% glycerol ) . Cell lysates were then resolved by SDS-PAGE . β-actin ( Fisher Lab , Hampton , NH ) was used as a loading control . Primary antibodies ( RUNX1: Abcam ab92336 , ELF5: Abcam ab77007 , CITED1: Abcam ab92550 , ERα: Santa Cruz Biotechnology sc-8002 , FOXA1: Santa Cruz Biotechnology sc-6553 ) were detected using HRP-conjugated anti-rabbit antibodies and visualized using enhanced chemiluminescence detection ( ECL reagents from Fisher Lab ) . Proliferation of T47D cells was determined by absorbance of alamarBlue , following manufacturer's protocol ( Invitrogen Lot155363SA ) . 1 × 105 cells were seeded in a 96-well plate and were measured after 3 or 5 days in culture . 1/10 volume of alamarBlue reagent was directly added to cells in culture medium , incubated for 4 hr at 37°C . Absorbance of alamarBlue was monitored at 570 nm , using 600 nm as a reference wavelength ( normalized to the 600 nm value ) . The results were reported as mean ± S . E . M . unless otherwise indicated , and Student's t tests were used to calculate statistical significance . The microarray expression profiling datasets generated in this manuscript have been deposited to the GEO database under the following accession numbers: GSE47375 ( for Runx1 ) and GSE47376 ( for normal MEC subsets ) or as SuperSeries GSE47377 .	Stem cells can develop into the many types of specialized cell found in the body . Several proteins regulate these transformations by switching on and off the expression of genes that are specific to different cell types . Disrupting these proteins can cause the development of cells to go awry and can lead to cancer . A protein called RUNX1 controls gene expression to direct the development of blood cells . Mutations in the gene encoding this protein have been linked to blood cancers and a particular type of breast cancer , which begins in the cells that line the ducts that carry milk towards the nipple . Mammary duct-lining cells develop from a pool of stem cells that produces breast tissue cells . Now van Bragt et al . have found that RUNX1 is expressed in the cells lining the ducts of the mammary glands , except those that produce milk . Deleting the gene for RUNX1 in mice reduced the number of duct-lining cells , especially a subgroup of cells that are the sensors for the hormone estrogen . Through experiments on breast cancer cells , van Bragt et al . found that RUNX1 is able to dictate the fate of duct-lining breast cells by controlling other protein regulators . RUNX1 boosts the activity of at least one regulator that encourages the cells to become duct-lining cells and represses another regulatory protein that turns cells into milk-producing cells . Next , van Bragt et al . found that , in mice lacking the gene for RUNX1 , reducing the amounts of certain proteins that normally suppress the formation of tumors restored the populations of estrogen-sensing duct-lining cells . This suggests that mutations in the gene encoding RUNX1 , coupled with the loss of a tumor-suppressing protein , may contribute to the development of cancer in the cells that line the breast ducts . The next challenge is to determine exactly how RUNX1 mutations work together with the loss of the tumor-suppressing protein to drive breast cancer development . This knowledge may translate into new approaches to prevent or treat this type of breast cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cancer", "biology" ]	2014	RUNX1, a transcription factor mutated in breast cancer, controls the fate of ER-positive mammary luminal cells
To form and maintain organized tissues , multicellular organisms orient their mitotic spindles relative to neighboring cells . A molecular complex scaffolded by the GK protein-interaction domain ( GKPID ) mediates spindle orientation in diverse animal taxa by linking microtubule motor proteins to a marker protein on the cell cortex localized by external cues . Here we illuminate how this complex evolved and commandeered control of spindle orientation from a more ancient mechanism . The complex was assembled through a series of molecular exploitation events , one of which – the evolution of GKPID’s capacity to bind the cortical marker protein – can be recapitulated by reintroducing a single historical substitution into the reconstructed ancestral GKPID . This change revealed and repurposed an ancient molecular surface that previously had a radically different function . We show how the physical simplicity of this binding interface enabled the evolution of a new protein function now essential to the biological complexity of many animals . The evolution of organized multicellularity is one of the most important and least understood transitions in the history of life ( Grosberg and Strathmann , 2007; Maynard Smith and Szathmary , 1995; Bonner , 1998; King , 2004 ) . Multicellularity – defined as the differentiation and spatial arrangement of cell types into functioning tissues within an integrated organism – evolved independently in several eukaryotic lineages , using unique mechanisms each time to drive the cellular functions necessary for tissue organization ( Rokas , 2008; De Smet and Beeckman , 2011; Trillo and Nedelcu , 2015 ) . Comparative analyses have established that many protein families involved in cell adhesion , signal transduction , and cell differentiation in modern animals first appeared in the genomes of unicellular eukaryotes that were progenitors of animals ( King , 2003; Nichols et al . , 2006; Richter and King , 2013; Rokas , 2008 ) . Virtually nothing is known , however , concerning the molecular mechanisms by which these proteins’ functions evolved . These events happened in the deep past , so horizontal comparisons between extant species are often not sufficient to establish the historical changes in protein sequence , function , or biophysical properties that caused them . Vertical evolutionary analysis using ancestral protein reconstruction – phylogenetic inference of ancestral sequences followed by gene synthesis , genetic manipulation , and experimental characterization – has proven to be an effective strategy for elucidating these questions ( Harms and Thornton , 2010; Harms and Thornton , 2013 ) . Here , we apply ancestral protein reconstruction to investigate the historical trajectory , timing , and mechanisms of evolution of a new protein function important to organized multicellularity in diverse animal phyla . For dividing animal cells to generate and maintain organized tissues , the mitotic spindle must be oriented relative to the position of surrounding cells ( Morin and Bellaïche , 2011; Gillies and Cabernard , 2011; Lu and Johnston , 2013; Cabernard and Doe , 2009; Williams et al . , 2011 ) . Cells that orient the spindle parallel to the epithelial plane , for example , expand the tissue; those that rotate it orthogonally to the plane escape the epithelium , as in epithelial-mesenchymal transitions during development ( Morin and Bellaïche , 2011; Gillies and Cabernard , 2011; Nakajima et al . , 2013 ) . Experiments have identified a protein complex that mediates robust positioning of the mitotic spindle by using a scaffolding protein to link the spindle’s astral microtubules to a molecular marker that is localized on the cell’s cortex by external signals ( Figure 1A ) ( Lu and Johnston , 2013; Johnston et al . , 2009; Siegrist and Doe , 2005; Siegrist , 2006 ) . The complex and its functions have been most extensively studied in Drosophila melanogaster neuroblasts , but it plays a similar role in birds and mammals ( Saadaoui et al . , 2014 ) and in other cell types , such as several kinds of epithelium ( Nakajima et al . , 2013; Saadaoui et al . , 2014; Bergstralh et al . , 2013; Bell et al . , 2015 ) . In this complex , the scaffold is the GK protein interaction domain ( GKPID ) of the protein Discs large ( Dlg ) , which binds microtubule-associated motor proteins , such as the kinesin-3 family member KHC-73 , and the Partner of Inscuteable protein ( Pins in insects , LGN in vertebrates ) . In neuroblasts , the complex is localized relative to the position of adjacent cells by the interaction of a transmembrane receptor – which receives local extracellular signals – with Pins , which in turn recruits GKPID , KHC-73 , and the spindle microtubules ( Yoshiura et al . , 2012 ) . In epithelia , in contrast , localization of the complex relative to surrounding cells appears to be mediated by Dlg itself; Pins is then recruited into the complex via interaction with Dlg’s GKPID . In addition to serving as a localized molecular mark in some cell types , Pins also brings other proteins to the complex , including Mud/Numa and its partners , which generate pulling forces and reinforce proper spindle orientation once it is established by the GKPID complex ( Lu and Johnston , 2013; Johnston et al . , 2009 ) . Other molecules and pathways may be important in spindle orientation in other kinds of cells ( Morin and Bellaïche , 2011; Gillies and Cabernard , 2011; Lu and Johnston , 2013 ) , and further work is required to comprehensively assess the generality of the GKPID complex’s role in spindle orientation across cell types and in the most basal animal lineages . Nevertheless , the fact that the GKPID-mediated complex orients the mitotic spindle in multiple cell types in both protostomes and deuterostomes suggests an ancient and essential role in the biology of complex animals . Indeed , compromising Dlg’s GKPID or other components of the Pins-Dlg-KHC-73 complex affects numerous tissues and cell types by causing impaired spindle orientation , tumorigenesis plasia , defects in cell polarity and differentiation , and developmental failures of tissue organization ( Nakajima et al . , 2013; Johnston et al . , 2009; Siegrist and Doe , 2005; Bergstralh et al . , 2013; Yoshiura et al . , 2012; Bilder , 2000; Woods , 1996 ) . 10 . 7554/eLife . 10147 . 003Figure 1 . Function and phylogeny of the guanylate kinase ( gk ) and GKPID protein family . ( A ) The GKPID of the protein Discs-large ( Dlg , blue ) serves as a scaffold for spindle orientation by physically linking the localized cortical protein Pins ( green ) to astral microtubules ( red ) via the motor protein KHC-73 ( black ) . ( B ) Reduced phylogeny of the protein family containing gk enzymes ( brown ) and protein-binding GKPIDs ( blue ) . Parentheses show the number of sequences in each clade . Reconstructed proteins Anc-gkdup ( the preduplication ancestor of gk enzymes and GKPIDs in animals/choanoflagellates ) , Anc-GK1PID and Anc-GK2PID ( the GKPID in the common ancestor of animals and choanoflagellates , and of animals , respectively ) are marked as circles with approximate likelihood ratio support . Scale bar indicates number of substitutions per site . For unreduced phylogeny , see Figure 1—figure supplement 1 . Characteristics of the reconstructed sequences are found in Figure 1—figure supplement 2 . For sequences analyzed , see Figure 1—source data 1 . For sequences and posterior probabilities of amino acid states , see Figure 1—source data 2 , Figure 1—source data 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 00310 . 7554/eLife . 10147 . 004Figure 1—source data 1 . Species and identifiers for sequences used in alignment and phylogenetic analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 00410 . 7554/eLife . 10147 . 005Figure 1—source data 2 . Posterior probability distribution of ancestral states for Ancgkdup . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 00510 . 7554/eLife . 10147 . 006Figure 1—source data 3 . Posterior probability distribution of ancestral states for AncGK1PID . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 00610 . 7554/eLife . 10147 . 007Figure 1—figure supplement 1 . Complete phylogeny of 224 guanylate kinase enzyme and GKPIDs . Nodes are labeled with approximate likelihood ratio supports , and branch lengths are in substitutions per site ( see scale bar ) . The tree is rooted on the bacterial GK enzymes . Major paralogs in the GKPID family , with the number of sequences included in each , are labeled . See Figure 1—source data 1 for species and accessions used . See Supplementary files 1 and 2 for alignment and tree file . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 00710 . 7554/eLife . 10147 . 008Figure 1—figure supplement 2 . Sequence characteristics of maximum likelihood reconstructions of Anc-gkdup and Anc-GK1PID . ( A , B ) The histogram shows the distribution over sites of posterior probability support for maximum likelihood amino acid states ( see Figure 1—source data 2 , Figure 1—source data 3 for full sequences and support ) . ( C ) Sequence similarity of ancestral sequences to extant gk enzymes and GKPIDs . The table shows the percent of residues identical between each pair of sequences . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 008 Little is known concerning the evolution of animal spindle orientation or the GKPID-mediated complex in particular . Dlg is a member of a larger family of membrane-associated multidomain proteins , all of which contain a GKPID and form protein complexes important to cell adhesion , neural synapse organization , and other functions ( Funke et al . , 2005 ) . The GKPID has been found only in animals , choanoflagellates , and Filasterea ( te Velthuis et al . , 2007; de Mendoza et al . , 2010 ) , but it is similar in both sequence and structure to the guanylate kinase ( gk ) enzymes , which are common to all life and regulate nucleotide homeostasis by catalyzing the transfer of phosphate groups from ATP to GMP ( Li et al . , 1996 ) . These observations suggest that Dlg’s GKPID may have evolved from an ancient gk enzyme ( Johnston et al . , 2011 ) , but this hypothesis is untested and its evolutionary implications are unexplored . For example , it is not known when the GKPID’s scaffolding functions first evolved , either in relation to the emergence of organized multicellularity or the origin of the other components of the spindle orientation complex; it is therefore unclear how the complex was assembled by evolution or what its role may have been in the emergence of spindle orientation and tissue organization . Some work has been done to identify amino acids that contribute to the functional differences between gk enzymes and GKPIDs in present-day organisms ( Johnston et al . , 2011 ) , but it is unknown whether those residues are historically relevant to the evolution of GKPID’s functions or whether they provide a necessary and sufficient explanation in the context of the ancestral protein to explain the emergence of the protein’s spindle-orienting functions . How an ancient gk enzyme gave rise to a protein domain with scaffolding/spindle orientation functions is also a striking model for the evolution of novel molecular functions . Most studies of protein evolution to date have focused on relatively subtle shifts in function , such as changes in relative ligand preference , in allosteric regulation , or in quantitative measures of activity ( Harms and Thornton , 2013 ) . GKPIDs and gk enzymes , however , have entirely different biochemical functions – specific protein binding and catalysis of a nucleotide substrate – suggesting the evolution of an entirely new function . Virtually nothing is known concerning the mechanisms and dynamics by which fundamentally novel protein functions evolve . We therefore used ancestral protein reconstruction to reconstruct the sequences of ancestral members of the protein family that contains the gk enzymes and GKPIDs . This strategy allowed us to trace their functional evolution through time and dissect the genetic and biophysical mechanisms that mediated the evolution of the GKPID’s new functions . To understand the context in which GKPIDs evolved , we also sought insight into the evolution of other components of the spindle orientation machinery – and spindle orientation itself – by examining their presence in the single-celled eukaryotes most closely related to animals . Understanding the historical process of protein evolution begins with a phylogeny . We first assembled a sequence alignment of the gk enzyme/GKPID family by searching publicly available databases and genomes . We recovered gk enzymes from taxa across the tree of life , suggesting a universal distribution , as previously reported ( te Velthuis et al . , 2007; de Mendoza et al . , 2010 ) . In contrast , GKPIDs are present in all animal genomes analyzed as well as in their closest unicellular relatives , the choanoflagellates and Filasterea; however , GKPIDs are absent from the genomes of all sequenced fungi and all other eukaryotic and prokaryotic lineages analyzed . We aligned 224 broadly sampled amino acid sequences and inferred the phylogeny of the gk enzyme/GKPID family using maximum likelihood phylogenetics , rooted using the bacterial gk enzymes as an outgroup ( Figure 1B , Figure 1—figure supplement 1 ) . All GKPIDs cluster together as a monophyletic group , with the gk enzymes forming a paraphyletic set of basal lineages . Within the GKPIDs , there are two major clades , one of which contains Dlg and closely related paralogs; the other contains other family members , which are involved in cell adhesion and numerous other processes . Choanoflagellate and Filasterean genomes each contain both a gk enzyme and a GKPID , with the latter proteins occupying a well-supported basal position sister to the metazoan Dlg-containing clade . This topology indicates that the gene that came to code for GKPIDs was generated by duplication of an ancient gk enzyme before the last common ancestor of Filozoa ( animals+choanoflagellates+Filasterea ) and after the split of Filozoa from the lineage leading to fungi , which contain no GKPIDs . Further gene duplications within the animals produced the diverse proteins that now contain GKPIDs ( Figure 1B , ref . [de Mendoza et al . , 2010] ) . To understand how and when the scaffolding and spindle-orienting functions of the GKPID evolved , we used maximum likelihood ( ML ) phylogenetics to reconstruct ancestral sequences at critical nodes on the tree . We focused on two key ancestral proteins: Anc-GK1PID , which represents the single GKPIDs from which GKPIDs in all Filozoan taxa descend ) and its progenitor , Anc-gkdup , which existed just before the gene duplication that split the gk enzymes from the GKPIDs . Most sites in Anc-gkdup were reconstructed with high confidence ( mean posterior probability per site 0 . 94 , with only 20 ambiguously reconstructed sites , defined as having a second plausible reconstruction with PP > 0 . 20 ) ; Anc-GK1PID was reconstructed with lower confidence ( mean PP = 0 . 77 , and 51 ambiguous sites , see Figure 1—figure supplement 2 and Figure 1—source data 2 and 3 ) To determine when GKPID’s functions evolved , we synthesized DNAs coding for the ancestral protein sequences , expressed and purified the proteins in cultured cells , and characterized their functions by 1 ) measuring guanylate kinase activity in vitro using a coupled enzyme assay , 2 ) assessing affinity for a labeled Pins peptide using a fluorescence anisotropy assay , and 3 ) characterizing spindle-orienting function using an assay of mitotic spindle geometry in cultured cells transfected with a GK domain of interest . In the latter assay , Drosophila S2 cells in which native Dlg is knocked down were transfected with Pins fused to the cell-adhesion protein Echinoid , which localizes a crescent of Pins to the area of contact between adjacent cells; if and only if a functional GKPID is cotransfected will the spindle align during mitosis at a right angle to the crescent , along the axis between the two cells ( Johnston et al . , 2009 ) . We found that Anc-gkdup is an active guanylate kinase enzyme , with a Michaelis constant ( KM ) comparable to that of the human enzyme , albeit with a slower kcat ( Figure 2A ) . It displays no measurable Pins binding and failed to orient the mitotic spindle in living cells ( Figure 2B–E ) . These data indicate that enzyme activity is , as predicted , the ancestral function of the family; further , the scaffolding functions associated with spindle orientation were not yet present , even in suboptimal form , when duplication of the gk enzyme gene gave rise to the locus leading to GKPIDs . 10 . 7554/eLife . 10147 . 009Figure 2 . Evolution of a novel spindle-orientation function in ancestral GKPID . ( A ) Anc-gkdup ( circles ) is an active nucleotide kinase in a coupled enzyme assay for the reaction shown; Anc-GK1PID ( boxes ) is inactive . Activity of the human gk enzyme ( triangles ) is shown for reference . Error bars show SEM for three replicates . ( B ) The more recent ancestral protein Anc-GK1PID ( boxes ) binds a 20 amino-acid peptide ( see methods ) from the Pins protein in a fluorescent anisotropy assay , but Anc-gkdup ( cirlcles ) does not . Pins binding by the GKPID of the Drosophila melanogaster Dlg protein ( triangles ) is shown for reference . Error bars show SEM for three replicates . ( C–F ) Evolution of spindle orientation function as assayed in cultured S2 cells that do not express endogenous Dlg protein . Cells were transfected with a GK construct ( C , –control: empty transfection vector; D , + control: GKPID from extant Drosophila Dlg ) and scored for alignment of the mitotic spindle ( red , tubulin , visualized immunocytochemically ) relative to the Pins cortical crescent ( green , a GFP-tagged Pins-Ecd fusion ) . In the example images for each experiment , two cells are shown , the bottom one of which is dividing . The angle of the mitotic spindle relative to a line bisecting the Pins crescent ( from 0° , precisely aligned , to 90° ) was recorded in many dividing cells; the radial histogram ( right ) shows the distribution of observed angles among all cells scored with a given genotype . Cells transfected with Anc-gkdup ( E ) do not display robust spindle orientation , but those transfected with Anc-GK1PID do ( F ) . SEM: Standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 00910 . 7554/eLife . 10147 . 010Figure 2—figure supplement 1 . Properties of ancestral protein AncGK2-PID . ( A ) Distribution over sites of posterior probabilities of ML amino acid states in AncGK2PID . Mean posterior probability = 0 . 87 . ( B ) Anc-GK2PID ( squares ) binds the Pins peptide ligand with high affinity in a fluorescence anisotropy assay . Binding by the Drosophila melanogaster Dlg GKPID is shown for comparison ( circles ) . Error bars who standard error of the mean of three replicates . ( C ) Anc-GK2PID is capable of orienting the mitotic spindle . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 01010 . 7554/eLife . 10147 . 011Figure 2—figure supplement 2 . Robustness of functional inferences about ancestral proteins to uncertainty about the sequence reconstruction . For each of Anc-gkdup and Anc-GK1PID , an alternate reconstruction ( Alt-All ) was synthesized , containing the next-best amino acid state at all sites with multiple plausible states ( ( PP>0 . 2 ) . The enzyme activity of Anc-gkdup in a coupled enzymatic assay for cofactor turnover ( panel A ) and the Pins-binding activity of Anc_GK1PID ( panel B ) in a fluorescence anisotropy assay are shown for both maximum likelihood and Alt-All reconstructions . Affinities and maximal velocities differ quantitatively , but the presence/absence of each property is robust to incorporation of uncertainty about the ancestral sequence ( see Figure 1F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 011 By the time of the Filozoan ancestor , however , the evolving GKPID had lost the ancestral enzyme activity entirely and gained de novo spindle-orienting functions . Specifically , we found that Anc-GK1PID has no detectable guanylate kinase activity , but it binds Pins with moderate affinity and is highly effective in orienting the mitotic spindle in cell culture ( Figure 2A , B , F ) . We also reconstructed Anc-GK2PID – the more recent progenitor of all Dlg proteins in the ancestral animal – and found that it too orients the mitotic spindle and binds Pins with even higher affinity , suggesting a subsequent fine-tuning of Pins-binding capacity ( Figure 2—figure supplement 1 ) . It is unlikely that the ML ancestral reconstruction is precisely correct at all sites , so it is important to determine whether our conclusions about the functions of Anc-gkdup and Anc-GK1PID are robust to uncertainty about their inferred sequences . We therefore constructed an alternative version of each ancestral protein , in which all plausible alternative amino acid states ( defined as those with posterior probability > 0 . 20 ) were introduced at once . These ‘Alt-All’ sequences represent the far edge of the cloud of plausible ancestral sequences , and they contain more differences from the ML reconstruction than the expected number of errors in the ML sequence ( Figure 1—figure supplement 2 ) . They therefore represent a conservative test of functional robustness to statistical uncertainty about the ancestral sequence . When assayed experimentally , the alternative version of Anc-gkdup , like the ML reconstruction , was an active gk enzyme that did not bind Pins , and the alternative version of Anc-GK1PID bound Pins , as did the ML sequence ( Figure 2—figure supplement 2 ) . These results indicate that both the inferred trajectory of functional evolution and the phylogenetic interval during which protein scaffolding activity first evolved are robust to statistical uncertainty about the precise ancestral sequences . Taken together , these findings indicate that the capacity of the GKPID to bind Pins and orient the mitotic spindle arose well before the evolution of animals or multicellularity itself . Indeed , these functions arose even before the divergence of the choanoflagellate and filasterean lineages and before the subsequent gene duplication that gave rise to the modern subgroups of GKPID-containing proteins , including Dlgs . The capacity of GKPIDs to carry out these functions therefore evolved before the evolution of organized multicellularity itself , because filastereans do not form organized colonies ( although cells from some species do aggregate [Sebé-Pedrós et al . , 2013] ) . This result is consistent with the hypothesis that the emergence of the GKPID’s derived functions in protein scaffolding contributed to the subsequent evolution of animal complexity . The evolution of GKPID’s peptide-binding functions could have conferred spindle orientation only if its binding partners were already present . To understand when the other key components of the spindle orientation complex evolved , we characterized the taxonomic distribution of orthologs of metazoan KHC-73 and Pins by searching protein sequence databases and then inferring the age of each protein by parsimony-based inference on the taxonomic tree of life . We found that KHC-73 orthologs are clearly present in animals , choanoflagellates , and Filasterea; fungal genomes do not contain a convincing ortholog of KHC-73 but do contain a closely related paralogous member of the kinesin-3 family ( Figure 3 , Figure 3—source data 1 ) . This observation indicates that the KHC-73 gene is as old as the Filozoan ancestor and originated during the same phylogenetic interval in which Anc-GK1PID evolved its novel functions . Pins orthologs have a more recent origin . The genome of Salpingoeca rosetta , a choanoflagellate that forms spatially organized spherical colonies ( Dayel et al . , 2011 ) , contains an ortholog of Pins , which is very similar in sequence and domain architecture to its metazoan orthologs ( Figure 3 , Figure 3—source data 1 , Figure 3—source data 2 ) . A Pins ortholog is also present in another choanoflagellate , Monosiga brevicollis , but none was detected in Fungi , Filasterea , or any lineages outside of choanoflagellates and animals . In Filasterea , for example , the most similar protein to D . melanogaster Pins has a different domain structure , has detectable sequence similarity in only one small portion of the protein , and , when used as a query in a reciprocal search of the D . melanogaster genome , returns a protein from an entirely different family as the best hit with significance score 10 orders of magnitude better than its match to Pins ( Figure 3—source data 1 ) . Indeed , a key domain of Pins – the GoLoco domain , a simple 23-amino acid sequence that mediates Pins’ contact with membrane-associated G-proteins and is therefore crucial to its cortical localization ( Smith and Prehoda , 2011 ) – exists only in animals and choanoflagellates . This observation establishes that Pins evolved before the split of choanoflagellates and animals and suggests that it evolved after a functional GKPID emerged in the Filozoan stem lineage . The association of GKPID with Pins in the spindle orientation complex therefore evolved through a stepwise process of molecular exploitation ( Bridgham , 2006 ) – a form of exaptation in which a newly evolved molecule recruits as a binding partner a more ancient molecule , which previously had different functions and a fortuitous but unrealized affinity for its new partner . In this case , GKPID's affinity for Pins apparently existed before the Pins protein came into existence . What could the functions of GKPID and KHC-73 have been before Pins evolved in the lineage leading to choanoflagellates and animals ? GKPID and KHC-73 may have associated with each other to orient the mitotic spindle relative to a cellular mark other than Pins . Alternatively , GKPID may have functioned in other processes executed by its descendants , such as cell adhesion; KHC-73 also has numerous functions as an intracellular motor protein that are independent of spindle orientation ( Hanada , 2000; Horiguchi , 2006 ) . In either case , when a suitable form of Pins evolved considerably later , GKPID already had the capacity to bind it and thus form a scaffold for spindle orientation . The experiments reported above establish when GKPID evolved its capacity to bind extant animal Pins protein and to organize spindle orientation in extant animal cells; they do not reveal when a Pins protein that could bind GKPID first evolved . Pins is too poorly conserved for its ancestral form to be reconstructed with confidence . We therefore evaluated this question indirectly by investigating whether choanoflagellate GKPID binds choanoflagellate Pins , as expected if the association of Pins and GKPID and its role in spindle orientation predate the metazoan-choanoflagellate ancestor . We purified the GKPID of the S . rosetta Dlg-like protein and measured its affinity for the linker peptide of the Pins protein – the region to which GKPID binds ( Johnston et al . , 2009 ) – from S . rosetta and from metazoans . 10 . 7554/eLife . 10147 . 012Figure 3 . Taxonomic distribution of proteins in the spindle orientation complex . ( A ) The genomes of the choanoflagellate Salpingoeca rosetta and the filasterean Capsaspora owczarzaki contain orthologs of human Dlg , Pins , and KHC-73 . The domain architecture of each protein is shown , as inferred using the SMART database . Each group of proteins are reciprocal best BLAST hits ( RBH ) to the human query protein shown . For details , see Figure 3—source data 1 . ( B ) Aligned sequences from the linker portion of Pins ( see panel A ) , which binds to Dlg . Colors highlight identical or biochemically conservative residues . Asterisk , phosphorylated or negatively charged residue 436 , which in the Drosophila melanogaster Pins protein anchors Dlg binding . For complete species names and accessions , see Figure 3—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 01210 . 7554/eLife . 10147 . 013Figure 3—source data 1 . Results of reciprocal Blast search of metazoan and nonmetazoan genomes for orthologs of GKPID , Pins and KHC-73 . Each row shows , for a query sequence from Drosophila melanogaster , the accession and e-score for the best-hit sequence in the target species , as well as the reciprocal best-hit and e-score when the target best-hit is used as a query against D . melanogaster . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 01310 . 7554/eLife . 10147 . 014Figure 3—source data 2 . Identifiers and species for Pins sequences shown in Figure 3B . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 014 We found that S . rosetta GKPID binds the D . melanogaster Pins linker peptide with moderate affinity ( Figure 4A ) , corroborating our finding that GKPID’s capacity to bind metazoan Pins originated before the common ancestor of animals and choanoflagellates . In contrast , S . rosetta GKPID does not detectably bind the S . rosetta Pins linker: in a simple pull-down assay using a glutathione-S-transferase/Pins fusion protein and His-tagged GKPID , we detected no association of the two S . rosetta proteins under conditions in which the orthologous pair of proteins from D . melanongaster shows robust binding , even though both S . rosetta proteins are expressed and soluble ( Figure 4B ) . That S . rosetta GKPID can bind the fruitfly’s Pins but not its own suggests that Pins evolved its capacity to bind GKPID after animals diverged from choanoflagellates . We cannot rule out the less parsimonious possibilities that Pins lost an ancient capacity to bind GKPID in choanoflagellates or that some unique and unknown mode of association between GKPID and Pins operates in S . rosetta , such as requiring a bridging protein or some post-translational modification . Still , our experiments indicate that the mechanism of GKPID’s association with Pins is not conserved between animals and choanoflagellates and may have evolved in the animal lineage after its divergence from choanoflagellates . Why S . rosetta GKPID can bind the Drosophila Pins linker but not its own is unclear; one possibility is that the surface of GKPID that fortuitously binds Pins is conserved in S . rosetta because it binds another structurally similar ligand , possibly an ancient one . 10 . 7554/eLife . 10147 . 015Figure 4 . Spindle orientation and its molecular components in choanoflagellates . ( A ) Purified GKPID from the Salpingoeca rosetta Dlg ortholog ( triangles ) binds fluorescently labeled Drosophila melanogaster Pins . Binding by Anc-GK1PID ( squares ) and lack of binding by Anc-gkdup ( circles ) are shown for comparison . Error bars , SEM for three replicates . ( B ) Purified GKPID of S . rosetta does not interact with Pins of S . rosetta; under the same conditions , the D . melanogaster orthologs of these proteins do bind . A GST-fusion of each Pins linker regions ( visible on the coomassie stained gel ) was incubated with an MBP-fusion of each species’ GKPID and detected by anti-MBP western blot . As previously found , binding of D . melanogaster Pins requires phosphorylation by Aurora A kinase; S . rosetta GKPID does not bind its own Pins whether or not the kinase is present . ( C , D ) Spindle orientation in colonial ( panel C , n = 12 ) and noncolonial ( panel D , n = 7 ) S . rosetta . For each condition , the image shows one representative colony; the cell at bottom center is mitotic , as evidenced by condensed DNA ( blue , DAPI ) without a defined nuclear envelope ( green , visualized using anti-nuclear pore complex ) . Red , mitotic spindle visualized using anti-tubulin . In colonial cells , the angle of the mitotic spindle ( solid white line ) was measured relative to a line perpendicular to the plane of the colony extending through the colony’s center ( dashed line ) . The histogram shows the distribution of spindle angles among all dividing S . rosetta cells measured , with 90° representing perfect alignment relative to the colony ring . In noncolonial cells , the spindle angle was measured relative to a line extending from the midpoint between the bases of the two duplicate flagella to the cell body’s centroid ( dotted line ) . An angle of 90° represents perfect alignment relative to the flagella . SEM: Standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 015 How does the history of the GKPID-Pins complex relate to the evolution of spindle orientation ? Spindle orientation itself is not a metazoan novelty; indeed , most eukaryotes – even unicellular ones – orient the mitotic spindle , but most appear to do so relative to the cell’s internal structure alone rather than in response to cues from adjacent cells ( Wang et al . , 2003 ) . We visualized the mitotic spindle in colonial S . rosetta using fluorescence microscopy and found that dividing S . rosetta cells appear to orient their mitotic spindles relative to the surface of the colony in a way that maintains the colony’s spherical geometry ( Figure 4C ) . If GKPID and Pins do not associate in choanoflagellates , how do colonial cells accomplish this ? S . rosetta cells form colonies in a flagellum-out fashion , so orienting the spindle relative to the colony surface also entails orienting it at a right angle to the axis of the flagellum and , more generally , to the apical-basal ( A-B ) axis of the cell . We found that individual dividing S . rosetta cells that are not organized into colonies also orient their spindles relative to the flagellum and the A-B axis ( Figure 4D ) . Because spindle orientation relative to the cellular axis occurs in both colonial and noncolonial S . rosetta cells , spindle orientation is likely to involve internal marks imposed by the cell’s polarity rather than cues from neighboring cells . Taken together , these findings suggest that spindle orientation mediated by the GKPID-Pins complex in response to external cues replaced a more ancient mode of spindle orientation along the stem lineage leading to animals . This more ancient mechanism may have involved the flagellar basal bodies . The deeper choanoflagellate-metazoan ancestor almost certainly had flagella ( Nielsen , 2008; Buss , 1988 ) . Previous research indicates that during choanoflagellate mitosis , the flagellar basal body duplicates; the daughter bodies migrate symmetrically away from the cell’s basal pole and then serve as microtubule organizing centers for assembling the spindle and orienting it perpendicular to the A-B axis ( Buss , 1988; Leadbeater , 2015 ) . A parsimonious hypothesis is therefore that spindle orientation in the choanoflagellate-animal ancestor was oriented relative to the A-B axis via the position of the flagellar basal bodies ( Buss , 1988 ) . This ancient mechanism would have been retained in choanoflagellates , allowing organized cell division in the context of flagellum-out spherical colonies . Along the branch leading to the metazoan ancestor , basal-body mediated orientation would have been replaced by the GKPID-Pins association , providing a mechanism for the spindle to be oriented relative to adjacent cells via an externally organized molecular mark as the flagellum was lost from many animal cell types . We next sought to identify the genetic mechanism by which the GKPID evolved its capacity to bind Pins and thus serve as a localized scaffold for spindle orientation . The potential candidate mutations are those that occurred during the phylogenetic interval between Anc-gkdup and Anc-GK1PID – the same branch on which the transition to GKPID function took place . Seventy-one amino acid replacements occurred along this branch , making identification of the functionally important changes an apparent challenge . To identify potentially causal substitutions from this large set of candidates , we used both structure-function information and the phylogenetic pattern of sequence conservation/divergence within and between gk enzymes and GKPIDs . Extant gk enzymes contain two nucleotide binding lobes connected by a flexible hinge region around a central catalytic core ( Figure 5A ) . In crystal structures of the gk enzyme in the absence of nucleotide substrate , the binding lobes are separated from each other in an open conformation ( Blaszczyk et al . , 2001 ) . Upon nucleotide binding , the lobes move inward and occupy a closed conformation , bringing GMP and its co-substrate ATP together and allowing catalysis to occur ( Sekulic et al . , 2002 ) . In contrast , the GKPID remains constitutively in the open conformation , and Pins – which is considerably larger than the enzyme’s nucleotide ligands – binds to the exposed surface of the guanylate-binding lobe ( Figure 5A ) ( Johnston et al . , 2012 ) . We reasoned that the substitutions that caused the functional transition from enzyme to scaffold might have affected residues in the GMP/Pins binding interface or , alternatively , in the hinge that determines the orientation of the lobes relative to each other , thus affecting the size , geometry , or accessibility of the ligand-binding cleft . 10 . 7554/eLife . 10147 . 016Figure 5 . Evolution of the binding interface and hinge during the evolution of AncGKPID spindle orientation functions . ( A ) All gk/GKPID family members share a common structural architecture , comprising a catalytic core , two binding lobes ( the GMP-binding domain , GBD , shown in dark hue , and the ATP-binding lid ) , and a flexible hinge region , which connects the GBD to the core and comprises two segments of contiguous residues ( magenta ) . Left: in gk enzymes bound to GMP ( red spheres ) , the lobes adopt a closed conformation , bringing GMP and ATP ( cyan spheres ) adjacent to each other in the core . Right: the GKPID has an open conformation , in which Pins ( green spheres ) binds to the surface of the GBD in the cleft between the two binding lobes . Structures shown are mouse gk enzyme ( brown , PDB 1LVG ) and the GKPID from rat Dlg1 ( blue , 3UAT ) . ( B ) Most residues in Anc-GK1PID that bind Pins ( blue boxes ) are unchanged from the ancestral state in Anc-gkdup . White surface , D . melanogaster Dlg GK1PID ( 3TVT ) . Green , Pins peptide . Ancestral and derived amino acid states are shown; residues with historical amino acid replacements between the two ancestral proteins are outlined . ( C ) In the hinge region , two historical substitutions ( outlined and colored , with side-chains shown as sticks ) were conserved in the ancestral state in extant enzymes and in a different state in extant GKPIDs . Colored bars above the sequence indicate position in the protein structure ( right ) . Hinge segments are shown in pink and the GMP-binding lobe in gray . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 016 Of the amino acid changes that occurred in these regions of the protein during the interval between Anc-gkdup and Anc-GK1PID , only five are conserved among descendant GKPIDs ( Figure 5B , C ) . To test these substitutions’ functional importance , we introduced the derived states individually into Anc-gkdup and characterized their effects on guanylate kinase activity , Pins binding , and spindle orientation . Remarkably , we found that either of two amino acid changes in the hinge is sufficient to confer the protein-binding function . Substitution s36P , located where the hinge meets the binding lobe , virtually abolished the catalytic activity of Anc-gkdup and established moderate-affinity Pins binding ( Figure 6A , B; lower and upper case residue symbols denote ancestral and derived states , respectively ) . In cultured cells , this single substitution also gave Anc-gkdup the capacity to robustly mediate spindle orientation ( Figure 6C ) . Substitution f33S , also in the hinge , had largely similar effects , conferring Pins binding and decreasing – but not abolishing – enzyme activity; the affinity of Anc-gkdup+f33S is almost identical to that of AncGKdup+s36P , suggesting that f33S may also confer spindle orientation capacity , although this possibility was not directly tested . Combining f33S with s36P did not further shift the protein's function beyond that caused by either change alone ( Figure 6A ) . In contrast , the three substitutions in the binding interface ( s34C , a73G and f75Y ) caused minor reductions in enzyme activity but did not confer even moderate Pins-binding ( Figure 6—figure supplement 1 ) . 10 . 7554/eLife . 10147 . 017Figure 6 . A simple genetic basis for the evolution of spindle orientation functions in the ancestral GKPID . ( A , B ) Introducing either of two historical substitutions s36P or f33S into Anc-gkdup confers Pins binding and reduces ancestral guanylate kinase activity . Error bars are SEM of 3 replicates . Combining s36P and f33S does not further affect the derived functions beyond the effects of the single substitutions . ( C ) Introducing s36P into Anc-gkdup is sufficient to confer the full capacity to drive orientation of the mitotic spindle ( compare Figure 1 ) . As shown in Figure 6—figure supplement 1 , historical substitutions in the binding interface do not recapitulate the evolution of Pins binding and the loss of gk activity . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 01710 . 7554/eLife . 10147 . 018Figure 6—figure supplement 1 . Introducing other historical substitutions into AncGKdu does not confer GKPID-like function . ( A ) The hinge substitution s34C ( purple ) does not abolish guanylate kinase activity in a coupled enzyme activity assay; Anc-gkdup is shown for comparison . ( B ) s34C ( purple ) does not confer substantial Pins-binding affinity on Anc-gkdup . Pins-binding activity relative to s36P ( orange ) in a fluorescence anisotropy assay . ( C , D ) Anc-GK1PID Pins interface substitutions a73G and f75Y reduce gunaylate kinase enzyme activity ( D ) but do not confer Pins binding ( E ) when introduced into Anc-gkdup . Error bars are standard error of the mean with three replicates . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 018 The genetic basis for the evolutionary origin of the GKPID’s spindle-orienting functions therefore appears to have been very simple . A single substitution – either s36P or f33S – was sufficient to recapitulate the evolution of the new functions that emerged during the evolution of the GKPID and to completely or partially abolish the ancestral function . The historical order of these amino acid changes is unknown . If s36P occurred first , it would have conferred on the protein all the major aspects of the functional transition , and f33S would have been subsequently inconsequential . If f33S occurred first , however , our data suggest that it would have generated a functionally hybrid intermediate that both bound Pins and retained some enzyme activity; s36P would then have completed the functional transition by abolishing the residual enzyme activity . The two causal historical substitutions that we identified are both reconstructed without any statistical ambiguity at the relevant nodes . They are not identical to the set of mutations found in a previous study to be important to GKPID functions based on a horizontal comparison and mutagenesis of extant proteins from D . melanogaster and S . cerevisiae ( Johnston et al . , 2011 ) . Replacement s36P was found in both studies , but the historical replacement f33S was not identified in the horizontal comparison . The other mutation identified in the horizontal comparison ( S34P ) never occurred during history , and we found that the historical amino acid replacement that did take place at this site ( s34C ) does not confer GKPID-like functions ( Figure 6—figure supplement 1 ) . How could single amino acid changes have caused an entirely new function to evolve ? Residue 33 does not contact the ligand , and residue 36 makes only a minor and apparently nonspecific contact to Pins at one end of the binding site ( Figure 7—figure supplement 1 ) . Overall , the residues that compose the Pins-binding surface are almost entirely conserved from Anc-gkdup to Anc-GK1PID ( Figures 5B , 7A , B ) , and the backbone structure of the binding lobe is almost identical between the crystal structures of extant GKPIDs , gk apo-enzymes , and a gk apo-enzyme containing mutation S36P ( Figure 7B , see also ref . Johnston , et al . , 2011 ) . Thus , the Pins binding surface appears to have been present , even before the interaction with Pins itself evolved . 10 . 7554/eLife . 10147 . 019Figure 7 . Evolution of GKPID’s new function by unveiling a latent protein-binding site . ( A ) The binding surface for Pins in GKPIDs is derived from the GMP-binding surface of gk enzymes . Homology models of Anc-gkdup ( left ) and Anc-GK1PID ( right ) are shown as white surface , with all side chains that contact either GMP or Pins as yellow sticks . Pink sticks show GMP; green ribbon shows Pins backbone , with the side chains of all Pins residues that contact the GK protein shown as sticks . The phosphate group on GMP and on Pins residue 436 are shown as orange and red sticks . Black dotted lines , protein-ligand hydrogen bonds . In the AncGK1PID structure , substitutions at sites in the binding interface are shaded red , including key substitution s36P . The binding modes of extant gk enzymes and GKPIDs are similar and support the same conclusions ( see Figure 7—figure supplement 1 ) . ( B ) The structure of the hinge and GMP/Pins-binding lobes is conserved between the Pins-bound GKPID ( blue , rat Dlg , 3UAT ) , the apo-gk enzyme ( brown , S . cerevisiae guanylate kinase 1EX6 ) , and the apo-gk-s36P mutant ( gray , 4F4J ) , all in the open conformation . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 01910 . 7554/eLife . 10147 . 020Figure 7—figure supplement 1 . Structural context of key historical mutations . ( A ) Location of historical hinge substitutions s36P and f33S . Residues Pro36 ( orange ) and Ser33 ( purple ) are shown on the crystal structure of the GKPID from rat Dlg ( 3UAT , white surface ) . Pins peptide ligand is shown in green . ( B , C ) Similarity of GMP binding site in extant guanylate kinase enzyme to Pins binding site in extant GKPID . The guanylate binding domains ( GBDs ) of yeast gk enzyme ( PDB 1GKY , panel B ) and of rat Dlg GKPID ( PDB 3UAT , panel C ) is shown as white surface , with all side chains that contact either GMP or Pins as yellow sticks . Pink sticks show GMP; green ribbon shows Pins backbone , with the side chains of all Pins residues that contact the GKPID protein shown as sticks . The phosphate group on GMP and on Pins residue S436 are indicated . Black dotted lines , protein-ligand hydrogen bonds . Key substitution s36P is highlighted in pink . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 020 Why would a latent Pins binding surface have been present ? Homology models of the ancestral proteins , as well as the structures of extant family members , indicate that the key portion of the Pins-binding surface of GKPID was derived without significant modification from the ancient surface that gk enzymes use to bind GMP ( Figure 7A , B , Figure 7—figure supplement 1 ) . This GMP-binding site could be repurposed for binding Pins because the two ligands – one a nucleotide , the other a peptide -- share a key structural feature: a negatively charged head flanked by a small hydrophobic region ( Figure 7A , B , ref . Johnston , et al . , 2012 ) . Specifically , the phosphate group at the head of GMP is anchored to the enzyme with hydrogen bonds to four clustered residues that form a positively charged pocket ( R40 , R43 , Y80 , Y82 ) ; in the GKPID , these same residues form hydrogen bonds to the phosphate group of Pins’ phospho-serine 436 . At the other end of GMP , the hydrophobic guanosine ring occupies a hydrophobic groove in the enzyme , and the same groove on the GKPID binds to a hydrophobic methionine side chain on Pins . Additional interactions beyond this common interface are required for Pins binding ( Zhu et al . , 2011 ) , but the physical interactions are relatively simple , including a series of small hydrophobic patches that bind to hydrophobic side chains on Pins and just two additional hydrogen bonds , both from backbone atoms on Pins to polar residues on the GKPID surface that were solvent-exposed in the enzyme . Thus , the fortuitous similarity between GMP and one portion of Pins , along with the fortuitous arrangement of hydrophobic patches near the ancestral enzyme’s GMP-binding site , made it possible for GKPID to evolve Pins binding by a very minor genetic modification . If the binding surface for Pins was already present in latent form in the Anc-gkdup enzyme , how did s36P or f33S confer Pins binding ? Several lines of evidence suggest that these substitutions altered the protein’s dynamics and/or increased the relative occupancy of a conformation in which the latent binding site is exposed for peptide binding . First , the hinge where these two residues are located ( Figures 5C , 7B ) is known to mediate the dynamic opening/closing of the binding lobes relative to each other ( Blaszczyk et al . , 2001 ) . Second , the degree to which the domain is open or closed appears to be essential for function . In gk enzymes , closing is critical for catalysis , because it brings the nucleotide substrates close together in the protected active site ( Bhabha et al . , 2013 ) ; in GKPIDs , however , open conformations appear to be required for Pins binding , because the Pins peptide is significantly larger and is therefore predicted to sterically clash with the GKPID when the lobes are close together . Third , substituting a proline at position 36 could restrict backbone dihedral angles in the hinge , possibly altering the dynamics of the hinge and/or the distribution of conformations it occupies . Replacing the bulky phenylalanine at position 36 could also change backbone dynamics , altering the occupancy of conformations in this region . Fourth , introducing a proline at residue 36 into extant gk enzymes has been shown to impede the GMP-induced closing motion , abolish enzyme activity , and to confer Pins binding ( Johnston et al . , 2011 ) . Because the effects of mutation s36P on the function of the ancestral gk enzyme are nearly identical to those it has on the extant enzyme , it is likely that similar biophysical mechanisms pertain in the two proteins . Fifth , engineering yet a third mutation in the hinge – a serine-to-proline replacement at site 34 – into extant gk enzymes has been previously found to reduce enzyme activity , confer spindle orientation , and impede conformational closing , further supporting a causal link between these phenomena ( Johnston et al . , 2011 ) . We therefore propose that the historical hinge substitutions s36P or f33S inhibited enzyme activity and conferred the novel scaffolding function on GKPID by restricting the ancestral enzyme’s dynamic hinge motions and/or changing its occupancy of conformations that have high affinity for Pins . This scenario is consistent with recent findings that changes in conformational occupancy may play an important role in the evolution of protein function ( Harms et al . , 2013; Tokuriki and Tawfik , 2009 ) . Subsequent substitutions apparently fine-tuned Pins-binding affinity , yielding the higher-affinity Anc-GK1PID and Anc-GK2PID . In addition to its likely effects on conformational occupancy and/or dynamics , Pro36 makes van der Waals contact to the Pins ligand ( Figure 7—figure supplement 1 ) , so it is possible that s36P may also have contributed to optimizing the latent Pins binding interface itself; in contrast , f33S does not contact the ligand and is therefore unlikely to have directly affected the interface . A central issue in evolutionary biology is how complex systems originate through the action of mutation , drift , and natural selection . Tissue organization , spindle orientation , and the GKPID complex itself are all examples of complexity , defined as the integrated functioning of a system made up of differentiated , interacting parts . The GKPID complex can orient the mitotic spindle because of specific interactions between its component molecules and with other molecules in the cell and its local environment . In turn , the cellular phenomonenon mediated by this complex – regular orientation of the plane of cell division relative adjacent cells – allows the development and maintenance of organized , differentiated tissues , and this phenomenon in turn makes possible a higher level of biological complexity– the multicellular organism – from a collection of individual cells . Understanding the evolution of complexity at the molecular level can therefore help to illuminate the evolution of macroscopic complexity , including functions that are now crucial to animal biology per se . Our work indicates that the GKPID complex was assembled stepwise through a process of molecular exploitation , in which old molecules with one function are recruited into a functional binding interaction with a newly evolved molecule . In this case , the GKPID , a duplicate of an ancient enzyme with an essential metabolic role in all life forms , already had the fortuitous capacity to bind the Pins protein , even before the latter protein appeared or subsequently acquired its relatively simple GKPID-binding linker motif ( Figure 8 ) . Once Pins did evolve this linker — along with its GoLoco motif , which interacts with G-protein complexes , which are also ancient ( de Mendoza et al . , 2014 ) — then a mechanism would have been assembled that could bring to specific locations on the cell cortex the GKPID and other proteins associated directly or indirectly with it , such as KHC-73 and astral microtubules , thus enabling externally-cued spindle orientation . Our analyses do not establish a complete history of the spindle orientation complex . Many key steps remain to be reconstructed , including how and when the interaction between GKPID and KHC-73 evolved , the mechanisms by which Pins’ acquired its linker and GoLoco sequences , and the relationship of these components to other molecular complexes and pathways involved in animal spindle orientation . Despite these knowledge gaps , our observations establish a broad overview of the history of the GKPID complex , provide a detailed mechanistic reconstruction of a key event , and point to the importance of reusing molecules – and specific surfaces within them – for fortuitous new purposes that have the potential to become biologically essential . 10 . 7554/eLife . 10147 . 021Figure 8 . Historical evolution of GKPID-mediated spindle orientation complex . The center portion shows the phylogeny of Metazoa and closely related taxa . The origin of cell differentiation and spatially organized tissues is marked . The left portion shows major events in the evolution of the components of the spindle orientation complex reconstructed in this study . Duplication of an ancestral gk enzyme ( brown ) and the key mutations that led to the origin of a GKPID ( blue ) that could bind other molecules in the complex are shown relative to the phylogeny’s time scale . The apparent date of origin of KHC-73 ( black ) and Pins ( green ) are also shown . Dotted green line shows the origin of Pins in a form not yet bound by GKPID . Solid green line shows GKPID-binding form . Horizontal lines indicate binding between proteins . The right portion shows a schematic of the spindle orientation machinery in metazoans , which allows orientation relative to external cues from nearby cells , as well as spindle orientation relative to the internal cell axis as marked by the flagella in both solitary and colonial choanoflagellates . DOI: http://dx . doi . org/10 . 7554/eLife . 10147 . 021 Other reported cases of molecular exploitation have involved recruiting new binding partners that are subtle variants of its parent’s ligand – such as steroid hormones with a modified functional group at a key position or a minor change in the hormone’s structure ( Bridgham , 2006; Harms et al . , 2013; Carroll et al . , 2008 ) . In contrast , GKPID acquired an entirely new function – from enzyme ancestor to protein-binding scaffold – and affinity for an entirely different class of macromolecule – from nucleotide to peptide , and it did so through as little as one historical change in amino acid sequence . The genetic simplicity of the evolutionary change in GKPID function is underscored by the fact that we found not one but two historical amino acid replacements from the relevant phylogenetic interval , either of which is sufficient to confer the GKPID’s derived functions on the ancestral enzyme . This finding indicates that GK acquired its new protein-binding function through a relatively simple , high-probability genetic path , rather than a long trajectory that required many specific mutations before the new function could be established . GKPID’s dramatic evolutionary transition in function could take place through such a simple genetic mechanism because of its biophysical architecture . The gk enzyme’s simple binding site for GMP can also be occupied by a simple two-residue motif on the Pins peptide , which fortuitously has similar surface properties . In addition , a series of small hydrophobic patches , which happen to be adjacent , was available to bind the hydrophobic portion of the Pins peptide and increase affinity . All that was required to confer the protein's new function was a single mutation that revealed this molecular surface , apparently by changing the protein’s conformational flexibility . In this way , the physical simplicity of an interaction between ancient molecules set the stage for the easy evolution of a novel molecular complex and , in turn , a cellular function that now plays an important role in the complex biology of multicellular animals . Annotated protein sequences of 224 guanylate kinases and GKPIDs were downloaded from UniPROTKB/TrEMBL , GenBank , the JGI genome browser , and Ensemble databases . Amino acid sequences were aligned using MUSCLE ( Edgar , 2004 ) , followed by manual curation and removal of lineage-specific indels . For species and accessions used , see Figure 1—source data 1 . Guanylate kinase sequences were trimmed to include only the active gk domain predicted by the Simple Modular Architecture Research Tool ( SMART ) ( Schultz et al . , 1998 ) . The phylogeny was inferred by ML using PhyML v2 . 4 . 5 ( Guindon et al . , 2010 ) and the WAG model with gamma-distributed rate variation and empirical state frequencies , which was selected using ProtTest software and the AIC criterion . Statistical support for each node was evaluated by obtaining the approximate likelihood ratio ( the likelihood of the best tree with the node divided by the likelihood of the best tree without the node ) and the chi-squared confidence statistic derived from that ratio ( Anisimova et al . , 2011 ) . Ancestral protein sequences and their posterior probability distributions were inferred by the ML/empirical Bayes method ( Yang et al . , 1995 ) , assuming the ML phylogeny and the best-fit model , using PAML v3 . 13 and Lazarus software ( Hanson-Smith et al . , 2010 ) . Average probabilities were calculated across all GK sites except those containing indels . Plausible alternative non-ML states were defined as those with posterior probability >0 . 20 . Alternate ancestral sequences at each node were prepared containing the ML state at all unambiguously reconstructed sites and the state with the second highest posterior probability at all ambiguously reconstructed sites . Sequences of reconstructed proteins Anc-gkdup , Anc-GK1PID , and Anc-GK2PID were deposited in Genbank with identifiers KP068002 , KP068003 , and KP068004 , respectively . Homology modeling was performed with MODELLER software using the automodel single comparison class . 1GKY and 1EX6 were used as the template for Anc-gkdup and 3UAT for Anc-GK1PID . For each protein , 10 homology models were prepared , and that with the lowest DOPE score was used as the best model ( Eswar et al . , 2006 ) . Choanoflagellate and filasterean genes orthologous to metazoan guanylate kinase , Dlg , Pins , and KHC-73 proteins were identified by using the NCBI BLAST tool using Homo sapiens or Drosophila melanogaster protein sequences as queries . For each search , the top hit in the target species was verified with a reciprocal BLAST search against the query genome . Domain architecture similarity was assessed using the SMART domain recognition tool ( Schultz et al . , 1998 ) . DNA sequences coding for reconstructed ancestral proteins and optimized for E . coli codon bias were synthesized ( Genscript ) and then inserted into pBH plasmid vector with a hexa-His tag for E . coli expression at 20°C . Protein purification was carried out using sequential NiNTA affinity , anion exchange , and size-exclusion chromatographies . All proteins eluted as predicted monomers from the size-exclusion column at purity >95% by Coomassie staining of an SDS–PAGE gel . Proteins were concentrated using Vivaspin concentrators ( Sigma-Aldrich ) , flash frozen in liquid nitrogen , and stored at −80°C in buffer ( 20 mM Tris , pH 7 . 5 , 150 mM NaCl , 1 mM DTT ) . We used a coupled enzyme assay , as described previously ( Agarwal et al . , 1978 ) , which quantifies release of ADP in the guanylate kinase-catalyzed reaction by coupling it to pyruvate kinase- and lactate dehydrogenase-catalyzed reactions and measuring the consequent oxidation of NADH by following the decrease in absorbance at 340 nm , which we measured on a Tecan Safire plate reader . Guanylate kinase enzyme was at 50–200 nM in assay buffer ( 100 mM Tris , pH 7 . 5 , 100 mM KCl , 10 mM MgCl2 , 1 . 5 mM sodium phosphoenolpyruvate , 300 mM NADH , 4 mM ATP , and 100 units pyruvate kinase and 100 units lactate dehydrogenase ) . Initial GMP concentrations ranged from 500 nM to 1 mM . The reaction was initiated by adding GMP and briefly mixing . Reactions were carried out at 30°C and measured 30 times at 15s intervals . Data were analyzed and plotted using GraphPad Prism assuming Michaelis-Menten kinetics . Reaction rates are plotted as initial rate of ADP production . Binding of D . melanogaster Pins was assayed by fluorescence anisotropy on a Tecan Sapphire plate reader equipped with automatic polarizers using default settings for anisotropy assays . A NH2-terminal FITC-labeled peptide ( GVRVRRQ ( pS ) MEQLDLIKITPD , Genscript ) of the fly Pins-Linker peptide ( 0 . 25 µM ) was incubated with increasing concentrations of GK protein in phospho-buffered saline solution with 1mM DTT . A one-site binding model was used to fit the data and infer binding affinity in Graphpad Prism . Attempts to measure affinity of the S . rosetta GKPID for the S . rosetta Pins linker using MBP pull-down experiments were unsuccessful under a variety of assay conditions due to loss of the immobilized MBP-Pins fusion protein upon addition of GKPID . Ancestral Pins peptides were not reconstructed because Anc-gkdup and Anc-GK1PID both correspond to gene duplications rather than speciation events; no nodes precisely corresponding to the same genomes would exist on a Pins phylogeny . Our inferences are therefore limited to the capacity of ancestral Anc-gkdup and Anc-GK1PID to bind to Pins peptides found in extant animals . For GST pull-downs , we expressed GST fused to the 20-residue Pins linker of S . rosetta ( PRGSKTGEVEQFTIDDSDD ) or D . melanogaster ( GVRVRRQSMEQLDLIKITPD ) and MBP-tagged GKPID proteins from the same species . Proteins were expressed separately in BL21 ( DE3 ) E . coli cultures and purified with glutathione or amylose agarose , respectively . All four proteins expressed in soluble form . The GST-Pins fusion was left attached to the resin and incubated with MBP-tagged GKPID , followed by three washes . Bound GKPID was eluted in SDS-PAGE loading buffer and visualized using Western electrophoresis and an anti-MBP antibody ( Qiagen ) ; unbound GKPID from the final wash was visualized by electrophoresis and Coommassie staining . Binding reactions were performed in the absence and presence of Aurora A kinase ( Calbiochem ) , which is necessary for phosphorylation of the D . melanogaster Pins linker at residue S436 and subsequent GKPID binding ( Johnston et al . , 2009 ) . Maintenance of S2 cells , construction of expression plasmids , and cell-adhesion/spindle orientation assays were performed as detailed previously ( Johnston et al . , 2011 ) . S2 cells were transfected with constructs coding for a FLAG- or HA-tagged GK protein and for a Pins-GFP-Echinoid fusion protein , using Effectene reagent ( Qiagen , Germantown , MD ) and 0 . 4–1 μg total DNA for 24–48 hr . Endogenous Dlg was knocked down using RNAi: transfected cells were incubated for 1 hr in serum-free media containing approximately 1 μg RNAi followed by 72 hr in normal growth media . Protein expression was induced by adding 500 μM CuSO4 for 24 hr . Cell-adhesion clustering was induced by constant rotation at approximately 175 rpm for 1–3 hr . Pins fusion protein was visualized by fluorescence ( excitation 488 nm , emission 509 nm ) . Mitotic spindles were visualized using rat anti-tubulin ( Abcam 1:500 ) and goat anti-rat conjugated to Alexa 555 ( Life Technologies , 1:500 ) . GK was visualized using mouse anti-FLAG or anti-HA ( Sigma 1:500 ) and chicken anti-mouse:Alexa 647 ( Life Technologies 1:200 ) ; histones were visualized using rabbit antiphospho-histone3 ( Upstate 1:8000 ) and donkey anti-rabbit:Alexa 488 ( Life Technologies , 1:500 ) . For immunostaining , clustered cells were fixed in 4% paraformaldehyde for 20 min , washed , and incubated with primary antibodies overnight at 4°C in buffer ( 0 . 01% saponin plus 0 . 1% albumin diluted in phosphate-buffered saline ) . Slides were subsequently washed and fluorescently linked secondary antibodies were added for 2 hr at room temperature ( RT ) . Finally , slides were again washed and mounted using Vectashield Hardset medium ( Vector Laboratories ) . All images were collected using a Biorad Radiance 2100 confocal microscope with a 1 . 4NA 60× objective . Spindle angles were measured in ~20 cells for each condition and their distribution displayed in a radial histogram . Growth medium was prepared in artificial sea water . S . rosetta cultures ( ATCC 50818 ) consisting primarily of chain colonies and slow swimmers were maintained by passaging 2 ml of culture into 18 ml fresh medium every day ( King et al . , 2009 ) . Rosette colonies were produced by inoculating S . rosetta chain colonies with Algoriphagus machipongonensis bacteria ( Dayel et al . , 2011 ) . Log phase S . rosetta cells were treated with 0 . 33 micromolar Nocodazole ( Sigma M1404 ) for 18 hr at RT . Cells were pelleted by centrifugation for 5 min at 2000 x g , and washed thrice in artificial sea water to remove drug . Cells were allowed to recover for 30 , 45 , or 60 min at RT before fixation . Approximately 0 . 1 ml of cells were applied to poly-l-lysine-coated 96-well plates and left to attach for 30 min . Cells were fixed for 5 min with 0 . 2 ml 6% acetone , and then for 20 min with 0 . 2 ml 4% formaldehyde . Acetone and formaldehyde were diluted in artificial seawater , pH 8 . 0 . Wells were washed gently four times with 1 ml washing buffer ( 100 mM PIPES at pH 6 . 9 , 1 mM EGTA , and 0 . 1 mM MgSO4 ) and incubated for 30 min in 1 ml blocking buffer ( washing buffer with 1% BSA , 0 . 3% Triton X-100 ) . Cells were incubated with primary antibodies diluted in 0 . 15 ml blocking buffer for 1 hr , washed four times with 0 . 2 ml of blocking buffer , and incubated for 1 hr in the dark with fluorescent secondary antibodies ( 1:1000 in blocking buffer , Alexa Fluor 488 goat anti-mouse , and Alexa Fluor 568 goat anti-rabbit; Invitrogen ) . Wells were washed thrice with washing buffer , blocked with 0 . 2 ml DAPI at 1 μg/ml for 5 min , and washed twice more . The following primary antibodies were used: Mouse monoclonal antibody against β-tubulin ( E7 , 1:100; Developmental Studies Hybridoma Bank ) and nuclear pore complexes ( 1:100 , Covance ab24609 ) . Images were taken with a 63× oil immersion objective on a Leica DMI6000 B inverted compound microscope and Leica DFC350 FX camera . The spindle angle for individual , non-colonial S . rosetta cells was measured from images of cells fixed and stained for tubulin and DNA as described above . Only mitotic cells with clear flagella , spindle poles , and cell body axis position were selected for analysis . We measured the acute angle between the spindle axis ( defined as a line connecting the two spindle poles ) and a line extending from the cell body's centroid to the midpoint between the duplicated flagella's basal bodies . The ImageJ software package was used to calculate the centroid and angle .	For billions of years , life on Earth was made up of single cells . In the lineage that led to animals – and independently in those that led to plants and to fungi – multicellular organisms evolved as cells began to specialize and arrange themselves into tissues and organs . Although the evolution of multicellularity is one of the most important events in the history of animal life , very little is known about the molecular mechanisms by which it took place . To form and maintain organized tissues , cells must coordinate how they divide relative to the position of their neighbours . One important aspect of this process is orientation of the mitotic spindle , a structure inside the dividing cell that distributes the chromosomes —and the genetic material they carry — between the daughter cells . When the spindle is not oriented properly , malformed tissues and cancer can result . In a diverse range of animals , the orientation of the spindle is controlled by an ancient scaffolding protein that links the spindle to “marker” proteins on the edge of the cell . Anderson et al . have now used a technique called ancestral protein reconstruction to investigate how this molecular complex evolved its ability to position the spindle . First , the amino acid sequences of the scaffolding protein’s ancient progenitors , which existed before the origin of the most primitive animals on Earth , were determined . Anderson et al . did this by computationally retracing the evolution of large numbers of present-day scaffolding protein sequences down the tree of life , into the deep past . Living cells were then made to produce the ancient proteins , allowing their properties to be experimentally examined . By experimentally dissecting successive ancestral versions of the scaffolding protein , Anderson et al . deduced how the molecular complex that it anchors came to control spindle orientation . This new ability evolved by a number of “molecular exploitation” events , which repurposed parts of the protein for new roles . The progenitor of the scaffolding protein was actually an enzyme , but the evolution of its spindle-orienting ability can be recapitulated by introducing a single amino acid change that happened many hundreds of millions of years ago . How could a single mutation have conferred such a dramatically new function ? Anderson et al . found that the ancient scaffolding protein uses the same part of its surface to bind to the spindle-orienting molecular marker as the ancient enzyme used to bind to its target substrate molecule , and the two partner molecules happen to share certain key chemical properties . This fortuitous resemblance between two unrelated molecules thus set the stage for the simple evolution of a function that is now essential to the complexity of multicellular animals .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "cell", "biology" ]	2016	Evolution of an ancient protein function involved in organized multicellularity in animals
Understanding how the brain computes eye position is essential to unraveling high-level visual functions such as eye movement planning , coordinate transformations and stability of spatial awareness . The lateral intraparietal area ( LIP ) is essential for this process . However , despite decades of research , its contribution to the eye position signal remains controversial . LIP neurons have recently been reported to inaccurately represent eye position during a saccadic eye movement , and to be too slow to support a role in high-level visual functions . We addressed this issue by predicting eye position and saccade direction from the responses of populations of LIP neurons . We found that both signals were accurately predicted before , during and after a saccade . Also , the dynamics of these signals support their contribution to visual functions . These findings provide a principled understanding of the coding of information in populations of neurons within an important node of the cortical network for visual-motor behaviors . The animals performed delayed memory saccades on a Cartesian grid whose nodes were spaced by 8 deg . The animals maintained fixation for 1000 ms at one randomly chosen location on the 3 × 3 grid . Next a target was flashed for 300 ms in one location randomly chosen from the eight neighboring locations , and the animals were required to maintain fixation during this epoch . The animals had to remember the target location for 1300 ms ± 200 ms ( drawn from a uniform distribution ) . Upon extinction of the fixation , they made a saccade to the remembered target location in complete darkness , thus acquiring a postsaccade fixation on a 5 × 5 grid . They maintained fixation there for 500 ms ( fixation I ) in the absence of visual input . A fixation light was subsequently turned on at the target location for another 500 ms , possibly triggering a small corrective saccade . Upon completion of this second fixation ( fixation II ) , the stimulus was removed and the animals were rewarded by a drop of water . We only considered correct trials . The animals completed ∼11 random blocks , for a total of ∼11 × ( ( 3 × 3 ) × 8 ) = 792 saccades . The eye position was tracked by an infrared camera , and the signal was sampled at 2 kHz ( EyeLink , SR Research , Ontario , Canada ) . All behavioral variables were instructed and monitored in real-time ( LabView , National Instruments , Austin , TX , USA ) . The delayed memory task allowed us to examine separately neuronal signals mediating sensory inputs ( vision ) , planning , and motor commands . Furthermore , we were able to study how pre- and postsaccade eye positions and saccade directions are represented because we extended the delayed memory task to cover a grid . One novelty of this task is that eye position and saccade directions were sampled exhaustively on a Cartesian grid , thus yielding a complete set of independent oculomotor variables . In contrast to previous single neurons studies where behaviors were tailored to best drive a given neuron ( e . g . , preferred and anti-preferred saccade direction ) , another novelty of the task resides in the fact that we used a fixed and finite set of oculomotor behaviors to drive neuronal populations as a whole . Two adult male rhesus monkeys ( Macaca mulatta , monkey S 14 . 3 kg and monkey P 13 . 7 kg ) were used in the experiments . All procedures were in accordance with the guidelines of the Caltech Institutional Animal Care and Use Committee and the National Institute of Health Guide for the Care and Use of Laboratory Animals . Each animal was implanted with a headpost and custom-made recording chamber under aseptic conditions using isoflurane anesthesia . Both animals received postoperative analgesics during postsurgical recovery . We used MRI scans as guides for the location of LIP . We recorded populations of LIP neurons using a 5-channel microdrive with movable quartz-platinum/tungsten microelectrodes ( Thomas Recording , Giessen , Germany ) . The neuronal data were filtered between 100 Hz and 8 kHz ( preamplifier , Plexon , Dallas , TX , USA ) , and then sampled at 40 kHz ( Multichannel Acquisition Processor , Plexon ) . Single units were sorted online using a dual-window discriminator based on the shape of the waveform . At the beginning of each recording session , we verified that each electrode tip was located in LIP by obtaining a vigorous sensory , planning and motor activity during a center-out delayed memory saccade task . Each animal was recorded in a different hemisphere . Hence , when combining the neuronal populations recorded in both animals , we were able to overcome the component of the sampling bias associated with the fact that LIP has contralateral response fields for visual stimuli and movement direction ( Quian Quiroga et al . , 2006 ) . To empirically quantify the component of the sampling inherent to the homogeneity of neurons in the sample , we acquired a different population of simultaneously recorded neurons for each session . We gathered empirical samples of simultaneously recorded neurons for three reasons . First , this allowed us to assess the sampling bias underlying the computation of the NPC . Second , the neurons recorded simultaneously were subject to exactly the same experimental conditions , thus eliminating session-to-session variability from the computation of the NPC . Third , simultaneously recorded neurons gave a more natural sample of the NPC because these neurons were involved—although to different degrees—in exactly the same behaviors within a session and across sessions . This is in stark contrast to most single neuron studies where the task was optimized to best drive the studied neuron . To address population coding in LIP , we studied the accuracy with which pre- and postsaccade eye positions and the saccade direction could be predicted from the population response . This neuronal population code ( NPC ) is a single summary metric that encompasses the response properties of multiple neurons across a range of oculomotor behaviors , unlike most classical measures of single-cell electrophysiology where each neuron and behavior are studied separately . The NPC describes the accuracy of cortical computations relevant to a given task within an area , and the information available in principle to a downstream neuron that gets synaptic inputs from this population . We used Bayesian inference to model the NPC by predicting the accuracy with which a behavior can be inferred from the population response taken to be the spike count of each neuron over a given temporal window ( Foldiak , 1993; Seung and Sompolinsky , 1993; Salinas and Abbott , 1994; Sanger , 1996; Oram et al . , 1998; Dayan and Abbott , 2001; Jaynes , 2003; Pouget et al . , 2003; Sanger , 2003; Averbeck and Lee , 2004; Brown et al . , 2004; Ma et al . , 2006; Graf et al . , 2011 ) . The prior p ( b ) on the behavior b , i . e . the probability of a given behavior , was computed from the number of occurrences of that behavior . It represents the behavioral history , and is entirely defined by the task ( the task has no error trials ) . The likelihood function p ( r\|b ) is the probability to obtain a population response ( spike count ) r given a behavior b , evaluated across behaviors . It is a description of the encoding process that addresses how a behavior is represented at the level of neuronal populations . We computed the likelihood by assuming that the N neurons were independent , yielding:p ( r\|b ) =∏i=1Np ( ri\|b ) Next , we used a parametric approximation for the likelihood function of each neuron . The simplest approximation is to assume that the spike counts of a neuron for a given behavior follow a Poisson distribution:p ( r\|b ) =μ ( b ) rr ! exp ( −μ ( b ) ) where μ ( b ) is the mean response across all trials corresponding to the behavior b . This model requires the empirical determination of one parameter: the mean response ( tuning curve ) . This model has been widely used , and was also extended to take correlations between neurons into account ( Sanger , 1996; Oram et al . , 1998; Ma et al . , 2006; Graf et al . , 2011 ) . Alternatively , a more complicated approximation is to assume that the spike count distribution can be approximated by a truncated Gaussian over positive integers:p ( r\|b ) =G ( r , μ , σ , b ) ∫0∞G ( r , μ , σ , b ) drwhere: G ( r , μ , σ , b ) =12πσ ( b ) 2exp ( − ( r−μ ( b ) ) 22σ ( b ) 2 ) This model requires the empirical determination of two parameters: the mean μ ( b ) and the variance σ ( b ) 2 of the response across all trials corresponding to the behavior b . Both approximations gave decoding accuracies ( see below ) that were strongly correlated ( r2 = 0 . 95 , p=0 ) . The posterior function p ( b\|r ) is the hallmark of decoding: it is the probability to obtain a behavior given a neuronal population response . By Bayes' theorem , it is proportional to the product of the likelihood function and the prior:p ( b\|r ) =p ( r\|b ) p ( b ) p ( r ) where p ( r ) is the partition function . In order to ensure good generalization ( and thus avoid overfitting ) , we evaluated the posterior function using a leave-one-out cross-validation scheme ( Duda et al . , 2001 ) . In other words , the trials that were used to evaluate the posterior were different from the trials used to compute the parameters of the model . The posterior function is a probability distribution that informs us how likely a behavior can be associated with a given neuronal population response . We considered the maximum of the posterior distribution as the estimate corresponding to a population activity ( MAP estimate ) . We determined the prediction accuracy of each behavior by computing the proportion of veridical MAP estimates evaluated across all trials , i . e . , the fraction of trials where true and estimated behaviors coincided ( exact estimates ) . Finally , the accuracy of the NPC was the average of the prediction accuracy of each behavior . The mean and standard error of the NPC were determined using 1000 bootstraps across all trials . Our task allowed us to study three oculomotor behaviors: the 3 × 3 = 9 presaccade eye positions , the eight saccade directions and the 5 × 5 = 25 postsaccade ( future ) eye positions . These behaviors were entangled in a single population response . In each trial , the animal executed one out of a total of 9 × 8 = 72 oculomotor behaviors . Each population response thus predicted one saccade direction and one presaccade eye position . Because presaccade eye position and saccade direction were independent , their respective NPCs were computed separately , for example the NPC for the saccade direction was determined using the fraction of correct saccade direction estimates across all trials . The postsaccade eye position was the vector sum of the presaccade eye position and the saccade direction . Multiple presaccade eye positions and saccade directions could yield the same postsaccade eye position . The postsaccade eye position was thus dependent on the presaccade eye position and the saccade direction . We explicitly modeled this dependency in the Bayesian framework by translating the vector sum of the behaviors into a marginalization of the posterior distributions: the posterior of the postsaccade eye position was the sum across the posteriors of all presaccade eye positions and saccade directions that yielded the same postsaccade eye position . In other words , the vector sum of the behaviors translated into a sum of their respective posteriors . To determine the time course of the NPC , we first aligned the spike times for each trial to either the target onset or the saccade onset . The saccade onset ( reaction time ) was determined offline based on the saccade velocity using a method adapted from ( Engbert and Kliegl , 2003 ) . For each neuron , we subsequently computed the spike counts over causal boxcar windows ( rectangular windows looking only in the past ) of length 250 ms sliding in 10 ms steps . We chose a length of 250 ms because this interval is short enough compared to the length of the epochs in the task , but long enough to have sufficient spike counts for decoding . For each time step , we then computed the prediction accuracy ( mean ± SEM from bootstrap estimates ) , yielding the time course of the NPC . The inference was thus done on intervals of identical length ( 250 ms ) across the entire task . To compute the decoding accuracy for a given task epoch ( which may vary in length ) , we computed the root mean square of the NPC time course between the events defining the task epoch ( mean ± SEM across bootstrap estimates ) . In other words , we computed the area under the time course normalized by the length of the epoch . This allowed us to compare the NPC across task epochs of different lengths in an unbiased fashion because the NPC computed directly from the spike counts collected over epochs of different lengths is dependent on the length of the epoch . We examined the possible sources of the NPC updating by computing the onsets and peaks of the time course of the NPC across the entire task . First , we smoothed the NPC time courses using a truncated Gaussians over a 250 ms window . We then compared for each time step the derivatives of the NPC time course before and after . We finally determined the onsets by maximizing the difference of the derivatives after and before the onset , and the peaks by maximizing the difference of the derivatives before and after the peak . To quantitatively rank the contributions of each neuron to the NPC , we developed a technique derived from machine learning ( Duda et al . , 2001; Guyon et al . , 2002 ) : Recursive Neuronal Elimination ( RNE ) . RNE finds the subset of most important neurons by iteratively removing neurons that least affect the accuracy of the NPC . We first pooled all recording sessions that had at least nine trials per oculomotor behavior , for a total of 325 independently recorded neurons . We then found the neuron ( s ) that when removed from the population maximized the decoding accuracy across the remaining neurons . We iteratively applied this procedure to the remaining neurons , thus creating a list of neurons ranked from least to most important . The decoding accuracy was averaged across the three behaviors because the same neuronal subset was used to predict each one of them . Also , we determined the decoding accuracies using the root mean square across the task epochs that best mediate each behavior: the memory , fixation and fixation II epochs for the saccade direction , pre- and postsaccade eye position respectively . RNE yields populations of neurons that are devoid of non-task related neurons and redundant neurons ( e . g . , neurons with similar tuning ) . It is thus ideally suited to study the dependency of prediction accuracy and population size in a principled manner that avoids using neuronal subsets defined by random pooling ( Bremmer et al . , 1998 ) . As such RNE can be used to assess the sparseness of the NPC . Also , RNE is a quantitative method to address sampling bias in large neuronal populations . Finally , once an ‘optimized’ subset is defined , the accuracy of the NPC for this subset can be used to correct the accuracy of the NPC for each empirical population by scaling its average decoding accuracy to mach the one of the ‘optimized’ subset .	Whenever we reach towards an object , we automatically use visual information to guide our movements and make any adjustments required . Visual feedback helps us to learn new motor skills , and ensures that our physical view of the world remains stable despite the fact that every eye movement causes the image on the retina to shift dramatically . However , such visual feedback is only useful because it can be compared with information on the position of the eyes , which is stored by the brain at all times . It is thought that one important structure where information on eye position is stored is an area towards the back of the brain called the lateral intraparietal cortex , but the exact contribution of this region has long been controversial . Graf and Andersen have now clarified the role of this area by studying monkeys as they performed an eye-movement task . Rhesus monkeys were trained to fixate on a particular location on a grid . A visual target was then flashed up briefly in another location and , after a short delay , the monkeys moved their eyes to the new location to earn a reward . As the monkeys performed the task , a group of electrodes recorded signals from multiple neurons within the lateral intraparietal cortex . This meant that Graf and Andersen could compare the neuronal responses of populations of neurons before , during , and after the movement . By studying neural populations , it was possible to accurately predict the direction in which a monkey was about to move his eyes , and also the initial and final eye positions . After a movement had occurred , the neurons also signaled the direction in which the monkey's eyes had been facing beforehand . Thus , the lateral intraparietal area stores both retrospective and forward-looking information about eye position and movement . The work of Graf and Andersen confirms that the LIP has a central role in eye movement functions , and also contributes more generally to our understanding of how behaviors are encoded at the level of populations of neurons . Such information could ultimately aid the development of neural prostheses to help patients with paralysis resulting from injury or neurodegeneration .	[ "Abstract", "Materials", "and", "methods" ]	[ "short", "report", "neuroscience" ]	2014	Inferring eye position from populations of lateral intraparietal neurons

Subsets and Splits

No community queries yet

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